issn 1751-9659 optical coherence tomography used for...

/48

www.ietdl.org

Published in IET Image ProcessingReceived on 12th February 2007Revised on 8th June 2007doi: 10.1049/iet-ipr:20070021

ISSN 1751-9659

Optical coherence tomography used forsecurity and fingerprint-sensing applicationsS. Chang1 Y. Cheng2 K.V. Larin2 Y. Mao1 S. Sherif1

C. Flueraru11Optics Group, Institute for Microstructural Sciences, National Research Council Canada, Ottawa, Ontario, Canada K1A 0R62Biomedical Engineering Program, University of Houston, Houston, TX, USAE-mail: [email protected]

Abstract: Optical coherence tomography (OCT) is an emerging technology for high-resolution cross-sectionalimaging of three-dimensional structures. In the past, OCT systems have been used mainly for medicalapplications, especially ophthalmological diagnostics. As the OCT system is capable of exploring the internalfeatures of an object, the authors apply OCT technology to document security and fingerprint-basedbiometrics by directly retrieving the two-dimensional information form of a multiple-layer information carrierand internal human body objects. Since a typical depth-resolution of an OCT system is of micrometre scale,an information carrier having a volume of 20 mm � 20 mm � 2 mm could contain 200 mega-pixel images.On other hand, the technologies used in conventional biometrics can be easily fooled and tampered with byusing artificial dummies, because these ID features are extracted only from the surface of the skin. Hencethe use of OCT to explore the internal biometrics becomes increasingly important.

1 IntroductionTraditional biometric technologies used for security andperson identification essentially deal with fingerprints,hand geometry and face images. However, because allthese technologies use external features of humanbody, they can be easily fooled and tampered bydistorting, modifying or counterfeiting these features.Therefore to extract the internal features of humanbody that are unique to an individual is becoming anew and important trend for biometrics. Except thewell known technologies for iris and retinarecognition, other versatile technologies for internalbiometrics have been currently developed, such asvein scan technology that identifies a person from thepatterns of the blood vessels in the hand [1–3]; skinpattern recognition technology, which measures thecharacteristic of an individual’s skin [4] and fingernailbed identification that is based on the distinctgroove spatial distribution of the epidermal structuredirectly beneath the fingernail [5]. However, all these

IET Image Process., 2008, Vol. 2, No. 1, pp. 48–58doi: 10.1049/iet-ipr:20070021

technologies are based on two-dimensional (2D)surface scanning to extract specific biometriccharacteristics and, thus, could be easily spoofed bydifferent fraudulent methods. The technology that canexplore the internal unique features of the humanbody becomes very attractive, for example, the multi-spectral fingerprint detection [6]. The multi-spectralimaging device captures subsurface information aboutthe finger that makes it difficult to spoof. Althoughthis type of sensor is further designed to gainsubsurface images, it still falls into the category of 2Dimaging. Because this spectral information is the resultof the integration of all the subspectral-informationsemitted from different depths, it does not have thecapability of exploring the cross-sectional structures ofan object.

For object/document identification, the mostprevalent technology is the use of 1D or 2D physicallyimprinted patterns, such as barcode and passportcode. The visibility of the imprinted pattern is

& Canadian Crown copyright 2008

& C

www.ietdl.org

vulnerable to counterfeiting. From the viewpoint ofsecurity, multiple-layered imprinted patterns with tinysizes and invisibility to the human eye, will be of greatimportance for the security applications.

In summary, a technology that has the capability toprobe the internal structure of multilayered tissues orpre-encoded information carriers will be a newpowerful tool for security and document identity anddifferent biometric applications. Optical coherencetomography (OCT) could be such a technologicalsolution.

2 OCT systemsOCT is a recently developed optical imaging techniquethat permits high-resolution cross-sectional imaging ofan object. The first OCT system was reported byHuang et al. in 1991 [7], and since then OCTtechnology has been attracting attention fromresearchers all around the world. For the past decade,more than 1000 papers have been published by thescientific community, and new OCT imaging theoriesand applications have been continuously developed andreported. A good survey book and a review articleare provided in [8, 9].

