fingerprint enhancement frequency domaingovind/cse666/fall2007/sharat... · 2007. 10. 29. · gabor...
TRANSCRIPT
FINGERPRINT ENHANCEMENTFrequency Domain
Outline
Fingerprint Image Enhancement Prior Related WorkProposed Algorithm: STFT AnalysisExperimental Evaluation
Need for Enhancement
What you see What you ‘think’ you see
Reality: What you usually get..
High contrast print Typical dry print
Low contrast print Typical Wet Print Creases
Faint print
Prior Related Work
ChallengesFingerprint image is non stationary (has dominant local orientation and frequency)General purpose image processing algorithms are not usefulTraditional operators and filters assume gaussian noise model‘Noise’ in fingerprint images consists mostly of ridge breaks
Contextual FiltersExisting techniques are based on ‘contextual’ filteringFilter parameters are adapted to each local neighborhoodFilters themselves may be spatial or Fourier domain basedFilter parameters in ‘unrecoverable’ regions can be interpolated based on its neighbors
Spatial Filtering(Yang et.al 1996, Greenberg et. Al 1999) proposed local anisotropic filteringFilter kernel adapts at each pixel location
( ) ( )⎭⎬⎫
⎩⎨⎧ −
+⋅−
−+=⊥
⊥
)().(
)()(exp)(),(
02
20
02
20
00 xnxx
xnxxxxVSxxf
σσρ
Parameters: radial extent of the filter: vector parallel to the ridge direction ridge direction: vector perpendicular to the ridge direction, , : shape parameters
In our case, S= -2, V = 10 , = 4, = 2
ρn⊥n
ρ )( 02 xσ
)( 02 xσ )( 0
2 x⊥σ)( 0
2 x⊥σ
Spatial Filtering (cont.)
Even Symmetric Kernel
Fourier spectrum showing the localization
( ) ( )[ ]yxi
xyxx
eeyxG 002
20
2
20
21),( νξβσ
π
πσβ+⎥
⎥⎦
⎤
⎢⎢⎣
⎡ −+
−−
=
Hong et al, 96/98 proposed the use of Gabor filters for enhancementGabor filter has the best joint space-frequency localizationThe filter is aligned with the direction of the ridgesDoes not handle high curvature regions well due to block wise approach. Angular and radial bandwidths are constant.
Fourier Domain FilteringSherlock et al 94, proposed the use of Fourier domain filtering The image is convolved with a filter bank of directionally selective filtersImage enhanced by selecting a linear combination of filter responsesHas high space complexity, requires estimation of core/delta locations
Watson et al. 94, proposed the use or ‘root filtering’ for enhancement.(Pseudo matched filter)Does not require the computation of orientation images
{ }{ }),(),(
),(),(),( 1
yxIFFTvuFvuFvuFFFTyxI k
enh
=
= −
Root Filtering
Fourier Domain Filtering
Traditional Approaches
Local Ridge SpacingF(x,y)
Projection Based Method
EnhancementFrequency/Spatial
Local Orientationθ(x,y)
Gradient Method 87] Witkinsand [Kaas
tan21
),(),(
),(),(2
1
22
⎟⎟⎠
⎞⎜⎜⎝
⎛=
−=
=
−
∈ ∈
∈ ∈
∑∑
∑∑
xx
xy
Wu Wvyxxx
Wu Wvyxxy
GG
vuGvuGG
vuGvuGG
θ
[Ratha et al 95]
Proposed Approach: Overview
STFT Analysis
Frequency Image
RegionMask
Orientation Image
CoherenceImage
Fourier domain Enhancement
Surface Wave ModelFingerprint ridges can be modeled as an oriented wave
[ { } ]
=
=+=
),(
),(
)sin()cos(2cos),(
yx
yx
f
yxfAyxiθ
θθπLocal ridge orientation
Local ridge frequency
Local Neighborhoods
Validity of the model
Surface wave
STFT Analysis
Fingerprint image is non stationary, so we require both space and frequency resolution: time frequency analysisSTFT in 1D
