vii. fingerprint recognitionshodhganga.inflibnet.ac.in/bitstream/10603/99462/14/14_chapter...

Fingerprint as Biometric: Feature, Application and Recognition

____________________________________________________________________________________

- 76 -

VII. Fingerprint Recognition

‘Discovery consists of looking at the same thing as everyone else and thinking something different’

- Albert Szent-Györgyi in IEEE (1985) Bridging the present and the future: IEEE Professional

Communication Society Conference Record, Williamsburg, Virginia, October 16-18, 1985. p. 14.

Fingerprints have long been used for personal identification. It is assumed that every single

person possesses unique fingerprints and hence the fingerprint matching is considered one of

the most reliable and effective techniques of person identification, therefore it is very widely

used. Not to forget to mention fingerprint recognition system is quite cheaper than many

other biometric identifiers. Injuries like cuts, burns and bruises can temporarily spoil quality

of fingerprints but when/if fully healed patterns will be restored in most cases. This property

makes fingerprints a very strikingly attractive biometric identifier. Until the recent

advancements of experimental automated systems, all fingerprint analysis had been done by

human beings. The need for automated systems becomes quite obvious when one realizes that

the Federal Bureau of Investigation (FBI) alone is called upon to classify and file thousands

of sets of fingerprint cards everyday. In recent years, many algorithms and models are given

to improve the accuracy of fingerprint recognition system. However, recognizing fingerprints

in poor quality images is still a very complex problem. The reliability of any automatic

fingerprint recognition system strongly relies on the precision obtained in the feature

extraction process. Extraction of appropriate features is one of the most important tasks for a

recognition system. Therefore a very important phase is noise removal as even a small

amount of noise can make the ridge structure different.

In this part, first we will do a literature survey to have an understanding of what kind of

works were done in past and recent past in this field, and then we will go through different

methods applied during this research towards fingerprint recognition. We will study about the

minutiae as a fingerprint recognition feature. This will be followed by a fingerprint

recognition model. Then we will see the effects of noise on a fingerprint. Next fingerprint

pre-processing and matching using correlation in frequency domain will be discussed. Next

soft computing based fingerprint recognition will be noted.

Then experimental results will be noted. First the minutiae based fingerprint matching will be

noted down. Then fingerprint recognition using both global and local structures will be

discussed. And then the fingerprint recognition based on pores will be discussed. Then soft

computing based fingerprint recognition will be studied. Later a new proposal to recognize a

fingerprint by divide and conquer method will be discussed. At last few other simplistic

methods will be noted down, such as Point Extraction Technique in Spatial-domain for

fingerprint recognition, Revolutionary Extended Spatial Point Extraction using Circular

Technique, and A Bar Code Design & Encoding for Fingerprints. This chapter is to be

concluded by a comparative study and noting down limitations of fingerprint systems. All the

sub-chapters can be perceived as standalone chapters.


____________________________________________________________________________________

- 77 -

VII.1. Literature Survey

VII.1.1. Introduction

Human fingertips contain ridges and valleys which together form distinctive patterns. These

patterns are fully developed under pregnancy and are permanent throughout whole lifetime.

Prints of those patterns are called fingerprints. Injuries like cuts, burns and bruises can

temporarily damage quality of fingerprints but when fully healed, patterns will be restored.

Figure 49 shows a sample fingerprint and figure 50 shows minutiae patterns.

Figure 49: Sample fingerprint.

Figure 50: Minutiae patterns.


____________________________________________________________________________________

- 78 -

Fingerprints have long been used for personal identification. It is believed that every person

possesses inimitable fingerprints and hence the fingerprint matching is considered one of the

most dependable techniques of people identification. A fingerprint image exhibits a

quasiperiodic pattern of ridges (darker regions) and valleys (lighter regions). The local

topological structures of this pattern together with their spatial relationships determine the

distinctiveness of a fingerprint. There are more than 100 different types of local ridge

structures that have been identified. Nevertheless, most of the automatic fingerprint

identification/verification systems adopt the model used by the Federal Bureau of

Investigation. The model relies on representing only the two most prominent structures: ridge

ending and ridge bifurcation, which are collectively called minutiae. Several methods of

automatic minutiae extraction from fingerprint images have been proposed in the literature.

VII.1.2. Fingerprint Recognition System

Theoretically, any human physiological or behavioural characteristic can be used to make

personal identification as long as it satisfies the following requirements [3], [91]:

(i) Universality (every person should have the characteristic),

(ii) Uniqueness (no two person should be the same in terms of the characteristic),

(iii)Permanence (the characteristic should be invariant with time), and

(iv) Collectability (the characteristic can be measured quantitatively).

Figure 51 shows the common workflow for fingerprint classification systems.

Figure 51: Common workflow for fingerprint classification systems.


____________________________________________________________________________________

- 79 -

VII.1.3. Fingerprints in Historical Context

Fingerprints have been found on ancient Babylonian clay tablets, seals, and pottery [73].

They have also been found on the walls of Egyptian tombs and on Minoan, Greek, and

Chinese pottery, as well as on bricks and tiles from ancient Babylon and Rome.

Some of these fingerprints were deposited unintentionally by the potters and masons as a

natural consequence of their work, and others were made in the process of adding decoration.

However, on some pottery, fingerprints have been impressed so deeply into the clay that they

were possibly intended to serve as an identifying mark by the maker.

Fingerprints were used as signatures in ancient Babylon in the second millennium BCE. In

order to protect against forgery, parties to a legal contract would impress their fingerprints

into a clay tablet on which the contract had been written.

By 246 BCE, Chinese officials were into impressing their fingerprints into the clay seals used

to seal documents. With the advent of silk and paper in China, parties to a legal contract

impressed their handprints on the document.

Sometime before 851 CE, an Arab merchant in China, Abu Zayd Hasan, witnessed Chinese

merchants using fingerprints to authenticate loans.

By 702, Japan allowed illiterate petitioners seeking a divorce to "sign" their petitions with a

fingerprint [73], [103], [104].

Although ancient peoples probably did not realize that fingerprints could uniquely identify

individuals [33], references from the age of the Babylonian king Hammurabi (1792-1750

BCE) indicate that law officials would take the fingerprints of people who had been arrested

[73].

During China's Qin Dynasty, records have shown that officials took hand prints, foot prints as

well as finger prints as evidence from a crime scene [32].

In China, around 300 CE, handprints were used as evidence in a trial for theft.

By 650, the Chinese historian Kia Kung-Yen remarked that fingerprints could be used as a

means of authentication [73].

In his Jami al-Tawarikh (Universal History), the Persian physician Rashid-al-Din Hamadani

(also known as "Rashideddin", 1247–1318) referred to the Chinese practice of identifying

people via their fingerprints [25], commenting: "Experience shows that no two individuals

have fingers exactly alike." In Persia at this time, government documents may have been

authenticated with thumbprints.


____________________________________________________________________________________

- 80 -

VII.1.4. Initial Days of Fingerprint Systems

In 1665, the Italian physician and a professor of anatomy at the University of Bologna,

Marcello Malpighi (1628–1694) briefly mentioned the existence of patterns of ridges and

sweat glands on the fingertips. He noted in his treatise; ridges, spirals and loops in

fingerprints. He made no mention of their value as a tool for individual identification [166].

In 1684, the English physician, botanist, and microscopist Nehemiah Grew (1641–1712)

published the first scientific paper to describe the ridge structure of the skin covering the

fingers and palms.

In 1685, the Dutch physician Govard Bidloo (1649–1713) published a book on anatomy

which also illustrated the ridge structure of the fingers.

A century later, in 1788, the German anatomist Johann Christoph Andreas Mayer (1747–1801) recognized that fingerprints are unique to each individual.

In 1823, John Evangelist Purkinje (1787–1869), a Czech physiologist and an anatomy

professor at the University of Breslau, published his thesis discussing nine fingerprint

patterns, but he did not mention any possibility of using fingerprints to identify people [102].

Some years later, the German anatomist Georg von Meissner (1829–1905) studied friction

ridges [102], [166].

And few years after, in 1858, Sir William James Herschel initiated fingerprinting in India.

The English first began using fingerprints in July of 1858, when Sir William James Herschel,

Chief Magistrate of the Hooghly district in Jungipoor, India, first used fingerprints on native

contracts [102]. On a whim, and without thought towards personal identification, Herschel

had Rajyadhar Konai, a local businessman; impress his hand print on a contract. In 1877 at

Hooghly (near Calcutta) Herschel instituted the use of fingerprints on contracts and deeds to

prevent the then-rampant repudiation of signatures [22] and he registered government

pensioners' fingerprints to prevent the collection of money by relatives after a pensioner's

death [23]. Herschel also fingerprinted prisoners upon sentencing to prevent various frauds

that were attempted in order to avoid serving a prison sentence.

In 1863, Professor Paul-Jean Coulier (1824–1890), professor for chemistry and hygiene at the

medical and pharmaceutical school of Val-de-Grâce military hospital in Paris, published his

observations that (latent) fingerprints can be developed on paper by iodine fuming, explains

how to preserve (fix) such developed impressions and mentioned the potential for identifying

suspects' fingerprints by use of a magnifying glass [30], [31].

In 1880, Dr. Henry Faulds, a Scottish surgeon in a Tokyo hospital, published an article in the

Scientific Journal, "Nature". He discussed fingerprints as a means of personal identification,

and the use of printers ink as a method for obtaining such fingerprints. He is also credited

with the first fingerprint identification of a greasy fingerprint left on an alcohol bottle.


____________________________________________________________________________________

- 81 -

Dr. Henry Faulds also established the first classification and was also the first to identify

fingerprints left on a vial [117]. Returning to the UK in 1886, he offered the concept to the

Metropolitan Police in London but it was dismissed at that time [118]. Faulds wrote to

Charles Darwin with a description of his method but, too old and ill to work on it, Darwin

gave the information to his cousin, Francis Galton, who was interested in anthropology.

Having been thus inspired to study fingerprints for ten years, Sir Francis Galton, a British

anthropologist, began his observations of fingerprints as a means of identification in the

1880's. In 1892, he published his book, "Finger Prints", establishing the individuality and

permanence of fingerprints. The book included the first classification system for fingerprints.

Galton published a detailed statistical model of fingerprint analysis and identification and

encouraged its use in forensic science in his book. He had calculated that the chance of a

"false positive" (two different individuals having the same fingerprints) was about 1 in 64

billion [125]. Galton's primary interest in fingerprints was as an aid in determining heredity

and racial background. While he soon discovered that fingerprints offered no firm clues to an

individual's intelligence or genetic history, he was able to scientifically prove what Herschel

and Faulds already suspected: that fingerprints do not change over the course of an

individual's lifetime, and that no two fingerprints are exactly the same. Galton identified the

characteristics by which fingerprints can be identified. These same characteristics (minutia)

are basically still in use today, and are often referred to as Galton's Details (or Galton Points).

In 1882, Gilbert Thompson of the U.S. Geological Survey in New Mexico used his own

thumb print on a document to prevent forgery. This is the first known use of fingerprints in

the United States.

Juan Vucetich, an Argentine chief police officer, created the first method of recording the

fingerprints of individuals on file, associating these fingerprints to the anthropometric system

of Alphonse Bertillon, who had created, in 1879, a system to identify individuals by

anthropometric photographs and associated quantitative descriptions. In 1891, he began the

first fingerprint files based on Galton pattern types. In 1892, after studying Galton's pattern

types, Vucetich set up the world's first fingerprint bureau. In 1892, Juan Vucetich made the

first criminal fingerprint identification. He was able to identify a woman by the name of

Rojas, who had murdered her two sons, and cut her own throat in an attempt to place blame

on another. Her bloody print was left on a door post, proving her identity as the murderer.

Francisca Rojas of Necochea, was found in a house with neck injuries, whilst her two sons

were found dead with their throats cut. Rojas accused a neighbour, but despite brutal

interrogation, this neighbour would not confess to the crimes. Inspector Alvarez, a colleague

of Vucetich, went to the scene and found a bloody thumb mark on a door. When it was

compared with Rojas' prints, it was found to be identical with her right thumb. She then

confessed to the murder of her sons [102].

In Mark Twain's book, "Life on the Mississippi", a murderer was identified by the use of

fingerprint identification. In a later book by Mark Twain, "Pudd'n Head Wilson", there was a

dramatic court trial on fingerprint identification.


____________________________________________________________________________________

- 82 -

VII.1.5. Early Fingerprint Recognition Systems

On 12 June 1897, the Council of the Governor General of India approved a committee report

that fingerprints should be used for classification of criminal records. Later that year, the

Calcutta (now Kolkata) Anthropometric Bureau became the world's first Fingerprint Bureau.

Working in the Calcutta Anthropometric Bureau, before it became the Fingerprint Bureau,

were Azizul Haque and Hem Chandra Bose. Haque and Bose were Indian fingerprint experts

who have been credited with the primary development of a fingerprint classification system

eventually named after their supervisor, Sir Edward Richard Henry [32], [115].

The Henry Classification System, co-devised by Haque and Bose, was accepted in England

and Wales when the first United Kingdom Fingerprint Bureau was founded in Scotland Yard,

the Metropolitan Police headquarters, London, in 1901.

Introduction of fingerprints for criminal identification in England and Wales, using Galton's

observations and revised by Sir Edward Richard Henry was in 1901. Thus began the Henry

Classification System, used even today in all English speaking countries.

The first year for the first known systematic use of fingerprint identification began in the

United States is 1902 [102]. In the United States, Dr. Henry P. DeForrest used fingerprinting

in the New York Civil Service in 1902,

By 1906, New York City Police Department Deputy Commissioner Joseph A. Faurot, an

expert in the Bertillon system and a finger print advocate at Police Headquarters, introduced

the fingerprinting of criminals to the United States.

The Scheffer case of 1902 is the first case of the identification, arrest and conviction of a

murderer based upon fingerprint evidence. Alphonse Bertillon identified the thief and

murderer Scheffer, who had previously been arrested and his fingerprints were filed some

months before, from the fingerprints found on a fractured glass showcase, after a theft in a

dentist's apartment where the dentist's employee was found dead. It was able to be proved in

court that the fingerprints had been made after the showcase was broken.

A year later, Alphonse Bertillon created a method of getting fingerprints off smooth surfaces

and took a further step in the advance of dactyloscopy.

The New York Civil Service Commission established the practice of fingerprinting applicants

to prevent them from having better qualified persons take their tests for them.

The New York state prison system began to use fingerprints for the identification of criminals

in 1903.

In 1904 the fingerprint system accelerated when the United States Penitentiary at

Leavenworth, Kansas, and the St. Louis, Missouri, Police Department both established

fingerprint bureaus. They were assisted by a Sergeant from Scotland Yard who had been on

duty at the St. Louis Exposition guarding the British Display.


____________________________________________________________________________________

- 83 -

1905 saw the use of fingerprints for the U.S. Army. Two years later the U.S. Navy started,

and was joined the next year by the Marine Corp.

During the next 25 years more and more law enforcement agencies join in the use of

fingerprints as a means of personal identification. Many of these agencies began sending

copies of their fingerprint cards to the National Bureau of Criminal Identification, which was

established by the International Association of Police Chiefs.

It was in 1918 when Edmond Locard wrote that if 12 points (Galton's Details) were the same

between two fingerprints, it would suffice as a positive identification. This is where the often

quoted ´12 points´ originated.

Be aware though, there is "NO" required number of points necessary for an identification.

Some countries have set their own standards which do include a minimum number of points

(not in India).

During the first quarter of the 20th century, more and more local police identification bureaus

established fingerprint systems.

The growing need and demand by police officials for a national repository and clearinghouse

for fingerprint records led to an Act of Congress in United States on July 1, 1921,

establishing the Identification Division of the FBI.

By 1946, the F.B.I. had processed 100 million fingerprint cards in manually maintained files;

and by 1971, 200 million cards. With the introduction of AFIS technology, the files were split

into computerized criminal files and manually maintained civil files. Many of the manual

files were duplicates though, the records actually represented somewhere in the

neighbourhood of 25 to 30 million criminals, and an unknown number of individuals in the

civil files.

By 1999, the FBI had planned to stop using paper fingerprint cards (at least for the newly

arriving civil fingerprints) inside their new Integrated AFIS (IAFIS) site at Clarksburg, WV.

IAFIS would initially have individual computerized fingerprint records for approximately 33

million criminals. Old paper fingerprint cards for the civil files are still manually maintained

in a warehouse facility in Fairmont, WV, USA.

Since the Gulf War, most military fingerprint enlistment cards (in USA) received have been

filed only alphabetically by name. The FBI hopes to someday classify and file these cards so

they can be of some value for unknown casualty (or amnesiac) identification (when no

passenger/victim list from a flight, etc., is known).

Currently, paper fingerprint cards are still in use and being processed for all identification

purposes [102].


____________________________________________________________________________________

- 84 -

VII.1.6. Gradual Advancements in Fingerprint Recognition System

In 2005 INTERPOL's Automated Fingerprint Identification System repository exceeds

50,000 sets fingerprints for important international criminal records from 184 member

countries. Over 170 countries have 24 x 7 interface ability with INTERPOL expert

fingerprint services. On 16 July 2010, the world's largest and oldest forensic science

organization (IAI) acknowledged advances in fingerprint science during the past three

decades, dropped the ban on qualified identification conclusions, and opened the door for

future validation of probability models involving finger/palm print comparisons [102].

In the “fingerprint Image Enhancement using Filtering Techniques”, of April 1995, by

Shlomo Greenberg, Mayer Aladjem, Daniel Kogan and Itshak Dimitrov [112], it has been

shown that, Extracting minutiae from fingerprint images is one of the most important

steps in automatic fingerprint identification and classification. Minutiae are local

discontinuities in the fingerprint pattern, mainly terminations and bifurcations. In this work

two methods for fingerprint image enhancement are done. The first one is carried out using

local histogram equalization, Wiener filtering, and image binarization. The second

method uses a unique anisotropic filter for direct gray scale enhancement. The results

achieved are compared with those obtained through some other methods. Both methods show

some improvement in the minutiae detection process in terms of either efficiency or time

required.

In “An Identity-Authentication System using Finger”, published in April 1997, by Anil K.

Jain, it has been seen that, Fingerprint verification is an important biometric technique for

personal identification [91]. In this paper, the design and implementation of a prototype

automatic identity-authentication system that uses fingerprint to authenticate the identity of

an individual is used. An improved minutiae-extraction algorithm is used that is faster and

more accurate than earlier algorithm. An alignment-based minutiae-matching algorithm has

been proposed. This algorithm is capable of finding the correspondences between input

minutiae and the stored template without resorting to exhaustive search and has the ability to

compensate adaptively for the nonlinear deformations and inexact transformations between

an input and a template. To establish an objective assessment of the system, both the

Michigan State University and the National Institute of Standards and Technology NIST 9

fingerprint data bases have been used to estimate the performance numbers. The experimental

results reveal that this system can achieve a good performance on these data bases. This paper

also demonstrated that this system satisfies the response-time requirement. A complete

authentication procedure, on average, takes about 1.4 seconds on a Sun ULTRA 1

workstation (it is expected to run as fast or faster on a 200 HMz Pentium.

From “Fingerprint Verification”, of 2006, By Lawrence O’ Gorman of Veridicom Inc.,

Chatham, NJ [135], it has been understood, the use of fingerprints for identification has

been employed in law enforcement for about a century. A much broader application of

fingerprints is for personal authentication, for instance to access a computer, a network, a

bank-machine, a car, or a home. The topic of this paper is fingerprint verification, where

"verification" implies a user matching a fingerprint against a single fingerprint


____________________________________________________________________________________

- 85 -

associated with the identity that the user claims. The following topics are covered: history,

image processing methods, enrolment and verification procedures, system security

considerations, recognition rate statistics, fingerprint capture devices, combination with other

biometrics, and the future of fingerprint verification.

In “Fingerprint Image Enhancement and Minutiae Extraction”, of July 2003, Raymond Thai

said that, Fingerprints are the oldest and most widely used form of biometric identification.

Despite the widespread use of fingerprints, there is little statistical theory on the uniqueness

of fingerprint minutiae [136]. A critical step in studying the statistics of fingerprint minutiae

is to reliably extract minutiae from the fingerprint images. However, fingerprint images are

rarely of perfect quality. They may be degraded and corrupted due to variations in skin and

impression conditions. Thus, image enhancement techniques are employed prior to minutiae

extraction to obtain a more reliable estimation of minutiae locations. In this dissertation,

discussion on the methodology and implementation of techniques for fingerprint image

enhancement and minutiae extraction is used. Experiments were done using a mixture of both

synthetic test images and real. Fingerprint images are then conducted to evaluate the

performance of the implemented techniques. In combination with these techniques,

preliminary results on the statistics of fingerprint images are then presented and discussed.

In the “Image Retrieval Based On Hierarchical Gabor Filters”, April 2005, by Tomasz

Andrysiak, Michal Chora, they were explaining, how Content Based Image Retrieval (CBIR)

is now a widely investigated issue that aims at allowing users of multimedia information

systems to automatically retrieve images coherent with a sample image. A way to achieve

this goal is the computation of image features such as the colour, texture, shape, and position

of objects within images, and the use of those features as query terms. Here Gabor filtration

properties are used in order to finding such appropriate features. The article presents

multichannel Gabor filtering and a hierarchical image representation. Then a salient

(characteristic) point detection algorithm is presented so that texture parameters are computed

only in a neighbourhood of salient points. The Gabor texture features are used as image

content descriptors and efficiently employ them to retrieve images [137].

In the paper “A Fingerprint Minutiae Recognition System Based on Genetic Algorithms”, April 18, 2005 by Jihad Jaam, Mohamed Rebaiaia and Ahmad Hasnah [138], we can see that,

a fingerprint verification system using two different modules: the automatic classification of

fingerprints which is based on the minutiae-matching algorithms and the verification-search

technique which is based on genetic n algorithms are used. This experiments on a large set

of fingerprint images show that approach is highly promising. Its performance shows a

great flexibility in recognizing a person very quickly with an error-prone of at most 1%.

In “A Fingerprint Recognition Algorithm Using Phase-Based Image Matching for Low-

Quality Fingerprints”, 2005, by Koichi Ito, Ayumi Morita , Takafumi Aoki, Tatsuo Higuchi, Hiroshi Nakajima, and Koji Kobayashi, they discussed about a major approach for fingerprint

recognition today, that is to extract minutiae from fingerprint images and to perform

fingerprint matching based on the number of corresponding minutiae pairings [139]. One of

the most difficult problems in fingerprint recognition has been that the recognition


____________________________________________________________________________________

- 86 -

performance is significantly influenced by fingertip surface condition, which may vary

depending on environmental or personal causes. Addressing this problem, this paper presents

a fingerprint recognition algorithm using phase-based image matching. The use of phase

components in 2D (two-dimensional) discrete Fourier transforms of fingerprint images makes

possible to achieve highly robust fingerprint recognition for low-quality fingerprints.

Experimental evaluation using a set of fingerprint images captured from fingertip with

difficult conditions (e.g., dry fingertip, rough fingertip, allergic-skin fingertip) demonstrates

an efficient recognition performance of the proposed algorithm compared with a typical

minutiae-based algorithm.

In “Fingerprint Recognition Using Zernike Moments”, April 2006, by Hasan Abdel Qader, Abdul Rahman Ramli, and Syed Al-Haddad, it has been discussed that, how a fingerprint

matching approach based on localizing the matching regions in fingerprint images is used

[140]. It shows the determination of the location of such Region Of Interest (ROI) using only

the information related to core points based on a novel feature vectors extracted for each

fingerprint image by Zernike Moment Invariant (ZMI) as the shape descriptor. The

Zernike Moments is selected as feature extractor due to its robustness to image noise,

geometrical invariants property and orthogonal property. These features are used to identify

corresponding ROI between two fingerprint impressions by computing the Euclidean distance

between feature vectors. The fingerprint matching invariance under translations, rotations

and scaling using Zemike Moment Invariants and the experimental results obtained from a

FVC2002 DB1 database confirm that the Zernike moment is able to match the fingerprint

images with high accuracy.

In “Fingerprint Enhancement Using STFT Analysis”, May 2006, by Sharat Chikkerur,

Alexander N. Cartwright Venu Govindaraju, it was shown that, contrary to popular belief,

despite decades of research in fingerprint, reliable fingerprint recognition is still an open

problem [141]. Extracting features out of poor quality prints is the most challenging problem

faced in this area. This paper introduces a new approach for fingerprint enhancement based

on Short Time Fourier Transform (STFT) Analysis. STFT is a well known technique in

signal processing to analyze non-stationary signals. Here the authors extend its application to

2D fingerprint images. The algorithm simultaneously estimates all the intrinsic properties of

the fingerprints such as the foreground region mask, local ridge orientation and local ridge

frequency. Furthermore they propose a probabilistic approach of robustly estimating these

parameters. They experimentally compared the proposed approach to other filtering

approaches in literature and showed that this technique performs favourably.

In “Fingerprint Pattern Recognition Using Distance Method Algorithm” (2006) it has been

discussed that biometrics is a popular approach for data security, for instance using

fingerprint for pattern recognition. There are a lot of algorithms for the purpose such as

Distance Method. Here, a new method was used to identify the Reference Point or Core

Point. Furthermore, the Centre of Gravity was defined on the fingerprint to provide an axis

for alignment with the Core Point. Then, by taking that as the x axis, the mean x and y

distances from the core point was used to scale the print. The distribution of the scaled

distances from the Core Point then was used as a fingerprint template and was stored as a


____________________________________________________________________________________

- 87 -

template master in a system database for next recognition. When someone wants to enter into

the system, the machine would compare his fingerprint with the template. If the system finds

the acceptance rate close to 1 and rejection rate nearby 0, the user would be allowed to enter

into the system; otherwise the system would reject him, even though he is the real member of

the system.

In the “Fingerprint Recognition Wavelet Domain”, Feb 2007 by Wan Azizun Wan Adnan, Lim Tze Siang & Salasiah Hitam, it has been said [134], that fingerprint technique is one of

the most reliable biometric technologies. And the importance of the pre-processing such as

smoothing, binarization in the fingerprint recognition, and also why thinning is needed was

discussed. Then, fingerprint minutiae feature is extracted. It was shows that some effectively

fast fingerprint identification algorithm (such as using Fast Fourier Transform, (FFT)) may

require as so much computation might be impractical. Wavelet based algorithm may be the

key to making a low cost fingerprint identification system that would operate on a small

computer. The authors present a fingerprint recognition system that can match the fingerprint

images based on features extracted in the wavelet transform domain. This study was

implemented based on MATLAB Software and their toolbox applications, such as Wavelet

and Image Processing Toolbox.

By “A Hierarchical Fingerprint Matching System” July 2009 by Abhishek Rawat, Fingerprint Recognition is a widely popular but a complex pattern recognition problem [133]. It is

difficult to design accurate algorithms capable of extracting salient features and matching

them in a robust way. The real challenge is matching fingerprint affected by: i) High

displacement/or rotation which results in smaller overlap between template and query

fingerprint, ii) Non-linear distortion caused by the finger plasticity, iii) Different pressure and

skin condition and iv) Feature extraction errors which may result in spurious or missing

features. The information contained in a fingerprint can be categorized into three Different

levels, namely, Level 1 (pattern), Level 2 (minutiae points), and Level 3 (pores and ridge

contours). Despite their discriminative power, the Level 3 features are barely used by the vast

majority of contemporary automated fingerprint authentication systems (AFAS) which rely

mostly on minutiae features. This is mainly because, most of these authentication systems are

equipped with 500 ppi (FBI’s standard of fingerprint resolution for AFAS) scanners, and reliably extracting “fine and detailed” Level 3 features require high resolution images. While this may have been the case with many older live-scan devices, the current devices are

capable of detecting a reasonable amount of level 3 details even at the relatively limited 500

ppi resolution. In this thesis the above mentioned problems have been addressed and a new

hierarchical matcher has been proposed. The hierarchical matcher utilizes Level 3 features

(pores and ridge contour) in conjunction with Level 2 features (minutiae) for matching. The

aim is to reduce the error rates, namely FAR (False Acceptance Rate) and FRR (False

Rejection Rate) in the existing minutiae based systems. The hierarchical matcher has been

tested on three diverse databases in public domain. The obtained results are promising and

verify the claim.

In the paper named “An Iterative Fingerprint Enhancement Algorithm Based on Accurate

Determination of Orientation Flow”, of July 2009, by Simant Dube, he proposed an algorithm


____________________________________________________________________________________

- 88 -

to enhance and binarize a fingerprint. The algorithm was based on accurate determination of

orientation flow of the ridges of the fingerprint image by computing variance of the

neighbourhood pixels around a pixel in deferent directions [132]. The iterative algorithm was

used which captures the mutual interdependence of orientation flow computation,

enhancement and binarization. That gives very good results on poor quality images.

Now, in “Fingerprint Recognition Using Minutia Score Matching”, 2009 by Ravi. J, K. B. Raja, Venu-gopal. K. R, it was claimed that fingerprint which is unique and permanent

throughout a person’s life is a popular biometric. And it was used to authenticate a person. A

minutiae matching approach is widely used for fingerprint recognition and can be classified

as ridge ending and ridge bifurcation. In this paper they projected Fingerprint Recognition

using Minutia Score Matching method (FRMSM). For Fingerprint thinning, the Block Filter

was used, which scaned the image at the boundary to preserves the quality of the image and

extract the minutiae from the thinned image. The false matching ratio is better compared to

the previously existing algorithm [111].

In “Iterative Fingerprint Enhancement with Matched Filtering and Quality Diffusion in

Spatial-Frequency Domain”, (of 2010) by Prawit Sutthiwichaiporn, Vutipong Areekul, and

Suksan Jirachaweng; they proposed fingerprint enhancement algorithm utilizing power

spectrum in spatial-frequency domain. The input fingerprint was partitioned and assessed as

high/low quality zone by using signal-to-noise ratio (SNR) approach. For high quality zone,

signal spectrum with noise suppression was used to shape an enhanced filter in frequency

domain [131]. Then, this algorithm was fed into neighbouring enhanced zone back in order to

repair unreliable low quality region. The proposed algorithm out-performed Gabor and STFT

approaches by fingerprint matching experiments on FVC2004 Db2 and Db3.

In the paper “Environmental Modeling with Fingerprint Sequences for Topological Global

Localization”, by Pierre Lamon, Adriana Tapus, Etienne Glauser, Nicola Tomatis and Roland Siegwart, a perception approach allowing for high distinctiveness was presented. The method

worked in accordance to the fingerprint concept. Such representation allowed using a very

flexible matching approach based on the minimum energy algorithm. The whole extraction

and matching approach was presented in details and viewed in a topological optic, where the

matching result can directly be used as observation function for a topological localization

approach. The experimentation section validated the fingerprint approach and presented

different set of experiments in order to explain practically the choice of different types of

features.

In this paper named, “Probabilistic Orientation Field Estimation for Fingerprint Enhancement and Verification”, by Kuang-chih Lee and Salil Prabhakar a novel probabilistic method was

used to estimate the orientation field in fingerprint images. Traditional approach based on the

smoothing of local image gradients usually generates unsatisfactory results in poor quality

regions of fingerprint images. It was showed how to improve the orientation field estimation

by first constructing a Markov Random Field (MRF) and then inferring the orientation field

from the MRF model. The MRF is made up of two components. The first component

incorporates a global mixture model of orientation field learned from training fingerprint


____________________________________________________________________________________

- 89 -

examples. The second component enforces a smoothness constraint over the orientation field

in the neighbouring regions. The proposed approach of improved fingerprint orientation field

was useful in fingerprint enhancement and minutiae extraction processes. It showed

remarkable improvement of fingerprint verification accuracy on a relatively large fingerprint

dataset.

In “Fingerprint Recognition”, by Andrew Ackerman, it has been discussed that fingerprint

matching is the process used to determine whether two sets of fingerprint ridge detail come

from the same finger. There exist multiple algorithms that do fingerprint matching in many

different ways. Some methods involve matching minutiae points between the two images,

while others look for similarities in the bigger structure of the fingerprint. In this project,

method for fingerprint matching based on minutiae matching was used. However, unlike

conventional minutiae matching algorithms this algorithm also took into account region and

line structures that exist between minutiae pairs. This allowed for more structural information

of the fingerprint to be accounted for thus resulting in stronger certainty of matching

minutiae. Also, since most of the region analysis was pre-processed it did not make the

algorithm slower. The algorithm for matching was not created; however, the process in which

the regional data is obtained was explained in this paper. Evidence from the testing of the

pre-processed images gave stronger assurance that using such data could lead to faster and

stronger matches.