Most OCT systems use point-scanning-basedtechnology, especially optical fibre-basedinterferometers. In time-domain OCT, depth scanningis achieved by the longitudinal translation of areference mirror, and a sectional image is obtained byusing scanning mirrors that laterally scan a focusedprobe beam on a test sample. Such three-axis scanningmakes the system relatively slow and cumbersome. Toincrease the acquisition speed and eliminate the needfor lateral scanning, parallel detection schemes havebeen investigated. Parallel OCT systems illuminate theentire 2D target and collect light from all pixelssimultaneously. These parallel OCT systems areusually called full-field OCT systems. A few OCTsystems working directly on 2D full-field imageswere recently reported [10–13]. The optical fibre-based OCT has high sensitivity with relatively lowframe rate, whereas the full-field OCT has fast imageacquisition ability with relatively low signal-to-noise ratio.

In the past decade, OCT systems have beendeveloped mainly for medical and biomedicalapplications, especially for ophthalmological,dermatology, dentistry and cardiology diagnostics [7–9]. These successful applications can be easilyextended to the internal biometrics, especially foriris/retina recognition, nailbed and skin patternidentification. Besides image recognition with the 2Dcross-sectional tomography, OCT can also performthree-dimensional (3D) shaping of the biometric

anadian Crown copyright 2008

object, say 3D profile of iris and fingerprint. Thesenew features make the biometrics system morediscriminability and more robust against fraudulency.To take the advantage of being capable to probe theinternal structure of an object, Chang et al. [13]developed a full-field OCT system for informationencoding and retrieving from a multiple-layerinformation carrier, info chip. Because OCT has theresolution of micrometre scale and the ability ofpeeling cross-sectional images from developed internalstructures of an object, it has potential applications indocuments security and object identification. As aninfo chip can be made by using non-scattering clearmaterials, the annoying signal/noise issue will nolonger be a major problem in this circumstance, sothat the specific hardware such as smart pixel device,lock-in detection apparatus and complicated softwaredesigned to deal with the scattering effects in digitalsignal processing can be greatly simplified. In addition,as it does not requires x–y axes scanning mechanism,the 2D parallel OCT could be the simplest and mostpractical imaging system for applications in multiple-layer information extraction.

The standard depth-resolution of an OCT system is atmicrometre level. If a 20 mm � 20 mm sampling areawith a 1024 � 1024 CCD array is used in the OCTsystem having 10 mm depth-resolution, an informationcarrier having a volume of 20 mm � 20 mm � 2 mmcould contain 200 mega-pixel images. Because of itstiny size and large information volume, theinformation carrier with its OCT retrieving systemwill have potential applications in documents securityand object identification. In addition, as both theinformation carrier and information coating can bemade by transparent material, the signal-to-noise ratiowill be improved dramatically.

As mentioned above, there are two basicconfigurations of OCT, which are illustrated inFigs. 1a and 1b. Both of these configurations areinterferometer-based systems, consisting of a lightsource, reference arm, sample arm and camera/photodetector. The light sources used in OCT systemsare broadband, low-coherence sources. The coherencelength of these light sources directly determinates thedepth-resolution of the OCT system. Fig. 1a is aMichelson free-space interferometer-based system usedfor full-field OCT. It is possible to produce largeworking area that is at centimetre level or higher.Fig. 1b shows a basic configuration of a time-domainfibre-optics-based OCT system. The optical source inthis system was a low-coherent superluminescent laserdiode with wavelength of 1310+ 15 nm. The imagescaptured were 450 � 450 pixels. In-depth scanningwas up to 2.2 mm (in air) whereas lateral scanningwas 2.4 mm.