STFT in 2D
dtetwtxX tjωτωτ −∞
∞−
∗∫ −= )()(),(
∫ ∫∞
∞−
∞
∞−
+−∗ −−= dxdyeyxwyxIX yxj )(212121
21),(),(),,,( ωωττωωττ
Parameter Estimation
Paradigm: The Fourier domain response can be viewed as a distribution of surface waves. Each term F(r, θ) corresponds to a surface wave of frequency 1/r and orientation θWe seek to find the most likely surface wave and hence estimate the dominant direction and frequencyWe can represent the Fourier spectrum in polar form as F(r,θ) The power spectrum is reduced to a joint probability density function using
The angular and frequency densities are given by marginal density functions
∫ ∫=
r
drdrF
rFrp
φ
φθ
θθ 2
2
),(
),(),(
∫=r
drrfp ),()( θθ ∫=θ
θθ drfrp ),()(
Ridge Orientation Image
⎟⎟⎟⎟
⎠
⎞
⎜⎜⎜⎜
⎝
⎛
⋅
⋅
=Ε∫
∫−
φ
φ
θθθ
θθθθ
dp
dp
)()2cos(
)()2sin(tan
21}{ 1
( )( ) ⎟⎟
⎠
⎞⎜⎜⎝
⎛∗∗
= −
),(),(2cos),(),(2sintan
21),(' 1
yxWyxOyxWyxOyxO
Region Mask
⎟⎟⎠
⎞⎜⎜⎝
⎛= ∫∫ θθ drdrFE
2),(log
The surface wave approximation does not hold in the background regionThe region mask is obtained by simple thresholding of the block energy image
Coherence Image
• Block processing is unreliable in regions of high curvature• Sherlock and Monro 94, relax filter parameters near the singular locations• Estimation of singular point is difficult in poor images!• We use an angular coherence measure proposed by Rao and Jain 90
WW
yxOyxOyxC Wji
×
−=∑∈),(
00
00
)),(cos()),(cos(),(
Frequency Image
∫ ⋅=Εr
drrprr )(}{∑ ∑
∑ ∑+
−=
+
−=
+
−=
+
−== 1
1
1
1
1
1
1
1
),(),(
),(),(),(),(' x
xu
y
yv
x
xu
y
yv
vuIvuW
vuIvuWvuFyxF
[Jain et al 00]
Enhancement
1994] al, et.[Sherlock frequency mean :}{r n,orientatiomean :}{bandwidth,angular :,bandwidth radial:
otherwise 0
if 2
)(cos)( )()(
)()(
)()(),(
c
2
2222
2
rEEr
Hrrrr
rrrH
HrHrH
c
BWBW
BWcBW
C
nc
nBW
nBW
r
r
==
⎪⎩
⎪⎨⎧ ≤−
−=
−+=
⋅=
φφφ
φφφφφφπ
φ
φφ
φ
φ
Additional Enhancement Results
Qualitative Comparison
Original Image Root Filtering
Qualitative Comparison(cont.)
Gabor Filter based Enhancement Proposed Approach
Objective Evaluation
• We evaluated the effect of enhancement on 800 images from FVC2002 DB3• The evaluation consists of 2800 genuine test and 4950 impostor tests• It can be seen that the matcher performance improves with enhancement
Feature Extraction
Outline
Minutia Feature ExtractionPrior Related WorkChain code contourExperimental Evaluation
Background
Minutiae represent local discontinuities in ridge flowMinutiae features are the most widely used fingerprint representationThere are several standards such as CBEFF (file format) and ANSI-NIST (interchange format) standards for minutiae based fingerprint representation
Minutiae extraction approaches may be broadly categorized intoBinarization based approachesDirect gray scale extraction
Prior Related Work
Binarization ApproachesMINDTCT,NIST NFIS, (Garris et. Al, 02)
Directionally adaptive binarizationTemplate matching is used to detect minutiaeGreedy approach to minutia detection leads to false positives.Extensive post processing is required to eliminate false positives
Adaptive Flow Orientation technique (Ratha et. al., 95)Binarization is performed by peak detection Peak detection leads to false positives in regions of poor ridge constrastThinning and morphological post processing shift minutia location.