The paper on the “Fast Fingerprint Recognition Using Spiral”, by Woon Ho Jung, gives an

idea of identification based on fingerprint and it being an active area of research in

biometrics. In this work a fast implementation of a recently developed fingerprint

identification algorithm based on wavelet packets was used. The various general and domain

specific code optimization techniques to efficiently implement the registration phase, which

takes as input a set of fingerprint images and produce an adapted packet tree and an

associated wavelet domain template were used. The code for the actual identification was

then generated automatically from a mathematical description of this packet tree using

SPIRAL. The optimization techniques used and present various benchmarks that

demonstrated the efficiency of the implementation was discussed.


____________________________________________________________________________________

- 90 -

VII.1.7. Classifications in Fingerprint Recognition Methods

If we classify the previous works done categorically, we can broadly categorize the works in

four main groups:

A. Heuristic Approach

i) Singular Point Based Method

ii) Global Ridge Structures

B. Structural Approach

i) Syntactic approaches

ii) Graph Matching

C. Neural Approach

D. Statistical Approach

The classification method is requiring the information from the entire fingerprint images and

it may be too restrictive for many applications. There are a lot of fingerprint classification

approaches that have been developed. Over the last thirty years, at least four major

approaches have been proposed by the researchers for classifying fingerprints. They are the

heuristic-based, structure-based, frequency-based and syntactic approached. The following

section discusses some of the general approaches used in fingerprint classification.

A. Heuristic Approach

i) Singular Point Based Method

Some of the rule-based approaches based on the singularity features, global ridge structure

information and sometimes the others used a combination of singularity and ridge features.

Henry introduced singular points because of their usefulness for classifying fingerprints. The

singularity features focus on detecting and extracting the number and location of core and

delta points in which it can be used to accurately classify fingerprints.

Though the core and delta points of fingerprints are not always necessary to be present, it still

appear as unique landmarks in fingerprint images that can be used as reference points for

classification process.

Fitz and Green proposed a Fourier transform method to locate the core points [59].

Rao and Black have been reported a syntactic method of extracting the singular points by

using tree grammar.

Karu and Jain have developed a six class fingerprint classification system based on heuristics

approach [6].

Another approach is using pruning based on neighbours and a relaxation method for noise

reduction to locate cores and deltas.


____________________________________________________________________________________

- 91 -

The Poincare Index is exploited to find type and position of the singular points and derives a

course classification. This approach is early proposed by Kawagoe and Tojo [85].

Srinivasan and Murthy used of structural information chose from the directional histogram of

the directional image of a fingerprint [38].

Zhang et al. suggested the corner detection method to find the region of singularities and grey

level of ridges was tracked to get the position of singular points.

Tong et al. proposed an indexing fingerprint based on location, direction, estimation and

correlation of singular points.

Liu Wei used the delta direction and the singularities to partition fingerprint classes.

In general, most of the current classification methods, whether structural methods or network-

based methods are based on the extraction of singular points in fingerprint image.

ii) Global Ridge Structures

Another technique that has been proposed for a heuristic classification system is based on the

representation of ridge structures. The ridge structures can be global features, which is these

features can often be reliably extracted even for noisy image.

Chong et al. used the global geometric shape of fingerprints to calculate the orientation of

fingerprint images. The classification system that proposed by the authors use the five

fingerprint classes.

A more robust technique is proposed by Hong and Jain. The authors introduced a rule based

classification algorithm that uses the number of singularities together with the global ridge

representation in fingerprint images. The combination of these two distinct features leads to a

better performance than that found in.

Ridge shape can also be incorporated with singularity information to aid classification. Jun Li

et al. combined an orientation and singularity information as an input features for fingerprint

classification.

Liu et al. used some curve features of ridgelines and employee the singular points.

B. Structural Approach

i) Syntactic approaches

A syntactic method is drawn between the structure of the input data’s features and the production rules. Every class has an associated set of rules which is defined as grammar that

describes how to build new sequence or sentences. In fingerprint classification system, each

fingerprint class would have a grammar that generates sequences corresponding to the class.


____________________________________________________________________________________

- 92 -

Recently, there are several approaches have been proposed by the researchers that do not

depend on singular points as the important aspects for fingerprint classification.

By Moayer and Fu, a syntactic approach was introduced for fingerprint classification. The

authors describe a class of context-free grammars as the fingerprint patterns, which is divided

into seven classes [106].

The same authors also experimented with other types of grammars such as stochastic

grammars and tree grammars.

Another approach proposed by Rao and Balck is based on the analysis of ridge line flow,

which is represented by a set of connected lines. These lines are labelled according to the

direction changes, thus obtaining a set of strings that are processes through grammars or

string-matching techniques to derive the final classification.

Chang and Fan developed a classification scheme that uses regular expressions to describe

the structure of fingerprint ridges.

In general, syntactic approaches require very complex grammars which require complicated

and unstable approaches due to the great diversity of fingerprint patterns. This method tends

to be robust in the presence of noise and can be designed to be invariant to translations and

rotations. However, syntactic method struggle with the large intra-class and small interclass

variations of fingerprint classes. In other words, the grammars that use for classification must

be able to recognize a wide variety of different sequences as being from the same class.

Besides that, it is able to differentiate very similar sequences from different classes.

ii) Graph Matching

Maio and Maltoni proposed a system that classifies fingerprints based on relational graphs

[69]. A model graph is created for each fingerprint class which has a typical structure of that

class. Further researches have been done on this technique. The authors used a template-

based matching method to guide the partitioning of the orientation field with the use of

dynamic masks. This approach is able to deal with partial fingerprints, where sometimes,

singular points are not available. The proposed method relies only on global structural

information and also can work on very noisy images. Relational graph on directional field is

introduced.

Relational graph approaches have interesting properties such as invariance to rotation and

displacement, and the possibility of handling partial fingerprints. It is not easy to robustly

partition the orientation image into homogeneous regions, especially in poor quality

fingerprint images. The relational graph for tented arches, left loops and right loops tend to

look similar. The strength of this approach is that the degree of similarity with three classes is

recorded and it is valuable discriminatory information. In state of this matter, it is beneficial

for relational graph approaches to exploit for continuous classification rather than forcing the

print into single arbitrary category.


____________________________________________________________________________________

- 93 -

C. Neural Approach

Most of the proposed neural network techniques are based on multilayer perceptrons and use

the orientation image information as an input features.

Kamijo presents an interesting pyramidal architecture constituted of several multilayer

perceptrons. Each of the perceptrons is trained to recognize fingerprints belonging to

different class.

Bowen used the location of the singularities together with orientation image for training the

two of disjoint neural networks.

Neural network computing on directional image is also adapted by few researchers.

NIST researchers used a multilayer perceptron for classification after reducing the

dimensionality of the feature vector. Later on the specific changes and optimizations in the

network architecture were introduced.

Neto and Borges developed a neural network classification system that uses of wavelet

features. These features are not very useful for fingerprint classification due to their

sensitivity to rotations and translations. The results from this approach are not very

impressive due to the limitations of the feature set.

A fuzzy-network classifier is used to classify fingerprints based on singularity features. The

features used include the number of core and delta points, the orientation of core points, the

relative position of core and delta points and the global direction of the orientation field.

Another fingerprint classification system used artificial neural networks.

Most of the neural network classification systems use the vectors from the orientation field as

the features for classification.

Dubravko used the homogeneity structure of fingerprint’s orientation field as the input vector for neural network classification system.

D. Statistical Approach

In statistical approaches, a fixed-size of numerical feature vector is derived from each

fingerprint and a general-purpose statistical classifier is used for the classification.

K-nearest neighbour is one of the most widely adopted statistical classifier that can be found.

The proposed method is based on Fourier transform for feature extraction.


____________________________________________________________________________________

- 94 -

VII.1.8. Fingerprint Recognition Steps

Figure 52 shows fingerprint matching steps and figure 53 shows the fingerprint authentication

steps.

Figure 52: Fingerprint matching steps.

Figure 53: Fingerprint authentication steps.

VII.1.8.1. Fingerprint Image Enhancement

The quality of the ridge structures in a fingerprint image is an important characteristic, as the

ridges carry the information of characteristic features required for minutiae extraction.

Ideally, in a well-defined fingerprint image, the ridges and valleys should alternate and flow

in locally constant direction. This regularity facilitates the detection of ridges and

consequently, allows minutiae to be precisely extracted from the thinned ridges. However, in

practice, a fingerprint image may not always be well defined due to elements of noise that

corrupt the clarity of the ridge structures. This corruption may occur due to variations in skin

and impression conditions such as scars, humidity, dirt, and non-uniform contact with the

fingerprint capture device [4]. Thus, image enhancement techniques are often employed to

reduce the noise and enhance the definition of ridges against valleys.

One of the most widely cited fingerprint enhancement techniques is the method employed by

Hong et al. [120], which is based on the convolution of the image with Gabor filters tuned to


____________________________________________________________________________________

- 95 -

the local ridge orientation and ridge frequency. The main stages of this algorithm include

normalisation, ridge orientation estimation, ridge frequency estimation and filtering.

The first step in this approach involves the normalization of the fingerprint image so that it

has a pre-specified mean and variance. Due to imperfections in the fingerprint image capture

process such as non-uniform ink intensity or non-uniform contact with the fingerprint capture

device, a fingerprint image may exhibit distorted levels of variation in grey-level values along

the ridges and valleys. Thus, normalization is used to reduce the effect of these variations,

which facilitates the subsequent image enhancement steps.

An orientation image is then calculated, which is a matrix of direction vectors representing

the ridge orientation at each location in the image. The widely employed gradient-based

approach is used to calculate the gradient [5], [144], [146], which makes use of the fact that

the orientation vector is orthogonal to the gradient. Firstly, the image is partitioned into

square blocks and the gradient is calculated for every pixel, in the x and y directions. The

orientation vector for each block can then be derived by performing an averaging operation

on all the vectors orthogonal to the gradient pixels in the block. Due to the presence of noise

and corrupted elements in the image, the ridge orientation may not always be correctly

determined. Given that the ridge orientation varies slowly in a local neighbourhood, the

orientation image is then smoothed using a low-pass filter to reduce the effect of outliers.

The next step in the image enhancement process is the estimation of the ridge frequency

image. The frequency image defines the local frequency of the ridges contained in the

fingerprint. Firstly, the image is divided into square blocks and an oriented window is

calculated for each block. For each block, an x-signature signal is constructed using the

ridges and valleys in the oriented window. The x-signature is the projection of all the grey

level values in the oriented window along a direction orthogonal to the ridge orientation.

Consequently, the projection forms a sinusoidal-shape wave in which the centre of a ridge

maps itself as a local minimum in the projected wave. The distance between consecutive

peaks in the x-signature can then be used to estimate the frequency of the ridges.

Fingerprint enhancement methods based on the Gabor filter have been widely used to

facilitate various fingerprint applications such as fingerprint matching [143], [149] and

fingerprint classification [129]. Gabor filters are band-pass filters that have both frequency-

selective and orientation-selective properties [148], which means the filters can be effectively

tuned to specific frequency and orientation values. One useful characteristic of fingerprints is

that they are known to have well defined local ridge orientation and ridge frequency.

Therefore, the enhancement algorithm takes advantage of this regularity of spatial structure

by applying Gabor filters that are tuned to match the local ridge orientation and frequency.

Based on the local orientation and ridge frequency around each pixel, the Gabor filter is

applied to each pixel location in the image. The effect is that the filter enhances the ridges

oriented in the direction of the local orientation, and decreases anything oriented differently.

Hence, the filter increases the contrast between the foreground ridges and the background,

whilst effectively reducing noise.


____________________________________________________________________________________

- 96 -

An alternative approach to enhancing the features in a fingerprint image is the technique

employed by Sherlock [145] called directional Fourier filtering. The previous approach was a

spatial domain technique that involves spatial convolution of the image with filters, which

can be computationally expensive. Alternatively operating in the frequency domain allows

one to efficiently convolve the fingerprint image with filters of full image size.

The image enhancement process begins by firstly computing the orientation image. In

contrast to the previous method, which estimates the ridge orientation using a continuous

range of directions, this method uses a set of only 16 directions to calculate the orientation.

An image window is centred at a point in the raw image, which is used to obtain a projection

of the local ridge information. The image window is then rotated in each of the 16 equally

spaced directions, and in each direction a projection along the window’s y axis is formed. The

projection with the maximum variance is used as the dominant orientation for that point in

the image. This process is then repeated for each pixel to form the orientation image.

Similar to the filtering stage applied by Hong et al. after the orientation image has been

computed, the raw image is then filtered using a set of band-pass filters tuned to match the

ridge orientation. The image is firstly converted from the spatial domain into the frequency

domain by application of the two-dimensional discrete Fourier transform. The Fourier image

is then filtered using a set of 16 Butterworth filters with each filter tuned to a particular

orientation. The number of directional filters corresponds to the set of directions used to

calculate the orientation image. After each directional filter has been independently applied to

the Fourier image, the inverse Fourier transform is used to convert each image back to the

spatial domain, thereby producing a set of directionally filtered images called pre-filtered

images.

The next step in the enhancement process is to construct the final filtered image using the

pixel values from the pre-filtered images. This requires the value of the ridge orientation at

each pixel in the raw image and the filtering direction of each pre-filtered image. Each point

in the final image is then computed by selecting, from the pre-filtered images the pixel value

whose filtering direction most closely matches the actual ridge orientation. The output of the

filtering stage is an enhanced version of the image that has been smoothed in the direction of

the ridges.

Lastly, local adaptive thresholding is applied to the directionally filtered image, which

produces the final enhanced binary image. This involves calculating the average of the grey-

level values within an image window at each pixel, and if the average is greater than the

threshold, then the pixel value is set to a binary value of one; otherwise, it is set to zero. The

grey-level image is converted to a binary image, as there are only two levels of interest, the

foreground ridges and the background valleys.

Overall, it can be seen that most techniques for fingerprint image enhancement are based on

filters that are tuned according to the local characteristics of fingerprint images. Both of the

examined techniques employ the ridge orientation information for tuning of the filter.

However, only the approach by Hong et al. takes into account the ridge frequency


____________________________________________________________________________________

- 97 -

information, as Sherlock’s approach assumes the ridge frequency to be constant. By using

both the orientation and ridge frequency information, it allows for accurate tuning of the

Gabor filter parameters, which consequently leads to better enhancement results. In this

dissertation both types of filtering approaches have been experimented to perform fingerprint

image enhancement.

VII.1.8.2. Minutiae Extraction and Image Post-processing

After a fingerprint image has been enhanced, the next step is to extract the minutiae from the

enhanced image. Following the extraction of minutiae, a final image post-processing stage is

performed to eliminate false minutiae.

VII.1.8.2.1. Minutiae extraction

The most commonly employed method of minutiae extraction is the Crossing Number (CN)

concept [144], [151], [152]. This method involves the use of the skeleton image where the

ridge flow pattern is eight-connected. The minutiae are extracted by scanning the local

neighbourhood of each ridge pixel in the image using a 3 X 3 window. The CN value is then

computed, which is defined as half the sum of the differences between pairs of adjacent

pixels in the eight-neighbourhood. Using the properties of the CN as shown in Table below,

the ridge pixel can then be classified as a ridge ending, bifurcation or non-minutiae point. For

example, a ridge pixel with a CN of one corresponds to a ridge ending, and a CN of three

corresponds to a bifurcation.

Table 3: Properties of the Crossing Number

Other authors such as Jain et al. [4] and Ratha et al. [5] have also performed minutiae

extraction using the skeleton image. Their approach involves using a 3 x 3 window to

examine the local neighbourhood of each ridge pixel in the image. A pixel is then classified

as a ridge ending if it has only one neighbouring ridge pixel in the window, and classified as

a bifurcation if it has three neighbouring ridge pixels. Consequently, it can be seen that this

approach is very similar to the Crossing Number method.

VII.1.8.2.2. Fingerprint Image Post-processing

False minutiae may be introduced into the image due to factors such as noisy images, and

image artefacts created by the thinning process. Hence, after the minutiae are extracted, it is


____________________________________________________________________________________

- 98 -

necessary to employ a post-processing stage in order to validate the minutiae. Figure below

illustrates some examples of false minutiae structures, which include the spur, hole, triangle

and spike structures [150]. It can be seen that the spur structure generates false ridge endings,

where as both the hole and triangle structures generate false bifurcations. The spike structure

creates a false bifurcation and a false ridge ending point.

Figure 54: Examples of typical false minutia structures.

The majority of the proposed approaches for image post-processing in literature are based on

a series of structural rules used to eliminate spurious minutiae.

One such approach is the one proposed by Ratha et al. [5], which performs the validation of

minutiae based on a set of heuristic rules. For example, a ridge ending point that is connected

to a bifurcation point, and is below a certain threshold distance is eliminated. This heuristic

rule corresponds to removal of the spike structure shown in Figure. Additional heuristic rules

are then used to eliminate other types of false minutiae. Furthermore, a boundary effect

treatment is applied where the minutiae below a certain distance from the boundary of the

foreground region are deleted.

A novel approach to the validation of minutiae is the post-processing algorithm proposed by

Tico and Kuosmanen [147]. Similar to the above techniques, this algorithm operates on the

skeleton image. However, rather than employing a different set of heuristics each time to

eliminate a specific type of false minutiae, this approach incorporates the validation of

different types of minutiae into a single algorithm. It tests the validity of each minutiae point

by scanning the skeleton image and examining the local neighbourhood around the minutiae.

The algorithm is then able to cancel out false minutiae based on the configuration of the ridge

pixels connected to the minutiae point.

VII.1.9. Conclusion

There have been many algorithms developed for extraction of both local and global

structures. Most algorithms found in the literature are somewhat tricky to implement and use

rather heuristic approach. The steadfastness of any automatic fingerprint recognition system

robustly relies on the exactness obtained in the extraction process. Extraction of proper

features is one of the most important tasks for a recognition system. The majority of the

fingerprint recognition systems rely on minutiae matching algorithms.


____________________________________________________________________________________

- 99 -

VII.2. Various Aspects of Minutia

as a Fingerprint Feature


Biometrics is one of the biggest tendencies in human identification. The fingerprint is the

most widely used biometric. Extracting minutiae from fingerprint images is one of the

most important steps in automatic fingerprint identification and classification. Minutiae

are local discontinuities in the fingerprint pattern, mainly terminations or end points and

bifurcations.

Fingerprints are the graphical flow-like ridges present on human fingers. Finger ridge

configurations do not change throughout the life of an individual except due to accidents such

as bruises and cuts on the fingertips. This property makes fingerprints a very attractive

biometric identifier. Fingerprint-based personal identification has been used for a very long

time [7]. Owning to their distinctiveness and stability, fingerprints are the most widely used

biometric features. Nowadays, most automatic fingerprint identification systems (AFIS) are

based on matching minutiae, which are local ridge characteristics in the fingerprint pattern.

Based on the features that the matching algorithms use, fingerprint matching can be classified

into image-based and graph-based matching.

Image-based matching [1] uses the entire gray scale fingerprint image as a template to match

against input fingerprint images. The primary shortcoming of this method is that matching

may be seriously affected by some factors such as contrast variation, image quality variation,

and distortion, which are inherent properties of fingerprint images. The reason for such

limitation lies in the fact that gray scale values of a fingerprint image are not stable features.

Graph-based matching [2], [63] represents the minutiae in the form of graphs. The high

computational complexity of graph matching hinders its implementation. To reduce the

computational complexity, matching the minutiae sets of template and input fingerprint

images can be done with point pattern matching. Several point pattern matching algorithms

have been proposed and commented in the literature [3], [4], [123], [164], [165].

Most of the fingerprint recognition systems first detect the minutiae in a fingerprint image

and then match the input image set with the template. A minutia is the unique, measurable

physical characteristics scanned as input and stored for matching by biometric systems. For

fingerprints, minutiae include the starting and ending points of ridges, bifurcations and ridge

junctions among other features. The most prevalent model for automated fingerprint

identification systems are based on minutiae.

A fingerprint is also represented in the form of graph whose nodes correspond to ridges and

edges represent ridge adjacency information. It is generally accepted that 8 to 17 distinct

minutiae matches between two fingerprints will conclude that the fingerprints match.


____________________________________________________________________________________

- 100 -

The first step in the approach involves binarization; converting the image to black and white

pixels whilst still retaining the distinct ridges. It is concluded that a global threshold is

inadequate in meeting these requirements. An adaptive threshold retains more information.

Median filtering helps to remove small noise segments present on the binarized image.

Without the application of a median filter, the noise would have made minutiae extraction

virtually impossible.

The thinning of the median filtered image results in the formation of spurs. These features

results in further false minutiae. Minutiae extraction yields many minutiae resulting from

fingerprint irregularities. The noise removal algorithms works particularly well at removing

the noise.

The purpose of this part is to discuss the minutia as a fingerprint feature to indentify people

from their fingerprints.

Figure 55: Ridge ending and ridge bifurcation.

VII.2.2. Definition and Types of Minutiae

A minutia is the unique, measurable physical characteristics scanned as input and stored for

matching by biometric systems. Minutiae (of fingerprints) include:

Ridge ending – the abrupt end of a ridge. Looks like this(-)

Ridge bifurcation – a single ridge that divides into two ridges

Short ridge, or independent ridge – a ridge that commences, travels a short distance and

then ends

Island – a single small ridge inside a short ridge or ridge ending that is not connected to

all other ridges

Ridge enclosure – a single ridge that bifurcates and reunites shortly afterward to continue

as a single ridge

Spur – a bifurcation with a short ridge branching off a longer ridge

Crossover or bridge – a short ridge that runs between two parallel ridges

Delta – a Y-shaped ridge meeting

Core – a U-turn in the ridge pattern


____________________________________________________________________________________

- 101 -

(a) (b)

(c)

Figure 56: Fingerprint image showing Minutiae.

VII.2.3. Components of Fingerprint Recognition System Using Minutiae

The following two modules are the main components of fingerprint recognition system using

minutiae:

Minutiae extraction: Minutiae are ridge endings or ridge bifurcations. Generally, if a perfect

segmentation can be obtained, then minutiae extraction is just a trivial task of extracting

singular points in a thinned ridge map. However, in practice, it is not always possible to

obtain a perfect ridge map. Some global heuristics need to be used to overcome this

limitation.

Minutiae matching: Minutiae matching, because of deformations in sensed fingerprints, is an

elastic matching of point patterns without knowing their correspondences beforehand.

Generally, finding the best match between two point patterns is intractable even if minutiae

are exactly located and no deformations exist between these two point patterns. The existence

of deformations makes the minutiae matching much more difficult.


____________________________________________________________________________________

- 102 -

VII.2.4. Fingerprint Pre-processing Techniques

Steps to extract minutiae as a feature are as follows:

VII.2.4.1. Binarization

Binarization is a method of transforming grayscale image pixels into either black or white

pixels by selecting a threshold. The process can be fulfilled using a multitude of techniques.

Binarization is relatively easy to achieve compared with other image processing techniques.

Unfortunately the global threshold technique sometimes proved to be troublesome in

determining appropriate thresholding levels. The resulting images obtained contain large

faded and large dark areas. Some globally binarized images are of an adequate standard. If

we Binarize using low pass filter then the method retains more of the information present in

the fingerprint than global threshold binarization.

(a) (b)

Figure 57: Thresholding

(a) binary image resulting from global thresholding.

(b) The histogram resulting form the original grayscale image.

Figure 58: An example of a poor image enhancement resulting from global binarization.


____________________________________________________________________________________

- 103 -

VII.2.4.2. Thresholding

The first technique considered focuses on finding the global threshold. The main black and

white pixel values of each image are determined. The pixel range in between these pixel

values is used to separate the black and white colours. Global binarization involves the

formulation of a histogram consisting of the number of pixels versus the pixel value.

Another method experimented with is called contrast enhancement binarization or Adaptive

Threshold Binarization. The method involves passing a low pass filter over the image and

using the resulting grayscale pixel number to discriminate between a black or white pixel.

The low pass filter does not process edges, one pixel wide in the image. Low-pass filtering

involves a spatial convolution process within a window.

VII.2.4.3. Median Filtering

Median filters calculate the average of pixel values in a pre-specified window size. The

central pixel is then assigned that value.

Median filtering removes a large majority of the noise. Noise is an unwanted perturbation to a

wanted signal. Image noise is generally regarded as an undesirable by-product of image

capture. The noise (small clusters of black) is averaged with its surroundings. After several

passes of the filter, the small clusters of noise disappear.

(a) (b)

Figure 59: The image after the (a) first and (b) seventh median filter

VII.2.4.4. Thinning

The aim of thinning is to reduce the fingerprint to lines one pixel wide. Thinning is a

morphological operation performed on binary images.


____________________________________________________________________________________

- 104 -

This is achieved by successive deletions of pixels from different sides of each image. Each of

the four sides (north, south, east, west) is eroded away according to some set template.

Eventually, the image being thinned will no longer possess any points which match the

deletion templates. This remaining image will be the thinned representation of the original

image.

False minutiae which are included in false minutiae structures, like spikes, holes, bridges,

ladder structures, and spurs are introduced to the fingerprint image, after thinning the original

image.

Figure 60: Effect of thinning: (a) Fingerprint , (b) Image after thinning.

Figure 61: Types of false minutiae structures.

From left to right and up to bottom we have: spike, bridge, hole, break, spur, and ladder structure. The false

minutiae generated by each structure are marked as (x) false ridge ending, and (o) false ridge bifurcation.


____________________________________________________________________________________

- 105 -

VII.2.4.5. Minutiae Detection

This type of method can only be performed on thinned images. It is known as crossing

number or connectivity number. The technique uses a sample window, 3 pixels by 3 pixels

wide to detect key features such as endpoints and bifurcation.

Minutiae detection is a trivial task. Without a loss of generality, it is assumed that if a pixel is

on a thinned ridge (eight-connected), then it has a value 1, and 0 otherwise.

Let (x, y) denote a pixel on a thinned ridge, and N0, N1, ..., N7 denote its eight neighbours.

A pixel (x, y) is

a ridge ending if = 1, and

a ridge bifurcation if > 2

However, the presence of undesired spikes and breaks present in a thinned ridge map may

lead to many spurious minutiae being detected. Therefore, before the minutiae detection, a

smoothing procedure is often applied to remove spikes and to join broken ridges. If several

minutiae form a cluster in a small region, then all of them except for the one nearest to the

cluster centre are removed.

The formula for the detection of key point is:

This is applied to the matrix:

Table 4: Minutiae and the corresponding crossing number.

P4 P3 P2

P5 P P1

P6 P7 P8


____________________________________________________________________________________

- 106 -

For each surviving minutia, the following parameters are needed to be recorded:

1) x-coordinate,

2) y-coordinate,

3) orientation which is defined as the local ridge orientation of the associated ridge, and

4) the associated ridge.

It is also needed to go through the alignment stage, where transformations such as translation,

rotation and scaling between an input and a template in the database are estimated and the

input minutiae are aligned with the template minutiae according to the estimated parameters.

Also we need to note that a false minutia is more affecting than missing minutiae.

VII.2.5. Directional Image

The directional image is usually used to derive the average direction of a small segment of

the image. The image is divided into 4 sub-directions. If required the image could be sorted

into a larger number of bins for greater segmentation.

The technique is used on the unprocessed gray-scale images. The darker pixels result in a

greater number in the corresponding sub-direction. A test area 16 pixels wide might be

selected. The actual test area depends on the dimensions of the entire image. The area should

include at least 1 ridge (dark area) and 1 valley.

Sub-direction = greatest [ (pixel_value-average)] ………………………….….. (2)

The technique gives similar results to the dominant ridge direction equation.

The technique for the classification of endpoint directions is easier to implement. The pixel

next to the endpoint determines the direction.

Unlike the bifurcation direction, endpoint directions are unidirectional.

(a) (b)

Figure 62: Directions of Minutiae:

(a) Possible Sub-directions of a bifurcation, (b) Possible directions of an endpoint.


____________________________________________________________________________________

- 107 -

VII.2.6. Feature Extraction

Many features will be extracted from each print. The co-ordinates of each minutia and the

type of it can be determined. The number of total minutiae is also recorded.

A fingerprint can have up to 80 minutiae. It is generally accepted as the same print if 8 to 17

points match. Some translation of the fingerprint will be acceptable, however rotation must

be minimized since no techniques have been implemented which specifically counteracts

rotation.

VII.2.7. Spur Removal

Spur removal helps to remove bifurcations and ends caused by thinning. If these points were

not removed, they would result in false endpoints and false bifurcations.

A compromise during spur removal must be met. Although it will remove some noise from

the thinned image, it will also move endpoints from their ‘real’ locations. This illustrates the dependency of filtering on the quality of the image. A quality image will require few or no

spur removal cycles.

When eroding spurs, normal endpoints are also eroded. This results in small variations of the

location of endpoints from their real locations.

VII.2.8. Minutiae Matching

Matching is a key operation in the fingerprint identification system. One of the most

important objectives of fingerprint systems is to achieve a high reliability in comparing the

input pattern with respect to the database pattern. Reliably matching fingerprint images is an

extremely difficult problem, mainly due to the large variability in different impressions of the

same finger (i.e., large intra-class variations).

The main factors responsible for the intra-class variations are: displacement, rotation, partial

overlap, non-linear distortion, variable pressure, changing skin condition, noise, and feature

extraction errors. Therefore, fingerprints from the same finger may sometimes look quite

different whereas fingerprints from different fingers may appear quite similar.

A minutia matching essentially consists of finding the alignment between the template and

the input minutiae sets that results in the maximum number of minutiae pairings. In Minutiae

based matching the similarity between the input and stored template are computed.

The implementation of a viable technique is quite difficult. Let’s consider the ideal. The

coordinates of both samples are identical. A simple coordinate matching algorithm would

suffice.


____________________________________________________________________________________

- 108 -

Figure 63: Identical matches of minutiae coordinates rarely match perfectly.

Unfortunately, this ideal situation rarely occurs. When a second print is recorded from the

same finger it is always misaligned from the original.

The elastic nature of skin means that some features may be stretched or warped, relative to

other sections of the print which retain their dimensions.

Noise will most likely occur, caused by applying too much pressure or smudging the print.

Slight rotation of the finger will also cause some features to vary from the original sample.

Figure 64: Sources of error in fingerprint recognition.

The searching algorithm must be flexible enough to allow some variance in coordinate

position. It must also attempt to distinguish the difference between real and false matches.

The main problem of the matching algorithm is false matches.


____________________________________________________________________________________

- 109 -

(a) (b)

(c)

Figure 65: Fingerprint Matching: (a) Fingerprint matching steps, (b) Fingerprint authentication steps,

(c) Fingerprint recognition using minutiae extraction technique.

Generally, an automatic fingerprint verification/identification is achieved with point pattern

matching (minutiae matching) instead of a pixel-wise matching or a ridge pattern matching of

fingerprint images. A number of point pattern matching algorithms have been proposed in the

literature [123], [162], [164], [165]. Because a general point matching problem is essentially

intractable, features associated with each point and their spatial properties such as the relative

distances between points are often used in these algorithms to reduce the exponential number

of search paths.

The relaxation approach [123] iteratively adjusts the confidence level of each corresponding

pair based on its consistency with other pairs until a certain criterion is satisfied.

Although a number of modified versions of this algorithm have been proposed to reduce the

matching complexity [164], these algorithms are inherently slow because of their iterative

nature.

The Hough transform-based approach proposed by Stockman et al. [163] converts point

pattern matching to a problem of detecting the highest peak in the Hough space of

transformation parameters. It discretizes the transformation parameter space and accumulates

evidence in the discretized space by deriving transformation parameters that relate two point

patterns using a substructure or feature matching technique. Karu and Jain [6] proposed a


____________________________________________________________________________________

- 110 -

hierarchical Hough transform-based registration algorithm which greatly reduced the size of

accumulator array by a multiresolution approach. However, if the number of minutiae point is

less than 30, then it is very difficult to accumulate enough evidence in the Hough transform

space for a reliable match.

Another approach to point matching is based on energy minimization. This approach defines

a cost function based on an initial set of possible correspondences and uses an appropriate

optimization algorithm such as genetic algorithm [162] and simulated annealing [165] to find

a possible suboptimal match. These methods tend to be very slow.

Recognition by alignment has received a great deal of attention during the past few years

[126], because it is simple in theory, efficient in discrimination, and fast in speed.

Alignment-based matching algorithm decomposes the minutiae matching into two stages:

1) Alignment stage, where transformations such as translation, rotation and scaling between

an input and a template in the database are estimated and the input minutiae are aligned with

the template minutiae according to the estimated parameters; and

2) Matching stage, where both the input minutiae and the template minutiae are converted to

polygons in the polar coordinate system and an elastic string matching algorithm is used to

match the resulting polygons.