IET Image Process., 2008, Vol. 2, No. 1, pp. 48–58/doi: 10.1049/iet-ipr:20070021

49

/50

www.ietdl.org

3 Multilayer informationextraction and documentidentificationBecause the OCT has micrometre-resolution and theability of peeling cross-sectional images from differentlayers of an object, this technology has promisingapplications in document security and objectidentification. The full-field OCT technology can beused to directly retrieve the 2D information prestoredin a multiple-layer information carrier, namely, info-chip, which has a tiny size and large informationvolume containing multilayer 2D data such astexts, images and so on. Because this informationcarrier can be made by low-scattering transparentmaterial, the signal-to-noise ratio will be improveddramatically.

OCT utilises a partial coherence light source andinterferometer to extract the cross-sectional images atdifferent depth. The interference image captured by acamera is given by

I(x, y) ¼ I0(x, y)þ Ai(x, y) sinf(x, y) (1)

Figure 1 Two types of OCT system

a Full-field OCTb Fibre-optics-based OCT


The above equation indicates that the image receivedby the camera, I(x, y), consists of three parts. One is thebackground image I0(x, y). Another part, Ai is thetomographic image of the ith layer and sinf itemrepresents the interference fringes. To extract thetomographic image hidden in background andinterference fringes, the traditional methods are basedon images obtained at different phase steps.

Chang et al. proposed an algorithm [13], in whichphase difference between two interference images, I1and I2, can be an arbitrary value w, and thetomography can be found by

Ai(x, y) ¼ ([I1(x, y)� I0(x, y)]2þ {½I2(x, y)� I0(x, y)]

� [I1(x, y)� I0(x, y)] cosw= sinw}2)1=2 (2)

However, (2) needs to know the phase-shift angle apriori. To solve the Ai(x, y) without knowing the valueof phase-shift, more phase-shift images have to bepresented. Two algorithms can be derived from wellknown four- or five-step phase-shift equations, whichare originally proposed to find the interference phasefunction f(x, y). The Carre equations [14, 15] offour-step phase-shift are given by

I1(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)� 3w]

I1(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)� w]

I1(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)þ w]

I1(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)þ 3w]

(3)

where w is a constant phase angle and 2w the phase shiftadded to each step. The tomographic image can besolved by

Ai(x, y)¼I1(x, y)� I4(x, y)þ I2(x, y)� I3(x, y)

8sinw cos2w

� �2�

þI1(x, y)þ I4(x, y)� I2(x, y)� I3(x, y)

cosw sin2w

� �2�1=2(4)

tgw¼3[I2(x, y)� I3(x, y)]� [I1(x, y)� I4(x, y)]

I1(x, y)� I4(x, y)þ I2(x, y)� I3(x, y)

� �1=2

(5)

Note when w ¼ n p/2, where n is an integer, a correcttable has to be used to avoid the presence of singularity[16].


& C

www.ietdl.org

As for the five-step method, Hariharan equations[14, 16]

I1(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)� 2w]

I2(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)� w]

I3(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)]

I4(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)þ w]

I5(x, y) ¼ I0(x, y)þ Ai(x, y) cos [f(x, y)þ 2w]

(6)

could be used to solve the tomography, which isexpressed as

Ai(x, y) ¼I2(x, y)� I4(x, y)

2 sinw

� �2�

þ2I3(x, y)� I5(x, y)� I1(x, y)

4 sin2 w

� �2�1=2(7)

where

tanw ¼2[I2(x, y)� I4(x, y)]

3I3(x, y)� I5(x, y)� I1(x, y)(8)

w can not be np.

Either a four- or five-step algorithm has to calculatethe angle of phase step for each pixel in the capturedinterference images. The computation is relativelyheavy, particularly when the image is large, sincealgorithms are pixel-based operations.