Peak Detection (Domeniconi 98)Fingerprints is treated as a 3D surfaceRidges are detected as peaks and saddle points on this surfaceUses Hessian matrix of the gradients to determine dominant directionsCannot compensate for ridge breaks
Direct Gray Scale Ridge Following Ridge Following (Maio and Maltoni 97, Jiang and Yau 01)
Based on ridge pursuitHas low computational complexity.Cannot handle poor contrast prints and images with poor ridge structure. Relies on a good orientation map for ridge pursuit
Binarization Method
Binarization Thinning Minutia DetectionAcquisition
Proposed Approach: Chain Code Contours
Provides a lossless description of the contour and also gives direction and curvature information.Translation and rotation invariantUsed in computer vision for encoding object boundariesUsed for character recognition (Madhavanth et. al 99)
Minutiae Detection using Chain Codes
Minutiae are encountered as points of ‘significant’ turn on the contourLeft turn: Ridge endingRight turn: Bifurcation ( )
( ) 0 : Left turn
0:Right turn
<×
>×
OUTIN
OUTIN
PPsign
PPsignrr
rr
Determining Turn Points
Th
OUTIN
OUTIN
PPPP
θθ
θ
>
•=
:t turnSignifican
)cos( rr
rr
⎟⎟⎠
⎞⎜⎜⎝
⎛=
+−=
−
X
Y
OUTIN
PP
PPP
1tan
2
φ
rrr
Post processing
• Feature Extraction errors• Missing minutiae • Spurious minutiae
• Spurious minutia can be removed using post processing• Heuristic rules:1. Merge minutiae that are a certain distance of each other and have similar angles2. Discard minutiae whose angles are inconsistent with ridge direction3. Discard all border minutia4. Discard opposing minutiae within certain distance of each other
Results
Results (cont.)
Experimental Evaluation
Test Data150 prints from FVC2002(DB1) were randomly selected for evaluation.Ground truth was established using a semi automated truthing tool. Results compared using NIST NFIS open source software.
MetricsProposed by Sherlock et. Al 94Sensitivity: Ability of the algorithm to detect true minutiaeSpecificity : Ability of the algorithm to avoid false positivesFlipped : Minutiae whose type has been exchanged
Truth Ground:N Exchanged, :E positives, False :FP Negatives, False :FN
,1,1NEFlipped
NFPySpecificit
NFNySensitivit =−=−=
Quantitative Analysis : Results
Examples
File Name NIST Proposed method
Actual TP FP M F TP FP M F
10_8.tif 18 16 8 2 1 17 0 1 1
11_6.tif 50 40 4 10 2 41 4 9 4
12_8.tif 29 22 5 7 3 22 3 7 1
13_6.tif 35 28 10 7 4 28 10 7 2
14_6.tif 44 34 12 10 6 37 13 7 5
15_7.tif 38 37 7 1 5 37 3 1 0
16_7.tif 41 35 12 6 5 36 8 5 8
17_6.tif 43 35 16 8 11 36 7 8 11
18_8.tif 34 31 7 3 4 32 6 2 1
19_7.tif 35 26 8 9 3 31 6 4 5
Results
Summary resultsCount TP(ANSI) > proposed : 40 of 150Count E(ANSI) < proposed : 40 of 150
Metric NIST Proposed
Sensitivity(%) 82.8 83.5
Specificity(%) 77.2 76.8
Flipped(%) 12.0 10.9
Sensitivity distribution Overall statistics
FINGERPRINT MATCHING
Outline
Fingerprint Image Enhancement Minutiae ExtractionMatching Algorithm
Prior Related WorkNew Representation: K-pletLocal Matching: Dynamic ProgrammingConsolidation: Coupled BFSExperimental Evaluation
Minutiae Based Matching
ChallengesMinutiae extraction is error prone is low quality imagesNot robust to non-linear distortion.Intra-user variation
Challenges: Non-linear Distortion
Challenges: Quality and Intra-user variance
Variation in quality
Intra-user variation
Prior Related Work
Minutiae based matching algorithms can be broadly categorized as
Global Matching : All points are aligned at onceImplicit Alignment: Alignment and matching done simultaneouslyExplicit Alignment: Has a separate pre-alignment stage
Local Matching: Only local neighborhoods are matchedFeatures are chosen to be rotation and translation invariantMore resilient to non-linear deformationChallenges in local matching approaches
Representation Local Matching Consolidation
Prior Related Work (cont.)