Alignment of Point Patterns:

Ideally, two sets of planar point patterns can be aligned completely by two corresponding

point pairs. A true alignment between two point patterns can be obtained by testing all

possible corresponding point pairs and selecting the optimal one.

However, due to the presence of noise and deformations, the input minutiae cannot always be

aligned exactly with respect to those of the templates. In order to accurately recover pose

transformations between two point patterns, a relatively large number of corresponding point

pairs need to be used.

This leads to a prohibitively large number of possible correspondences to be tested.

Therefore, an alignment by corresponding point pairs is not practical even though it is

feasible.

It is well known that corresponding curve segments are capable of aligning two point patterns

with a high accuracy in the presence of noise and deformations. Each minutia in a fingerprint

is associated with a ridge. It is clear that a true alignment can be achieved by aligning

corresponding ridges.

During the minutiae detection stage, when a minutia is extracted and recorded, the ridge on

which it resides can also be recorded to achieve this.


____________________________________________________________________________________

- 111 -

Figure 66: Alignment of the input ridge and the template ridge.

Aligned Point Pattern Matching:

If two identical point patterns are exactly aligned with each other, each pair of corresponding

points is completely coincident. In such a case, point pattern matching can be simply

achieved by counting the number of overlapping pairs.

However, in practice, such a situation is not encountered.

On the one hand, the error in determining and localizing minutiae hinders the alignment

algorithm to recover the relative pose transformation exactly, while on the other hand,

alignment scheme mentioned above does not model the nonlinear deformation of fingerprints

which is an inherent property of fingerprint impressions.

With the existence of such a nonlinear deformation, it is impossible to exactly recover the

position of each input minutia with respect to its corresponding minutia in the template.

Therefore, the aligned point pattern matching algorithm needs to be elastic which means that

it should be capable of tolerating, to some extent, the deformations due to inexact extraction

of minutia positions and nonlinear deformations.

Usually, such an elastic matching can be achieved by placing a bounding box around each

template minutia, which specifies all the possible positions of the corresponding input

minutia with respect to the template minutia, and restricting the corresponding minutia in the

input image to be within this box [5].

This method does not provide a satisfactory performance in practice, because local

deformations may be small while the accumulated global deformations can be quite large.


____________________________________________________________________________________

- 112 -

VII.2.9. Match Probability

The total number of minutiae in a sample image can be used to equate a match factor. The

similarity equation is as follows:

M =((Nm*Nm)/(N1*N2)) ……………………………………………………………... (3)

where, N1 and N2 are the number of bifurcations in fingerprint 1 and 2 respectively. Nm is the

maximum number of matches acquired when the print was compared to all prints in the

database set.

A match will have similarity M greater than 0.9 while non-matching pairs will have a

similarity measure less than 0.3.

VII.2.10. Approaches May Be Taken to Improve the Accuracy

1) Count the number of ridges between each point - It would provide many more constrains

with which to limit the amount of false matches. By implementing this extra small step a

security system will be possible.

2) Use another method to align fingerprint images - A direct match search (same minutia in

same coordinates) could be applied to compare images. The centre feature (whirl, loop etc.)

of the fingerprint can be found (possibly using the Dominant Ridge Direction) to align the

image.

3) Sort fingerprints into classes (whirls, loops etc.) then only search through the sample

fingerprints category - This will increase processing time and reduce the probability of a false

match.

4) Obtain multiple samples - Obtaining multiple samples means the sample print can be

judged by the number of picture matches, as well as the number of pixel matches.

5) Reduce search area - One possible way of reducing the search area is to only look to the

left, right, above or below the minutia under analysis. Theoretically it should reduce the

amount of false matches (reduction in acceptance area).

VII.2.11. Discussion and Conclusion

Recent interest in automatic fingerprint classification system has inspired many groups to

conduct researches in this area.

All fingerprint images in database need to classify according to the pre-defined classification

criteria. This is great importance in order to overcome the accuracy and the identification

speed problems. A number of approaches have been applied for about several years that

differ in the features used to describe the important of classifying fingerprint image.


____________________________________________________________________________________

- 113 -

However, potential ways to improve the algorithms especially on pre-processing steps are

still needed to be studied.

Different soft computing approaches can be applied for calculating minutiae. By introducing

soft computing tools we can add intelligence to the recognition system, so that the system can

tell the likelihood of the particular image to be on a particular database and other intelligent

features can be introduced.

There have been many algorithms developed for extraction of minutiae. Most algorithms

found in the literature are somewhat difficult to implement and use a rather heuristic

approach.

The reliability of any automatic fingerprint recognition system strongly relies on the precision

obtained in the extraction process. Extraction of appropriate features is one of the most

important tasks for a recognition system.


____________________________________________________________________________________

- 114 -

VII.3. Fingerprint Recognition Method


Fingerprint Recognition is one of the most popular and reliable personal biometric

identification methods. Here let’s discuss about a fingerprint recognition model.

The pre-processing part includes steps to acquire binarized and skeletonized ridges, which are

needed for feature point extraction. Feature points (minutiae) such as endpoints, bifurcations,

and core point are then extracted, followed by false minutiae elimination. Human fingerprints

are rich in details, the so-called minutiae, which can be used as identification marks for

fingerprint verification.

Fingerprints are imprints formed by friction ridges of the skin and thumbs. They have long

been used for identification because of their immutability and individuality. Immutability

refers to the permanent and unchanging character of the pattern on each finger. Individuality

refers to the uniqueness of ridge details across individuals; the probability that two

fingerprints are alike is about 1 in 1.9x1015.

However, manual fingerprint verification/identification is so tedious, time consuming and

expensive that is incapable of meeting today’s increasing performance requirements. An

automatic fingerprint identification system is widely adopted in many applications such as

building or area security and ATM machines.

This recognition approach is based on minutiae located in a fingerprint.

VII.3.2. Approach

Most automatic systems for fingerprint comparison are based on minutiae matching [111].

Minutiae are local discontinuities in the fingerprint pattern. A total of almost 150 different

minutiae types have been identified. In practice only ridge ending and ridge bifurcation [12]

minutiae types are used in fingerprint recognition [109].

Many known algorithms have been developed for minutiae extraction based on orientation

and gradients of the orientation fields of the ridges. We can adopt the method used by Leung,

where minutiae are extracted using feed-forward artificial neural networks.

The building blocks of a fingerprint recognition system are: Image acquisition, Edge

detection, Thinning, Feature extractor, Classifier.


____________________________________________________________________________________

- 115 -

VII.3.2.1. Image Acquisition

The first stage of any vision system is the image acquisition stage. Image acquisition is

hardware dependent. A number of methods are used to acquire fingerprints. Among them, the

inked impression method remains the most popular one. Inkless fingerprint scanners are also

present eliminating the intermediate digitization process [40].

2D Image Input: The basic two-dimensional image is a monochrome (grayscale) image

which has been digitized. An image can be described as a two-dimensional light intensity

function f(x,y) where x and y are spatial coordinates and the value of f at any point (x, y) is

proportional to the brightness or grey value of the image at that point. For computational

purposes, we may think of a digital image as a two-dimensional array where x and y index an

image point. Each element in the array is called a pixel (picture element).

3D Image Input : A 3D image containing has many advantages over its 2D counterpart:

• 2D images give only limited information the physical shape and size of an object in a scene. • 3D images express the geometry in terms of three-dimensional coordinates. e.g Size (and

shape) of an object in a scene can be straightforwardly computed from its three-dimensional

coordinates.

VII.3.2.2. Edge Detection

An edge is the boundary between two regions with relatively distinct gray level properties.

The idea underlying most edge-detection techniques is on the computation of a local

derivative operator such as Sobel operators [113]. In practice, the set of pixels obtained from

the edge detection algorithm seldom characterizes a boundary completely because of noise,

breaks in the boundary and other effects that introduce spurious intensity discontinuities.

Thus, edge detection algorithms typically are followed by linking and other boundary

detection procedures designed to assemble edge pixels into meaningful boundaries.

VII.3.2.3. Thinning

Thinning [109] is a morphological operation that successively erodes away the foreground

pixels until they are one pixel wide. A standard thinning algorithm is employed, which

performs the thinning operation using two sub-iterations. This algorithm is accessible in

MATLAB via the `thin' operation under the ‘bwmorph’ function. Each sub-iteration begins

by examining the neighbourhood of each pixel in the binary image, and based on a particular

set of pixel-deletion criteria, it checks whether the pixel can be deleted or not. These sub-

iterations continue until no more pixels can be deleted. The application of the thinning

algorithm to a fingerprint image preserves the connectivity of the ridge structures while

forming a skeletonised version of the binary image. This skeleton image is then used in the

subsequent extraction of minutiae. An important approach to representing the structural shape


____________________________________________________________________________________

- 116 -

of a plane region is to reduce it to a graph. This reduction may be accomplished by obtaining

the skeleton of the region via thinning (also called skeletonizing) algorithm.

The thinning algorithm while deleting unwanted edge points should not:

• Remove end points,

• Break connectedness,

• Cause excessive erosion of the region.

VII.3.2.4. Feature Extraction

Extraction of appropriate features is one of the most important tasks for a recognition system.

The feature extraction method used in will be explained below. A multilayer perceptron [114]

(MLP) [105] of three layers is trained to detect the minutiae in the thinned fingerprint image

of size 300x300. The first layer of the network has nine neurons associated with the

components of the input vector. The hidden layer has five neurons and the output layer has

one neuron. The network is trained to output a “1” when the input window in centered on a

minutiae and a “0” when it is not.

This approach needs to state the number of epochs needed for convergence as well as the

training time for the two methods. Once the network is trained, the next step is to input the

prototype fingerprint images to extract the minutiae. The fingerprint image is scanned using a

3x3 window given.

VII.3.2.5. Classifier

After scanning the entire fingerprint image, the resulting output is a binary image revealing

the location of minutiae. In order to prevent any falsely reported output and select

“significant” minutiae, two more rules are added to enhance the robustness of the algorithm:

1) At those potential minutiae detected points, they are re-examined by increasing the

window size by 5x5 and scanning the output image.

2) If two or more minutiae are too close together (few pixels away) all of them are ignored.

To insure translation, rotation and scale in variance, the following operations are to be

performed:

• The Euclidean distance d(i) from each minutiae detected point to the centre is calculated.

The referencing of the distance data to the centre point guarantees the property of

positional invariance.

• The data will be sorted in ascending order from d(0) to d(N), where N is the number of

detected minutiae points, assuring rotational invariance.

• The data is then normalized to unity by shortest distance d (0), i.e: dnorm(i) = d(0)/d(i);

This will assure scale invariance property.


____________________________________________________________________________________

- 117 -

In the algorithm described, the centre of the fingerprint image was used to calculate the

Euclidean distance between the centre and the feature point. Usually, the centre or reference

point of the fingerprint image is what is called the “core” point. A core point, is located at the

approximate centre, is defined as the topmost point on the innermost upwardly curving

ridgeline.

The human fingerprint is comprised of various types of ridge patterns, traditionally classified

according to the decades-old Henry system: left loop, right loop, arch, whorl, and tented arch.

Loops make up nearly 2/3 of all fingerprints, whorls are nearly 1/3, and perhaps 5-10% are

arches.

Many singularity point detection algorithms were investigated to locate core points. For

simplicity let’s assume that the core point is located at the centre of the fingerprint image.

After extracting the location of the minutiae for the prototype fingerprint images, the

calculated distances will be stored in the database along with the ID or name of the person to

whom each fingerprint belongs.

The last phase is the verification phase where testing fingerprint image:

1. is inputted to the system,

2. minutiae are extracted,

3. Minutiae matching: comparing the distances extracted minutiae to the one stored in the

database

4. Identify the person: State the results obtained (i.e: recognition rate).

VII.3.3. Conclusion

This approach is based on minutiae located in a fingerprint. Further new recognition method

to be implemented based on a different approach, i.e frequency content and ridge orientation

of a fingerprint.


obtained in the minutiae extraction process. The minutiae based matching is highly sensible,

as, if the finger is moved even a little bit that gives us a different set of minutiae.


____________________________________________________________________________________

- 118 -

VII.4. Noise Related Issues


obtained in the feature extraction process. Extraction of appropriate features is one of the

most important tasks for a recognition system. Therefore a very important phase is noise

removal as even a small amount of noise can make the ridge structures different.

Noise is the random variation of brightness or colour information in images produced by the

sensor and circuitry of a scanner or digital camera. Image noise can also originate in film

grain and in the unavoidable shot noise of an ideal photon detector. Image noise is generally

regarded as an undesirable by-product of image capture.

Although these unwanted fluctuations became known as "noise" by analogy with unwanted

sound, they are inaudible and actually beneficial in some applications, such as dithering.

Two terms needed to understand for understanding the properties of noise are MSE & PSNR.

MSE: In statistics, the mean squared error (MSE) of an estimator is one of many ways to

quantify the difference between values implied by a kernel density estimator and the true

values of the quantity being estimated. MSE is a risk function, corresponding to the expected

value of the squared error loss or quadratic loss. MSE measures the average of the squares of

the errors. For images mean squared error (MSE) for two m×n monochrome

images I and K where one of the images is considered a noisy approximation of the other is

defined as:

MSE = 2

……………………………………….. (4)

PSNR: The phrase peak signal-to-noise ratio, often abbreviated PSNR, is an engineering

term for the ratio between the maximum possible power of a signal and the power of

corrupting noise that affects the fidelity of its representation. Because many signals have a

very wide dynamic range, PSNR is usually expressed in terms of

the logarithmic decibel scale.

The PSNR is defined as:

PSNR = 10.log10 ( / MSE)

= 20.log10 (MAX I / ) …………...………………………………… (5)

Typical values for the PSNR in lossy image and video compression are between 30 and

50 dB, where higher is better.


____________________________________________________________________________________

- 119 -

VII.4.1. De-noising of Fingerprint for False Minutiae Elimination

VII.4.1.1. Introduction

Most of the current fingerprint identification and verification systems perform fingerprint

matching based on different attributes of the minutia details present in fingerprints. The

minutiae (i.e. ridge endings and ridge bifurcations) are usually detected in the thinned binary

image of the fingerprint. Due to the presence of noise as well as the use of different pre-

processing stages the thinned binary image contains a large number of false minutiae which

may highly decrease the matching performances of the system. A new algorithm of

fingerprint image post-processing is proposed here. The algorithm operates onto the thinned

binary image of the fingerprint, in order to eliminate the false minutiae. The proposed

algorithm is able to detect and cancel the minutiae associated with most of the false minutia

structures which may be encountered in the thinned fingerprint image.

Figure 67: Fingerprint Minutiae.

VII.4.1.2. Background

Fingerprints have long been used for personal identification. It is assumed that every person

possesses unique fingerprints [121] and hence the fingerprint matching is considered one of

the most reliable techniques of people identification. A fingerprint image exhibits a

quasiperiodic pattern of ridges (darker regions) and valleys (lighter regions). The local

topological structures of this pattern together with their spatial relationships determine the

uniqueness of a fingerprint. There are more than 100 different types of local ridge structures

that have been identified [121]. Nevertheless, most of the automatic fingerprint identification

/ verification systems adopt the model used by the Federal Bureau of Investigation (FBI)

[17]. The model relies on representing only the two most prominent structures: ridge ending

and ridge bifurcation, which are collectively called minutiae.


____________________________________________________________________________________

- 120 -

Several methods of automatic minutiae extraction from fingerprint images have been

proposed in the literature [3], [5], [37], [106]. Although rather different from one other, most

of the proposed methods perform first a segmentation of the fingerprint ridges followed then

by a ridge thinning process which reduces the width of each ridge to one pixel. The candidate

minutiae are located in those pixels of the thinned binary image where the number of

outgoing branches is different of 2. Due to different types of noise which may be present in

the fingerprint image (e.g. under-inking, over-inking, scars or excessively worn prints) as

well as due to the segmentation and thinning processes, a large number of false minutiae are

encountered among the candidate minutiae detected in the thinned image. Therefore a post-

processing stage of minutiae purification is required before fingerprint matching. The false

minutiae may be identified in the thinned binary image either as part of false minutia

structures (e.g. spikes, bridges, holes, breaks, spurs, ladder structures) or at the boundary of

the image region where the fingerprint pattern is located (boundary effect). The problems of

fingerprint image post-processing for false minutiae elimination have been addressed before

by different authors. Ratha, Chen, and Jain [5] use three heuristic rules in order to eliminate

ridge breaks, spikes and boundary effect. Hung [39] proposes a set of algorithms for detecting

and removing spurs, holes and bridges. Hung's algorithms use the duality between the ridge

and valley structures whose thinned versions are represented as graphs. A combined

statistical and structural approach is proposed by Xiao and Raafat [150] in order to remove

ridge breaks, and false bifurcations. The post-processing algorithm proposed in this

experiment is able to detect and cancel noises which are included in the false minutia

structures like spikes, holes, bridges, ladder structures, and spurs.

The organization of the part is as follows. This part serves as the first section. The following

section will briefly review the most important types of false minutia structures as well as

some of the main reasons these structures occur in the thinned ridge map image. The

proposed algorithm is presented in the next. The experimental results are shown in last and

finally some concluding remarks are presented.

VII.4.1.3. False Minutia Structures

The most common types of false minutia structures which may be encountered into a thinned

fingerprint image are shown in Figure 68. Each such structure generates two or more false

minutiae. The spike structure generates two false minutiae and may occur when thinning a

non-smooth ridge. The bridge and ladder structures usually occur between close ridges. Very

wide ridges may generate hole structures and very wide valleys may generate spurs. The

presence of scars in the fingerprint may determine ridge breaks in the thinned ridge map

image. In addition a large number of false minutiae are always detected close to the boundary

of the region of interest (boundary effect). The boundary effect is treated by cancelling all

minutiae which are below a certain distance to the boundary of the fingerprint pattern. It is of

importance to mention that false minutia structures may also be caused by different image

processing operations used to obtain the thinned ridge map image from the original gray-

scale fingerprint image.


____________________________________________________________________________________

- 121 -

Figure 68: Types of false minutia structures. From left to right and up to bottom we have: spike, bridge, hole,

break, spur, and ladder structure. The false minutiae generated by each structure are marked as (x) false ridge

ending, and (o) false ridge bifurcation.

VII.4.1.4. The Proposed Work

This proposed work is done by the Median Filter and the Sobel Operator.

Median filter: Neighbourhood averaging can suppress isolated out-of-range noise, but the

side effect is that it also blurs sudden changes (corresponding to high spatial frequencies)

such as sharp edges. The median filter is an effective method that can suppress isolated noise

without blurring sharp edges. Specifically, the median filter replaces a pixel by the median of

all pixels in the neighbourhood:

y[m,n] = median {x[i,j], (i,j) ϵ w} ………………... ………………………………….…… (6)

Where, w represents a neighbourhood centred around location (m,n) in the image.

1D median filter: Consider a 1x5 window sliding over a 1D array (either horizontal or

vertical) of pixels. Assume the five pixels currently inside the windows are:


____________________________________________________________________________________

- 122 -

where, the middle pixel with value 200 is an isolated out-of-range noise. The median of these

five values can be found by sorting the values (in either ascending or descending order). The

middle value is the median:

The original pixel value 200 is replaced by the median 110.

How to get rid of noise in the form of horizontal line across the image using 1D median filter

can be the question.

2D median filter: The window of a 2D median filter can be of any central symmetric shape,

a round disc, a square, a rectangle, or a cross. The pixel at the centre will be replaced by the

median of all pixel values inside the window.

Programming issues:

Sorting is necessary for finding the median of a set of values. There exist various sorting

algorithms with complexity of O(n log2n). However, in this case, as the number of pixels is

quite limited, a simple sorting method with complexity O(n2) can be used.

The code segment next sorts an array of k elements:

for (i=0; I <k-1; i++

for (j=i+1; j<k; j++)

if (bin[i] > bin[j])

{w = bin [i]; bin[i] = bin [j]; bin [j] = w;}

The median filter is also a sliding-window spatial filter, but it replaces the centre value in the

window with the median of all the pixel values in the window. As for the mean filter, the

kernel is usually square but can be of any shape.

An example of median filtering of a single 3x3 window of values is shown next.

unfiltered values

6 2 0

3 97 4

19 3 10

in order: 0, 2, 3, 3, 4, 6, 10, 15, 97


____________________________________________________________________________________

- 123 -

median filtered

* * *

* 4 *

* * *

Centre value (previously 97) is replaced by the median of all nine values (4).

Note that for the first example, the median filter would also return a value of 5, since the

ordered values are 1, 2, 3, 4, 5, 6, 7, 8, 9.

For the second example, though, the mean filter returns the value 16 since the sum of the nine

values in the window is 144 and 144 / 9 = 16.

This illustrates one of the celebrated features of the median filter: its ability to remove

'impulse' noise (outlying values, either high or low).

The median filter is also widely claimed to be 'edge-preserving' since it theoretically

preserves step edges without blurring.

However, in the presence of noise it does blur edges in images slightly.

Sobel Operator: The Sobel operator is used in image processing, particularly within edge

detection algorithms.

Technically, it is a discrete differentiation operator, computing an approximation of the

gradient of the image intensity function.

At each point in the image, the result of the Sobel operator is either the corresponding

gradient vector or the norm of this vector.

The Sobel operator is based on convolving the image with a small, separable, and integer

valued filter in horizontal and vertical direction and is therefore relatively inexpensive in

terms of computations.

On the other hand, the gradient approximation which it produces is relatively crude, in

particular for high frequency variations in the image.

In simple terms, the operator calculates the gradient of the image intensity at each point,

giving the direction of the largest possible increase from light to dark and the rate of change

in that direction.

The result therefore shows how "abruptly" or "smoothly" the image changes at that point and

therefore how likely it is that that part of the image represents an edge, as well as how that

edge is likely to be oriented.


____________________________________________________________________________________

- 124 -

In practice, the magnitude (likelihood of an edge) calculation is more reliable and easier to

interpret than the direction calculation.

Mathematically, the gradient of a two-variable function (here the image intensity function) is

at each image point a 2D vector with the components given by the derivatives in the

horizontal and vertical directions. At each image point, the gradient vector points in the

direction of largest possible intensity increase, and the length of the gradient vector

corresponds to the rate of change in that direction.

This implies that the result of the Sobel operator at an image point which is in a region of

constant image intensity is a zero vector and at a point on an edge is a vector which points

across the edge, from darker to brighter values.

Mathematically, the operator uses two 3×3 kernels which are convolved with the original

image to calculate approximations of the derivatives - one for horizontal changes, and one for

vertical.

If we define A as the source image, and Gx and Gy are two images which at each point

contain the horizontal and vertical derivative approximations, the computations are as

follows:

Gy = * A ……………………………………………………………… (7)

and

Gx = * A ..……………………………………………………………... (8)

where, * denotes the 2-dimensional convolution operation.

The x-coordinate is here defined as increasing in the "right"-direction, and the y-coordinate is

defined as increasing in the "down"-direction.

At each point in the image, the resulting gradient approximations can be combined to give the

gradient magnitude, using:

G = ….……………………………….………........................ (9)

Using this information, we can also calculate the gradient's direction:

Θ = arctan (Gy/Gx) ………………………………………………….……… (10)

where, for example, Θ is 0 for a vertical edge which is darker on the left side.

The Sobel operator represents a rather inaccurate approximation of the image gradient, but is

still of sufficient quality to be of practical use in many applications.

http://en.wikipedia.org/wiki/Convolution

http://en.wikipedia.org/wiki/Convolution


____________________________________________________________________________________

- 125 -

More precisely, it uses intensity values only in a 3×3 region around each image point to

approximate the corresponding image gradient, and it uses only integer values for the

coefficients which weight the image intensities to produce the gradient approximation.

As a consequence of its definition, the Sobel operator can be implemented by simple means

in both hardware and software; only eight image points around a point are needed to compute

the corresponding result and only integer arithmetic is needed to compute the gradient vector

approximation.

Furthermore, the two discrete filters described above are both separable:

= ………………...………………………………… (11)

and

= .…………………………………………………….. (12)

and the two derivatives Gx and Gy can therefore be computed as:

Gx = * A) ………………...…………………………………….. (13)

and

Gy = * A) ……..………………………………………………..…... (14)

In certain implementations, this separable computation may be advantageous since it implies

fewer arithmetic computations for each image point.

Applying convolution K to pixel group P can be represented in pseudo code as:

N(x,y) = Sum of { K(i,j).P(x-i,y-j)}, for i, j running from -1 to 1.

N(x,y) represents the new matrix resulted after applying the Convolution K to P, where P is

pixel matrix.


____________________________________________________________________________________

- 126 -

VII.4.1.5. Proposed Algorithm

A. Structure for image is declared.

B. Read a noise image file. The image element is stored in a matrix, say Gray.

C. A noise removed file is obtained.

D. So, following operation has performed.

i. For each: x=2 to row of image matrix.

For each: y=2 to column of image matrix.

A temp (say) array have taken which stores 25 element for each x=i & y=i using (ii).

ii. For one value of x and y:

For each i=-2 to 2

For each j=-2 to 2

Temp array will assign 25 elements.

Go to step (iii).

iii. For one value of x and y, temp[25] elements have sorted and then temp[12] is

assigned for sorting to gray[x][y].

Go to step (i).

iv. x > row of image matrix and

y > column of image matrix.

Then Stop.

E. A file <name> is obtained by using the operation:-

i. For each x=0 to row of an image matrix

For each x=0 to row of an image matrix.

(ii) operation occurred.

ii. For each i = -1 to 1,

For each j = -1 to 1, a value sum is obtained such that

Sum = sum + gray [x+i][y+i]*sob[i+1][j+1]

Where,

Sob[3][3] = [-1, 0, 1

-2, 0, 2

-1, 0, 1]

Filter mask is obtained.

iii. The value of sum is scaled to 255 or 0.

iv. Go to step(i).

v. x > row of image matrix and

y > column of image matrix.

Then Stop.


____________________________________________________________________________________

- 127 -

VII.4.1.6. Experimental Results

(a) (b)

Figure 69: Experimental Result of Noise Removal:

(a) Input image with noise, (b) output image after noise removal.

VII.4.1.7. Conclusion

A new algorithm of fingerprint post-processing has been proposed here.

The proposed algorithm is able to detect and reduce the noise and is able to do image

sharpening associated with spikes, holes, bridges, ladder structures and spurs.

The experimental results revel that the proposed algorithm cancels a large number of false

minutiae and in particular better performances are achieved in cancelling false ridge

bifurcations.

Also works were done to develop more robust techniques for detecting and cancelling false

ridge endings associated with ridge break structures.


____________________________________________________________________________________

- 128 -

VII.4.2. Algorithm for Introducing Noise in Image



matching based on different attributes of the minutia details present in fingerprints. The

minutiae (i.e. ridge endings and ridge bifurcations) are usually detected in the thinned binary

image of the fingerprint.

Due to the presence of noise as well as the use of different pre-processing stages the thinned

binary image contains a large number of false minutiae which may highly decrease the

matching performances of the system.

A new algorithm of fingerprint image for introduce noise is proposed here.

The algorithm operates onto the image of the fingerprint, in order to introduce the noise. The

proposed algorithm is able to introduce noise but it will be percentage-wise according to the

user’s choice and it can be used for security purpose.

A fingerprint image exhibits a quasiperiodic pattern of ridges (darker regions) and valleys

(lighter regions). The local topological structures of this pattern together with their spatial

relationships determine the uniqueness of a fingerprint. There are more than 100 different

types of local ridge structures that have been identified. Nevertheless, most of the automatic

fingerprint identification / verification systems adopt the model used by the Federal Bureau

of Investigation. The model relies on representing only the two most prominent structures:

ridge ending and ridge bifurcation, which are collectively called minutiae.

Several methods of automatic minutiae extraction from fingerprint images have been

proposed in the literature. Although rather different from one other, most of the proposed

methods perform first a segmentation of the fingerprint ridges followed then by a ridge

thinning process which reduces the width of each ridge to one pixel. The candidate minutiae

are located in those pixels of the thinned binary image where the number of outgoing

branches is different of two.

Due to different types of noise which may be present in the fingerprint image (e.g. under-

inking, over-inking, scars or excessively worn prints) as well as due to the segmentation and

thinning processes, a large number of false minutiae are encountered among the candidate

minutiae detected in the thinned image. Therefore a post-processing stage of minutiae

purification is required before fingerprint matching. The false minutiae may be identified in

the thinned binary image either as part of false minutia structures (e.g. spikes, bridges, holes,

breaks, spurs, ladder structures) or at the boundary of the image region where the fingerprint

pattern is located (boundary effect).

The problems of fingerprint image post-processing for false minutiae elimination have been

addressed before by different authors. Ratha, Chen, and Jain use three heuristic rules in order


____________________________________________________________________________________

- 129 -

to eliminate ridge breaks, spikes and boundary effect [5]. Hung proposes a set of algorithms

for detecting and removing spurs, holes and bridges [39]. Hung's algorithms use the duality

between the ridge and valley structures whose thinned versions are represented as graphs. A

combined statistical and structural approach is proposed by Xiao and Raafat in order to

remove ridge breaks, and false bifurcations [150].

The algorithm proposed here is able to introduce noises which are included in the false

minutia structures like spikes, holes, bridges, ladder structures, and spurs. The experimental

results are shown in last and finally some concluding remarks are presented.

VII.4.2.2. Image Noise

It is the random variation of brightness or colour information in images produced by the

sensor and circuitry of a scanner or digital camera. Image noise can also originate in film

grain and in the unavoidable shot noise of an ideal photon detector.

Image noise is generally regarded as an undesirable by-product of image capture. Although

these unwanted fluctuations became known as "noise" by analogy with unwanted sound, they

are inaudible and actually beneficial in some applications, such as dithering.

Figure 70: Noisy Fingerprint


____________________________________________________________________________________

- 130 -

Type of Noise:-

i) Amplifier noise (Gaussian noise): The standard model of amplifier noise is additive,

Gaussian, independent at each pixel and independent of the signal intensity, caused primarily

by Johnson–Nyquist noise (thermal noise), including that which comes from the reset noise

of capacitors ("kTC noise"). In colour cameras where more amplification is used in the blue

colour channel than in the green or red channel, there can be more noise in the blue channel.

Amplifier noise is a major part of the "read noise" of an image sensor, that is, of the constant

noise level in dark areas of the image.

ii) Salt-and-pepper noise: Fat-tail distributed or "impulsive" noise is sometimes called salt-

and-pepper noise or spike noise. An image containing salt-and-pepper noise will have dark

pixels in bright regions and bright pixels in dark regions. This type of noise can be caused by

dead pixels, analogue-to-digital converter errors, bit errors in transmission, etc. This can be

eliminated in large part by using dark frame subtraction and by interpolating around

dark/bright pixels.

iii) Shot noise: The dominant noise in the lighter parts of an image from an image sensor is

typically that caused by statistical quantum fluctuations, that is, variation in the number of

photons sensed at a given exposure level; this noise is known as photon shot noise. Shot noise

has a root-mean-square value proportional to the square root of the image intensity, and the

noises at different pixels are independent of one another. Shot noise follows a Poisson

distribution, which is usually not very different from Gaussian. In addition to photon shot

noise, there can be additional shot noise from the dark leakage current in the image sensor;

this noise is sometimes known as "dark shot noise" or "dark-current shot noise". Dark current

is greatest at "hot pixels" within the image sensor; the variable dark charge of normal and hot

pixels can be subtracted off (using "dark frame subtraction"), leaving only the shot noise, or

random component, of the leakage; if dark-frame subtraction is not done, or if the exposure

time is long enough that the hot pixel charge exceeds the linear charge capacity, the noise

will be more than just shot noise, and hot pixels appear as salt-and-pepper noise.

iv) Quantization noise (uniform noise): The noise caused by quantizing the pixels of a

sensed image to a number of discrete levels is known as quantization noise; it has an

approximately uniform distribution, and can be signal dependent, though it will be signal

independent if other noise sources are big enough to cause dithering, or if dithering is

explicitly applied.

v) Film grain: The grain of photographic film is a signal-dependent noise, related to shot

noise. That is, if film grains are uniformly distributed (equal number per area), and if each

grain has an equal and independent probability of developing to a dark silver grain after

absorbing photons, then the number of such dark grains in an area will be random with a

binomial distribution; in areas where the probability is low, this distribution will be close to

the classic Poisson distribution of shot noise; nevertheless a simple Gaussian distribution is

often used as an accurate enough model. Film grain is usually regarded as a nearly isotropic

(non-oriented) noise source, and is made worse by the distribution of silver halide grains in

the film also being random.