Chang et al. proposed a derivative-based algorithm,which does not need to handle the trigonometricfunctions. This algorithm is about three times fasterthan the traditional four- or five-step algorithms [17].In fact, (1) of an OCT interferometer is a function oftime variable when the interferometer scans axiallythrough a sample

I(x, y, t) ¼ I0(x, y)þ Ai(x, y, t) sin [vt

þ f(x, y)] (9)

where I0 (x, y) is a constant background image andv ¼ 2pf, f represents the central frequency of thelight source. When the light source is a low-coherence source, as is the case in most of the OCTdevices, (9) becomes

I(x, y, t) ¼ I0(x, y)þ Cr(x, y, t)Ai(x, y, t)

sin [vtþ f(x, y)] (10)

where Cr(x, y, t) is the envelop of autocorrelation of thelight source. As is well known, Cr(x, y, t) changes as a


Gaussian distribution with axis moving and determinesthe coherent length Dc, that is, the axis-resolution ofan OCT system. However, when a small shift d ismade, d � Dc, C(x, y, t) can be considered as 1 andthe tomography of the current layer Ai(x, y, t) can beconsidered as constant, that is, Ai(x, y, t) ¼ Ai(x, y).In this assumption, the first, second and thirdderivatives of I(x, y, t) with respect to scanning t canbe expressed by

I0t(x, y, t) ¼ vAi(x, y) cos [vtþ f(x, y)] (11)

I00t (x, y, t) ¼ �v2Ai(x, y) sin [vtþ f(x, y)] (12)

I000t (x, y, t) ¼ �v3Ai(x, y) cos [vtþ f(x, y)] (13)

respectively. Combining (11)–(13) by

I002t � I000t � I0t ¼ v4A2i (x, y) sin2 [vtþ f(x, y)]

þv4A2i (x, y) cos2 [vtþ f(x, y)]

¼ v4A2i (x, y)

we have

Ai(x, y) ¼ [(I002t � I000t � I0t)=v4]1=2 (14)

Because multiple phase-shifted images are taken in atime sequence, the derivatives have to beenexpressed by discrete differences of four sequentialimages I1(x, y) � I4(x, y). Thus, the first, second andthird difference of I(x, y, t) can be given by

I01 ¼ I2(x, y)� I1(x, y)

I001 ¼ I3(x, y)� 2I2(x, y)� I1(x, y)

I0001 ¼ I4(x, y)� 3I3(x, y)þ 3I2(x, y)� I1(x, y)

Therefore (14) becomes

Ai(x, y) ¼ C{[I3(x, y)� 2I2(x, y)� I1(x, y)]2

� [I4(x, y)� 3I3(x, y)þ 3I2(x, y)� I1(x, y)]

� [I2(x, y)� I1(x, y)]}1=2 (15)

where C ¼ 1/v4. Concerning that C is a constant andto avoid presence of complex value, a practicalalgorithm is proposed as

Ai(x, y) ¼ j[(I3(x, y)� 2I2(x, y� I1(x, y)]2

� {I4(x, y)� 3[I3(x, y)� I2(x, y)]� I1(x, y)}

� [I2(x, y)� I1(x, y)]j1=2 (16)


51

/52

www.ietdl.org

Hence the tomography of ith layer of the sample canbe solved without finding the phase-shift angle.

Full-field OCT system used in the experiments isshown in Fig. 1a, which is a Michelson interferometerequipped with a vision and motion controller. Thelight source is a superluminescent diode (SLD), whosecentral wavelength and bandwidth are 830 and 15 nm,respectively. Its coherence length, that is, depth-resolution of the OCT system, is 20 mm. The areacamera used in the experiment is a product ofPULNiX model TM-7CN. It has 768 � 494 pixelswith a sensing area of 6.41 mm � 4.89 mm. Eachpixel has 256 (8 bit) grey levels. The lateral-resolutionof the OCT system is determined basically by themagnification of the image lens and resolution of thecamera. As the pixel-width of the CCD sensor is8.3 mm and magnification of the imaging lens is 2�,the system lateral-resolution will be 4.15 mm. Thenumerical aperture of the imaging system is about0.1. A computer is used to control the acquisitiontime of the camera and the scanning step of thereference mirror, which has a minimum moving stepof 38 nm. A four-layer info carrier is used as thesample that contains two layers of text patterns and