Global MatchingPoint correspondences not known : combinatorial problem
Relaxation approach (Ranade and Rosenfield 93)Likelihood of each match is either decreased or increased at each iteration based on compatibility of rest of the pointsIterative approach makes it too slow to be practical
Generalized Hough Transform (Ratha et al. 96)All possible transformation represented as a quantized search spaceSearches for the most optimal transform in the search spaceVery fast
Ridge Alignment (Jain et al. 97)Performs explicit alignment before matchingEach minutiae is associated with its ridge (represented by a curve)The alignment is based on ridge correspondenceGlobal matching is then performed using string edit distance
Deformation ModelsRigid Transformation
Transformation parameters:
Affine Transformation
Non-rigid TransformationBazen and Garez (2002) used TPS to find arbitrary transformation. Point correspondences(pre-alignment) is still required
⎥⎦
⎤⎢⎣
⎡ΔΔ
+⎥⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡ −=⎥
⎦
⎤⎢⎣
⎡yx
yx
yx
)cos()sin()sin()cos(
''
θθθθ
[ ]θ,, yx ΔΔ
⎥⎦
⎤⎢⎣
⎡ΔΔ
+⎥⎦
⎤⎢⎣
⎡⎥⎦
⎤⎢⎣
⎡=⎥
⎦
⎤⎢⎣
⎡yx
yx
aaaa
yx
2221
1211
''
Local Matching
(Jiang and Yau 00)11 dimensional local features derived from reference minutiae and two closest neighborsBest match is used only for explicit alignment
(Jea and Govindaraju 04)5 dimesional features Si (ri0, ri1, φi0, φi1, δi) derived from two closest neighborsAlignment is still required
(Ratha et al. 00)‘Star’ representation derived from all minutiae within a particular radiusConsolidation by checking consistency
(Garris et. al 03: BOZORTH3)Line featuresConsolidation by linking consisting matches
iM
1N
0N
iδ
0iϕ
1iϕ
0ir
1ir
Proposed Algorithm
Representation K-PletFeatures invariant to rotation and translationLocal relationship formally represented by a directed graph
Local MatchingPosed as a string alignment problem and solved by dynamic programmingMatches all neighbors simultaneously
ConsolidationCoupled Breadth First SearchBreadth first search is used to propagate the matchesSimilar to human verificationNo explicit alignment required at any stage
Neighborhood Representation: K-plet
K-plet
r
Θ
Φ
The ‘Graphical’ View
Local Matching
All local neighbors have to be matched simultaneously. Greedy approach does not work when conflicts occurThese can solved by finding the alignment through optimization process such as by solving a string alignment problem
1Tm
2Tm
1Im
2Im
r
Example of alignment: S (acbcdb) – (ac__bcdb)T (cadbd) - (_cadb_d_)
Trivial solution requires exponential timeEach match is given a cost. Alignment solved through recurrence relation
⎪⎩
⎪⎨
⎧
⊗+−⊗+−
+−−=
])[,(]1,[)],[(],1[
])[],[(]1,1[max],[
jtjiDisjiD
jtisjiDjiD
σσσ
{ }( )
0])[,(,0)],[(otherwise