____________________________________________________________________________________

- 131 -

vi) Non-isotropic noise: Some noise sources show up with a significant orientation in

images. For example, image sensors are sometimes subject to row noise or column noise. In

film, scratches are an example of non-isotropic noise.

VII.4.2.3. Proposed Work


matching based on different attributes of the minutia details present in fingerprints.

The minutiae (i.e. ridge endings and ridge bifurcations) are usually detected in the thinned

binary image of the fingerprint.

Due to the presence of noise as well as the use of different pre-processing stages the thinned

binary image contains a large number of false minutiae which may highly decrease the

matching performances of the system.

A new algorithm of fingerprint image for introducing noise is proposed here. The algorithm

operates onto the image of the fingerprint, in order to introduce the noise.

The proposed algorithm is able to introduce noise but it will be percentage-wise according to

the user’s choice and it can be used for security purpose.

VII.4.2.4. Proposed Algorithm

Step1: take an input image say “inimage.pgm” in inmatrix.

Step2: take an output image say “outimage.pgm” in outimatrix.

Step3: store the value of inmatrix to outmatrix.

Step4: enter percentage for introducing noise.

Step5: now call a function (outmatrix, row, col, per).

Step6: function (outmatrix, row, col, per) which calculates the amount of noise to each

element.

Step7: for completion of step 6, call another function “binadd (int x)”; the function binadd

adjust the corresponding element of pixel in the range of 0-255 using binary value concept.


____________________________________________________________________________________

- 132 -

VII.4.2.5. Experimental Results

Figure 71: Results


A new algorithm has been proposed here. The proposed algorithm is able to introduce the

noise according to the users’ choice. It can be used for security purpose, by encrypting the

image with induced noise at the sending end and decrypting at the receiving end.


____________________________________________________________________________________

- 133 -

VII.5. Fingerprints in Frequency Domain

Different properties of fingerprints in frequency domain are also the present focus of my

research. Some surveys are done and also some implementations and experiments are done.

VII.5.1. Fingerprints Pre-processing in Frequency Domain


Fingerprints are the oldest and most widely used form of biometric identification. A critical

step in studying the fingerprint minutiae is to reliably extract minutiae from the fingerprint

images. However, fingerprint images are rarely of perfect quality. They may be degraded and

corrupted due to variations in skin and impression conditions. Thus, image enhancement

techniques are employed prior to minutiae extraction to obtain a more reliable estimation of

minutiae locations.

In electronics, control systems engineering, and statistics, frequency domain is a term used to

describe the domain for analysis of mathematical functions or signals with respect to

frequency, rather than time. The reason for doing the filtering in the frequency domain is

generally because it is computationally faster to perform two 2D Fourier transforms and a

filter multiplication, than to perform a convolution in the image (spatial) domain. This is

particularly so as the filter size increases.

Figure 72: Image of varying frequencies.

Much signal processing is done in a mathematical space known as the frequency domain. In

order to represent data in the frequency domain, some transform is necessary. Perhaps the


____________________________________________________________________________________

- 134 -

most studied one is the Fourier transform. Any signal is composed of different frequencies.

This applies to 1-dimensional signals such as an audio signal going to a speaker or a 2-

dimensional signal such as an image. The spatial frequency of an image refers to the rate at

which the pixel intensities change. Figure 72 shows an image consisting of different

frequencies.

The high frequencies are concentrated around the axes dividing the image into quadrants.

High frequencies are noted by concentrations of large amplitude swings in the small

checkerboard pattern. The corners have lower frequencies. Low spatial frequencies are noted

by large areas of nearly constant values.

The easiest way to determine the frequency composition of signals is to inspect that signal in

the frequency domain. The frequency domain shows the magnitude of different frequency

components. A simple example of a Fourier transform is a cosine wave. Many different

transforms are used in image, due to its wide range of applications in image processing, the

Fourier transform is one of the most popular. It operates on a continuous function of infinite

length. It is also possible to transform image data from the frequency domain back to the

spatial domain. This is done with an inverse Fourier transform.

VII.5.1.2. Why Frequency Domain

Frequency domain allows for techniques which could be used to determine the stability of the

system. Also, these techniques can be used in conjunction with the S-domain (Laplace

transform) which gives more insight to the stability of the system, transient response, and

steady state response. For a filter which has a large spatial extent, the frequency domain

techniques will be generally faster. For an input with N elements and a filter with M elements

where M is less than or equal to N, the spatial domain techniques generally perform on the

order of N*M calculations and the frequency domain techniques perform N*log(N)

calculations.

VII.5.1.3. Types of Frequency Domain

Frequency domain analysis can be one of the following types: (These are the most common

transforms and the fields in which they are used)

Fourier series – repetitive signals, oscillating systems

Fourier transform – nonrepetitive signals, transients

Laplace transform – electronic circuits and control systems

Wavelet transform – digital image processing, signal compression

Z transform – discrete signals, digital signal processing

Gaussian Filter

Gabor filter.


____________________________________________________________________________________

- 135 -

VII.5.1.4. Fourier Transform

The Fourier Transform is an important image processing tool which is used to decompose an

image into its sine and cosine components. The output of the transformation represents the

image in the Fourier or frequency domain, while the input image is the spatial domain

equivalent. In the Fourier domain image, each point represents a particular frequency

contained in the spatial domain image. The Fourier Transform is used in a wide range of

applications, such as image analysis, image filtering, image reconstruction and image

compression. For digital images the Discrete Fourier Transform (DFT) is used.

The DFT is the sampled Fourier Transform and therefore does not contain all frequencies

forming an image, but only a set of samples which is large enough to fully describe the

spatial domain image. The number of frequencies corresponds to the number of pixels in the

spatial domain image. For a square image of size N×N, the two-dimensional DFT is given by:

F (k,l) = e–i2п (ki/N + lj/N)

….……………………………………..…….. (15)

where f(a,b) is the image in the spatial domain and the exponential term is the basis function

corresponding to each point F(k,l) in the Fourier space. The equation can be interpreted as:

the value of each point F(k,l) is obtained by multiplying the spatial image with the

corresponding base function and summing the result.

The basis functions are sine and cosine waves with increasing frequencies, i.e. F(0,0)

represents the DC-component of the image which corresponds to the average brightness and

F(N-1,N-1) represents the highest frequency.

In a similar way, the Fourier image can be re-transformed to the spatial domain. The inverse

Fourier transform is given by:

f(a,b) = ei2п (ka/N + lb/N)

……………………………………….….. (16)

To obtain the result for the above equations, a double sum has to be calculated for each image

point. However, because the Fourier Transform is separable, it can be written as:

F(k,l) = e-i2п lb/N

……...…………………………...…………………… (17)

where,

P(k,b) = e-i2п ka/N

….………………………………………….……. (18)

Using these two formulas, the spatial domain image is first transformed into an intermediate

image using N one-dimensional Fourier Transforms. This intermediate image is then

transformed into the final image, again using N one-dimensional Fourier Transforms.

Expressing the two-dimensional Fourier Transform in terms of a series of 2N one-

dimensional transforms decreases the number of required computations. Even with these

computational savings, the ordinary one-dimensional DFT has N2 complexity. This can be

reduced to N log2N if we employ the Fast Fourier Transform (FFT) to compute the one-


____________________________________________________________________________________

- 136 -

dimensional DFTs. This is a significant improvement, in particular for large images. There

are various forms of the FFT and most of them restrict the size of the input image that may be

transformed, often to N= 2n where n is an integer.

VII.5.1.5. What Do Frequencies Mean in an Image

If an image has large values at high frequency components then the data is changing rapidly

on a short distance scale. e.g. a page of text. If the image has large low frequency components

then the large scale features of the picture are more important. e.g. a single fairly simple

object which occupies most of the image.

For colour images, the measure (now a 2D matrix) of the frequency content is with regard to

colour/chrominance: this shows if values are changing rapidly or slowly. Where the fraction

or value in the frequency matrix is low, the colour is changing gradually. Now the human eye

is insensitive to gradual changes in colour and sensitive to intensity. So we can ignore gradual

changes in colour and throw away data without the human eye noticing.

VII.5.1.6. How Transforming into the Frequency Domain Helps

Any function can be decomposed into purely sinusoidal components (sine waves of different

size/shape) which when added together make up the original signal. Thus transforming a

signal into the frequency domain allows us to see what sine waves make up the signal. More

complex signals will give more complex graphs but the idea is exactly the same. The graph of

the frequency domain is called the frequency spectrum.

VII.5.1.7. Fourier Transform and Image Processing

The Fourier Transform produces a complex number valued output image which can be

displayed with two images, either with the real and imaginary part or with magnitude and

phase. In image processing, often only the magnitude of the Fourier Transform is displayed,

as it contains most of the information of the geometric structure of the spatial domain image.

However, if we want to re-transform the Fourier image into the correct spatial domain after

some processing in the frequency domain, we must make sure to preserve both magnitude

and phase of the Fourier image.

The Fourier domain image has a much greater range than the image in the spatial domain.

Hence, to be sufficiently accurate, its values are usually calculated and stored in float values.

The Fourier Transform is used if we want to access the geometric characteristics of a spatial

domain image. Because the image in the Fourier domain is decomposed into its sinusoidal


____________________________________________________________________________________

- 137 -

components, it is easy to examine or process certain frequencies of the image, thus

influencing the geometric structure in the spatial domain.

The result shows that the image contains components of all frequencies, but that their

magnitude gets smaller for higher frequencies. Hence, low frequencies contain more image

information than the higher ones. The transform image also tells us that there are two

dominating directions in the Fourier image, one passing vertically and one horizontally

through the centre. These originate from the regular patterns in the background of the original

image.

VII.5.1.8. High Pass and Low Pass Filters

Applying a low pass filter in the frequency domain means zeroing all frequency components

above a cutoff frequency.

Applying a high pass filter frequency domain is the opposite of the low pass filter, that is, all

the frequencies below some cutoff radius are removed. It is a device that passes high

frequencies and attenuates (i.e., reduces the amplitude of) frequencies lower than its cutoff

frequency. A high-pass filter is usually modelled as a linear time-invariant system.

All the frequency filters used for low pass filter can be used for high pass filtering as well.

Apparently higher noise levels are false, and the graphs are auto scaled and thus the field only

appears larger because of the removal of the low frequency components.

VII.5.1.9. Steps of Frequency Domain and Filtering in Frequency Domain

(Fourier Transform)

With the frequency domain techniques, care needs to be taken in choosing the size of the

input region: certain sizes (those that have small prime factors) are handled much more

efficiently.

By default, this is done automatically by adding additional elements so that the input can be

handled efficiently and then stripping these additional elements off before writing the result.

To change this behaviour, modification needed for the edge handling and output size inputs.

Frequency domain techniques treat the right edge (after padding) as contiguous with the left

edge; the top and bottom are handled similarly. This can be problematic when the values

along the edges are very different.

Pre-processing the data to remove background trends which cause the edges to be different or

padding the data with additional elements so that the number of elements added is greater

than the approximate spatial extent of the filter are two possible approaches to mitigate this

problem.


____________________________________________________________________________________

- 138 -

Filtering in the frequency domain is a common image and signal processing technique. It can

smooth, sharpen, de-blur, and restore some images. There are three basic steps to frequency

domain filtering:

The image must be transformed from the spatial domain into the frequency domain using

the Fast Fourier transform.

The resulting complex image must be multiplied by a filter (that usually has only real

values).

The filtered image must be transformed back to the spatial domain.

These steps can be implemented using the Fast Fourier transform function FFT. A -1

argument to the FFT function indicates a transformation from the spatial to the frequency

domain, and a 1 indicates the reverse transformation.

The general form of the command is written like this:

filtered_image = FFT(FFT(image, -1) * filter, 1) …………………………………….... (19)

where image is either a vector or a two-dimensional image and filter is a vector or two-

dimensional array designed to filter out certain frequencies in the image. For 2D FFT, this

output is fed as input.

VII.5.1.10. Needs of Pre-processing

The performance of a fingerprint recognition system basically depends upon the quality of

the input image. Since the images acquired with different kinds of sensors are not of the

perfect quality and so they can’t be used directly for the matching. Therefore to ensure the

accurate working of the system the image is first enhanced.

VII.5.1.11. Experiments and Results

A given function or signal can be converted between the time and frequency domains with a

pair of mathematical operators called a transform. An example is the Fourier transform,

which decomposes a function into the sum of a (potentially infinite) number of sine wave

frequency components. The 'spectrum' of frequency components is the frequency domain

representation of the signal. The inverse Fourier transform converts the frequency domain

function back to a time function.

A spectrum analyzer is the tool commonly used to visualize real-world signals in the

frequency domain. Let’s have a look how different pictures were used in frequency domain

and check out the outputs after high pass and low pass filtering in frequency domain.

As the current interest mainly lies in fingerprints, after checking the outputs of low pass

filtering, we could see the fingerprint picture becomes blurred, making it difficult to


____________________________________________________________________________________

- 139 -

distinguish between ridge and valley. Here are the results of some filtering in frequency

domain.

Figure 73: Lena image

Figure 74: Frequency spectrum of Lena image

Figure 75: Output of the Lena image after high pass filtering


____________________________________________________________________________________

- 140 -

Figure 76: Sample fingerprint

Figure 77: Frequency spectrum of the fingerprint image

Figure 78: Frequency spectrum of the fingerprint image after ideal high pass filtering

Figure 79: Output of the fingerprint image after high pass filtering


____________________________________________________________________________________

- 141 -

Figure 80: Frequency spectrum of the fingerprint image after high pass filtering with a radius of 5

Figure 81: Output of the fingerprint image after high pass filtering with a radius of 5

Figure 82: Output of the fingerprint image after high pass filtering with a radius of 50

Figure 83: output of the fingerprint image after ideal low pass filtering

VII.5.1.12. Discussion and Conclusion

By high pass filtering, better noise removal was achieved and the picture became

smoothened. We can use the frequency domain filtering for noise removal of fingerprint

image; it is a part of image pre-processing of fingerprint recognition. Also we can check if it

can be used for better edge detection of fingerprints.


____________________________________________________________________________________

- 142 -

VII.5.2. Fingerprint Matching using Correlation


Fingerprints have long been used for personal identification. It is assumed that every person

possesses unique fingerprints and hence the fingerprint matching is considered one of the

most reliable techniques of people identification. To perform fingerprint matching based on

the number of corresponding minutia pairings, has been in use for quite sometimes. But this

technique is not very efficient for recognizing the low quality fingerprints. To overcome this

problem, some researchers suggest the correlation technique which provides better result. Use

of correlation-based methods is increasing day by-day in the field of biometrics as it provides

better results.

Automatic fingerprint recognition systems (AFRS) have been nowadays widely used in

personal identification applications such as access control.

Roughly speaking, there are three types of fingerprint matching methods: minutia-based,

correlation-based, and image-based.

In minutia-based approaches, minutiae (i.e. endings and bifurcations of fingerprint ridges) are

extracted and matched to measure the similarity between fingerprints. These minutia-based

methods are now the most widely used ones.

Different from the minutia-based approaches, both correlation-based and image-based

methods compare fingerprints in a holistic way.

The image-based methods first generate a feature vector from each fingerprint image and

then compute their similarity based on the feature vectors.

The correlation-based methods spatially or in frequency domain correlate two fingerprint

images to compute the similarity between them.

No matter what kind of fingerprint matchers are used, the fingerprint images usually have to

be aligned when matching them.

VII.5.2.2. Frequency Domain

Frequency domain is a term used to describe the domain for analysis of mathematical

functions or signals with respect to frequency, rather than time.

A given function or signal can be converted between the time and frequency domains with a

pair of mathematical operators called a transform.


____________________________________________________________________________________

- 143 -

An example is the Fourier transform, which decomposes a function into the sum of a

(potentially infinite) number of sine wave frequency components. The 'spectrum' of

frequency components is the frequency domain representation of the signal.

A spectrum analyzer is the tool commonly used to visualize real-world signals in the

frequency domain.

VII.5.2.3. Correlation

A coefficient of correlation is a mathematical measure of how much one number can be

expected to be influenced by changes in another. It is closely related to covariance.

Digital Image Correlation is a full-field image analysis method, based on grey value digital

images that can determine the contour and the displacements of an object under load in three

dimensions.

Digital image correlation (DIC) techniques have been increasing in popularity, especially in

micro- and nano-scale mechanical testing applications due to its relative ease of

implementation and use. Advances in computer technology and digital cameras have been the

enabling technologies for this method and while white-light optics has been the predominant

approach, DIC can be and has been extended to almost any imaging technology. Digital

Image Correlation and Tracking (DIC/DDIT) is an optical method that employs tracking &

image registration techniques for accurate 2D and 3D measurements of changes in images.

Correlation is an important measure; it is used to find the position of a target object in a given

image. In frequency domain approach, we generally use FFT (Fast Fourier Transform) for

correlation method. Any image is completely described by its two-dimensional (2-D) Fast

Fourier Transform (FFT) in frequency domain.

The correlation is best method for template matching.

VII.5.2.4. Fingerprint

Human fingertips contain ridges and valleys which together forms distinctive patterns. These

patterns are fully developed under pregnancy and are permanent throughout whole lifetime.

Prints of those patterns are called fingerprints. Injuries like cuts, burns and bruises can

temporarily damage quality of fingerprints but when fully healed, patterns will be restored.

A fingerprint image exhibits a quasiperiodic pattern of ridges (darker regions) and valleys

(lighter regions). The local topological structures of this pattern together with their spatial

relationships determine the uniqueness of a fingerprint.

There are more than 100 different types of local ridge structures that have been identified.


____________________________________________________________________________________

- 144 -

Nevertheless, most of the automatic fingerprint identification / verification systems adopt the

model used by the Federal Bureau of Investigation. The model relies on representing only the

two most prominent structures: ridge ending and ridge bifurcation, which are collectively

called minutiae.

Figure 84: Different fingerprint patterns

VII.5.2.5. Fingerprint Matching

Large volumes of fingerprints are collected and stored everyday in a wide range of

applications including forensics, access control, and driver license registration. An automatic

recognition of people based on fingerprints requires that the input fingerprint be matched

with a large number of fingerprints in a database. To reduce the search time and

computational complexity, it is desirable to classify these fingerprints in an accurate and

consistent manner so that the input fingerprint is required to be matched only with a subset of

the fingerprints in the database.

Fingerprint classification is a technique to assign a fingerprint into one of the several pre-

specified types already established in the literature which can provide an indexing

mechanism. Fingerprint classification can be viewed as a coarse level matching of the

fingerprints. An input fingerprint is first matched at a coarse level to one of the pre-specified

types and then, at a finer level, it is compared to the subset of the database containing that

type of fingerprints only.

Here it was aimed to try to develop an algorithm to classify fingerprints into four classes,

namely, whorl, right loop, left loop, and arch.

Fingerprint matching techniques can be placed into categories like: minutiae-based and

correlation based. Minutiae-based techniques first find minutiae points and then map their

relative placement on the finger. However, there are some difficulties when using this

approach. It is difficult to extract the minutiae points accurately when the fingerprint is of low

quality. Also this method does not take into account the global pattern of ridges and furrows.

The correlation-based method is able to overcome some of the difficulties of the minutiae-

based approach. However, it has some of its own shortcomings. Correlation-based


____________________________________________________________________________________

- 145 -

techniques require the precise location of a registration point and are affected by image

translation and rotation.

VII.5.2.6. The Algorithm for Correlation Theorem in Spatial Domain

In spatial domain, SAD is a pre-processing technique for correlation method. It is one of the

approaches which gives satisfactorily result in template matching. Here the template must

have same size as in the provided image. This method is normally implemented by first

picking out a part of the search image to use as a template: We will call the search image S(x,

y), where (x, y) represent the coordinates of each pixel in the search image. Let’s call the

template T(xt , yt), where (xt , yt) represent the coordinates of each pixel in the template. Then

let’s simply move the centre (or the origin) of the template T(xt , yt) over each (x, y) point in

the search image and calculate the sum of products between the coefficients in S(x, y) and

T(xt , yt) over the whole area spanned by the template.

As all possible positions of the template with respect to the search image are considered, the

position with the highest score is the best position. This method is sometimes referred to as

'Linear Spatial Filtering' and the template is called a filter mask.

VII.5.2.7. The Algorithm for Correlation Theorem in Frequency Domain

Template matching using correlation basically uses the Correlation theorem:

f (x, y) o h(x, y) F (u, v) H*(u, v) ……………………………………………….…… (20)

Where f(x,y) is the search image having N x M sizes and h(x,y) is the template image having

K x L sizes.

It consists of the following steps:

1. Multiply the search and the template image by (-1) x+y centre transformed.

2. Compute F (u,v), the FFT of the Zero padded search image from (1).

3. Compute H (u,v), the FFT of the Zero padded template from (1).

4. Multiply F (u,v) by H* (u,v) (Conjugate of H (u,v)).

5. Compute the inverse FFT of the result in (4).

6. Obtain the real part of the result in (5).

7. Multiply the result in (6) by (-1) x+y.


____________________________________________________________________________________

- 146 -

VII.5.2.8. Fingerprint Template Matching

Steps:

Find effective area

Match with stored template

For each template find the point where the maximum matching is spotted

For each template find the value at that point (different point for each template)

Compare the values and get the point with the maximum value

Store the point and also the value

Find the point where the maximum matching is found amongst all the templates

The fingerprint pattern will be of the type matched

VII.5.2.9. Experiments

Tried with spatial domain, need exact match for good result, means millions of templates,

so it is not feasible

Frequency domain gives a better result.

If the database is small, experiments on spatial domain, as well as frequency domain

correlation gives direct fingerprint identification (/verification) result, though it’s not much feasible for spatial domain if the database is large, (position invariant features were

must to be incorporated for achieving this type recognition system).

A set of 100 different templates were used for matching (in frequency domain) and hence

classification.


Correlation-based techniques require the precise location of a registration point and are

affected by image translation and rotation.


____________________________________________________________________________________

- 147 -

VII.6. Soft Computing for Identification

of Fingerprint Image

(SCIFI)


Identity is to establish the identity of a person, or to ascertain the origin, nature, or definitive

characteristics of a particular person. To uniquely recognize humans based upon one or more

intrinsic physical or behavioural traits is called as Biometrics which is a recent trend. The

fingerprint (the graphical flow-like ridges present on human fingers) is the most widely used

biometric.

Nowadays, most automatic fingerprint identification systems (AFIS) are based on matching

minutiae, which are local ridge characteristics in the fingerprint pattern. It is the most popular

and extensively used method. Minutia are specific points in a finger image, however,

potential ways to improve the algorithms especially on pre-processing steps are still needed

to be improved.

Here, we will see how we can try to implement an intelligent fingerprint recognition system

using soft computing approaches. In this part different soft computing tools and how they can

be used in different phases of fingerprint recognition is to be discussed.

The intrinsic fuzziness of biometric features (these are not crisp measurements) suggests that

a soft-computing approach towards designing the processes of the pattern-match biometric

features could lead to nearly 100% authentication. For example, humans never fail to

recognize a well-known person, even in prohibitive conditions. They process biometric

features fuzzily and neurally.




All fingerprint images in database need to classify according to the pre-defined classification

criteria. Recent interest in automatic fingerprint classification system has inspired many

groups to conduct researches in this area. This is great importance in order to overcome the

accuracy and the identification speed problems. A number of approaches have been applied

for about several years that differ in the features used to describe the important of classifying

fingerprint image.

Most of the fingerprint recognition systems rely on minutiae matching algorithms, and there

are many algorithms developed for extraction of minutia. Most algorithms found in the

literature are somewhat difficult to implement and use a heuristic approach. However,


____________________________________________________________________________________

- 148 -

potential ways to improve the algorithms especially on pre-processing steps still needed to be

studied, as a 100% accurate fingerprint recognition system is yet to be implemented.

Minutia is the unique, measurable physical characteristics scanned as input and stored for

matching by biometric systems. For fingerprints, minutiae include the starting and ending

points of ridges, bifurcations and ridge junctions among other features.

There are many different algorithms being used to get the job of calculating minutia

accomplished. By introducing soft computing tools we can add intelligence to the recognition

system, and flexibility to the system (and also so that the system can tell the likelihood of the

particular image to be on a particular database, this aspect is still under consideration for

probable future work) and other intelligent features can also be introduced.

Here soft computing tools namely fuzzy logic and neural network and other tools like

wavelets, and how they can be used in fingerprint recognition have been discussed.

A B C D

Figure 85: AFIS Pattern Types: (A) Arch; (B) Left Loop; (C) Right Loop; (D) Whorl

The most used fingerprint recognition systems in present day rely on minutiae matching

algorithms. Although minutiae based techniques are widely used because of their temporal

performances they do not perform so well on low quality images and in the case of partial

fingerprint they might not be used at all. Therefore, when comparing partial input fingerprints

to pre-stored templates, a different approach is needed. Soft computing can help to achieve

better result for these types of case.

Different soft computing tools can be applied in different phases of pre-processing, i.e.

denoising, classification, searching, matching, customized searching and filtering.

For extracting features, for minutia, rotational invariant features are very important, and

whether soft computing tools can be useful in this phase is under the scope of this experiment

too.

For classification, the features can be fed in to neural network, we can, say for example, train

the network with 10% data and classify with 20% data.


____________________________________________________________________________________

- 149 -

VII.6.2. Features of Fingerprints

The ridges on fingers are created during embryo development in response to pressures that

form patterns that can be classified by print examiners. These ridges are also referred to as

friction ridges. They provide a relatively rough surface area, making it possible to grasp and

hold on to objects with ease. Each ridge contains at least one pore, which is connected to a

sweat gland below the skin. The sweat gland helps to remove waste from the ridge area as

well as to maintain a relatively constant temperature through evaporation. The sweat

produced is also the source of deposits for latent prints, i.e., those finger images that remain

on a surface after it has been touched. In addition to water, the sweat contains trace elements

of oil and some minerals. These latent impressions remain on the contact surface after it is

touched. The condition of the surface, e.g., if it is shiny or porous, affects how much of the

sweat remains on the surface. Environmental factors such as heat and humidity also influence

how long the latent print will remain. Latent prints are left by everyone on almost every kind

of surface. Virtually anytime an object is touched by the body, it retains some of the body’s sweat. In the case of finger and palm images, these can form a unique combination of ridges,

ridge endings, bifurcations, core and delta locations, and other image characteristics. For a

simple exhibit of a latent print, let’s take a clear, colourless glass and wipe the exterior surface to remove any foreign material, now let’s put some water into the glass, next, let’s hold the glass of water and take a drink and put the glass down and release it from hand. Now

looking at the glass, the images that can be seen on the glass are latent prints. Fingerprints

have been compared to topographical maps. The contour lines of the maps are similar to the

friction ridges of fingerprints, which consist of ridge endings, bifurcations, and dots. They

generate a flow that can be identified as a pattern. They do not appreciably change over time.

Unlike contour lines, however, friction ridges remain relatively uniform in their spatial

distances and are rarely featureless. In contrast, the contour lines on a topographical map

appear more closely together to indicate a sharp change in elevation, and relatively large

spaces between lines indicate that the surface has only a gradual slope. The fingerprint image

contains a great amount of information that contributes to the uniqueness of the fingerprint

image, particularly to someone trained to look for it. For example, the friction ridges flow

around a centre area. If this were a topographical map, it would be interpreted as a mountain

or hill. The point at which the ridges form would be the top of that hill. The change in

elevation is constant and the ridges are uniformly distant from each other. The white areas are

creases or scars. The introduction of scars does not negate the value of the fingerprint image.

In some instances, it might even aid in the identification, depending on the size and location

of the scar.

Image capturing is the first and a very important part of fingerprint recognition. A good

fingerprint capture includes three levels of ridge details. Level 1 detail includes the general

ridge flow and pattern configuration. Level 1 detail is not sufficient for individualization but

can be used for exclusion. It may include information enabling orientation, core and delta

location, and distinction of finger versus palm. Level 2 detail includes formations, defined as

a ridge ending, bifurcation, dot, or combinations thereof. The information of Level 2 detail

enables individualization. Level 3 detail includes all dimensional attributes of a ridge, such as


____________________________________________________________________________________

- 150 -

ridge path deviation, width, shape, pores, edge contour, incipient ridges, breaks, creases,

scars, and other permanent details. Level 1 features can be used to categorized fingerprints in

to major pattern types such as arc, loop and whorl. Level 2 and level 3 features can be used to

establish a fingerprint individuality or uniqueness. High level features can usually be

extracted only if the fingerprint image resolution is high. i.e. level 3 feature extraction

requires images with more than 500ppi resolution. The characteristics of an ideal image for

an AFIS search are: it should be a clear image, rolled from one nail edge to the other, using

even pressure that results in an image in which the ridge shapes, deviations, and pore

locations can be distinguished.

The advantage with AFIS is that features such as ridge endings, bifurcations, and ridge flows

can be extracted electronically by a coder in just a few seconds. These same features can be

extracted identically time after time. AFIS systems can be used to search multiple fingers. For

(the legendary) tenprint identification purposes, this may be accomplished by using two

fingers. In most instances, the information from the patterns of all ten fingers and two finger

images is sufficient for identification. In addition to the images, other biographical

information, such as sex, may be used to reduce the need to search the entire database. Using

two fingers does more than just double the changes of making identification. Since each of

the finger images is coded and is launched in a separate search, the results should come back

with the target as the first, and perhaps, only candidate. The synergy of two fingers from the

same individual supports the opportunities for identification. If all finger images on file had

even clear level 2 detail this would certainly happen. In tenprint processing, some AFIS

systems use images of the two index fingers and some use the two thumbs. There are at least

two arguments for using thumbs. The first is that the thumbs offer more surface area than the

index finger, producing a larger print image. The second argument is that if the search on

thumbs produces no identification, the record can be sent to the Integrated Automated

Fingerprint Identification System (IAFIS) or another AFIS system where it can be searched

against the database of index fingers. As a result, a search for the record would have been run

on both the thumbs and the index fingers to obtain identification. One of the selling points of

AFIS is accuracy. Its vendors claim a high degree of accuracy under certain conditions,

which include good-quality inked impressions on the database and good-quality impressions

on the inquiry.

With AFIS, pattern recognition software, as well as examiners, classifies images by one of

four pattern types (whorl, arch, right slant loop, left slant loop, (Figure 85). There are no

secondary classifications; there are no complicated rules. AFIS coders can determine both the

pattern type and minutiae placement. In addition, with AFIS systems, fingerprint cards may

or may not physically exist. If they do exist, they may be retained at an off-site facility. The

images from the electronic cards can be displayed on a screen or printed onto card stock.

AFIS examiners no longer have to wait for a physical tenprint card; the card information is as

near as a computer terminal connected to the AFIS. The cards cannot be misfiled or subject to

deterioration due to the heat and humidity found in an office building. The information can be

virtually retrieved, reviewed, and returned. There is no wasted paper and no file cabinets that

must be searched through taking up space. This is one of the major advantages of AFIS.


____________________________________________________________________________________

- 151 -

VII.6.3. Soft Computing Tools

Wavelets: A wavelet is a wave-like oscillation with an amplitude that starts out at zero,

increases, and then decreases back to zero. It can typically be visualized as a "brief

oscillation" like one might see recorded by a seismograph or heart monitor. Generally,

wavelets are purposefully crafted to have specific properties that make them useful for signal

processing. Wavelets can be combined, using a "shift, multiply and sum" technique called

convolution, with portions of an unknown signal to extract information from the unknown

signal.

As wavelets are a mathematical tool they can be used to extract information from many

different kinds of data, including - but certainly not limited to - audio signals and images.

Sets of wavelets are generally needed to analyze data fully.

A use of wavelet is that of smoothing/denoising data based on wavelet coefficient

thresholding, also called wavelet shrinkage. By adaptively thresholding the wavelet

coefficients that correspond to undesired frequency components smoothing and/or denoising

operations can be performed. And then the output of this process can be fed into the system

as input for further processing.

Fuzzy logic: Fuzzy logic is a form of many-valued logic derived from fuzzy set theory to

deal with reasoning that is fluid or approximate rather than fixed and exact. In contrast with

"crisp logic", where binary sets have two-valued logic, fuzzy logic variables may have a truth

value that ranges in degree between 0 and 1.

In simple words we can say fuzzy logic is a super set of conventional (boolean) logic that has

been extended to handle the concept of partial truth -- the truth values between completely

true and completely false. Furthermore, when linguistic variables are used, these degrees may

be managed by specific functions.