Figure 2 4-layer tomography extraction

a 4-layer sampleb Directly reflected image of ac Tomographic images of the first layerd Tomographic images of the second layere Tomographic images of the third layerf Tomographic images of the fourth layer


two layers of fingerprints, as shown in Fig. 2a. Fourslide glasses are used as the substrates, whoserefractive index is n ¼ 1.58. A layer of letters isdeposited on the glass with material of HFO2(n ¼ 1.85). The thickness of the coating is about onequarter wavelength, around 200 nm. The view area ofthe sample is about 15 mm � 15 mm. Fig. 2b showsthe image of the sample directly viewed by thecamera. All the four images are fused together andthe fingerprint patterns are almost impossible torecognise. Figs. 2c–2f provide the tomographic imagesextracted from layers one to four, respectively.

The SLD used in experiment is about $5000, which isthe most expensive part in our OCT system. Recently,we have used halogen bulb as the light source in a full-field OCT. Because its bandwidth is 200 nm, the depth-resolution could be about 0.9 mm. That means thepotential information content in depth is about 20times larger than the SLD light source. Anotherbenefit is the cost: a halogen lamp is only about $10.However, the tuning up of this system is extremelydifficult, if done manually. An automatic detectionmethod must be introduced to find the exactinterference position. Fig. 3 provides a set of imagesderived from a halogen-light-based OCT system. Twoglass wafers with different text, ‘NRC’ and ‘OCT’,on their surface are placed together. This systemsuccessfully extracts each layer with a depth-resolutionaround 1 mm. A problem with this broadband OCT isthe dispersion. When more layers are bonded, acompensation procedure has to be considered.

4 2D and 3D fingerprintrecognition with OCTComparing with other biometric identificationtechniques, such as iris recognition, face recognitionand hand-geometry verification methods, fingerprintrecognition owns several merits, which make it themost popular. First, the fingerprint recognition hashigh permanence, since fingerprints form in the fetalstage and remain structurally unchanged throughoutthe person’s life. Secondly, it has high distinctiveness– even identical twins possess different fingerprints. Inaddition to high universality, a majority of thepopulation have legible fingerprints, more people thanhave licences and passports. Last but not the least,fingerprint acquisition is non-intrusive and it requiresno training, which shows high acceptability. However,with less than $10 worth of household supplies,artificial fingerprint gummies can be made and easilyspoof the fingerprint system [18]. An enhancedfingerprint recognition system needs to be developedto deal with these fraudulent methods.


& C

www.ietdl.org

During the past years, valuable improvements havebeen achieved by several scientific groups to enhancethe surface-scanning-based fingerprint system. Makinguse of a fast fingerprint enhancement algorithm,which could improve clarity of the fingerprint surfacestructures and based on the estimated local ridgeorientation and frequency, the verification accuracyand the false reject rate (FRR) could be improved[19]. However, the improvement in the fingerprintrecognition field was focused on obtaining a lowerFRR and false accept rate, which made nocontribution to detect whether artificial fingerprintswere presented. Here we applied OCT techniques todetect and identify artificial material commonly usedfor spoofing the surface-scanning-based biometricsystems.

Figure 3 Images derived from halogen-light-based OCT

a Direct imaging of two text layersb Tomography of the letter ‘N’ from the first layerDepth change in white area is 1 mmc Tomography of the letter ‘R’ from the first layerDepth change in white area is 1 mmd Tomography of the letter ‘C’ from the first layerDepth change in white area is 1 mme Tomography of the text extracted in the first layerf Tomography of the text extracted in the second layer


As we mentioned above, the OCT technique hasunique ability of in-depth and lateral scanning tocapture 2D images with resolution up to a fewmicrons. In this way, by analysing OCT imagesadditional artificial layers above the real finger couldbe identified. Our previous study [20] has alsodemonstrated that the OCT can be potentially appliedto identify artificial fingerprints, which could be madeof various artificial materials. In addition, artificialfingerprint dummy and real human tissue could beeffectively distinguished by autocorrelation analysis ofspeckle-noise of OCT images and signals [21].Therefore artificial materials used to make fingerprintscan be potentially detected in the new generation ofOCT-enhanced fingerprint systems.