bounds within if ])[],[()(
=⊗=⊗⎩⎨⎧
=++−
jtisMISMATCHeMATCHjtis
dCdrCdC r
σσ
σφθ φθ
Graphical Matching: Coupled BFS
Coupled BFS
Graphical Matching: Coupled BFS
Coupled BFS
Graphical Matching: Coupled BFS
Graphical Matching: Coupled BFS
Graphical Matching: Coupled BFS
Experimental Evaluation800 prints from FVC2002(DB1)2800 genuine tests,4950 impostor testsCompared with BOZORTH 3
Error RatesBOZORTH3: 3.6% EER, 5.0% FMR100Proposed: 1.5% EER, 1.65% FMR100
Matching Algorithm 2
http://www.cubs.buffalo.edu
Correspondence Problem
1Tm
2Tm
1Im
2Im
r
a) Non-linear distortion
b) Non-trivial correspondence
c) Global consistency
d) Dynamic thresholds
e) No singular feature registration
Brute-force MatchingChoose any pair of minutiae x and y from the images Q and R(Assume they are aligned)Convert minutiae in Q into polar coords w.r.t. to xConvert minutiae in R into polar coords w.r.t. to y
Compute COST matrix of minutiae pairs (A,B)
COST (rx, ry, φx, φy, δx, δy)O(N2) (Worst case)
Use the MCF to find best correspondence and total costO(nm2) == O(n3) (m = no. of edges; m ~ n)
Repeat for all pair alignments x and y
Choose the alignment <x,y> and correspondence with the lowest cos
Complexity: O(N2).(N2 + N3) = O(N5)
r0
r1
φ0
φ1
δ0δ1
A
BO
[Capacity constraint]
[Skew symmetry]
[Flow conservation]
Minimum Cost Flow (MCF)
s t
1
2
4
3 5
16 (5.8)
13 (4.0)
5 (4.6)
10 (6.0)
8 (3.3)
5 (5.8)15 (4.6)
25 (7.2)
6 (6.8)
10 (4.5)
),(),(,, vuwvufVvu <=∈∀),(),(,, vufuvfVvu −=∈∀
∑∈
=−∈∀Vv
vuftsVu 0),(},,{
Ford-Fulkerson algorithm [1962] O(nm2)Cycle-canceling algorithm [Orlin et al., 2000] O(m (m + n log n) log (nU))
U is the maximum capacity of links (1 in our case)n = number of nodes; m = number of edges
s t
Nq nodes from Q Nr nodes from R
1 (cij)
1 (0) 1 (0)
MCF and Fingerprint Matching
Links with cost less than a threshold are added to network.
Output:
Best correspondence
And its cost
Input:
COST matrix
Similarity Scores
Only number of matched minutiae (n) is not reliable
Use NN to combine the following
Number of matched minutiae (n)Numbers of features in overlapping parts of the convex hulls formed by the matched minutiae: Oq, Or
Cost from MCF
Traditionally: Alternate:rqNN
n2
rq NNn+2
15.88
1.96
1.58
Min.TER
N/A
2.49
3.32
Min.TER
N/A
N/A
1.88
Min.TER
19.96
5.46
6.55
Min.TER
9.28N/AN/A11.11FVC2002 DB3
1.161.57N/A3.37FVC2002 DB2
1.062.131.014.67FVC2002 DB1
EEREEREEREERDatabases
K-pletsmatch
Genuine test: 138ms
Imposter test: 136ms
Triplet match
Genuine test: 179ms
Imposter test: 7ms
Brute-Force+MCF
Bozorth3
Experimental Results
Adaptive Matching SystemMinimize Alignment Step
Matching path adapts to number of minutiae features
No
Nr > αNq > α
Triangle feature matching
Template database
Result
Yes
Yes
Yes
No
No
Triangle feature matching
Has a match?