Because fingerprint patterns are fuzzy in nature and ridge endings are changed easily by

scares, using only ridge bifurcation as fingerprints minutiae and also design a fuzzy feature

image encoder by using cone membership function to represent the structure of ridge

bifurcation features extracted from fingerprint can be useful.

Integrating the fuzzy encoder with back-propagation neural network (BPNN) as a recognizer

which has variable fault tolerances for fingerprint recognition can help recognition in huge.

Experimental results show that the proposed fingerprint recognition system has the capability

to be robust, reliable and rapid.

Neural Network: The term neural network was traditionally used to refer to a network or

circuit of biological neurons. The modern usage of the term often refers to artificial neural

networks, which are composed of artificial neurons or nodes. Artificial neural networks are

composed of interconnecting artificial neurons (programming constructs that mimic the

properties of biological neurons).


____________________________________________________________________________________

- 152 -

Artificial neural networks may either be used to gain an understanding of biological neural

networks, or for solving artificial intelligence problems without necessarily creating a model

of a real biological system. The real, biological nervous system is highly complex and

includes some features that may seem superfluous based on an understanding of artificial

networks.

We can feed a fingerprint image to the neural network and get the hierarchical classification.

Backpropagation Neural Network: Backpropagation learning algorithm can be divided into

two phases: propagation and weight update.

Phase 1: Propagation. Each propagation involves the following steps:

Forward propagation of a training pattern's input through the neural network in order to

generate the propagation's output activations.

Back propagation of the propagation's output activations through the neural network

using the training pattern's target in order to generate the deltas of all output and hidden

neurons.

Phase 2: Weight update. For each weight-synapse:

Multiply its output delta and input activation to get the gradient of the weight.

Bring the weight in the opposite direction of the gradient by subtracting a ratio of it from

the weight.

This ratio influences the speed and quality of learning; it is called the learning rate. The sign

of the gradient of a weight indicates where the error is increasing, this is why the weight must

be updated in the opposite direction. Repeat the phase 1 and 2 until the performance of the

network is good enough.

Actual network is for a 3-layer network with only one hidden layer, and one input and one

output layer.

Neuro-Fuzzy Network: In the field of artificial intelligence, neuro-fuzzy refers to

combinations of artificial neural networks and fuzzy logic. Human brain recognition system

for biometrics works neuro-fuzzily.

Evolutionary Algorithm: In artificial intelligence, an evolutionary algorithm (EA) is a

subset of evolutionary computation, a generic population-based metaheuristic optimization

algorithm.

Evolutionary optimization methods can be used for searching the database.


____________________________________________________________________________________

- 153 -

VII.6.4. Background

AFIS can be made more accurate using Soft Computing tools. Soft computing can help

cross many boundaries in fingerprinting systems.

The most used fingerprint recognition system in current days depends on minutia extraction.

Different soft computing tools can be used in different phases of fingerprint extraction and

matching.

Conventional computing or often called as hard computing, requires a precisely stated

analytical model and often a lot of computation time. Many analytical models are valid for

ideal cases, and real world problems exist in a non-ideal environment. Soft computing differs

from conventional (hard) computing in that, unlike hard computing, it is tolerant of

imprecision, uncertainty, partial truth, and approximation. In effect, the role model for soft

computing is the human mind. The guiding principle of soft computing is: exploit the

tolerance for imprecision, uncertainty, partial truth, and approximation to achieve tractability,

robustness and low solution cost. Soft computing may be viewed as a foundation component

for the emerging field of conceptual intelligence. Few soft computing tools are: Fuzzy

Systems, Neural Networks, Evolutionary Computation, Machine Learning and Probabilistic

Reasoning.

Noise is a big issue in fingerprint extraction and matching. Wavelets can be used in removing

noise from the fingerprint image. In the minutia extraction phase only extracting minutia

doesn’t help to get enough information about the minutia, so aligning minutia becomes a

necessary step. Fuzzy Logic (or Fuzzy Techniques) can be used in minutia alignment. For the

hierarchical classification of fingerprints Neural Network can be used, we can set four

classes for hierarchical classification i.e. arch, left loop, right loop and whorl. In artificial

intelligence, an Evolutionary Algorithm (EA) is a subset of evolutionary computation, a

generic population-based meta-heuristic optimization algorithm. Evolutionary optimization

methods can be used for searching the database.

In one of project proposals, it was proposed to use back propagation neural network for

fingerprint identification with fuzzy techniques incorporated after hierarchical classification

and minutia detection. For extracting features, for minutia, rotational invariant features are

very important, and soft computing tools can be useful in this phase.

The term neural network was traditionally used to refer to a network or circuit of biological

neurons. The modern usage of the term often refers to artificial neural networks, which are

composed of artificial neurons or nodes. Artificial neural networks are composed of

interconnecting artificial neurons (programming constructs that mimic the properties of

biological neurons).

Artificial neural networks may either be used to gain an understanding of biological neural

networks, or for solving artificial intelligence problems without necessarily creating a model

of a real biological system. The real, biological nervous system is highly complex and


____________________________________________________________________________________

- 154 -

includes some features that may seem superfluous based on an understanding of artificial

networks.

Real life applications and the tasks to which artificial neural networks responds best include

classification, including pattern and sequence recognition; novelty detection and sequential

decision making. Figure 86 shows geometrical interpretation of pattern classification and

decision regions for different perceptron networks.

Figure 86: Neural Networks

Back-propagation is a common method of teaching artificial neural networks how to perform

a given task. It is a supervised learning method, and is a generalization of the delta rule. It

requires a teacher that knows, or can calculate, the desired output for any input in the training

set. It is most useful for feed-forward networks (networks that have no feedback, or simply,

that have no connections that loop). The term is an abbreviation for "backward propagation of

errors". Back-propagation requires that the activation function used by the artificial neurons

(or "nodes") be differentiable.

Fuzzy logic is a form of many-valued logic derived from fuzzy set theory to deal with

reasoning that is fluid or approximate rather than fixed and exact. In contrast with "crisp

logic", where binary sets have two-valued logic, fuzzy logic variables may have a truth value

that ranges in degree between 0 and 1. In simple words we can say fuzzy logic is a super set

of conventional (boolean) logic that has been extended to handle the concept of partial truth--

the truth values between completely true and completely false. Furthermore, when linguistic

variables are used, these degrees may be managed by specific functions.


____________________________________________________________________________________

- 155 -

(a) (b)

Figure 87: Simplified view of a 3 layer feed-forward artificial neural network

Because fingerprint patterns are fuzzy in nature, integrating the fuzzy encoder with back-

propagation neural network (BPNN) as a recognizer which has variable fault tolerances for

fingerprint recognition can help recognition in huge.

In the field of artificial intelligence, neuro-fuzzy refers to combinations of artificial neural

networks and fuzzy logic. Human brain recognition system for biometrics works neuro-

fuzzily.

Neuro-fuzzy hybridization results in a hybrid intelligent system that synergizes these two

techniques by combining the human-like reasoning style of fuzzy systems with the learning

and connectionist structure of neural networks. Neuro-fuzzy hybridization is widely termed

as Fuzzy Neural Network (FNN) or Neuro-Fuzzy System (NFS) in the literature.

Both neural networks and fuzzy systems have some things in common. They can be used for

solving a problem (e.g. pattern recognition) if there does not exist any mathematical model of

the given problem. They solely do have certain disadvantages (along with some advantages)

which almost completely disappear by combining both concepts.

Neural networks can only come into play if the problem is expressed by a sufficient amount

of observed examples. These observations are used to train the black box. On the one hand no

prior knowledge about the problem needs to be given. On the other hand, however, it is not

straightforward to extract comprehensible rules from the neural network's structure.

On the contrary, a fuzzy system demands linguistic rules instead of learning examples as

prior knowledge. Furthermore the input and output variables have to be described

linguistically. If the knowledge is incomplete, wrong or contradictory, then the fuzzy system


____________________________________________________________________________________

- 156 -

must be tuned. Since there is not any formal approach for it, the tuning is performed in a

heuristic way. This is usually very time consuming and error-prone.

Compared to a common neural network, connection weights and propagation and activation

functions of fuzzy neural networks differ a lot. Although there are many different approaches

to model a fuzzy neural network most of them agree on certain characteristics such as the

following:

A neuro-fuzzy system based on an underlying fuzzy system is trained by means of a data-

driven learning method derived from neural network theory. This heuristic only takes into

account local information to cause local changes in the fundamental fuzzy system.

It can be represented as a set of fuzzy rules at any time of the learning process, i.e.,

before, during and after. Thus the system might be initialized with or without prior

knowledge in terms of fuzzy rules.

The learning procedure is constrained to ensure the semantic properties of the underlying

fuzzy system.

A neuro-fuzzy system approximates a n-dimensional unknown function which is partly

represented by training examples. Fuzzy rules can thus be interpreted as vague prototypes

of the training data.

A neuro-fuzzy system is represented as special three-layer feed forward neural network as

it is shown in Figure 88.

o The first layer corresponds to the input variables.

o The second layer symbolizes the fuzzy rules.

o The third layer represents the output variables.

o The fuzzy sets are converted as (fuzzy) connection weights.

o Some approaches also use five layers where the fuzzy sets are encoded in the units of

the second and fourth layer, respectively. However, these models can be transformed

into a three-layer architecture.

Figure 88: The architecture of a neuro-fuzzy system


____________________________________________________________________________________

- 157 -

VII.6.5. Working Principle

The most critical step in the procedure of minutia extraction is to avoid computing false

minutiae caused by noise in the scanned fingerprint image. To overcome this problem,

backtrack control is executed on each feature pattern before it is validated. This checks that

each of the three branches of the bifurcation is significantly long. The following soft features

are extracted from fingerprints:

total area,

mean intensity.

Total area is measured as the ratio between the total pixels available on the fingerprint-

scanning device and the total pixels of the captured fingerprint image that have a value higher

than the estimated peak-noise level. Mean intensity is measured as the sum of the intensity of

all the pixels with a value higher than the estimated average intensity of the background

noise. Both these fingerprint features are related to the way the person approaches contact

with the fingerprint sensor.

A minutia detected in a fingerprint image can be characterized by a list of attributes that

include the minutia position, the direction and the type of minutia (ending or bifurcation).

Representation of fingerprint pattern thus comprises the attributes of all detected minutia in a

minutia set. And this becomes a point pattern. In the ideal case 2 identical point patterns are

exactly same.

Fingerprint verification is the task of counting the number of specially overlapping pairs

between two minutia sets. Genetic algorithm can be used in this phase. The correct minutiae

extraction is very important in an automatic fingerprint identification system. However, the

presence of noise in poor-quality images will cause many extraction faults, such as the

dropping of true minutiae and inclusion of false minutiae. Nowadays, most fingerprint

identification systems are based on precise mathematical models, but they cannot handle such

faults properly. As we know, human beings are good at recognizing fingerprint pattern,

therefore, a human-like method is needed to be applied. This idea suggests the arrangement

of an adaptive Fuzzy logic and Neural Network method which has variable fault tolerance.

Experimental results have shown that this type fingerprint identification method has the

potential to be robust, reliable and rapid.

Fingerprint classification algorithm can use Artificial Neural Networks (ANN). Fingerprints

are classified into four categories: arches, left loops, right loops and whorls (figure 85). The

algorithm extracts a string of symbols using the block directional image of a fingerprint,

which represents the set of structural features for this image. The statistical feature of the

pattern is computed, for this string and its Euclidean Distance Measures (EDM) are

computed. This system can use a multilayer artificial neural network composed of four sub-

networks one for each class. Images with poor quality are to be rejected. Classification is an

important step towards fingerprint recognition. In the classification stage, fingerprints are

usually associated to one of the four classes “A”, “L”, “R”, “W”. The aim is to reduce the number of comparisons that are necessary for recognition.


____________________________________________________________________________________

- 158 -

VII.6.6. Proposal

There are many methods in the literature for fingerprint recognition using minutia as a

feature. But there is not much work done using soft computing tool such as neural network.

A new method of fingerprint recognition system is being proposed here where the minutia

will be taken as feature to train the neural network system. And then finally add fuzzy logic

to make the system more flexible.

The method is as follows:

Cut effective area of a fingerprint image

Block-wise divide the image

Extract minutia of each block

Calculate the number of minutia for each block and gradient for each minutia

Use this number and the respective gradient values as a feature to train the back

propagation neural network

Use 40% data (fingerprint image of a same person of same finger) to train the system

Later on incorporate fuzzy logic with this system to decrease error rate, use fuzzy logic in

input as well as output layer, and so making the system at least a 5 layer network (input fuzzy

layer, input of actual neural network, hidden layer, output layer of neural network, output

layer of fuzzy logic)

As a three layer neural network may not be useful enough to map a fingerprint image

perfectly to a certain user, future plan is to implement a four (or even more) layer neural

network.

Figure 89: Simplified version of proposed algorithm


____________________________________________________________________________________

- 159 -

VII.7. Fingerprint Recognition on

Minutia Based Approach

(FR-MBA)



most broadly used biometric. However considering the automatic fingerprint recognition an

entirely solved problem is a widespread mistake. The most popular and extensively used

method is the minutiae-based method. This part summarizes a simple procedure for pre-

processing and extracting minutiae from digital fingerprint images.


configurations do not change throughout the life of an individual (except due to accidents

such as bruises and cuts on the fingertips). This property makes fingerprints a very attractive

biometric identifier. Fingerprint-based personal identification has been used for a very long

time [121]. Owning to their distinctiveness and stability, fingerprints are the most widely

used biometric features.


minutiae, which are local ridge characteristics in the fingerprint pattern. The two most

prominent minutiae types are ridge ending and ridge bifurcation. Based on the features that

the matching algorithms use, fingerprint matching can be classified into image-based and

graph-based matching.

Image-based matching [122] uses the entire gray scale fingerprint image as a template to

match against input fingerprint images. Graph-based matching [2], [63] represents the

minutiae in the form of graphs.

Fingerprint system can be separated into two categories Verification and Identification.

Verification system authenticates a person’s identity by comparing the captured biometric characteristic with its own biometric template(s) pre-stored in the system. It conducts one-to-

one comparison to determine whether the identity claimed by the individual is true. A

verification system either rejects or accepts the submitted claim of identity.

Identification system recognizes an individual by searching the entire template database for a

match. It conducts one-to-many comparisons to establish the identity of the individual. In an

identification system, the system establishes a subject’s identity (or fails if the subject is not enrolled in the system database) without the subject having to claim an identity.

In order to implement a successful algorithm of this nature, it is necessary to understand the

topology of a fingerprint. A fingerprint consists of many ridges and valleys that run next to

each other, ridges are shown in black and valleys are shown in white. The ridges bend in such


____________________________________________________________________________________

- 160 -

ways as to form both local and global structures; either of which can be used to identify the

fingerprint. The global level structures consist of many ridges that form arches, loops, whirls

and other more detailed classifications, as shown in Figure 94. Global features shape a special

pattern of ridge and valleys. On the other hand, the local level structures, called minutiae, are

further classified as either endpoints or bifurcations. Minutiae are also given an associated

position and direction, as shown in Figure 90 and in Figure 92.

This procedure is mainly based on minutiae, as well as on global level structure for finding a

reference point.

Figure 90: Types of fingerprint minutiae and their respective directions.

(a) an endpoint, (b) a bifurcation.

In addition, scanned fingerprints are subject to distortions that must also be taken into

account including rotation, translation, non-linear scaling and extraneous or missing minutiae

between matching fingerprints. This creates difficulty in the matching phase because it causes

the minutiae to differ between two identical fingerprints.

Figure 93 shows a fingerprint image with extracted minutiae.

VII.7.2. Background

Most approaches to recognizing a fingerprint involve five basic stages:

(i) acquisition, where the image is obtained from hardware or a file;

(ii) pre-processing, which may include thinning, noise reduction, image enhancements and

error correction;

(iii) structural extraction, where global and local structures may be found;

(iv) post-processing, where the structures are converted into a more useful format;

and then matching, where fingerprints are compared against a database.

These stages are shown in Figure 91.

The method chosen for acquisition of a fingerprint image depends on many different factors,

including the cost and reliability of an input device.


____________________________________________________________________________________

- 161 -

VII.7.3. Components of Fingerprint Recognition System Using Minutia

The minutiae provide the details of the ridge-valley structures, like ridge-endings and

bifurcations. Minutiae are, for instance, used for fingerprint matching, which is a one-to-one

comparison of two fingerprints. In this approach currently one to one pixel matching for

minutia extraction have been used.

Minutiae extraction: Minutiae are ridge endings or ridge bifurcations. Generally, if a perfect

segmentation can be obtained, then minutia extraction is just a trivial task of extracting

singular points in a thinned ridge map.

However, in practice, it is not always possible to obtain a perfect ridge map.

Some global heuristics need to be used to overcome this limitation.

Minutia matching: Minutia matching, because of deformations in sensed fingerprints, is an

elastic matching of point patterns without knowing their correspondences beforehand.

Generally, finding the best match between two point patterns is intractable even if minutiae

are exactly located and no deformations exist between these two point patterns.

The existence of deformations makes the minutia matching much more difficult.

VII.7.4. Algorithm Used for Recognition using Minutia Based Matching

Noise reduction

Image normalization

Selection of the interest region

Binarization

Thinning

Minutiae extraction

Cancellation of improper minutiae

Matching


____________________________________________________________________________________

- 162 -

Figure 91: Stages of the fingerprint recognition process.

(a) (b)

Figure 92: Fingerprint images showing Minutiae.

Figure 93: (a) Original fingerprint and (b) Detected minutiae.

Arch Tented Arch Left Loop Right Loop Whorl

Figure 94: Fingerprint Patterns.


____________________________________________________________________________________

- 163 -

VII.7.5. Pre-processing

This is an essential part of fingerprint recognition. In this step the image is made ready for the

actual matching. The input of this phase is the original fingerprint image and the final output

of this step is the minutiae set of that image. The proposed algorithm for pre-processing is as

followed.

VII.7.5.1. Fingerprint Image Enhancement

There are three types of degradations affect the quality of the fingerprint image. The ridges

get some gaps; parallel ridges connected due to noise; and natural effect to the finger like

cuts, wrinkles and injuries. The Fingerprint enhancement is anticipated to improve the

contrast between ridges and valleys and reduce noises in the fingerprint images. High quality

fingerprint image is very important for fingerprint verification or identification to work

properly. In real life, the quality of the fingerprint image is affected by noise like:

smudgy area created by over-inked area,

breaks in ridges created by under-inked area,

changing the positional characteristics of fingerprint features due to skin resilient nature,

dry skin leads to fragmented and low contrast ridges,

wounds may cause ridge discontinuities,

sweat on fingerprints also leads to smudge marks and connects parallel ridges.

VII.7.5.2. Noise Reduction

Noise is an unwanted perturbation to a wanted signal. Image noise is generally regarded as an

undesirable by-product of image capture. Noise reduction is the process of removing noise

from a picture (here it is fingerprint image). Different types of filtering methods like median

filter, global and adaptive thresholding were checked, tried and applied to reduce the noise.

VII.7.5.3. Image Normalization

The objective of this stage is to decrease the dynamic range of the gray scale between ridges

and valleys of the image in order to facilitate the processing of the following stages.

VII.7.5.4. Selection of the Interest Region

Since the image has background noise, the algorithm may generate minutiae outside the

fingerprint area. So selection of the interest area is one important step. This step is carried in

few phases:


____________________________________________________________________________________

- 164 -

(a) divide the image into blocks,

(b) find the average gradient of each block,

(c) find the position of the image where the average gradient of two successive blocks has the

zero crossing and the maximum absolute value,

(d) take the approximate middle point of these two particular blocks as the reference point,

(e) crop out a suitable region around this reference point.

The elementary orientations in the image are given by the gradient vector [Gx(x,y) Gy(x,y)]T,

which is defined as:

= sign(Gx) I(x,y)

= sign(∂I(x,y)/∂x) …………………………………………..… (21)

where I(x,y) represents the gray-scale image. This equation is used to calculate the

elementary orientations in the image by the gradient vector.

VII.7.5.5. Binarization

In the pre-processing stage, the image is converted from greyscale to black and white. This is

done by calculating the average background intensity and subtracting this value from the

greyscale image. Next a greyscale threshold (basic global and adaptive thresholding) is

calculated so pixels above this value become black, and the ones below become white.

Figure 95: (a) Original Fingerprint and (b) Binarized Fingerprint.

VII.7.5.6. Thinning

Next the ridges must be thinned to a width of one-pixel. In this step two consecutive fast

parallel thinning algorithms are applied, in order to reduce to a single pixel width, of the

width of the ridges of the binary image. These operations are necessary to simplify the

subsequent structural analysis of the image for the extraction of the fingerprint minutiae. The

thinning must be performed without modifying the original ridge structure of the image.


____________________________________________________________________________________

- 165 -

During this process, the algorithms can neither miscalculate beginnings, endings and/or

bifurcation of the ridges, nor break ridges structures. Figure 96 shows the thinned image of

the binarized image.

Figure 96: (a) Binarized Fingerprint and (b) Image after thinning.

VII.7.5.7. Minutiae Extraction

In the last stage, the minutiae from the thinned image are extracted, obtaining accordingly the

fingerprint biometric pattern. This process involves the determination of:

i) whether a pixel, belongs to a ridge or not and,

ii) if so, whether it is a bifurcation, a beginning or an ending point, obtaining thus a group of

candidate minutiae.

Next, all points at the border of the interest region are removed.

VII.7.5.8. Cancellation of Improper Minutiae

This is an important step of minutiae based fingerprint reorganization system. In this step, the

improper minutia which are mainly result of spurious noise of input image, are cancelled.

VII.7.6. Matching

Matching is a key operation in the current fingerprint identification system. One of the most




same finger (i.e., large intra-class variations). The main factors responsible for the intra-class

variations are:

displacement,

rotation,


____________________________________________________________________________________

- 166 -

partial overlap,

non-linear distortion,

variable pressure,

changing skin condition,

noise,

feature extraction errors.

Therefore, fingerprints from the same finger may sometimes look quite different whereas

fingerprints from different fingers may appear quite similar.

The method employed in the research was minutiae based matching. A minutia matching

essentially consists of finding the alignment between the template and the input minutiae sets,

that results in the maximum number of minutiae pairings. In Minutiae based matching the

similarity between the input and stored template are computed.

Figure 97: (a) Fingerprint matching steps and (b) Fingerprint authentication steps.

Figure 98: Matching; Verification and Identification


____________________________________________________________________________________

- 167 -

VII.7.7. Experimental Result

This Fingerprint Recognition System works for two types of matching. One fingerprint image

is fed into the system to check:

(a) whether it belongs to a particular entry of database and if so then for which entry/entries

of the database (Identification; one to n matching), or

(b) if it confirms to be fingerprint of a particular person (Verification; one to one matching).

This project further automatically adds new fingerprint to the database, if the fingerprint to be

matched does not exist in the database previously, in a separate section, for probable future

use.

This proposed method is based on pixel to pixel matching for minutia.

In this method the fingerprint image is being cropped with respect to a particular point

(reference point) of the image; this cropped area is called the region of interest. In this

system, the region of interest is taken as the 68 x 68 pixel block around the reference point.

Here, gradient calculation method was used to calculate the reference point. The reference

point is being calculated by calculating the average gradient of 8 x 8 pixel block of the

fingerprint image [41], [76]. For the maximum value of the average gradient of two

successive blocks that has the zero crossing, the middle point of the successive blocks is

taken as the reference point.

(a) (b)

Figure 99: Original picture and the calculated region to be cropped.

Currently this algorithm was tested on a small database (almost noise free dummy database of

100 entries). And the output obtained confirms the desired output as we can check visually.

(a) (b) (c)

Figure 100: (a) Original picture, (b) Cropped region, (c) Extracted minutiae


____________________________________________________________________________________

- 168 -

Figure 100 shows the original image, and the interest region and the minutiae extracted from

the cropped region.

Here a threshold value was kept for checking the number of minutiae matched. By changing

the threshold value we can get different rate of acceptance and rejection.

Figure 101(b), Figure 101(c) and Figure 101(d) are distorted images obtained from Figure

101(a); (b) is accepted but (c) & (d) are not accepted because of high degradation.

(a) (b) (c) (d)

Figure 101: (a) Fingerprint to be matched with, (b) Accepted little distorted image,

(c) Rejected more distorted image, (d) Rejected.

VII.7.8. Discussion


structures. Most algorithms found in the literature are somewhat difficult to implement.

This experiment was done using C language, in LINUX environment. Also MATLAB 7.1

was used in Windows Vista environment for implementing some parts.

Here in this proposed method, 8 x 8 pixel block was used for gradient calculation. And 68 x

68 pixel values around the reference point was taken, as the method was tried on dummy

database. For larger database having larger fingerprint image sizes, the pixel values can be

suitably changed.

For noisy database, it was found that the 4 x 4 pixel block for gradient calculation was giving

a better result.

It was also seen that this gradient approach was not suitable for all kinds of fingerprints, and

further attributes were also required in order to accomplish the matching. In future work, this

shortcoming is to be tried to solve.

VII.7.9. Future Scope and Conclusion

In this part, minutiae extraction based fingerprint detection was applied with gradient

detection as a step, to find the reference point.


____________________________________________________________________________________

- 169 -

The singular point detection method can be applied as a step to cluster the fingerprint images

into five major groups (i.e. arch, tented-arch, left loop, right loop, whorl), and then this

minutiae extraction based method can be applied on the clusters to achieve a hierarchical

fingerprint detection algorithm.

If the image is not noisy, it calculates the reference point to be the singular point.

Clustering the fingerprint images in five major groups is quite easy if it is done manually by

visual checking, but implementing an automated system for this is quite a hard job. This

approach is incorporated in next phases.

The minutiae based matching is highly sensible, as, if the finger is moved even a little bit that

gives us a different set of minutiae. In next part, position invariant features are applied.

Different soft computing approaches for calculating the reference point, as well as minutiae,

for better result was also being investigated, tried and tested.

This method is not yet generalised, finer approach will be taken in future.

Also this method was not tested on large database.


____________________________________________________________________________________

- 170 -

VII.8. Fingerprint Recognition using

Global (Singular Point) and Local (Minutia)

Structures

(FR-GLS)


The pattern of the ridges and valleys on the human fingertips forms the fingerprint images.

Analyzing this pattern at different levels reveals different types of features that are, global

feature and local feature.

Global features shape a special pattern of ridge and valleys, called singularities or Singular

Point (SP) and the important points are the core and the delta.

The core is defined as the topmost centred point on the inner most ridges and a delta is

defined as the centre point where three different directions flows meet. The SP provides

important information for fingerprint classification, fingerprint matching and fingerprint

alignment.

Local features or the so-called minutiae are an important feature for fingerprint matching.

This present procedure is mainly based on minutiae, as well as on global level structure for

finding a reference point by which alignment of two template is to be accomplished.

Figure 102: (a) Original fingerprint and (b) Detected minutiae.


____________________________________________________________________________________

- 171 -

Possible directions of an endpoint Possible sub-directions of a bifurcation

Figure 103: Types of fingerprint minutiae and their respective directions. (a) an endpoint, (b) a bifurcation.

Figure 104: Stages of the fingerprint recognition process.


____________________________________________________________________________________

- 172 -

In addition, scanned fingerprints are subjected to distortions that must also be taken into

account including rotation, translation, non-linear scaling and extraneous or missing minutiae

between matching fingerprints. This creates difficulty in the matching phase because it causes

the minutiae to differ between two identical fingerprints.

Figure 102 shows a fingerprint image with extracted minutiae. Most approaches to

recognizing a fingerprint involve five basic stages; these stages are shown in Figure 104.

Figure 105: Common workflow for fingerprint classification systems.

VII.8.2. Approach to Extract Singular Point

In figure 107(a), a fingerprint is depicted. The information carrying features in a fingerprint

are the line structures, called ridges and valleys. In this figure, the ridges are black and the

valleys are white. It is possible to identify two levels of detail in a fingerprint. The directional

field (DF), shown in Figure 107(b), describes the coarse structure, or basic shape, of a

fingerprint. It is defined as the local orientation of the ridge valley structures.

The DF is, in principle, perpendicular to the gradients. However, the gradients are

orientations at pixel scale, while the DF describes the orientation of the ridge-valley

structures, which is a much coarser scale. Therefore, the DF can be derived from the


____________________________________________________________________________________

- 173 -

gradients by performing some averaging operation on the gradients, involving pixels in some

neighborhood [81].

Various methods used to estimate the DF from a fingerprint are known from literature. They

include matched-filter approaches [4], [77], [78], methods based on the high-frequency power

in three dimensions [74], 2- dimensional spectral estimation methods [78], and micropatterns

that can be considered binary gradients [26]. These approaches do not provide as much

accuracy as gradient based methods, mainly because of the limited number of fixed possible

orientations. This is especially important when using the DF for tasks like tracing flow lines.

The gradient-based method was introduced by M. Kass and A. Witkin [76] and adopted by

many researchers [4], [5], [82], [84]. The elementary orientations in the image are given by

the gradient vector [Gx(x,y) Gy(x,y)]T, which is defined as: = sign (Gx) I(x,y)

= sign (∂I(x,y)/∂x) …………………………………..… (22)

where, I(x,y) represents the gray-scale image.

The first element of the gradient vector has been chosen to always be positive. The reason for

this choice is that in the DF, which is perpendicular to the gradient, opposite directions

indicate equivalent orientations.

(a) (b) (c)

Figure 106: Minutiae examples:

(a) Fingerprint image showing Minutiae, (b) Minutiae,

(c) Types of false minutia structures. From left to right and up to bottom we have: spike, bridge, hole,

break, spur, and ladder structure. The false minutiae generated by each structure are marked as (x) false

ridge ending, and (o) false ridge bifurcation.


____________________________________________________________________________________

- 174 -

Figure 107: Examples of a fingerprint, its directional field and its singular points:

(a) fingerprint, (b) directional field, and (c) singular points

Next phase is the extraction of the SPs, which are the points in a fingerprint where the DF is

discontinuous. In Fig. 108, two segments of the fingerprint of Fig. 107 are shown, one

containing a core and one containing a delta. The SPs are somewhere in the center of the

segments. However, they cannot be located more accurately than within the width of one

ridge-valley structure in the gray-value fingerprint.

Figure 108: Segments of a fingerprint that contain a singular point: (a) Core and (b) delta.

Figure 109: Directional fields: (a) Core and (b) delta.


____________________________________________________________________________________

- 175 -

VII.8.3. Approach to Extract Minutia

A minutia is the unique, measurable physical characteristics scanned as input and stored for

matching by biometric systems.

Minutiae include:

Ridge ending – the abrupt end of a ridge. Looks like this(-)

Ridge bifurcation – a single ridge that divides into two ridges

Short ridge, or independent ridge – a ridge that commences, travels a short distance

and then ends

Island – a single small ridge inside a short ridge or ridge ending that is not connected

to all other ridges

Ridge enclosure – a single ridge that bifurcates and reunites shortly afterward to

continue as a single ridge

Spur – a bifurcation with a short ridge branching off a longer ridge

Crossover or bridge – a short ridge that runs between two parallel ridges

Delta – a Y-shaped ridge meeting

Core – a U-turn in the ridge pattern



comparison of two fingerprints. In this current approach currently one to one pixel matching

for minutia extraction have been used.

The following two modules are the main components of fingerprint recognition system using

minutia:

Minutiae extraction. Minutiae are ridge endings or ridge bifurcations. Generally, if a

perfect segmentation can be obtained, then minutia extraction is just a trivial task of

extracting singular points in a thinned ridge map. However, in practice, it is not always

possible to obtain a perfect ridge map. Some global heuristics need to be used to

overcome this limitation.

Minutia matching. Minutia matching, because of deformations in sensed fingerprints, is

an elastic matching of point patterns without knowing their correspondences beforehand.

Generally, finding the best match between two point patterns is intractable even if

minutiae are exactly located and no deformations exist between these two point patterns.

The existence of deformations makes the minutia matching much more difficult.

Image Acquisition is the 1st phase of the system.

Pre-processing is an essential part of fingerprint recognition. In this step the image is made

ready for the actual matching. The input of this phase is the original fingerprint image and the

final output of this step is the minutiae of that image.

Proposed algorithm for pre-processing is as followed:

Fingerprint image enhancement

Noise reduction


____________________________________________________________________________________

- 176 -

Image normalization


Binarization

Thinning

Minutiae extraction

Cancellation of improper minutiae

VII.8.4. Fingerprint Image Enhancement

There are many types of degradations that affect the quality of the fingerprint image. The

ridges get some gaps; parallel ridges get connected due to noise and natural effect to the

finger like cuts, wrinkles and injuries. The Fingerprint enhancement is anticipated to improve

the contrast between ridges and valleys and reduce noises in the fingerprint images.