Fig. 4a shows typical cross-sectional 2D OCT image,which exposes artificial material layer placed over realhuman tissue. From this figure, one can see thatcharacteristic layers of human skin (stratum corneum,epidermis and dermis) are shifted down because of the

Figure 4 OCT image obtained from artificial fingerprintdummy placed over a real finger and its correspondingsignal curve

a OCT image obtained from artificial fingerprint dummy placedover a real fingerb Corresponding OCT signal curve


53

/54

www.ietdl.org

presence of an additional layer corresponding to thematerial. Gross picture of this artificial fingerprintdummy is shown in Fig. 5b. The thickness of thedummy was about 150 mm. Although the ridges andvalleys of the artificial fingerprint are tiny andtransparent, all of them could be clearly detected inOCT 2D image (Fig. 4a). Additionally, the artificialfingerprint layer can be further exposed and analysedfrom the related OCT signal curve shown in Fig. 4b.Each pixel in Fig. 4a could be converted to anequivalent intensity value and hence the whole 2Dimage can be seen as a 2D intensity value matrix with450 � 450 components. The OCT signal curvereflects the general profile of how the light goesthrough absorbing/scattering objects and how thephotons reflect back [3]. The two high peaks stand forthe surfaces of the artificial and the real fingerprintlayers, respectively. After the second peak, which isthe stratum corneum of the skin, typical OCT signalscorresponding to epidermis and dermis can be seen.Thus, both the 2D OCT image and correspondingsignal curve clearly show presence of an artificialfingerprint layer placed above the real human finger.


The female mould for the artificial fingerprintdummy, which was made of plasticene (DixonTiconderoga Company, Mexico), is shown in Fig. 4a.The corresponding male mould from the femalemould, which is also called artificial fingerprintdummy, is made of household cement (ITW DevconCorp., Mass.) and is illustrated in Fig. 4b. Testingexperiments between a commercially availablefingerprint scanning system (Microsoft FingerprintReader, Model: 1033, Redmond, WA) and the OCTsystem were carried out. First the real fingerprintpatterns, such as thumbs, forefingers, middle fingersand ring fingers, from both hands of a volunteer, wereregistered into the fingerprint scanning system using acomputer. Secondly the same fingers were used toprepare the corresponding artificial fingerprintdummies. Thirdly the dummies were placed overanother person’s real finger and were tested by boththe commercial fingerprint reader and the OCTsystem. This dummy in Fig. 4b spoofed the fingerprintscanning system mentioned above continuously(n ¼ 10). However, at the same time the OCTsystem could recognise the artificial fraudulent dummy

Figure 5 Photographs and OCT images for artificial and real fingerprints

a Female mould made of plasticeneb Male mould (artificial fingerprint dummy) made from the female mouldc Reconstructed surface profile of an artificial fingerprint based on 100 OCT images


& C

www.ietdl.org

every time successfully and provided typical detectionimages as shown in Figs. 3a and 3b. This dummy wasused to bypass a commercially available fingerprintscanning system.

As we mentioned earlier, OCT technique is capableof capturing 3D whole fingerprint patterns. The


reconstructed surface fingerprint profile of theartificial fingerprint dummy (Fig. 5b) is shown inFig. 5c (by using VolView2.0, Kitware Inc. software).This image consists of �100 slices of 2D OCTimages. The interval between each neighbourhood 2Dimage is 50 mm and the 3D scanning orientation isperpendicular in the respect to the 2D image plane.