Brute-force matching
Nq < αNr <α
Query fingerprint
Minutiae extraction
Nq Nr
Triangle Feature Matching
Secondary FeaturesTriplets of MinutiaeS (r0, r1, φ0, φ1, δ0, δ1)
Derived only form minutiae
No ridge countNo minutiae type
Purely localized featureDoes not rely on global landmark
Transformation invariant
No pre-alignment needed
(a) (b)
0δ
1δ
0ϕ
1ϕ0r
1r
0M
1MCM
Triangle Feature MatchingSingle Alignment Step
Generate COST matrix of triangles in Q and R O(n2)Find correspondence of triangles and cost using MCF O(n3)Remove globally inconsistent matchesIdentify the max bin in histogram (10o)Average the orientation differences of triangle pairs from the max bin and neighbors (10o, 20o, 350o) = 0.78o
Same could be done for translation
Move back to minutiae from trianglesConvert ALL minutiae to polar coordsusing best matched triangle McShift all orientations by average orientation Compute the COST matrix O(n2)Use MCF to obtain correspondence and cost O(n3)Complexity is O(n3) + O(n2) + O(n3) + O(n2) = O(n3)
False Matched TrianglesGood Matched TrianglesTriangle: central minutiae Mc
OD Distrubtion
0
2
4
6
8
10
12
0 30 60 90 120
150
180
210
240
270
300
330
Degree
OD=0.7865°
15.88
1.96
1.58
Min.TER
N/A
2.49
3.32
Min.TER
N/A
N/A
1.88
Min.TER
19.96
5.46
6.55
Min.TER
9.28N/AN/A11.11FVC2002 DB3
1.161.57N/A3.37FVC2002 DB2
1.062.131.014.67FVC2002 DB1
EEREEREEREERDatabases
K-pletsmatch
Genuine test: 138ms
Imposter test: 136ms
Triplet match
Genuine test: 179ms
Imposter test: 7ms
Brute-ForceBozorth3
Experimental Results
Experimental Results (Partial Fingerprints)Adaptive Matching System
0
10
20
30
40
50
60
70
10 20 30 40 50 60 70 80 90
Image size (%)
Err
or ra
te (%
)
TER EER
FVC2002 DB1:
110 fingers – 8 impressions
(2 normal)
1 x 5 % area = 550 samples
FRR rate
7 x 550 samples tested for genuine matches
Use 1,3-8 impressions are ref
Use 2nd impression for query
FAR rate
Use any 2 of 5 samples as query
Use all 1st impressions as ref
Note: values are taken from individualROCs
Improve Robustness of triangle featuresUse notion of clusters of minutiaeImprove speed and global consistency by indexing reference templates
K-plet Matching with Secondary Features
MOD
Sensitivity of Triangle Features
(a) (b)
(c) (d)
O
A
BC
OB
A
C
O
A
BC x
O
A
C
(a) (b) (c)
O
A
BC
O
A
BC x
O
A
C
Sensitivity of Triangles to Minutiae Extraction
Associate K triangles
Verification in clustersAnalysts’ Examination Method
Additional K Triangle Features
S= (r0, r1, φ0, φ1, δ0, δ1)
r0
r1
φ0
φ1
δ0
δ1
A
BC
D
O
SOABSOCASODASOBCSODBSOCD
features.secondary Max. 2KC
Remove secondary features that are too wide (close to 180°) or too narrow (close to 0°)
K is decided by the size of the fingerprint
If Minutiae > 30Set k as 6
If Minutiae < 20Set k as 10
If Minutiae in [20, 30]Set k as 7
Triangle IndexingNeighboring minutia labeled with two quadrant tags (bins) clockwise.
Total bins is 32 (4 x 8)
SOAB has index labelQ0Q2, Q1Q2, Q0Q3, and Q1Q3
Each triangle in reference is in 2-4 binsEach query triangle can have 1 bin
92% reduction in matches (580-47)/580 8% of edges in COST matrix input
Avg. Triangles on a fingerprint
580
Avg. Triangles in a bin 47
Min. Triangles in a bin 4
Max. Triangles in a bin 102
A
B
O
Q0Q1
Q2
Q3Q4
Q5
Q6
Q7
si <•••>
s <•••>s <•••>
s <•••>s <•••>s <•••>s <•••>
Label: Q0Q1
s <•••>s <•••>
Label: Q3Q6
s <•••>s <•••>
s <•••>s <•••>
Label: Q7Q2
s <•••>s <•••>s <•••>
Query fingerprint
Reference fingerprint
Indexing bins
with label Q3Q6
Triangle Matching
Coupled K-plet AlgorithmFind lists of matched triangles
Construct the cost matrix using the binning scheme (8% of edges)Apply MCF to obtain matched triangles: Lr and Lq.