High quality fingerprint image is very important for fingerprint verification or identification

to work properly. In real life, the quality of the fingerprint image is affected by noise like

smudgy area created by over-inked area, breaks in ridges created by under-inked area,

changing the positional characteristics of fingerprint features due to skin resilient in nature,

dry skin leads to fragmented and low contrast ridges, wounds may cause ridge discontinuities

and sweat on fingerprints also leads to smudge marks and connects parallel ridges.

VII.8.5. Noise Reduction & Image Normalization

Image noise is generally regarded as an undesirable by-product of image capture. Noise

reduction is the process of removing noise from a picture (here it is fingerprint image).

The objective of normalization is to decrease the dynamic range of the gray scale between

ridges and valleys of the image in order to facilitate the processing of the following stages.

VII.8.6. Selection of the Interest Region


fingerprint area. So selection of the interest area is one important step. This step is carried in

few phases: (a) divide the image into blocks, (b) find the average gradient of each block, (c)

find the position of the image where the average gradient of two successive blocks has the

zero crossing and the maximum absolute value, (d) take the approximate middle point of

these two particular blocks as the reference point, (e) crop out a suitable region around this

reference point. Equation No. 22 is used to calculate the elementary orientations in the image

by the gradient vector.


____________________________________________________________________________________

- 177 -

Most authors process fingerprints block-wise for calculating DF [3], [35]. This means that the

directional field is not calculated for all pixels individually. Instead, the average DF is

calculated in blocks of, for instance, in this present approach 8 by 8 pixel block was used.

Gradient calculation method was used here to calculate the reference point. The reference

point was calculated by calculating the average gradient of 8 x 8 pixel block of the fingerprint

image [26], [76].

Figure 110: Gray-scale coded directional field and coherence, (a) Directional field and (b) Coherence.

VII.8.7. Binarization



greyscale image. Next, greyscale threshold (basic global and adaptive thresholding) is

calculated so pixels above this value become black, and the ones below become white [42].

Figure 111: (a) Original Fingerprint, (b) Binarized Fingerprint.


____________________________________________________________________________________

- 178 -

VII.8.8. Thinning


parallel thinning algorithms are applied, in order to reduce to a single pixel width, from the

width of the ridges in the binary image.

These operations are necessary to simplify the subsequent structural analysis of the image for

the extraction of the fingerprint minutiae.

The thinning must be performed without modifying the original ridge structure of the image.

During this process, the algorithms cannot miscalculate beginnings, endings and or

bifurcation of the ridges, neither ridges can be broken.

Figure 112 shows the thinned image of the binarized image.


VII.8.9. Minutiae Extraction


fingerprint biometric pattern.

This process involves the determination of:

whether a pixel, belongs to a ridge or not and,

if so, whether it is a bifurcation, a beginning or an ending point, obtaining thus a group of

candidate minutiae.



____________________________________________________________________________________

- 179 -

VII.8.10. Cancellation of Improper Minutiae

This is an important step of minutiae based fingerprint reorganization system. In this step, the

improper minutiae which are mainly result of spurious noise of input image, are cancelled.

Figure 113: (a) Image after thinning and imperfection removal, and (b) Minutiae pattern.

VII.8.11. Matching

The method employed in the research was both SP and minutiae based matching. A minutia

matching essentially consists of finding the alignment between the template and the input

minutiae sets, that results in the maximum number of minutiae pairings. In Minutiae based

matching the similarity between the input and the stored template are computed.

Figure 114: Fingerprint authentication steps.


____________________________________________________________________________________

- 180 -



is fed into the system to check (a) whether it belongs to a particular entry of database and if

so then for which entry/entries of the database (Identification), or (b) if it confirms to be

fingerprint of a particular person (Verification). This project further automatically adds new

fingerprint to the database, if the fingerprint to be matched does not exist in the database

previously. The proposed method is based on pixel to pixel matching for minutia. In this

method the fingerprint image is being cropped with respect to a particular point (reference

point) of the image; this cropped area is called the region of interest. In this system, the

region of interest is taken as the 68 x 68 pixel block around the reference point. At first the

noise was removed by median filtering followed by basic adaptive global thresholding.

Fingerprints were processed block-wise for calculating DF. This means that the directional

field is not calculated for all pixels individually. Instead, the average DF is calculated in

blocks of, for instance, in this approach 8 by 8 pixel block. Gradient calculation method was

used to calculate the reference point. The reference point is being calculated by calculating

the average gradient of 8 x 8 pixel block of the fingerprint image. For the maximum value of

the average gradient of two successive blocks that has the zero crossing, the middle point of

the successive blocks is taken as the reference point. Currently this algorithm was tested on a

small database (almost noise free dummy database of 100 entries). And the output confirms

the desired output as checked visually.

(a) (b)

Figure 115: Original picture and the calculated region to be cropped.

(a) (b) (c)

Figure 116: (a) Original picture, (b) Cropped region, and (c) Extracted minutiae.


____________________________________________________________________________________

- 181 -


the threshold value we can obtain different rate of acceptance and rejection. Figure 117(b)

and Figure 117(c) are two distorted images obtained from Figure 117(a); (b) is accepted but

(c) is not because of its high degradation.

(a) (b) (c)

Figure 117: (a) Fingerprint to be matched with,

(b) Accepted little distorted image, (c) Rejected more distorted image.

(a) (b) (c)

(d) (e) (f)

Figure 118: Result of pre-processing steps.

(a) original image, (b) image after noise reduction and normalization, (c) region of interest,

(d) cropped image, (e) thinned image, (f) extracted minutia.


____________________________________________________________________________________

- 182 -



structures.

In this method, minutiae extraction based fingerprint detection was applied with gradient



68 pixel values around the reference point was taken, as the method was applied on dummy

database, for larger database having larger fingerprint image sizes, the pixel values can be

suitably changed.

For noisy database, it was found that the 4 x 4 pixel block for gradient calculation was giving

a better result. It was also seen that this gradient approach was not suitable for all kinds of

fingerprints, and further attributes were also required in order to accomplish the matching. In

future work, this shortcoming is to be solved.

In this method, minutiae extraction based fingerprint detection was applied with gradient

detection as a step, to find the reference point. The singular point detection method can be

applied as a step to cluster the fingerprint images into five major groups (i.e. arch, tented-

arch, left loop, right loop, whorl), and then minutiae extraction based method can be applied

on the clusters to achieve a hierarchical fingerprint detection algorithm. This approach is to

be incorporated next.

This experiment was done using C language, in LINUX (Dolphin) environment. Also to

make use of the inbuilt functions, MATLAB was used. MATLAB 7.1 was used in Windows

7 Home Basic, but these OS and Software were incompatible so was creating problems, so

finally, MATLAB 6 was used. The system used for programming was based on Intel®

Core™ Duo processor (T650 @ 2.10 Ghz), with a 2.00GB RAM and 32-bit Operating

System.


visual checking, but implementing an automated system for this is quite a hard job. Also

investigating different soft computing approaches for calculating the reference point, as well

as minutiae, for better result, is still going on. The plan was to implement an intelligent

system for fingerprint recognition using soft computing approach (discussed previously).

This method is not yet generalised, finer approach is to be taken. Also this method was not

tested on large database. If the image is not much noisy, it calculates the reference point to be

the singular point. Position invariant features are to be applied next.



important tasks for a recognition system. The minutiae based matching is highly sensible, as,

if the finger is moved even a little bit that gives a different set of minutiae.


____________________________________________________________________________________

- 183 -

Figure 119: Sources of error in fingerprint recognition.

Figure 120: Identical matches of minutia coordinates rarely match perfectly.

Figure 121: Alignment of the input ridge and the template ridge.


____________________________________________________________________________________

- 184 -

From the study it can be said that, most of the fingerprint recognition systems rely on

minutiae matching algorithms. Although minutiae based techniques are widely used because

of their temporal performances, they do not perform so well on low quality images and in the

case of partial fingerprint they might not be useful at all. Therefore, when comparing partial

input fingerprints to pre-stored templates, a different approach is needed. Soft computing can

help to achieve better result for these types of case. For extracting features, for minutia,

rotational invariant features are also very important.

Recent interest in automatic fingerprint classification system has inspired many groups to

conduct researches in this area. All fingerprint images in database needed to be classified

according to the pre-defined classification criteria. This is of great importance in order to

overcome the accuracy and the identification speed problems. A number of approaches have

been applied for about several years that differ in the features used to describe the importance

of classifying fingerprint images. But, potential ways to improve the algorithms especially on

pre-processing steps are still needed to be studied and experimented.

From studies, it was found that there are various techniques and features that have been used

to classify a fingerprint image. There are still open research opportunities in this field that are

related to the performance of a system that rejects a high percentage of its input.

The future study of fingerprint classification systems might be the use combinations of

features. The best features for classification are claimed to be fingerprint singularities.

However, this method can be difficult when extracting core and delta points due to noisy

images. Ideas for future study will give better insights on how we can improve the quality of

fingerprint images in pre- processing step.

Works were also there to attempt the problem with poor quality and the presence of noise in

fingerprint images. The system must be design to be robust when dealing with the quality of

fingerprint images and it will give better performance in fingerprint classification system.


____________________________________________________________________________________

- 185 -

VII.9. Fingerprint Recognition using a

Reference Point and Pores

(FR-RPP)



most widely used biometric. Fingerprint-based personal identification has been used for a

very long time [121]. However considering the automatic fingerprint recognition a

completely solved problem is a common mistake. There are many different algorithms being

used to get this accomplished. The global level structures consist of many ridges to form

some specific shape like arch, loop, and whorl. Local level structures are called minutia,

which further are classified as either endpoints or bifurcations. Either of which (global or

local features) can be used to identify the fingerprint. The most popular and extensively used

method is the minutia-based method.

In this method pores are being used as the fingerprint feature for recognition.

The fingerprint is a duplicate of a fingertip epidermis, when a person touches a smooth

surface, the fingertip epidermis characteristic is transferred to the surface. The pattern of the

ridges and valleys on the human fingertips forms the fingerprint images. Analyzing this

pattern at different levels reveals different types of features that are, global feature and local

feature. Global features shape a special pattern of ridge and valleys, called singularities or

Singular Point (SP) and the important points are the core and the delta; the core defined as the

top most point on the inner most ridge and the delta defined as the centre point where three

different directions flows meet. The SP provides important information for fingerprint

classification, fingerprint matching and fingerprint alignment. Local features, the so-called

minutia are an important (and the most extensively used) feature for fingerprint matching.

Fingerprint patterns are full of ridges and valleys. The information of the ridge structures can

be treated as three levels. At the coarse level, the number and the relative positions of

singular points, including cores and deltas, are concerned for classification. At the fine level,

the minutiae, a group of ridge endings and bifurcations, are used as the features for matching.

Between the above two levels, the middle level also contains important information,

including local ridge orientation (LRO) and local ridge frequency (LRF). Conventionally,

only the structures of LROs are used to find the singular points for classification or to

enhance ridge structures for minutiae extraction.

The first step in an identification system is often continuous classification of fingerprints.

This reduces the partition of the database to be searched for matches. To facilitate high-

performance classification, algorithms for accurate singular-point estimation are needed.

Singular point detection is a critical process for both fingerprint matching and fingerprint


____________________________________________________________________________________

- 186 -

classification. The process of singular points detection must be fast and robust; otherwise, the

performance of the whole fingerprint recognition system would be influenced heavily.

In high level fingerprint classification algorithms, extracting the number and precise location

of singular points (SP), namely core and delta points are of great importance. According to

the number and location of these robust characteristics, fingerprints can be classified in to

five main groups; arch, tented arch, right loop, left loop, and whorl.

Using high-level classification process can efficiently reduces the search area in large

fingerprint databases and therefore speeds up the subsequent matching algorithm. There are

four main approaches to allocate SPs [129]. 1) Methods based on mathematical model

representation of fingerprint, 2) Methods based on statistical approaches, 3) Methods based

on different frequency transforms and 4) Methods based on fingerprint structures. Some

approaches combine several types of the above mentioned methods and make a new

combined system. Singular points detection is the most challenging and important process in

biometrics fingerprint verification and identification systems. Singular points are used for

fingerprint classification, fingerprint matching and fingerprint alignment.


minutia, which are local ridge characteristics in the fingerprint pattern. The two most

prominent minutiae types are ridge ending and ridge bifurcation. Based on the features that

the matching algorithms use, fingerprint matching can be classified into image-based and

graph-based matching. Image-based matching [122] uses the entire gray scale fingerprint

image as a template to match against input fingerprint images. Graph-based matching [2],

[63] represents the minutiae in the form of graphs. The high computational complexity of

graph matching hinders its implementation. To reduce the computational complexity,

matching the minutiae sets of template and input fingerprint images can be done with point

pattern matching [3], [4], [67], [123], [127].

In order to implement a successful algorithm, it is necessary to understand the topology of a

fingerprint. A fingerprint consists of many ridges and valleys that run next to each other,

ridges are shown in black and valleys are shown in white. The ridges bend in such ways as to

form both local and global structures; either of which can be used to identify the fingerprint.

The global level structures consist of many ridges that form arches, loops, whirls and other

more detailed classifications, as shown in Figure 122. Global features shape a special pattern

of ridge and valleys. On the other hand, the local level structures, called minutiae, are further

classified as either endpoints or bifurcations. Other than usual minutiae there are sweat pores

in the fingerprint which can also be used for fingerprint matching.

One of the most popular biometric traits, fingerprints are widely used in personal

authentication, particularly with the availability of a variety of fingerprint acquisition devices

and the advent of thousands of advanced fingerprint recognition algorithms. Such algorithms

make use of distinctive fingerprint features that can usually be classified at three levels of

detail [73], as shown in Fig. 125 and referred to as level 1, level 2, and level 3. Level-1

features are the macro details of fingerprints, such as singular points and global ridge


____________________________________________________________________________________

- 187 -

patterns, e.g., deltas and cores (indicated by triangles in Fig. 125). They are not very

distinctive and are thus mainly used for fingerprint classification rather than recognition. The

level-2 features (rectangles) primarily refer to the Galton features or minutiae, namely, ridge

endings and bifurcations. Level-2 features are the most distinctive and stable features, which

are used in almost all automated fingerprint recognition systems (AFRSs) [69], [73], [79] and

can reliably be extracted from low-resolution fingerprint images (~500 dpi). A resolution of

500 dpi is also the standard fingerprint resolution of the Federal Bureau of Investigation

(FBI) for AFRSs using minutiae [80]. Level-3 features (circles) are often defined as the

dimensional attributes of the ridges and include sweat pores, ridge contours, and ridge edge

features, all of which provide quantitative data supporting more accurate and robust

fingerprint recognition.

Among these features, pores have most extensively been studied [80], [88–100] and are

considered to be reliably available only at a resolution higher than 500 dpi.

This current experimental procedure is based on pores, as well as on global level structure for

finding a reference point by which alignment and matching of two templates is to be

accomplished. This is a new approach.

Arch Tented Arch

Left Loop Right Loop Whorl

Figure 122: Fingerprint Patterns.


____________________________________________________________________________________

- 188 -

Figure 123: Fingerprint image showing different features.

Figure 124: Stages of the fingerprint recognition process using pores.

Figure 125: Three levels of fingerprint features.


____________________________________________________________________________________

- 189 -

VII.9.2. AFRS

Automatic fingerprint recognition systems (AFRS) have been nowadays widely used in

personal identification applications such as access control. Roughly speaking, there are three

types of fingerprint matching methods: minutia-based, correlation-based, and image-based. In

minutia-based approaches, minutiae (i.e. endings and bifurcations of fingerprint ridges) are

extracted and matched to measure the similarity between fingerprints. These minutia-based

methods are now the most widely used ones.

Different from the minutia-based approaches, both correlation-based and image-based

methods compare fingerprints in a holistic way. The correlation-based methods spatially

correlate two fingerprint images to compute the similarity between them, while the image-

based methods first generate a feature vector from each fingerprint image and then compute

their similarity based on the feature vectors.

No matter what kind of fingerprint matchers are used, the fingerprint images usually have to

be aligned when matching them. So another important aspect of fingerprint matching is the

fingerprint alignment methods, though in this method, the alignment methods were not

considered.

In order to further improve the accuracy of AFRS, people are now exploring more features in

addition to minutiae on fingerprints. The recently developed high resolution fingerprint

scanners make it possible to reliably extract level-3 features such as pores. Pores have been

used as useful supplementary features for a long time in forensic applications. Researchers

have also studied the benefit of including pores in AFRS and validated the feasibility of pore

based AFRS.

Using pores in AFRS has two advantages. First, pores are more difficult to be damaged or

mimicked than usually taken minutiae i.e. ridge endings and bifurcations. Second, pores are

abundant on fingerprints. Even a small fingerprint fragment could have a number of pores

(Fig. 123). Therefore, pores are particularly useful in high resolution partial fingerprint

recognition where the number of minutiae is very limited. As it can be seen in a good quality

fingerprint image, the friction ridges are dotted with small circular openings – the sweat

pores. The perspiration is exuded from these.

VII.9.3. Approach to Extract Singular Point

In Figure 123, a fingerprint is depicted. The information carrying features in a fingerprint are

the line structures, called ridges and valleys. In this figure, the ridges are black and the

valleys are white. It is possible to identify two levels of detail in a fingerprint. The directional

field (DF) describes the coarse structure, or basic shape, of a fingerprint. It is defined as the

local orientation of the ridge valley structures. The DF is, in principle, perpendicular to the

gradients. However, the gradients are orientations at pixel scale, while the DF describes the

orientation of the ridge-valley structures, which is a much coarser scale. Therefore, the DF


____________________________________________________________________________________

- 190 -

can be derived from the gradients by performing some averaging operation on the gradients,

involving pixels in some neighbourhood [81].

Various methods used to estimate the DF from a fingerprint are known including matched-

filter approaches [4], [77], [78], methods based on the high-frequency power in three

dimensions [74], 2- dimensional spectral estimation methods [78], and micro-patterns that

can be considered binary gradients [42].

These approaches do not provide as much accuracy as gradient based methods, mainly

because of the limited number of fixed possible orientations. This is especially important

when using the DF for tasks like tracing flow lines. The gradient-based method was

introduced in 1987, by M. Kass and A. Witkin [76] and was adopted by many researchers [4],

[5], [82], [84]. The elementary orientations in the image are given by the gradient vector

[Gx(x,y) Gy(x,y)]T, which is defined as: = sign (Gx) I(x,y) = sign (∂I(x,y)/∂x) …………………. (23)


The first element of the gradient vector has been chosen to always be positive. The reason for

this choice is that in the DF, which is perpendicular to the gradient, opposite directions

indicate equivalent orientations. Next phase is the extraction of the SPs, which are the points

in a fingerprint where the DF is discontinuous.


(a) fingerprint, (b) directional field, and (c) singular points.

VII.9.4. Approach to Extract Pores

Pores, also known as sweat pores, are located on finger ridges. They are formed in the sixth

month of gestation due to the sweat-gland ducts reaching the surface of the epidermis. Once

the pores are formed, they are fixed on the ridges and there can be between 9 to 18 pores

along a centimetre of ridge [69]. A pore can be visualized as open on one print, but as closed

on the other print depending on pressure and whether it is exuding perspiration. Pore

detection, without a loss of generality, might seem to be a trivial task.


____________________________________________________________________________________

- 191 -

In this method, surrounding of each ridge pixel is checked to check if it is surrounded by

other ridge pixel or no, if it is not surrounded by other ridge pixel then it denotes a presence

of pore. For each pore the central pixel is stored. It is also needed to go through the Alignment

stage, where transformations such as translation, rotation and scaling between an input and a

template in the database are estimated and the input pores are aligned with the template pores

according to the estimated parameters [3].

VII.9.5. Image Acquisition





VII.9.6. Pre-processing



of this step is the pore dataset of that image. This proposed algorithm for pre-processing is as

followed:

Noise reduction

Image normalization


Pore extraction.

(a) (b) (c)

Figure 127: Pre-processing steps.

(a) original image, (b) image after noise reduction and normalization, (c) region of interest.


____________________________________________________________________________________

- 192 -

VII.9.7. Matching



input pattern with respect to the database pattern.

Reliably matching fingerprint images is an extremely difficult problem, mainly due to the

large variability in different impressions of the same finger (i.e., large intra-class variations).

The main factors responsible for the intra-class variations are: displacement, rotation, partial

overlap, non-linear distortion, variable pressure, changing skin condition, noise, and feature

extraction errors. So fingerprints from the same finger may sometimes look quite different

whereas fingerprints from different fingers may appear quite similar.

The method employed in the research was both SP and pore based matching. A pore

matching essentially consists of finding the alignment between the template and the input

pore sets that results in the maximum number of pore pairings. In pore based matching the

similarity between the input and the stored template is computed.

VII.9.8. Future Scope

In this experiment, pore extraction based fingerprint detection was applied with gradient

detection as a step, to find the reference point. The singular point detection method can be

applied as a step to cluster the fingerprint images into five major groups, and then pore

extraction based method can be applied on the clusters to achieve a hierarchical fingerprint

detection algorithm. This approach is to be tried to be incorporated in future.

Also different soft computing approaches are being investigated for calculating the reference

point, as well as pores. By introducing soft computing tools we can add intelligence to the

recognition system.

A more effective fingerprint recognition system might be implemented using singular points,

pores and minutiae.



structures. Most algorithms found in the literature are somewhat difficult to implement and

use a heuristic approach. The reliability of any automatic fingerprint recognition system

strongly relies on the precision obtained in the extraction process. Extraction of appropriate

features is one of the most important tasks for a recognition system.


____________________________________________________________________________________

- 193 -

VII.10. Fingerprint Recognition

By Classification Using Neural Network

And Matching Using Minutia

(FR-NNMM)


Biometrics is one of the biggest tendencies in human identification. Though fingerprint is the

most widely used biometric, considering the automatic fingerprint recognition a completely

solved problem is a common mistake.

The global level structures consist of many ridges to form some specific shape such as arch,

loop, and whorl. Local level structures are called minutiae, which further classified as either

endpoints or bifurcations. Either of which can be used to identify the fingerprint, and this

approach uses both methods. Soft computing was used here for classification of fingerprints.


configurations do not change throughout the life of an individual except due to accidents such

as bruises and cuts on the fingertips. This property makes fingerprints a very striking

biometric identifier. Fingerprint-based personal recognition has been used for a very long

time [121]. Owning to their uniqueness, constancy, robustness, and convenience, fingerprints

is the most commonly used biometric feature.

The fingerprint is a duplicate of a fingertip epidermis; when a person touches a smooth

surface, the fingertip epidermis characteristic gets transferred to the surface. The pattern of

the ridges and valleys on the human fingertips forms the fingerprint images. Analyzing this

pattern at different levels reveals different types of features that are, global feature and local

feature. Global features shape a special pattern of ridge and valleys, called singularities or

Singular Point (SP) and the important points are the core and the delta. The core defined as

the topmost point on the inner most ridges and a delta defined as the centre point where three

different directions flows meet. The SP provides important information for fingerprint

classification, fingerprint matching and fingerprint alignment. Local features, the so-called

minutia are an important feature for fingerprint matching.

Fingerprint patterns are full of ridges and valleys. The information of the ridge structures can

be treated as three levels. At the coarse level, the number and the relative positions of

singular points, including cores and deltas, are concerned for classification. At the fine level,

the minutiae, a group of ridge endings and bifurcations, are used as the features for matching.

Between the above two levels, the middle level also contains important information,

including local ridge orientation (LRO) and local ridge frequency (LRF). Conventionally,


____________________________________________________________________________________

- 194 -

only the structures of LROs are used to find the singular points for classification or to

enhance ridge structures for minutiae extraction.

There are more than 100 different types of local ridge structures that have been identified.

Although other approaches are possible, like, for instance, the hashing technique in the

minutiae domain, the first step in an identification system is often continuous classification of

fingerprints. This reduces the partition of the database to be searched for matches. To

facilitate high-performance classification, algorithms for accurate singular-point estimation

are needed. Singular point detection is a critical process for both fingerprint matching and

fingerprint classification. The process of singular points detection must be fast and robust;

otherwise, the performance of the whole fingerprint recognition system would be influenced

heavily.

In high level fingerprint classification algorithms, extracting the number and precise location

of singular points (SP), namely core and delta points are of great importance. According to

the number and location of these robust characteristics, fingerprints can be classified in to

main groups; arch, right loop, left loop, and whorl.

Using high-level classification process can efficiently reduces the search area in large

fingerprint databases and therefore speeds up the subsequent matching algorithm.

Singular points detection is the most challenging and important process in biometrics

fingerprint verification and identification systems. Singular points are used for fingerprint

classification, fingerprint matching and fingerprint alignment.

Soft computing approach was used to classify fingerprints. The feature of Self Organizing

Map to provide topologically preserved mapping from input to output spaces, makes it

suitable to be used for classification.


minutiae, which are local ridge characteristics in the fingerprint pattern. The two most

prominent minutia types are ridge ending and ridge bifurcation. Based on the features that the

matching algorithms use, fingerprint matching can be classified into image-based and graph-

based matching. Image-based matching [122] uses the entire gray scale fingerprint image as a

template to match against input fingerprint images. The primary shortcoming of this method

is that matching may be seriously affected by some factors such as contrast variation, image

quality variation, and distortion, which are inherent properties of fingerprint images. The

reason for such limitation lies in the fact that gray scale values of a fingerprint image are not

stable features. Graph-based matching [2], [63] represents the minutiae in the form of graphs.

The high computational complexity of graph matching hinders its implementation. To reduce

the computational complexity, matching the minutiae sets of template and input fingerprint

images can be done with point pattern matching. Several point pattern matching algorithms

have been proposed and commented in the literature [3], [4], [67], [123], [127].

Fingerprint system can be separated into two categories Verification and Identification.

Verification system authenticates a person’s identity by comparing the captured biometric


____________________________________________________________________________________

- 195 -

characteristic with its own biometric template(s) pre-stored in the system. It conducts one-to-

one comparison to determine whether the identity claimed by the individual is true. A

verification system either rejects or accepts the submitted claim of identity. Identification

system recognizes an individual by searching the entire template database for a match. It

conducts one-to-many comparisons to establish the identity of the individual. In an

identification system, the system establishes a subject’s identity (or fails if the subject is not enrolled in the system database) with/without the subject having to claim an identity.

In order to implement a successful algorithm for fingerprint recognition system, it is

necessary to understand the topology of a fingerprint. A fingerprint consists of many ridges

and valleys that run next to each other, ridges are shown in black and valleys are shown in

white. The ridges bend in such ways as to form both local and global structures; either of

which can be used to identify the fingerprint. The global level structures consist of many

ridges that form arches, loops, whirls and other more detailed classifications, as shown in

Figure 129. Global features shape a special pattern of ridge and valleys. On the other hand,

the local level structures, called minutiae, are further classified as either endpoints or

bifurcations. Minutiae are also given an associated position and direction, as shown in Figure

131.

Self Organizing Map was used for classification and the matching procedure was based on

minutia, as well as on global level structure for finding a reference point by which alignment

of two template was to be accomplished.

In addition, scanned fingerprints are subjected to distortions; including rotation, translation,

non-linear scaling and extraneous or missing minutiae between matching fingerprints, that

must also be taken into account. This creates difficulty in the matching phase because it

causes the minutiae to differ between two identical fingerprints. Position invariant features to

facilitate the recognition were introduced.

Most approaches to recognizing a fingerprint involve five basic stages: (i) acquisition, (ii)

pre-processing, (iii) structural extraction, (iv) post-processing, and (v) and then matching.

Figure 128: A fingerprint sample.


____________________________________________________________________________________

- 196 -

(b) (c) (d)

Figure 129: Fingerprint types: (a) Arch, (b) Left Loop, (c) Right Loop, (d) Whorl.

Figure 130: Fingerprint image showing different ridge features.

Figure 131: Types of fingerprint minutiae and their respective directions.

(a) an endpoint, (b) a bifurcation.

Figure 132: (a) Original fingerprint. (b) Detected minutia.


____________________________________________________________________________________

- 197 -

VII.10.2. Classification

Classification is the problem of identifying to which of a set of categories (sub-populations) a

new observation belongs, on the basis of a training set of data containing observations (or

instances) whose category membership is known.

We can use soft computing tools to classify fingerprints, and map them into major groups

(i.e. arch, left loop, right loop, and whorl).

VII.10.2.1. Soft Computing

Conventional computing or often called as hard computing, requires a precisely stated

analytical model and often a lot of computation time. Many analytical models are valid for

ideal cases, and real world problems exist in a non-ideal environment.

Soft computing differs from conventional (hard) computing in that, unlike hard computing, it

is tolerant of imprecision, uncertainty, partial truth, and approximation. In effect, the role

model for soft computing is the human mind. The guiding principle of soft computing is:

exploit the tolerance for imprecision, uncertainty, partial truth, and approximation to achieve

tractability, robustness and low solution cost. Soft computing may be viewed as a foundation

component for the emerging field of conceptual intelligence. Few soft computing tools are:

Fuzzy Systems, Neural Networks, Evolutionary Computation, Machine Learning and

Probabilistic Reasoning.

VII.10.2.2. Neural Network

The term neural network was traditionally used to refer to a network or circuit of biological

neurons. The modern usage of the term often refers to artificial neural networks, which are

composed of artificial neurons or nodes. Artificial neural networks are composed of

interconnecting artificial neurons (programming constructs that mimic the properties of

biological neurons).

For many years, artificial neural networks (ANNs) have been studied and used to model

information processing systems based on or inspired by biological neural structures. They not

only can provide solutions with improved performance when compared with traditional

problem-solving methods, but also give a deeper understanding of human cognitive abilities.

Among various existing neural network architectures and learning algorithms, Kohonen’s Self Organizing Map (SOM) is one of the most popular neural network models.

Developed for an associative memory model, it is an unsupervised learning algorithm with a

simple structure and computational form.


____________________________________________________________________________________

- 198 -

Self-organization in general is a fundamental pattern recognition process, in which intrinsic

inter- and intra-pattern relationships among the stimuli and responses are learnt without the

presence of a potentially biased or subjective external influence.

The SOM can provide topologically preserved mapping from input to output spaces. It is

mainly used for data clustering and feature mapping. The learning process involves updating

network architecture and connection weights so that a network can efficiently perform a

specific classification/clustering task.

Figure 133: Kohonen’s self-organizing map model.

The input is connected to every cell in the postsynaptic sheet (the map). The learning makes the map localized,

in other words different local fields will respond to different ranges of inputs. The lateral excitation and

inhibition connections are emulated by a mathematical modification, namely local sharing, to the learning

mechanism (so there are no actual connections between cells – grey lines are used to indicate these virtual

connections)

VII.10.2.3. The SOM Algorithm

The SOM uses a set of neurons, often arranged in a 2-D rectangular or hexagonal grid, to

form a discrete topological mapping of an input space, X ∈ Rn.

At the start of the learning, all the weights {w1,w2, ...,wM} are initialized to small random

numbers. wi is the weight vector associated to neuron i and is a vector of the same dimension

– n – of the input, M is the total number of neurons, and let ri be the location vector of neuron

i on the grid.


____________________________________________________________________________________

- 199 -

Then the algorithm repeats the steps shown in Algorithm, where η(ν, k, t) is the

neighbourhood function, and Ω is the set of neuron indexes.

Although one can use the original stepped or top-hat type of neighbourhood function (one

when the neuron is within the neighbourhood; zero otherwise), a Gaussian form is often used

in practice – more specifically…

η (ν, k, t) = exp – …………………………...…………………………... (24)

with σ representing the effective range of the neighbourhood, and is often decreasing with

time.

The coefficients {α(t), t ≥ 0}, termed the ‘adaptation gain’, or ‘learning rate’, are scalar-valued, decrease monotonically, and satisfy:

(i) 0 < α(t) < 1;

(ii) lim t→∞ Σ α(t)→∞;

(iii) lim t→∞ Σ α2(t) < ∞;

The SOM algorithm vector-quantizes or clusters the input space and produces a map which

preserves topology.

It can also be and has been used for classification. In this case, the map is trained on

examples of known categories.

The nodes are then classified or labelled so that the map can be used to classify unseen

samples.

VII.10.3. Approach to Extract Reference Region

In Figure 122, a fingerprint is depicted. The information carrying features in a fingerprint are

the line structures, called ridges and valleys. In this figure, the ridges are black and the

valleys are white.