Figure 6 Images of a dummy fingerprint obtained by a full-field OCT system

a– f OCT images extracted from the outer surfaces. Layer distance: 50 mmg– l OCT images extracted from inner surfaces. Layer distance: 20 mm


55

/56

www.ietdl.org

The typical ridges and valleys on the artificial dummycan be seen from this image, which demonstrates thatthe OCT technique can provide the depth andlongitude image information and capture surfacefingerprints.

As the artificial fingerprints are normally made bytranslucent materials, the full-field OCT becomes apowerful tool to fast and effectively recognise thosedummies. Full field OCT can detect both surfaces of adummy: fingerprint surface and non-print surface,which do not exist in a real finger. In addition, a full-field OCT can explore the internal structure withinthese two surfaces, which is also different to a realfinger. Fig. 6 shows another set of cross-sectionalimages obtained from an artificial dummy by a full-field OCT system. During the depth scanning, twoobvious surfaces of the dummy were found thatexhibited different features. The outer surface shows asmooth 2D curve, Figs. 6a–6f, which does not existin the real fingerprint; however, the inner surfaceshows segmented fingerprints at different layers,Figs. 6g–6l.


The summation of Figs. 6a–6f is given by Fig. 7a,which is a bright area without any fingerprints in it.(As the image sampling separation is sort of toobroad, 40 mm, so that the black fringes appear whenthey are overlapped.) Fig. 7b shows the summation ofthese segmented fingerprints, as shown in Figs. 6g–6l.Fig. 7c demonstrates the summation of all thosetomographies, which completely destroys thefingerprint and is totally different to the imagecaptured by a common 2D camera used in afingerprint recognition system Fig. 7d.

Fig. 8 provides another set of rotated 3D volumedata. The red parts show the internal structure of thedummy, which are totally different to the internaltissues of a real finger. The presence of them and thepatterns of those two surfaces prove that the object isan artificial fingerprint dummy.

5 Discussion and conclusionsOCT provides a new powerful tool for the applicationsof security and document identification. By extending

Figure 7 Images of a dummy

a 2D image of the outer surface of the dummy obtained from OCT imagesb 2D image of the inner surface of the dummy obtained from OCT imagesc Summation of above images a and bd Direct imaging of the dummy by a camera used in fingerprint recognition device


& C

www.ietdl.org

the existing biometrics that are based on surface scan ofexternal features, the OCT system can probe andextract the internal features of multilayered objectsand tissues, and is more robust against tampering andcounterfeiting. In this study, we demonstrated that theOCT technique could be successfully applied fordetection of artificial materials commonly used tomake fraudulent fingerprints. Overall, our resultsdemonstrated that (i) current commercial fingerprintsystems have security flaws and could be easilyspoofed by fingerprint dummies; (ii) high-resolution

Figure 8 Three rotated images of the 3D volume data of adummy


OCT 2D images and corresponding signal curvescould clearly present the artificial materials at alltimes and (iii) OCT is capable of providing high-resolution 3D images for security identificationreference. Our future research will be focused onincreasing the OCT 3D image acquisition rate andapplying the current pattern recognition method inprocessing OCT images (2D and 3D) to betterdistinguish not only the fraud/real person but also theartificial fingerprint layer, which can spoof the currentfingerprint reader, in order to better enhance thesecurity systems.

As for the document security, it is possible to extractthe cross-sectional 2D information, for example text,image, phase-relief, micro 3D profiles and so on,fused in multiple-layers. Because the OCT system hasthe depth-resolution at micrometre level, theinformation carrier could be fabricated in a tiny size.By bonding all the information together in a tinycarrier and then covering it by a special wavelengthwindow, an info chip is created. To date, there is noassociated technology to peel them layer-by-layer,except the OCT system.