Identify clusters Form cliques within Lr and Lq
BFS: Start with any pair of corresponding mr and mqRemove mr and mq from Lr and LqBreadth First Extended MatchAdd matched mr and mq into matched queues Qr and Qq
Find the no. of matched minutiae (score) in the BFS sequence
If there are unvisited minutiae in lists LRepeat the BFS procedure and find new score
Return with best sequence in terms of number i ti
Link Clusters
B
Cb
c
Aa0 1
2
345
67
89
1011
12
02
3
1
4
678
9
1011
12
II RR
Link the seedsTo form Cliques
Link the seedsTo form Cliques
Let k =4
Lq [A,B,C]
Returned by MCF
A(0,1,2,4)
B(12,11,9,10)
C( …)
4(5,6,7,8)
Coupled Breadth-First Matching
B
Cb
c
Aa0 1
2
345
67
89
1011
12
02
3
1
4
678
9
1011
12
II RR
B
Cb
c
Aa0 1
2
345
67
8 910
11
12
02
3
1
46
78
9 1011
12
II RR
A a A0124 a0124
Coupled Breadth-First Search Matching (2)
B
Cb
c
Aa0 1
2
345
67
8 910
11
12
02
3
1
46
78
9 1011
12
II RR
B
Cb
c
Aa0 1
2
345
67
8 910
11
12
02
3
1
46
78
9 1011
12
II RR
A,0,1,2,4,B A,0,1,2,4,b A,0,1,2,4,B,7,8,6,12,11,9,10
a,0,1,2,4,b,7,8,6,12,11,9,10
Extended Matching Result
B
Cb
c
Aa0 1
2
3456
78 9
1011
12
02
3
1
46
78
9 1011
12
II RR
B
Cb
c
Aa0 1
2
3456
78 9
1011
12
02
3
1
46
789
101112
II RR
Without Seed Linking Seed Linking
A,0,1,2,4,B,7,8,6,3,12,11,9,10
14 matched minutiae
A,0,1,2,4,7,8,6,3
9 matched minutiae
15.88
1.96
1.58
Min.TER
N/A
2.49
3.32
Min.TER
N/A
N/A
1.88
Min.TER
19.96
5.46
6.55
Min.TER
9.28N/AN/A11.11FVC2002 DB3
1.161.57N/A3.37FVC2002 DB2
1.062.131.014.67FVC2002 DB1
EEREEREEREERDatabases
K-pletsmatch
Genuine test: 138ms
Imposter test: 136ms
Triplet match
Genuine test: 179ms
Imposter test: 7ms
Brute-ForceBozorth3
Experimental Results
Experimental Results
10−4
10−3
10−2
10−1
100
10−0.04
10−0.03
10−0.02
10−0.01
100
False accept rate
Gen
uine
acc
ept r
ate
ROC
Bozorth3ProposedMethod
10−4
10−3
10−2
10−1
100
10−0.04
10−0.03
10−0.02
10−0.01
100
False accept rate
Gen
uine
acc
ept r
ate
ROC
Bozorth3ProposedMethod
FVC2002 DB1 FVC2002 DB2
5.163.167.934.7211.778.28211.03146.4150
10.176.1611.176.2619.6912.35188.70130.9040
16.209.7917.199.5533.1825.63163.36113.3130
30.9220.9728.0216.1268.4043.90133.3092.4220
Min. TER(%)
EER(%)Min. TER(%)
EER(%)Min.TER(%)
EER(%)
Coupled K-plets MatchTriplet MatchBozroth3Avg. Width(pixels)
Avg. Height(pixels)
Sizes(%)
Partial Fingerprint Experimental Results
Conclusion
ContributionsNew Fingerprint Image Enhancement using STFT Analysis.
Simultaneously estimates all intrinsic imagesIncreases recognition rate of existing matchers
New Feature Extraction Algorithm using Chain code ContourObviates need for thinningPerforms favorably with NIST feature extractor
New Graph based matching algorithm Robust to non linear distortionFormal technique for propagating local matchesPerforms better than NIST BOZORTH3 matcher over FVC DB1 database