____________________________________________________________________________________

- 200 -

It is possible to identify two levels of detail in a fingerprint. The directional field (DF), shown

in Figure 134(b), describes the coarse structure, or basic shape, of a fingerprint. It is defined

as the local orientation of the ridge valley structures.

The DF is, in principle, perpendicular to the gradients. However, the gradients are

orientations at pixel scale, while the DF describes the orientation of the ridge-valley

structures, which is a much coarser scale. Therefore, the DF can be derived from the

gradients by performing some averaging operation on the gradients, involving pixels in some

neighborhood [81].


(a) fingerprint, (b) directional field, and (c) singular points.

Figure 135: Fingerprint patterns: arch, loop, and whorl; and core and delta fingerprint landmarks.


____________________________________________________________________________________

- 201 -

Various methods used to estimate the DF from a fingerprint are known from literature. They

include matched-filter approaches [4], [77], [78], methods based on the high-frequency power

in three dimensions [74], 2- dimensional spectral estimation methods [78], and micropatterns

that can be considered binary gradients [41]. These approaches do not provide as much

accuracy as gradient based methods, mainly because of the limited number of fixed possible

orientations.

The elementary orientations in the image are given by the gradient vector [Gx(x,y) Gy(x,y)]T,

which is defined as:

= sign (Gx) I(x,y)

= sign (∂I(x,y)/∂x) …………………………….………... (25)


This equation is used to calculate the elementary orientations in the image by the gradient

vector. The first element of the gradient vector has been chosen to always be positive. The

reason for this choice is that in the DF, which is perpendicular to the gradient, opposite

directions indicate equivalent orientations.

Since the image has background noise, the minutia extraction algorithm may generate

minutiae outside the fingerprint area. So selection of the interest area is one important step.

This step is carried in few phases:

divide the image into blocks,

find the average gradient of each block,



take the approximate middle point of these two particular blocks as the reference point,

crop out a suitable region around this reference point.

VII.10.4. Approach to Extract Minutia



comparison of two fingerprints.

Minutia detection is a trivial task.

Without a loss of generality, we can assume that if a pixel is on a thinned ridge (eight-

connected), then it has a value 1, and 0 otherwise.


____________________________________________________________________________________

- 202 -

Let (x, y) denote a pixel on a thinned ridge, and N0, N1, ..., N7 denote its eight neighbours.

A pixel (x, y) is

a ridge ending if = 1, and

a ridge bifurcation if > 2

However, the presence of undesired spikes and breaks present in a thinned ridge map may

lead to many spurious minutiae being detected. Therefore, before the minutia detection, a

smoothing procedure is needed to remove spikes and to join broken ridges.

VII.10.4.1. Image Acquisition





The basic two-dimensional image is a monochrome (greyscale) image which has been

digitized. Image is described as a two-dimensional light intensity function f(x,y) where x and

y are spatial coordinates and the value of f at any point (x, y) is proportional to the brightness

or grey value of the image at that point.

A digitized image is one where, spatial and greyscale values have been made discrete,

intensity measured across a regularly spaced grid in x and y directions, intensities sampled to

8 bits (256 values).

The method chosen for acquisition of a fingerprint image depends on many different factors,

including the cost and reliability of an input device.

VII.10.4.2. Pre-processing



of this step is the minutiae of that image. The proposed algorithm for pre-processing is as

followed.

VII.10.4.2.1. Fingerprint Image Enhancement

Degradations affect the quality of the fingerprint image. The ridges get some gaps; parallel

ridges connected due to noise and natural effect to the finger like cuts, wrinkles and injuries.

The Fingerprint enhancement is anticipated to improve the contrast between ridges and

valleys and reduce noises in the fingerprint images.


____________________________________________________________________________________

- 203 -

High quality fingerprint image is very important for fingerprint verification or identification

to work properly. In real life, the quality of the fingerprint image is affected by noise; like

smudgy area created by over-inked area, breaks in ridges created by under-inked area,

changing the positional characteristics of fingerprint features due to skin being resilient in

nature, dry skin leading to fragmented and low contrast ridges, wounds causing ridge

discontinuities and sweat on fingerprints also leads to smudge marks and connects parallel

ridges.

VII.10.4.2.2. Noise Reduction

Noise is an unwanted perturbation to a wanted signal. Image noise is generally regarded as an

undesirable by-product of image capture. Noise reduction is the process of removing noise

from a picture (here it is the fingerprint image). We can use different types of filtering

methods like median filter, global and adaptive thresholding to reduce the noise [42].

VII.10.4.2.3. Image Normalization

The objective of this stage is to decrease the dynamic range of the gray scale between ridges

and valleys of the image in order to facilitate the processing of the following steps. The

processing of fingerprint normalization can reduce the variance in gray-level values along

ridges and valleys by means of adjust the gray-level values to the predefined constant mean

and variance. And normalization can remove the influences of sensor noise and gray-level

deformation.

Let I(i,j) denote the gray-level value of pixel (i,j) in acquired image, the size of fingerprint

image is m x n , M and V are the estimated mean and variance of input fingerprint image,

respectively, and N(i, j) denote the normalized gray-level value at pixel (i, j). The normalized

image is then defined as follows:

N(i,j)= …………...…………… (26)

where, M0, and V0 are the expected mean and variance values, respectively. Normalization is

a pixel-wise operation and does not change the ridge and valley structures.

VII.10.4.2.4. Selection of the Interest Region


fingerprint area. So selection of the interest area is one important step.

This step is carried in few phases:


____________________________________________________________________________________

- 204 -

divide the image into blocks,

find the average gradient of each block,



take the approximate middle point of these two particular blocks as the reference point,

crop out a suitable region around this reference point.

Most authors process fingerprints block-wise for calculating Directional Field (DF) [3], [35].

This means that the directional field is not calculated for all pixels individually. Instead, the

average DF is calculated in blocks of, for instance, in this approach 8 by 8 pixel block was

used. Gradient calculation method was used to calculate the reference point. The reference


fingerprint image [41], [76].

VII.10.4.2.5. Binarization



greyscale image. Next greyscale threshold (basic global and adaptive thresholding) is

calculated so pixels above this value become black, and the ones below become white [42].

VII.10.4.2.6. Thinning


parallel thinning algorithms are applied, in order to reduce the width of the ridges in the

binary image to a single pixel. These operations are necessary to simplify the subsequent

structural analysis of the image for the extraction of the fingerprint minutiae. The thinning

must be performed without modifying the original ridge structure of the image. During this

process, the algorithms cannot miscalculate beginnings, endings and or bifurcation of the

ridges, neither ridges can be broken.

VII.10.4.2.7. Minutia Extraction


fingerprint biometric pattern. This process involves the determination of:

whether a pixel, belongs to a ridge or not and,

if so, whether it is a bifurcation, a beginning or an ending point, obtaining thus a group of

candidate minutiae.



____________________________________________________________________________________

- 205 -

VII.10.4.2.8. Cancellation of Improper Minutia

This is an important step of minutia based fingerprint reorganization system. In this step, the

improper minutiae which are mainly result of spurious noise of input image, are cancelled.

VII.10.4.2.9. Position Invariant Feature of Minutia

Ideally, two sets of planar point patterns can be aligned completely by two corresponding

point pairs. A true alignment between two point patterns can be obtained by testing all

possible corresponding point pairs and selecting the optimal one.

The position invariant features are added here by aligning the fingerprints, by matching the

two nearest minutiae of the reference point, and then translating the image such a way that

the third nearest minutia matches, in case there is no match the fingerprints are different.

VII.10.5. Hierarchical Classification

SOM was used in this approach to do the hierarchical classification (to major groups like

loop, whorl, arch and then classify the fingerprints further down, to left loop and right loop,

plain arch and tented arch, etc and then classify further down to actual fingerprint which is a

class).

40% of the data was used to train the system. The features used were the minutia, the number,

location and direction of it.

If two fingerprints don’t belong to the same class, no further steps are needed to be

performed.

VII.10.6. Matching





same finger (i.e., large intra-class variations). The main factors responsible for the intra-class

variations are: displacement, rotation, partial overlap, non-linear distortion, variable pressure,

changing skin condition, noise, and feature extraction errors. Therefore, fingerprints from the

same finger may sometimes look quite different whereas fingerprints from different fingers

may appear quite similar. A minutia matching essentially consists of finding the alignment

between the template and the input minutiae sets that result in the maximum number of

minutiae pairings. In Minutia based matching the similarity between the input and stored

template are computed.


____________________________________________________________________________________

- 206 -

Figure 136: Fingerprint matching steps.



is fed into the system to check

whether it belongs to a particular entry of the database and if so then for which

entry/entries of the database (Identification), or

if it confirms to be fingerprint of a particular person (Verification).

This project further automatically adds new fingerprint to the database, if the fingerprint to be

matched does not exist in the database previously, for probable future use.

At first the noise was removed by median filtering followed by basic adaptive global

thresholding.

The proposed method is based on pixel to pixel matching for minutiae.

In this method the fingerprint image is being cropped with respect to a particular point

(reference point) of the image; this cropped area is called the region of interest. In this

system, the region of interest is taken as the 68 x 68 pixel block around the reference point.

Fingerprints were processed block-wise for calculating DF. This means that the directional

field is not calculated for all pixels individually. Instead, the average DF is calculated in

blocks, for instance, in this approach 8 by 8 pixel block was used.

And gradient calculation method was used to calculate the reference point. The reference


fingerprint image. For the maximum value of the average gradient of two successive blocks

that has the zero crossing, the middle point of the successive blocks is taken as the reference

point.


____________________________________________________________________________________

- 207 -


the threshold value we may get different rate of acceptance and rejection.

Steps used:

Classification, using SOM

Extraction of Reference Region using Gradient

Extraction of minutiae and matching minutia

Result:

Accuracy in Recognition Rate: 96%

False Acceptance Rate: 1%

False Rejection Rate: 3%

(a) (b) (c)

(d) (e) (f)

Figure 137: Result of pre-processing steps.

(a) original image, (b) image after noise reduction and normalization, (c) region of interest,

(d) cropped image, (e) thinned image, (f) extracted minutiae.


____________________________________________________________________________________

- 208 -


The reliability of any automated fingerprint based recognition system strongly relies on the

precision obtained in the extraction process. Extraction of appropriate features is one of the

most important and also difficult tasks for a recognition system. There have been many

algorithms developed for extraction of both local and global structures. Most algorithms

found in the literature are somewhat difficult to implement and use a rather heuristic

approach.


68 pixel values were taken around the reference point. But for a database having larger

fingerprint image sizes, the pixel values can be suitably changed.

For noisy database, it was seen that the 4 x 4 pixel block for gradient calculation is giving us

a better result.

It is also seen that this gradient approach is not suitable for all kinds of fingerprints, and

further attributes are also required in order to accomplish the matching. In future work, this

shortcoming is to be solved.

In this experiment, minutia extraction based fingerprint detection was applied with gradient


Just like in this part, SOM algorithm is used for classification, the singular point detection

method can also be applied as a step to cluster the fingerprint images into major groups (i.e.

arch, tented-arch, left loop, right loop, whorl), and then minutia extraction based method can

be applied on the clusters to achieve a hierarchical fingerprint detection algorithm.


visual checking, but implementing an automated system for this is quite a hard job.

The implementation was done using MATLAB 6, on a system based on Intel® Core™ Duo processor (T650 @ 2.10 Ghz), with a 2.00GB RAM and 32-bit Operating System (Windows

7 Home Basic).


____________________________________________________________________________________

- 209 -

VII.11. Fingerprint Recognition by

Divide & Conquer Method

(FR-DC)

Figure 138: A sample fingerprint.


The recent tendency for human recognition is biometrics. In today’s world, fingerprints, due to its stability and uniqueness, is the most widely used biometrics. Though it is the most

ancient as well as most widely researched biometric technology, a 100% accurate

fingerprinting system is still a myth. Here a new method is being proposed for fingerprint

recognition system, namely divide and conquer method, where a fingerprint is to be cropped

and merged with the database fingerprint and a matching will confirm the

verification/identification. Here frequency domain enhancement is used for pre-processing

the image, and soft computing tool SOM is used for classification of fingerprints. The final

matching is the most used feature of fingerprints, the minutia based.


____________________________________________________________________________________

- 210 -

VII.11.2. Background

Biometrics is the recent trend. Fingerprints are the most widely used biometric now-a-days

[7]. But this idea is ancient. Traces of fingerprints were found even in early day cave

drawings, so we can safely say, even ancient people were drawn to finger ridge patterns.

These patterns are formed during the foetal development, and remain unchanged throughout

the life span of an individual. Injuries like cuts, burns can temporarily damage fingertips but

when/if fully healed, these patterns are reformed back to original pattern. Thus it is claimed

that fingerprints is a very suitable biometric feature. Its stability, reliability, uniqueness and

comparatively cheap recognition procedure amongst all biometrics makes fingerprint

recognition a success. Fingerprints can be used to identify individuals for

private/commercial/forensic purposes.

Figure 139: Fingerprint.

Fingerprints are not entirely genetic characteristic, though people from same family might

have similar patterns, but the overall shape is different for everyone, even for identical twins.

The ultimate shape of finger ridge patterns are believed to be influenced by environmental

factors during pregnancy, like, nutrition, blood pressure, position in the womb, growth rate,

etc.

Figure 138 shows a sample fingerprint. Fingerprints are the graphical flow-like ridges present

on human fingertip. The fingerprint is a duplicate of a fingertip epidermis; when a person

touches a smooth surface, the fingerprint epidermis characteristic is transferred to the surface.

The pattern of the ridges and valleys on human fingertips forms the fingerprint images.

Analyzing this pattern at different levels reveals different types of features that are, global

feature (singular points – core & delta) and local feature (minutia – bifurcations & ridge

endings).


____________________________________________________________________________________

- 211 -

Figure 140: Fingerprint types.

Figure 141: Fingerprint types features.

Figure 142: Few common fingerprint minutia types.


____________________________________________________________________________________

- 212 -

VII.11.3. Working Principle

Fingerprint recognition system uses two types of matching: graph based matching [2], [63],

and image based matching [1]. There are three different levels of features for fingerprints,

they are used for recognition. The most widely used feature is minutia. It is the discontinuity

of a fingerprint ridge structure. There are many minutia types recognized (figure 142), but the

most widely used ones are ridge ending and bifurcation (figure 143). In this project these two

types of minutiae are used for matching. The matching is done for either identification or

verification (figure 145).

Ideally, fingerprints are enhanced for better performance after image acquisition; here

frequency domain based filtering is used for fingerprint image enhancement. Next phase is to

find the effective region by essentially finding a reference point; here middle point of the

effective image area is used for this purpose (this middle point finding is a key task in this

proposed method, but we can also use gradient based method to find reference point, then

crop the effective region, and find the middle point of the region, which is essentially the

reference point itself). Then matching features are extracted; here the matching feature is the

minutia. Simultaneously it’s classified in few pre-defined classes, like arch, loop, and whorl

(figure 147); here SOM is used for achieving this. And then finally the actual matching is

carried out. The steps are shown in figure 144. The programming was done using MATLAB

7.1.

Figure 143: Fingerprint minutia: ridge ending & bifurcation.


____________________________________________________________________________________

- 213 -

Figure 144: Fingerprint recognition steps.

Figure 145: Verification & Identification.


____________________________________________________________________________________

- 214 -

Figure 146: Fingerprint enhanced in frequency domain.

Figure 147: Fingerprint classification.


____________________________________________________________________________________

- 215 -

VII.11.4. Classification

Classification is the problem of identifying to which of a set of categories a new observation

belongs, on the basis of a training set of data containing observations whose category

membership is known [110]. We can use soft computing tools to classify fingerprints, and

then map them into major groups (i.e. arch, loop and whorl). The role model for soft

computing is the human mind. Among various existing neural network architectures and

learning algorithms, Kohonen’s self organizing map (SOM) is one of the most popular neural network models. Developed for an associative memory model, it’s an unsupervised learning algorithm. Self-organization in general is a fundamental pattern recognition process in which

intrinsic inter- and intra- pattern relationships among stimuli and responses are learnt without

the presence of potentially biased or subjective external influence. In SOM, The input is

connected to every cell in the postsynaptic sheet (the map); the learning makes the map

localized, in other words different local fields will respond to different ranges of inputs; the

lateral excitation and inhibition connections are emulated by a mathematical modification,

namely local sharing, to the learning mechanism (so there are no actual connections between

cells – grey lines are used to indicate these virtual connections). The SOM can provide

topologically preserved mapping from input to output without spaces (figure 148), it’s used for data clustering and feature mapping; the learning process involves updating network

architecture and connection weights so that a network can efficiently perform a specific

classification/clustering task. In our method, minutia features are used as SOM feature to

classify a fingerprint into pre-defined classes.

Figure 148: Kohonen’s self-organizing map model.


____________________________________________________________________________________

- 216 -

VII.11.5. Minutia Related Issues

Most of the fingerprint recognition systems first detect minutiae in the image and then match

input image with the template. A minutia is the unique measurable physical characteristic

scanned as input and stored for matching by recognition systems. The two main modules are

minutiae extraction & minutia matching.

The co-ordinates, type of minutia and total number of minutia can be determined, stored and

used for minutia based matching. The method detects key features (bifurcations & endpoints).

It is a trivial task. The presence of undesired spikes and breaks present in a thinned image

may lead to spurious (false) minutiae being detected; false minutia structures including

spikes, spurs, holes, bridges etc might get introduced in the pre-processing steps, and removal

of these is very much a needed operation. We need to note that a false minutia is more

affecting than a missing minutia. If several minutiae form a cluster in a small region, then all

of them, except for the one nearest to the cluster centre are removed.

A minutia matching essentially consists of finding the alignment between the template and

the input minutiae sets that result in the maximum number of minutia pairings. But when a

second print is recorded from the same finger, it is always misaligned from the original, and

slight rotation of the finger also causes some features to vary from the original sample. Two

sets of planar point patterns can be aligned by two corresponding point pairs. The position

invariant features are added here by aligning the fingerprints, by matching the two nearest

minutiae of the reference point (middle point), and then translating the image such a way that

the third nearest minutia matches, in case there is no match the fingerprints are different.

(a) (b)

Figure 149: (a) Sources of error in fingerprint recognition,

(b) Types of false minutia structures. From left to right and up to bottom we have: spike, bridge, hole, break,

spur, and ladder structure. The false minutiae generated by each structure are marked as (x) false ridge ending,

and (o) false ridge bifurcation.


____________________________________________________________________________________

- 217 -

VII.11.6. Matching

Matching is the key operation in the minutia based system (figure 150). Generally an

automated recognition system is achieved with point pattern matching (minutia matching)

instead of pixel-wise matching. Here, pixel to pixel minutia matching is used.

Figure 150: Fingerprint matching.

VII.11.7. Database Related Issue

A database is a structured collection of data; the data are typically organized to model

relevant aspects of reality in a way that supports processes requiring this information. The

term database system implies that data are managed to some level of quality (accuracy,

availability, usability, resilience) and this in turn often implies the use of a general-purpose

database management system (DBMS).

This fingerprint recognition system works for two types of matching (figure 145); one

fingerprint is fed to the system to check (a) whether it belongs to a particular database entry

and if so then which entry/entries (identification), or (b) if it confirms to be a fingerprint of a

particular person (Verification).

This project further automatically adds new fingerprints to the database, if the fingerprint to

be matched doesn’t exist in the database previously (in a separate cluster), for possible future use.


____________________________________________________________________________________

- 218 -

VII.11.8. Algorithm

Steps used in this method:

Take input image

Enhance the image (noise reduction, image normalization)

Find the reference point and crop the interest region, and find the middle point

Extract minutiae (binarization, thinning, extraction of minutiae and removal of improper &

false minutiae)

Use SOM to classify the image (arch, loop, whorl)

If input image and template image (database image with which the comparison is being

carried on) belong to same class, proceed, else show “not matched” (for verification) or go to next template image (for identification)

Cut both the input image and the template image in halves using the middle point; fuse the

left side of input image with right side of template image; and right side of the input image

with the left side of the template image

Do pixel-wise matching for minutia matching, if 80% minutiae match, a recognition match

occurs.

VII.11.9. Model Analysis

Figure 151: Fingerprint preprocessing:

Original fingerprint Binarized fingerprint Finding 4 extreme points Finding middle point.


____________________________________________________________________________________

- 219 -

Figure 152: Cropped Input image (image after finding the reference point, cropping and thinning).

Figure 153: (a) Input image with detected minutia, (b) Left half of input image, (c) Right half of input image.

Set 1 Set 2

Figure 154: Set 1(a) Left half of input image & Set 1(b) Right half of template image (with detected minutia); to

be fused to make a new image, to be compared with the new image made by fusing

Set 2(a) right half of input image with Set 2(b) left half of template image.


____________________________________________________________________________________

- 220 -

(a) (b)

Set 1

(a) (b)

Set 2

Figure 155: Finding reference point by gradient based method:

Set 1(a) original picture, Set 1(b) cropped region; Set 2(a) original picture, Set 2(b) cropped region.

(a) (b) (c)

Figure 156: Diving the picture: (a) cropped image, (b) Left half of input image, (c) Right half of input image,

(but dividing the image in half before minutia detection is not to be performed).


This fingerprint recognition system gave us 96% accuracy. There are still scopes for

improvements. As human brain recognizes biometrics neuro-fuzzily, whether a neuro-fuzzy

system can help to achieve better accuracy is under experiments. We can use soft computing

for minutia detection for better result. This system doesn’t work for extremely noisy or partially corrupted fingerprint database. The growth of finger, in case of growing from child

to adult, is also not considered in this proposal. There is one distinguished advantage of this

algorithm; minor accidental cases would be managed, such as if after taking the enrollment

copy of the fingerprint, in case if any accident happened resulting in cut marks in finger then

false rejection would occur during the fingerprint recognition; but by the use of this new

method this problem would be solved as it works with 80% minutia match, and missing

minutiae would be ignored; and this proposed approach works on partial fingerprint images.


____________________________________________________________________________________

- 221 -

VII.12. Point Extraction Technique

in Spatial-domain

for Fingerprint Recognition

(PETS)


Here a new proposal is made for a new algorithm to make a simpler approach to Fingerprint

Recognition, to reduce False Rejection due to accident and to reduce the problem due to

shrinking of finger due to winter season or water, namely Point Extraction Technique in

Spatial-domain for fingerprint recognition (PETS).

Fingerprint is one of the wonder in the biometric and forensic science. Day by day it was

groomed and turned into digital application. But for some technical problems it is now in

little back foot. But let’s introduce a new mechanism to store the corresponding data and

some reference points. It has a low space complexity, simple logical fundamental side to

understand and high distinguishing factor.

After crossing the thresholding of old traditional and complex mechanisms, a totally new

mechanism is being introduced here, which does not bother with types of curve, pattern

recognition, special features like core, delta, bifurcation points and all of these kind of

complex things. It deals only with the shape of the finger and takes the border ridges as the

reference to the fingerprint. The database contains a small array and a special serial marking

for each array.

This is an approach to make fingerprint recognition technology more ease of use in our daily

life for security, forensic, and other biometrics purposes.

There are many kinds of used technologies for fingerprint recognition, but in all cases some

basic problems during collecting the fingerprint remain. Basically maximum problems arise

at the time of scanning the fingerprint (recently most fingerprint scanners take atleast 3 copies

of same finger and if they seem same then only register a fingerprint input).

Let’s discuss here only about two types of drawbacks which can be overcome by this new

method. Firstly, after taking the enrolment copy of the fingerprint, in case there was kind of

any accident happened and any cut mark be found in finger then False Rejection would occur.

But by the use of this new method this problem will be solved greatly as it works only with

the border area and middle point of the fingerprint.

Secondly, after taking the enrolment copy of the fingerprint, in the winter season the fingers

might shrink and False Rejection might happen. But by using this new method this problem

would be solved.


____________________________________________________________________________________

- 222 -

VII.12.2. Proposed Work

Here let’s make a new experiment where data storing is much more efficient than other old

technologies, with low space complexity, simple data structure and simple fundamental

logics.

It works with eight directional searching algorithm to calculate the distance from the middle

point, according to a rectangular border with respect to most Left, Right, Top and Bottom

points of that image, to the every 45 degree cross section of the border ridges of that image.

VII.12.3. Algorithm

Step 1:- Take an image of fingerprint through a scanner.

Step 2:- Convert that image to a .pgm file in ASCII mode.

Step 3:- Use a filter to remove the noise.

Step 4:- Use thresholding to binarize the image.

Step 5:- Calculate the four extreme points of the picture for border. i.e. Left, Right, Top,

Bottom.

Step 6:- Make a rectangle using the four extreme points and calculate the middle point.

Step 7:- Find the eight extreme points every after 45 degree cross section.

Step 8:- Calculate the distance between middle point and eight extreme points.

Step 9:- Form an array containing these eight distances.

Step 10:- Keep that array with a reference name or number in database.

Figure 157: Matrix representation of the algorithm.


____________________________________________________________________________________

- 223 -

VII.12.4. A Model Result after Implementation of P.E.T.S.

Figure 158: PHASE –I

A. ORIGINAL IMAGE B. AFTER FILTERING AND BINARIZATION

A B C

Figure 159: PHASE–II

A. FIND THE EXTREME LEFT, RIGHT, TOP, AND BOTTOM POINT.

T:- TOP; L:-LEFT; B:-BOTTOM; R:-RIGHT.

B. DRAW THE RECTANGULAR BORDER & FIND THE MIDDLE POINT. M:- MIDDLE POINT.

C. 8 EXTREME POINTS FOR EACH CROSS SECTIONS.

Figure 160: PHASE–III

A. DISTANCE FROM MIDDLE POINT TO 8 POINTS, B. THE DISTANCES IN AN ARRAY.


____________________________________________________________________________________

- 224 -

VII.12.5. Results

IMAGE1 IMAGE 2 IMAGE 3 IMAGE 4 IMAGE 5

IMAGE 6 IMAGE 7 IMAGE 8 IMAGE 9 IMAGE 10

Figure 161: INPUT Fingerprint of 10 Different Persons.

Figure 162: BAR GRAPH For Inter Class Difference.

The above bar graph, which is according to the resultant table, represents that each and every

one image has a unique value as a distinguished factor. Thus the aim through this experiment

about inter class difference is successfully achieved.


____________________________________________________________________________________

- 225 -

pos_1 pos_2 pos_3

pos_4 pos_5 pos_6

Figure 163: Set of fingerprint of person -1 in different position.

Figure 164: Bar graph for intra class difference.


____________________________________________________________________________________

- 226 -

pos_1 pos_2 pos_3

pos_4 pos_5 pos_6

Figure 165: Set of fingerprint of person-2 in different position.

Figure 166: Bar graph for intra class difference.

The above bar graph, which is according to the resultant table, represents that the set of same

fingerprint images has a unique value as a distinguished factor. But in some cases it can be

observed that there are some little value miss-match for the same images. This is due to

scanning problem, and in some cases false rejection occurred. But in the maximum cases the

experiment about intra class difference is successfully done.


____________________________________________________________________________________

- 227 -

VII.12.6. Advantages and Disadvantages

There are some advantages of this algorithm such as the output of this technology occupy

much less space for an image, it will take less time to store and check that image. This

algorithm is easy to implement. Some minor accidental cases will be managed. Shrinking of

finger will be managed. There will be no question about huge False Acceptance case.

But there is a big question about False Rejection as a result of fault during taking the image

of fingerprint. And also growth of a finger cannot be considered.

VII.12.7. Conclusion

This new approach to fingerprint (recognition digitally) was proposed to rescue the lost glory

of this technology. This approach takes a very small space in database to store the

corresponding data for an image. This has high efficiency about distinguished factors and

with less complexity than other technologies. It is a very simple approach but without any

complexity of curvature, pattern recognition.

Actually this type of approach might make this digital fingerprint recognition technology

more simple and easy to use in our real life.

The programming was done using C language in LINUX environment. The database was

created with the fingerprints of the department students captured.

This experiment was done with a small database. And finer approach is planned to be taken

in future.


____________________________________________________________________________________

- 228 -

VII.13. Revolutionary Extended Spatial Point

Extraction using Circular Technique

(RESPECT)


Let’s propose here a new algorithm to make a simpler approach to Fingerprint Recognition,

to reduce False Rejection due to accident and to reduce the problem due to shrinking of finger

due to winter season or water, namely Revolutionary Extended Spatial Point Extraction using

Circular Technique for fingerprint recognition (R.E.S.P.E.C.T).

This work is inspired by the previous work namely Point Extraction Technique in Spatial-

domain for Fingerprint Recognition (PETS).

In that experiment a new technology with low space complexity, simple data structure and

simple fundamental logics was proposed.

That method worked with eight directional searching algorithm to calculate the distance from

the middle point, according to a rectangular border with respect to most Left, Right, Top and

Bottom points of that image, to the every 45 degree cross section of the border ridges of that

image.

VII.13.2. Proposed Work

Here let’s introduce a project proposal about fingerprint recognition technique with the help

of the previous work namely PETS.

The method is going to cover the fingerprint image with a circle, with the help of the middle

point, according to a rectangular effective area, calculated using the top-most, bottom-most,

right-most and left-most points of the fingerprint image.

VII.13.3. Algorithm


Step 2:- Use a filter to remove the noise.

Step 3:- Use thresholding to binarize the image.

Step 4:- Do the thinning of the fingerprint image.


____________________________________________________________________________________

- 229 -


Bottom.


Step 7:- Calculate the lowest distance between the middle point and the extreme four points,

to be used as the radius of the circle to be drawn.

Step 8:- Using the calculated middle point, draw a circle and store the distance of the points

where the circle cuts the thinned fingerprint image, from the middle point in a dynamic array.


VII.13.4. A MODEL Analysis of RESPECT

PHASE -I:- (step1,2,3)

Figure 167: (a) Original image and (b) After filtering and binarization.

PHASE-II:- (step 4)



____________________________________________________________________________________

- 230 -

PHASE III:- (step 5,6)

(a) (b)

Figure 169: Pre-processing Steps:

(a) Find the extreme left, right, top, and bottom point. T:- top; L:-left; B:-bottom; R:-right.

(b) Draw the rectangular border & find the middle point. M:- middle point.

PHASE-IV:- (step 7,8)

(a) (b)

Figure 170: (a) Circle drawn, and (b) Reference points noted.


____________________________________________________________________________________

- 231 -

PHASE-V:- (step 8,9)

Figure 171: Distances From Middle Point.

Figure 172: Dynamic Array of Distances.

VII.13.5. Advantages and Disadvantages

There are some advantages of this algorithm such as the output of this technology occupy

much less space for an image, it will take less time to store and check that image. This

algorithm is easy to implement. Some minor accidental cases will be managed. Shrinking of

finger will be managed. There will be no question about False Acceptance case.

But there is a big question about False Rejection as a result of fault during taking the image

of fingerprint. And also growth of a finger cannot be considered.


This new approach to fingerprint (recognition digitally) will rescue the lost glory of this

technology. This approach takes a very small space in database to store the corresponding

data for an image. This has high efficiency about distinguished factors and with less

complexity than other technologies, it is a very simple approach but without any complexity

of curvature, pattern recognition. Actually this approach might make this digital fingerprint

recognition technology more simple and easy to use in our real life.


____________________________________________________________________________________

- 232 -

VII.14. A Bar Code Design and Encoding for

Fingerprints

(ABCDEF)


Fingerprints are an effective choice for recognition. Fingerprints are graphical flow-like

ridges present on human fingers or fingertips. An individual's fingerprints are unique thus

giving all humans an identification barcode itself. Here, a Barcode approach for Fingerprint

Recognition systems is being presented.

Human fingertips enclose ridges and valleys which altogether forms distinctive patterns [3],

[4]. A fingerprint image exhibits a pattern of ridges (darker regions) and valleys (lighter

regions). The local topological structures of this pattern added with the spatial relationships

determine the uniqueness of a fingerprint.


structures. There are many different types of local ridge structures that have been identified

[121].

Most of the automatic fingerprint identification/verification systems adopt the model used by

the FBI [17].

Figure 173: Fingerprints.

The approaches to recognizing a fingerprint involve five basic stages:

(i) acquisition, where the image is obtained from hardware or a file;


____________________________________________________________________________________

- 233 -

(ii) pre-processing, which may include noise reduction, image enhancements and error

correction;

(iii) structural extraction, where global and local structures may be found;

(iv) post-processing, where the structures are converted into a more useful format;

(v) and then matching, where fingerprints are compared against a database.

Most of the fingerprinting system depends on minutiae (that are ridge endings and

bifurcations) based system. This one is the project of a fast and simple fingerprint

identification system based on barcode system.