6 References

[1] TOH KA, ENG HL, CHOO YS, ET AL.: ‘Identity verificationthrough palm vein and crease texture’. Proc. Advances inBiometrics, 2006, pp. 546–553

[2] WANG LY, LEEDHAM G: ‘A thermal hand vein patternverification system’, Lect. Notes Comput. Sci., 2005, 3687,pp. 58–65

[3] MIURA N, NAGASAKA A, MIYATAKE T: ‘Feature extraction offinger-vein patterns based on repeated line tracking andits application to personal identification’, Mach. Vis. Appl.,2004, 15, (4), pp. 194–203

[4] LI ZJJ, FERRER M, TRAVIESO ACM, ET AL.: ‘Biometric based onridges of palm skin over the head of the secondmetacarpal bone’, Electron. Lett., 2006, 42, (7),pp. 391–393

[5] Emerging Biometric Technologiges. GlobalSecurity:available at http://www.globalsecurity.org/security/systems/emerging.htm, accessed January 2007

[6] NIXON KA, ROWE RK: ‘Multispectral fingerprintimaging for spoof detection’, Proc. SPIE, 2005, 5779,pp. 214–225

[7] HUANG D, SWANSON EA, LIN CP, ET AL.: ‘Optical coherencetomography’, Science, 1991, 254, pp. 1178–1181


57

/58

www.ietdl.org

[8] FERCHER AF, DREXLER W, HITZENBERGER CK, ET AL.: ‘Opticalcoherence tomography—principles and applications’, Rep.Prog. Phys., 2003, 66, pp. 239–303

[9] BOUMA BE, TEARNEY GJ: ‘Handbook of opticalcoherence tomography’ (Marcel Dekker, New York,2002)

[10] BEAUREPAIRE EA, BOCCARA C, LEBEC M, ET AL.: ‘Full-fieldoptical coherence microscopy’, Opt. Lett., 1998, 23,pp. 244–246

[11] BOURQUIN S, SEITZ P, SALATHE RP: ‘Optical coherencetopography based on a two-dimensional smart detectorarray’, Opt. Lett., 2001, 26, pp. 512–514

[12] AKIBA M, CHAN KP, TANNO N: ‘Full-field optical coherencetomography by two-dimensional heterodyne detectionwith a pair of CCD cameras’, Opt. Lett., 2003, 28,pp. 816–818

[13] CHANG S, LIU X, CAI X, ET AL.: ‘Full-field optical coherencetomography and its application to multiple-layer 2DInformation retrieving’, Opt. Commun., 2005, 246,pp. 579–585

[14] >GREIVENKAMP JE, BRUNING JH: ‘Optical Shop Testing’(John Wiley and Sons, New York, 1992, 2nd edn.), Chap. 14


[15] CARRE P: ‘Installation et utilization du comparateurphotoelectrique et interferentiel du Bureau Internationaldes Poids et Mesures’, Metrologia, 1966, 2, (13), pp. 13–23

[16] HARIHARAN P, OREBM BF, EIJU T: ‘Digital phase-shiftinginterferometry: a simple error-compensating phasecalculation algorithm’, Appl. Opt., 1987, 26, pp. 2504–2506

[17] CHANG S, CAI X, FLUERARU C: ‘An efficient algorithm usedfor full-field optical coherence tomography’, Opt. LasersEng., 2007, 45, pp. 1170–1176

[18] MATSUMOTO T, MATSUMOTO H, YAMADA K, ET AL.: ‘Impact ofartificial ‘gummy’ fingers on fingerprint systems’. Proc.SPIE 4677, 2002, pp. 275–289

[19] SUZUKI H, YAMAGUCHI M, YACHIDA M, ET AL.: ‘Experimentalevaluation of fingerprint verification system based ondouble random phase encoding’, Opt. Express, 2006, 14,pp. 1755–1766

[20] MANAPURAM RK, GHOSN M, LARIN KV: ‘Identification ofartificial fingerprints using optical coherence tomographytechnique’, Asian J. Phys., 2006, 15, pp. 15–27

[21] CHENG Y, LARIN KV: ‘Artificial fingerprint recognition byusing optical coherence tomography with autocorrelationanalysis’, Appl. Opt., 2006, 45, pp. 9238–9245


issn 1751-9659 optical coherence tomography used for...

Documents