Presenting here a method for fingerprint recognition, by designing and encoding a barcode.

Not much work is done in literature by this method which can give better time and space

complexity values. This is an approach to make fingerprint recognition technology more ease

of use in our daily life for security, forensic, and other biometrics purposes.


obtained in the feature extraction process. Extraction of appropriate features is one of the

most important tasks for a recognition system. Therefore a very important phase is noise

removal for this barcoding system, as even a small amount of noise can make the ridge

structure different.

Figure 174 shows a sample fingerprint and figure 175 shows different fingerprint types.

Figure 174: A fingerprint sample.


____________________________________________________________________________________

- 234 -

(a) (b)

(c) (d)

Figure 175: Fingerprint types: (a) Whorl, (b) Arch, (c) Right Loop, (d) Left Loop.

VII.14.2. Fingerprint Ridge

In Figure 176, a fingerprint is shown along-with the different ridge features. The information

carrying features in a fingerprint are the line structures, called ridges and valleys. In this

figure, the ridges are black and the valleys are white.


____________________________________________________________________________________

- 235 -

Figure 176: Fingerprint image showcasing different ridge features.

Figure 177: Sample barcode.

VII.14.3. Barcode

A barcode is an optical machine-readable representation of data relating to the object to

which it is attached. Originally barcodes systematically represented data by varying the

widths and spacings of parallel lines (so it’s linear i.e. one-dimensional).

Figure 177 shows a barcode sample.

VII.14.4. Fingerprint Barcode

A proposal of a method of expressing a fingerprint by the means of barcode is presented here.

The barcode obtained from a fingerprint would look like something in figure 179.


____________________________________________________________________________________

- 236 -

(a)

(b)

(c)

Figure 178: Steps of general fingerprint recognition system.

Image Capture

Image Pre-processing &

Feature Extraction

Matching Stored Pattern

Enrollment Stage

Authentication Stage

Matching Score

Fingerprint

Enrollment

Fingerprint

Enrollment

Minutiae

Extraction

Minutiae

Extraction

Stored

Template

Minutiae

Matching

Acquisition

Image is Obtained from Input Device or File

Pre-Processing

Image Enhancement

Finding Reference Point & Interest Region

Conversion to Black and White

Structural Extraction

Local and/or Global Structures are Extracted from the Image

Post-Processing

Extracted Data is Converted to More Useful Format

Matching

Fingerprints are Compared against a Database


____________________________________________________________________________________

- 237 -

Figure 179: Fingerprint BARCODE.

VII.14.5. Algorithm


Step 2:- Remove noise.

Step 3:- Binarize the image.


Bottom.


Step 6:- Draw an imaginary line horizontally using the middle point.

Step 7:- Check if the imaginary line crosses the original fingerprint ridges, and store the

values as 1 if and where it cuts the line or as 0 if it does not. This way we can get an array of

1s and 0s.



____________________________________________________________________________________

- 238 -

VII.14.6. Discussion

For a better result we need to remove pores from the fingerprint before applying the

algorithm (As it can be seen in a good quality fingerprint image, the friction ridges are dotted

with small circular openings – the sweat pores, the perspiration is exuded from these).

Noise is an unwanted perturbation to a wanted signal. It is the random variation of brightness

or colour information in images, produced by the sensor and circuitry of a scanner or digital

camera. Image noise is often regarded as an undesirable by-product of image capture. Noise

reduction is the process of removing noise from a picture, and more than often it is a must to

do process for image processing (in the pre-processing phase) [40], [42].

Here a different type of approach was used to remove noise:

firstly apply Median filter and then

the Sobel operator on the obtained image to remove noise for the original fingerprint

image (figure 180).

We can take any other reference point (such as singular point) and not the middle point, but it

has to be the same for each image. For an even better space complexity we can save the total

number of 1s in the array followed by total number of 0s and not the 1s and 0s.

We can also save the total number of black strips as a feature (in the sample, figure 179, it is

25). Note in the barcode the strips always start and end with the black one.

We can also save the pixel values of black strips only for an even better space complexity,

but in this project both black and white strips were saved.

To have a fixed number of array elements for each fingerprint image, we can crop a fixed

length area around the reference point and perform the operation on that area.

Observations showed that, for the trial database, for upto 15% of angular displacement the

obtained barcode remains the same, thus helping in incorporating position invariant features.

If a person gets a scar in the finger after first storing his fingerprint in the database, later on

his fingerprint may not match with the data value stored earlier giving a false rejection.

One point to be noted is that it was overlooked in the experiment: the issue of a finger

growing; as the person ages. Though the finger ridge pattern remains same, but the size of the

finger gets bigger, so a scaling is needed before performing the recognition steps.


____________________________________________________________________________________

- 239 -

VII.14.7. Model Analysis

(a) (b)

Figure 180: (a) Input image with noise, and (b) the image after the noise removal

Figure 181: (a) Original Fingerprint, and (b) Binarized Fingerprint.


____________________________________________________________________________________

- 240 -

Figure 182: Fingerprint BARCODE-ing intermediate state.

The barcode obtained for the sample fingerprint (Figure 179) is as follows:

10 11 20 5 18 6 15 5 12 3 11 4 6 9 11 7 12 9 15 3 7 10 16 5 14 9 12 4 10 8 13 8 13 8 17 7 17 12 17 7 22 12 16 13 10 13 13 12 7

Where the numbers in black (bold and underlined ones) represents the total number of black

pixels (1s) followed by gray number represents the number of white pixels (0s).

Here a random point is taken as the reference point, for illustration purpose.



structures. Most algorithms found in the literature are not only difficult to implement, but

they also use heuristic approach.

The reliability and usability of any automatic fingerprint recognition system very much relies

on the precision obtained in the feature extraction process. Extraction of appropriate features

is one of the most important tasks for a recognition system.

Therefore, a very important phase is noise removal for this proposed barcoding system, as

even a small amount of noise can make the ridge structure different, affecting the whole

barcoding system.

Injuries like cuts, burns and bruises can temporarily spoil quality of fingerprints but when/if

fully healed patterns will be restored in most cases. This property makes fingerprints a very

strikingly attractive biometric identifier.


____________________________________________________________________________________

- 241 -

VII.15. Comparative Analysis of All the

Proposed Methods

The following table compares between all the methods experimented during this dissertation.

Initially an already existing Fingerprint Recognition Method (using minutia) was

experimented, and then this method (FR-M) was compared with other methods proposed

during this research work. There were eight proposed methods, namely, Fingerprint

Recognition using Minutia Based Approach (FR-MBA) (Experiments done on Dummy

Database), Fingerprint Recognition using Global and Local Structures (FR-GLS), Fingerprint

Recognition using a Reference Point and Pores (FR-RPP), Fingerprint Recognition By

Classification using Neural Network and Matching using Minutia (FR-NNMM), Fingerprint

Recognition by Divide & Conquer Method (FR-DC), Point Extraction Technique in Spatial-

domain for fingerprint recognition (PETS), Revolutionary Extended Spatial Point Extraction

using Circular Technique (RESPECT) and A Bar Code Design & Encoding for Fingerprints

(ABCDEF).

Table 5: Comparison Table of Different Approaches towards Fingerprint Recognition.

Method Recognition Rate False Acceptance

Rate (FAR)

False Rejection

Rate (FRR)

FR-M 90% - -

FR-MBA 90% - -

FR-GLS 92% 5% 3%

FR-RPP 91% 5% 4%

FR-NNMM 96% 1% 3%

FR-DC 96% 2% 2%

PETS 94% 5% 1%

RESPECT 95% 4% 1%

ABCDEF 96% 2% 2%

During this study, few different approaches were tried towards fingerprint recognition.

From the table we can see that there is scope of improvements of the algorithms used. The

proposed Soft Computing for Identification of Fingerprint Image (SCIFI) gave a better result

with approximate 97% Accuracy. This method can further be made more accurate.

The following graphs show the comparative results.


____________________________________________________________________________________

- 242 -

Table 6: Comparison Table of Recognition Rates of Different Approaches.

Table 7: Comparison Table of FAR and FRR.

Table 8: Comparison of Different Approaches towards Fingerprint Recognition.

86

88

90

92

94

96

98

Recognition Rate (%)

Recognition Rate (%)

0

1

2

3

4

5

6

FR-GLS FR-RPP FR-NNMM FR-DC PETS RESPECT ABCDEF SCIFI

Comparison of False Acceptance (FAR) & False Rejection (FRR)

FAR (%) FRR (%)

0

20

40

60

80

100

120

FR-GLS FR-RPP FR-NNMM FR-DC PETS RESPECT ABCDEF SCIFI

Comparison of Methods

Recognition Rate (%) FAR (%) FRR (%)


____________________________________________________________________________________

- 243 -

VII.16. Comparative Study with Previous Works

in Fingerprint Recognition


In this chapter different proposed methods are compared numerically with some of previous

works defining the state of the art in the field of fingerprint recognition.

Many different approaches were experimented during this dissertation. Few approaches were

inspired by the traditional methods where as other totally different approaches were proposed

and experimented as well. Undoubtedly, the existence of a completely different but

equivalent (or almost equivalent) solution increases the insight into the problem's nature.

All the proposed methods were compared in tabular form in the previous chapter (Chapter

VII. 15.).

The proposed methods which are comparable with some previously done good works in this

field are discussed here.

VII.16.2. Minutia Based Method

A. K. Jain, L. Hong, and R. Bolle’s “On-line fingerprint verification” operates on two stages, (i) minutia extraction and (ii) minutia matching [3]. This method gives > 99% recognition

rate on average.

When the minutia based Fingerprint Recognition was experimented during the dissertation

90% accuracy was achieved.

Table 9: Comparison of Average Number of False Singular Points using Different Methods

Method Approached Accuracy

On-line fingerprint verification [3] > 99%

Fingerprint recognition experimented in this

dissertation

~ 90%

The low accuracy was due to poor quality fingerprint images. Also the automated fingerprint

ridge orientation of the image was not carried out.

More robust fingerprint pre processing was done in later approached methods. Fingerprint

invariant features were also incorporated in the system later on giving a better result

(compared later in this chapter).


____________________________________________________________________________________

- 244 -

VII.16.3. Singular Point Extraction

In the paper “Systematic Methods for the Computation of the Directional Fields and Singular

Points of Fingerprints,” by Asker M. Bazen and Sabih H. Gerez [41], the experiments were done on singular points. The subject of the paper was the estimation of a high resolution

directional field of fingerprints. A method, based on principal component analysis, was

proposed. The method computed the direction in any pixel location and coherence. This

method provided the same results as the “averaged square-gradient method” that is known from literature. This paper also dealt with singular point detection. An algorithm based on the

Poincare index was proposed that extracts singular points from the high-resolution directional

field.

Let’s compare the result with Fingerprint Recognition using Global and Local Structures

(FR-GLS) method proposed in this thesis.

Table 10: Comparison of Average Number of False Singular Points using Different Methods

Method used Average number of false SPs (%)

No segmentation, the whole image is taken as

fingerprint region [41]

15.4

Manual segmentation 0.8

High resolution segmentation algorithm that uses the

coherence estimate as feature and morphological

operators to smooth the segmentation result [41]

0.8

High resolution segmentation algorithm that uses the

coherence, the mean, and the variance of the

fingerprint image as features and morphological

operators to smooth the segmentation result [41]

0.5

Proposed Method in this thesis, gradient based method

on calculated region of interest

2.0

From the table it can be deduced that the singular point detection method used in this thesis

gives satisfactory result compared to the state of the art method.

VII.16.4. Pore Based Matching

In “Pores and ridges: High-resolution fingerprint matching using level-3 features,” A. Jain, Y. Chen, and M. Demirkus [80], it was established that fingerprint friction ridge details are

generally described in a hierarchical order at three levels, namely, Level 1 (pattern), Level 2

(minutiae points) and Level 3 (pores and ridge shape). Most Automated Fingerprint

Identification Systems (AFIS) employ only Level 1 and Level 2 features. A matcher that

utilizes Level 3 features, including pores and ridge contours were proposed by A. Jain, Y.

Chen, and M. Demirkus [80]. It was observed that matching results based on Level 3 features

alone is very comparable to that of Level 2 features and EER (Equal Error Rate) values were


____________________________________________________________________________________

- 245 -

reduced (relatively ~20%) when Level 3 features are employed in combination with Level 1

and 2 features.

EER of Level 2 matcher (Minutiae -based) = 4.2%

EER of Level 3 matcher (Pore-based) = 4.9%

The pore based matching mechanism proposed in this thesis gives a 91% accurate result. It

was also noted in this thesis too that level 1 and level 2 features combined with level 3 feature

gives a more accurate result. The low accuracy was due to poor quality fingerprint images.

In the next chapter (Chapter VII.17.) it will be noted the different types of limitations that a

fingerprint recognition system faces.

VII.16.5. Hybrid Matching System

In general, the hybrid matcher performs better than a minutiae-based fingerprint matching

system. In Ross, A., Jain, A., and Reisman, J., “A hybrid fingerprint matcher,” it was stated that the most fingerprint matching systems rely on the distribution of minutiae on the

fingertip to represent and match fingerprints [143]. While the ridge flow pattern is generally

used for classifying fingerprints, it is seldom used for matching. This paper described a

hybrid fingerprint matching scheme that uses both minutiae and ridge flow information to

represent and match fingerprints. The Genuine Accept Rate (GAR) of the hybrid matcher

proposed in that paper observed to be ~10% higher than that of a minutiae-based system at

low False Acceptance Rate (FAR) values. The hybrid technique proposed [143] outperformed

the minutiae based scheme over a wide range of FAR values. For example, at a FAR of 0:1%,

the GAR of the minutiae matcher was ~67%, while that of the hybrid matcher was ~84%.

The Equal Error Rate (EER) of the hybrid technique was observed to be ~4%.

During this thesis two types of hybrid matcher were experimented, namely Fingerprint

Recognition By Classification using Neural Network and Matching using Minutia (FR-

NNMM), Fingerprint Recognition by Divide & Conquer Method (FR-DC), in both cases the

accuracy rate were ~96%. %. Neuro-Fuzzy based approaches were experimented and the

present work could be extended to make the system more robust using these and other Soft

Computing algorithms.


Some of proposed methods are compared here with some of previous good works in the field.

The existence of a completely different but equivalent or almost equivalent solution increases

the insight into the problem's nature. It was noted that poor quality fingerprint images give

low accuracy rate. There are many other limitations as well. Soft Computing based

approaches were experimented too, which is not a much experimented technique towards

fingerprint recognition. Future plan is to make the system more accurate.


____________________________________________________________________________________

- 246 -

VII.17. Limitations of Fingerprint Recognition

Systems


The current tendency for human recognition is biometrics. In today’s world, fingerprints, due to its stability and inimitability, is the most extensively used biometrics. Though it is the most

ancient as well as most commonly researched biometric technology, there is still a lot of

scope of improvements.

Fingerprints can be used to identify individuals for private and commercial purposes.

Fingerprint recognition system is now viable for technology to be used in the matching of

fingerprints in real time. A fingerprint is often used for biometric identification in criminal

investigations. The use of fingerprints as a means of identifying individuals has been used for

centuries. Different biometric features can uniquely identify a person unless there are

identical twins. In case of identical twins many biometrics fail to distinguish them as separate

person, but fingerprint still can distinguish as two different persons. Fingerprint is the most

balanced biometric system. Its efficiency and comparative cheap identification system makes

fingerprinting system the most advantageous biometrics. It is assumed that every single

person possesses unique fingerprints [121] and hence the fingerprint matching is considered

as one of the most reliable and effective techniques of person identification.

Figure 183: Various Biometrics.


____________________________________________________________________________________

- 247 -

Biometrics (that is any human physiological or behavioural characteristic having

Universality, Uniqueness, Permanence, and Collectability) is not a sure-shot authentication

system, like a password, which can always perform (but passwords can be forgettable, able to

be stolen, guessable, etc which is not possible for biometrics, as it is human feature/nature).

Biometric features are analogue information. Therefore, they are subject to variability when

captured by biometric scanning devices (fingerprint sensors, microphones, cameras, and so

forth). Such features can be digitalized, but the data will nevertheless be characterized

fuzzily. There are a few extra limitations of fingerprints as well. Along with all common

disadvantages of biometric system on general (such as it’s not a sure-fire authentication

method like a password; or the technical vulnerability associated with the fingerprinting

system; or a possible disliking of common people towards touching a system, or the fear of

getting traced or being treated like criminals), there are noted cases of criminals to remove

fingers to open biometric locks. Also fingerprints might get deformed depending on the

environment, like the fingerprints of the people working in chemical industries are often

affected. If a person gets a scar in the finger after first storing his fingerprint in the database,

later on his fingerprint may not match with the data value stored earlier giving a false

rejection. Also we need to take into consideration the effects of weather and water on

fingertips. And we cannot overlook the effects of skin diseases on fingertips, thus affecting

the fingerprints. Other than these drawbacks of the fingerprinting systems, another limitation

is, fingerprints get resized with age, though the structure remains same but the proper and

perfect scaling of finger images cannot always be performed automatically as/if/when

required for automated fingerprint recognition system. Another limitation is, even if we

overlook all the above stated disadvantages, a 100% technically accurate fingerprint

recognition system is still a myth. There are other limitations of fingerprint systems as well.

Here in this part we will be discussing the limitations of fingerprints in general. Limitations

of fingerprints biometric are reviewed in the part with illustrative examples.

VII.17.2. Fingerprints

Fingerprints are the most widely used biometric now-a-days [2]. But this idea is ancient.

Traces of fingerprints were found even in early day cave drawings, so we can safely say, even

ancient people were drawn to finger ridge patterns. These patterns are formed during the

foetal development, and remain unchanged throughout the life span of an individual. Injuries

like cuts, burns can temporarily damage fingertips but when/if fully healed, these patterns are

reformed back to original pattern. Thus it is claimed that fingerprints is a very suitable

biometric feature. Its stability, reliability, uniqueness and comparatively cheap recognition

procedure amongst all the biometrics make fingerprint recognition a success.

Fingerprints are the graphical flow-like ridges present on human fingertip. The fingerprint is

a duplicate of a fingertip epidermis; when a person touches a smooth surface, the fingerprint

epidermis characteristic is transferred to the surface. The pattern of the ridges and valleys on

human fingertips forms the fingerprint images. Analyzing this pattern at different levels


____________________________________________________________________________________

- 248 -

reveals different types of features that are, global feature (singular points – core & delta) and

local feature (minutia – bifurcations & ridge endings).

Figure 184: Comparison table of different Biometrics.

VII.17.3. Shortcomings of Biometrics

Biometrics still has many hurdles to get by in order to become present and common far and

wide. Problems that face biometric growth is the fact that the cost of identification devices

are, presently, much too high and “people are hesitant to trust giving a ‘piece of themselves’ to a machine”. Another problem is that biometrics has always been used in the case of criminals, and when we start using these identification technologies on innocent civilians, it

gives the innocent civilians a presumption of guilt. Perhaps the strongest argument against

implementing biometrics into our everyday lives is that people would have to enter the

information into machines, and people make mistakes. In a world where a person’s name would be tied to nothing but that person’s biometric fingerprint, a mix-up could be disastrous

and place false guilt on someone. The shortcomings of biometrics can be specified as:

The reliability of any biometric identification depends on ensuring that the signal

acquired and compared has actually been recorded from a live body part of the person to

be identified and is not a manufactured template.

Police have at times misused biometric information.


____________________________________________________________________________________

- 249 -

Another disadvantage is that a determined pirate can steal the biometric information if it's

stored on a computer.

There are many other drawbacks to these systems, such as, biometrics requires a lot of

data to be kept on a person, these systems are not always reliable as human beings change

over time. And if someone is ill; eyes puffy, voice hoarse or fingers are rough from

labouring for example, it may be more difficult for the machinery to identify the person

accurately.

Every time someone uses biometrics, he/she is being tracked by a database bringing up a

range of privacy issues.

As with other identification infrastructure (national residents databases, ID cards, etc.),

civil rights activists have voiced concerns that biometrics recognition technology might

help governments to track individuals beyond their will.

An objection has been raised in some cultures against fingerprint scanners, where a finger

has to touch a surface.

The most effective disadvantages are the expense and technical complexities of

biometrics systems. And if the system is strict enough, in case of emergency too (or in

case of data loss due to accidental reasons) it may not give access permission. Like for

example there is fire caught in an office and the door is not opening due to loss of data

stored due to fire, or the wires connecting the devices got mutilated and the gates are all

biometric operated, or the biometric lock itself got spoiled; then how will the people

inside survive! Well, we do need to give importance to security, but we also need to

consider safety measures. And safety also includes not letting unauthorized persons to go

through access control without a checking even in case of emergency.

VII.17.4. Drawbacks of Fingerprints

A person can manually damage (or distort) fingertip area (like say via micro surgery).

There are noted cases of criminals to remove fingers to open biometric locks.

Many commercially available recognition systems are easily fooled by presenting a high-

quality image instead of placing a real fingerprint, which makes such devices unsuitable

for unsupervised applications, such as door access-control systems.

The fingertips and hence the fingerprints of those people working in chemical industries

are often affected. Therefore these companies should not use the finger print mode of

authentication.

There is influence of accidents; if a person gets a scar in the finger after first storing his

fingerprint in the database, later on his fingerprint may not match with the data value

stored earlier giving a false rejection.

After taking the enrolment copy of the fingerprint, in the winter season the fingers may

shrink and false rejection may happen.

Also effect of water on finger ridge cannot be neglected.

Influence of skin diseases can also be noted here, in which fingertips can be temporarily

or permanently damaged.


____________________________________________________________________________________

- 250 -

Also it is often overlooked, the issue of a finger growing; as the person ages. Though the

finger ridge pattern remains same, but the size of the finger gets changed, so an

appropriate scaling is needed before performing the recognition steps.

In fingerprint recognition system, complexities of curvature, pattern recognition, etc often

become very difficult to handle.

Figure 185: Fingerprint Recognition steps.

VII.17.5. Issues Related With Fingerprint Recognition System

Recognition of fingerprints faces lots of hurdles, let’s list few of them.

Recognition is susceptible to poor image quality, with associated failure to enroll rates.

There have been many algorithms developed for extraction of local and global structures.

Most algorithms found in the literature are somewhat difficult to implement and use a

rather heuristic approach.

There are many different approaches towards fingerprint recognition experimented, but in

all cases some basic problems during collecting the fingerprint remain. Basically

maximum problems arise at the time of scanning the fingerprint, and that is the starting

phase of series of problems.

Matching is a key operation of most of the current fingerprint identification systems. One

of the most important objectives of fingerprint systems is to achieve a high reliability in

comparing the input pattern with respect to the database pattern. Reliably matching

fingerprint images is an extremely difficult problem, mainly due to the large variability in

different impressions of the same finger (i.e., large intra-class variations). The main


____________________________________________________________________________________

- 251 -

factors responsible for the intra-class variations are: displacement, rotation, partial

overlap, non-linear distortion, variable pressure, changing skin condition, noise, and

feature extraction errors. So fingerprints from the same finger may sometimes look quite

different whereas fingerprints from different fingers may appear quite similar.

Figure 186: Fingerprinting System.

Scanned fingerprints are subject to distortions that must also be taken into account;

including rotation, translation, non-linear scaling and extraneous or missing minutiae (the

most widely used fingerprint feature for recognition, the discontinuity of fingerprint

ridges) between matching fingerprints. This creates difficulty in the matching phase

because it causes the minutiae to differ between two identical fingerprints.

(a) (b)

Figure 187: Minutia alignment:

(a) Identical matches of minutiae coordinates rarely match perfectly.

(b) Sources of error in fingerprint recognition.

Most of the fingerprint recognition systems rely on minutiae matching algorithms.

Although minutiae based techniques are widely used, because of their temporal

performances they do not perform so well on low quality images and in the case of partial

fingerprint they might not be used at all. Therefore, when comparing partial input

fingerprints to pre-stored templates, a different approach is needed.


____________________________________________________________________________________

- 252 -

The performance of a fingerprint recognition system basically depends upon the quality

of the input image. Since the images acquired with different kinds of sensors are not of

the perfect quality they can’t be used directly for the matching. Therefore to ensure the accurate working of the system the image is first enhanced.

Figure 188: Original Image Enhanced Image.

During pre-processing, false minutiae like spikes, holes, bridges, ladder structures, and

spurs are included in the image. False minutiae structures are introduced to the fingerprint

image after thinning the original image. The presence of undesired spikes and breaks

present in a thinned ridge map may lead to many spurious minutiae being detected. Also

we need to note that a false minutia is more affecting than missing minutiae. Spur

removal helps to remove bifurcations and ends caused by thinning. If these points were

not removed, they would result in false endpoints and false bifurcations. A compromise

during spur removal must be met. Although it will remove some noise from the thinned

image, it will also move endpoints from their ‘real’ locations. This illustrates the

dependency of filtering on the quality of the image. A quality image will require few or

no spur removal cycles. When eroding spurs, normal endpoints are also eroded. This

results in small variations of the location of endpoints from their real locations.

Figure 189: False Minutiae.

False minutia structures, which may be encountered into a thinned fingerprint generates

two or more false minutiae.

o The spike structure generates two false minutiae and may occur when thinning a non-

smooth ridge.


____________________________________________________________________________________

- 253 -

o The bridge and ladder structures usually occur between close ridges.

o Very wide ridges may generate hole structures and very wide valleys may generate

spurs.

o The presence of scars in the fingerprint may determine ridge breaks in the thinned

ridge map image.

o In addition a large number of false minutiae are always detected close to the boundary

of the region of interest (boundary effect). The boundary effect is treated by

cancelling all minutiae which are below a certain distance to the boundary of the

fingerprint pattern.

It is of importance to mention that false minutia structures may also be caused by

different image processing operations used to obtain the thinned ridge map image from

the original gray-scale fingerprint image.

Figure 190: Types of false minutiae structures.

From left to right and up to bottom we have: spike, bridge, hole, break, spur, and ladder structure.

Classic pattern-matching methods may achieve very high accuracy in false acceptance

rate (FAR) and false rejection rate (FRR) but are still not good enough to attain a 100%

accurate acceptance.

A fingerprint can have up to 80 minutiae or more. It is generally accepted as the same

print if 8 to 17 points match. Some translation of the fingerprint will be acceptable,

however rotation must be minimized since no techniques have been implemented (with

much success) which specifically counteracts rotation.

The first step in an identification system is often continuous classification of fingerprints.

This reduces the partition of the database to be searched for matches. To facilitate high-

performance classification, algorithms for accurate singular-point estimation are needed.

Singular point detection is a critical process for both fingerprint matching and fingerprint

classification. The process of singular points detection must be fast and robust; otherwise,


____________________________________________________________________________________

- 254 -

the performance of the whole fingerprint recognition system would be influenced heavily.

Using high-level classification process can efficiently reduce the search area in large

fingerprint databases and therefore speeds up the subsequent matching algorithm. There

are four main approaches to allocate SPs [129]:

1. Methods based on mathematical model representation of fingerprint: Because of the

complexity of fingerprint patterns, representation of an accurate model for these

images is a difficult task [24], [145].

2. Methods based on statistical approaches: In these methods although the usage of

histogram has reduced the noise effect, however they cannot adapt themselves to

different image characteristics [38].

3. Methods based on different frequency transforms like Fourier Transform, Gabor

Transform: These methods are not efficient enough because of working in the

frequency domain. But some have claimed to obtain fair results [59]. Disadvantages

of this method often occur with the image size, which has to be of 2N. Some

fingerprint alignment problems remain in these types of recognition.

4. Methods based on fingerprint structures: These are usually well applied approaches,

which have been tested successfully on large databases [35], [60], [61].

Some approaches combine several types of the above mentioned methods and make a

new combined system [62]. But singular points detection remains the one of the most

challenging and important processes in biometrics fingerprint verification and

identification systems.

Figure 191: Fingerprint Classification System.


____________________________________________________________________________________

- 255 -

All fingerprint images in database needed to be classified according to the pre-defined

classification criteria, in order to overcome the accuracy and the identification speed

problems. A number of approaches have been applied for several years that differ in the

features used to describe the importance of classifying fingerprint image. However, a

potential way to improve the algorithms especially on pre-processing steps is still needed

to be improved. From the study, it was found that though there are various techniques and

features that have been used to classify a fingerprint image, there are still open research

opportunities in this field that is related to the performance of a system that rejects a very

high percentage of its input. The future study of fingerprint classification systems might

be the use of combinations of features. The best features for classification are claimed to

be fingerprint singularities. However, this method can be difficult when extracting core

and delta points due to noisy images.

The singular point detection method can be applied as a step to cluster the fingerprint

images into five major groups (i.e. arch, tented-arch, left loop, right loop, whorl), and

then minutiae extraction based method can be applied on the clusters to achieve a

hierarchical fingerprint detection algorithm. Clustering the fingerprint images in five

major groups is quite easy if it is done manually by visual checking, but implementing an

automated system for this purpose is quite a hard job.

Based on the features that the matching algorithms use, fingerprint matching can be

classified into image-based and graph-based matching. Image-based matching [122] uses

the entire gray scale fingerprint image as a template to match against input fingerprint

images. The primary shortcoming of this method is that matching may be seriously

affected by some factors such as contrast variation, image quality variation, and

distortion, which are inherent properties of fingerprint images. The reason for such

limitation lies in the fact that gray scale values of a fingerprint image are not stable

features. Graph-based matching [1], [63] represents the minutiae in the form of graphs.

The high computational complexity of graph matching hinders its implementation.

Among all the fingerprint features, pores have most extensively been studied, but are

considered to be reliably available only at a resolution higher than 500 dpi.

A minutia matching essentially consists of finding the alignment between the template

and the input minutiae sets that results in the maximum number of minutiae pairings. In

minutiae based matching the similarity between the input and stored template are

computed. The implementation of a viable technique is quite difficult. Let’s consider the ideal. The coordinates of both samples are identical. A simple coordinate matching

algorithm would suffice. Unfortunately, this ideal situation rarely occurs. When a second

print is recorded from the same finger it is always misaligned from the original. Skin has

an elastic nature, it means that some features may be stretched or warped, relative to other

sections of the print which retain their dimensions. Noise will most likely occur, caused

by applying too much pressure or smudging the print. A small amount of noise can make

the ridge structure different, affecting the whole recognition system. Slight rotation of the

finger will also cause some features to vary from the original sample. The searching

algorithm must be flexible enough to allow some variance in coordinate position. It must


____________________________________________________________________________________

- 256 -

also attempt to distinguish the difference between real and false matches. The main

problem of the matching algorithm is false matches.

Generally, an automatic fingerprint verification/identification is achieved with point

pattern matching (minutiae matching) instead of a pixel-wise matching or a ridge pattern

matching of fingerprint images. A number of point pattern matching algorithms have

been proposed in the literature [123], [162], [164], [165]. The relaxation approach [123]

iteratively adjusts the confidence level of each corresponding pair based on its

consistency with other pairs until a certain criterion is satisfied. Although a number of

modified versions of this algorithm have been proposed to reduce the matching

complexity [164], these algorithms are inherently slow because of their iterative nature.


There are different social issues associated with fingerprinting system, other than that there

are many technical shortcomings.




There are types of degradations that affect the quality of the fingerprint image. The ridges get

some gaps; parallel ridges get connected due to noise; natural effects to the fingers like cuts,

wrinkles and injuries; do make changes in fingerprints. High quality fingerprint image is very

important for fingerprint verification or identification to work properly. In real life, the

quality of the fingerprint image is affected by noise; like smudgy area created by over-inked

area, breaks in ridges created by under-inked area, changing the positional characteristics of

fingerprint features due to skin (resilient in nature), dry skin leading to fragmented and low

contrast ridges, ridge discontinuities caused by wounds, and sweat on fingerprints also leads

to smudge marks and connects parallel ridges.

The (most widely used) minutiae based matching is highly sensible, as, if the finger is moved

even a little bit that gives us a different set of minutiae.

Noise is the random variation of brightness or colour information in images produced by the

sensor and circuitry of a scanner or digital camera. Due to the presence of noise, as well as

the use of different pre-processing steps, the thinned binary image contains a large number of

false minutiae which may highly decrease the matching performances of the system.

Ideas for future study might give better insights on how we can improve the quality of

fingerprint images in pre-processing step. A system must be designed to be robust when

dealing with the quality of fingerprint images and it will give better performance in

fingerprint classification and matching system.

vii. fingerprint recognitionshodhganga.inflibnet.ac.in/bitstream/10603/99462/14/14_chapter...

Documents