Wavelet Based Modern Finger Knuckle Authentication

Procedia Computer Science 70 ( 2015 ) 649 – 657

1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Organizing Committee of ICECCS 2015 doi: 10.1016/j.procs.2015.10.101

ScienceDirect

4

International Conference on Eco-friendly Computing and Communication Systems

Wavelet Based Modern Finger Knuckle Authentication

Sujata Kulkarni

, R.D.Raut

, P.K.Dakhole

a

Research Scholar, Department of electronics Engineering ,YCCE, Electronics, Nagpur, India b

C.I.C Research cell, S.G.B.Amaravati University, Amravati, India

Abstract

Biometrics is a prominent technology for accurate and safe detection of claim identity. This paper explores new biometrics trait named finger knuckle (FK) for authentication. Finger Knuckle has unique bending and makes a distinctive biometric identifier. We use middle part of back surface of finger for recognition. The system consists of proposed prototype finger knuckle capturing device, own FK images acquired from this device and Kekre Wavelet Transform based feature for matching. Finger knuckle image capturing device i s mo d i f i e d s t r u c t u r al ly using SolidWorks13. F eature extraction is based on Kekre Wavelet Transform which gives energy coefficient as unique features for matching. Experimental tests are performed on r i g h t index and middle finger knuckle of 50 users of own FK image data base acquired from proposed device and standard Hong Kong Polytechnic University (Poly U) database. Proposed Kekre Wavelet Transform (KWT) algorithm is tested on both FK database shows improved recognition accuracy of 90% as compared to conventional wavelet based method. The experimental results shows that wavelet based fusion of local a n d global features shows improved recognition accuracy of 92.5%, which is more than integration of local features only. The local features namely local orientation, local phase, and phase congruency are integrated with one global feature of Fourier transform coefficient

© 2014 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of organizing committee of the International Conference on Eco-friendly Computing and Communication Systems (ICECCS 2015).

Keywords: Finger knuckle (FK); FK device; Kekre Wavelet Transform; authentication; error equal rate; accuracy

1. Introduction

Biometric recognition or simply biometric refers to use of distinctive anatomical and behavioral characteristics for automatically recognizing an individual. Automatic human identification has become an important issue in today’s global information society. Due to increasing security concerns, large number of systems currently required © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

positive identification before allowing an individual to use their services. During the last decade there has been a steady research effort toward providing user friendly and reliable methodologies for access to facilities, research and services. Automatic biometric systems have emerged as a more reliable alternative to the traditional personal identification systems. Different techniques have been developed each of them having its own advantages and disadvantages 1, 2.

Now a day’s authentication is based on the unique physiological and behavioral characteristics of human being. It is generally accepted that physical traits like iris, fingerprints, finger knuckle, finger vein, DNA finger print can uniquely define each member of large population which makes them suitable for large scale identification5.Reason of attraction of such traits is social acceptance and easy to use. Finger knuckle (FK) is user centric, contactless and unrestricted access control. Its texture and statistical features are available and easily extracted. It is independent to any behavioral aspect. No stigma of potential criminal investigation is associated with this approach 3, 4. Proposed system uses finger knuckle as biometric trait for recognition.

Lin Zhang et al.6developed FKP image acquisition device and created FK image database [Hong Kong Polytechnic University]. Authors implemented different techniques for features extraction from the F inger K nuckle P rint [FKP]. The promising performance of F K P based personal authentication system is reported in7. Loris Nanni et al.proposed Radom Transform and Haar Wavelet for feature extraction and showed approximately zero error equal rate for finger knuckle recognition system8. Scale Invariant Feature Transform (SIFT) and the Speeded up Robust Features (SURF) techniques are implemented and proved best recognition in9. Lin Zhang et al. proposed local and global feature extraction and proved that integration of local and global features gives better accuracy than integration of local features only14. H.B.Kekre et al. used Kekre and Haar Wavelet for finger knuckle feature extraction and tested on readymade database (Hong-Kong University). The performance of two wavelets demonstrate performance index of 80% 9, 11.

This paper highlights prototype finger knuckle (FK) acquisition device, formation of finger knuckle database of different classes, and feature extraction. Proposed finger knuckle recognition is based on Kekre Wavelet Transform. The rest of the paper is organized as follow. Sub section 1.1 describe block diagram of proposed authentication system. Sub section 1.2 and 1.3 describes proposed prototype FK acquisition device and formation of own FK image data base, sub section 1.4 and 1.5 describe Kekre Wavelet Transform (KWT) and feature extraction using Kekre Wavelet Transform (KWT), Sub section 1.6 discuss matching module, Section 2 presents fusion of local and global features, Section 3 deals results and discussion on performance of proposed FK device, proposed algorithm and algorithm based on integration of local and global features. We conclude in Section 4.

1.1. Proposed system

Proposed finger knuckle recognition shown in figure1 comprises the finger knuckle acquisition device, formation of own finger knuckle database, generation of Kekre wavelet Transform based feature vector and matching unit.

Genuine user/ Imposter

user

Fig. 1.Block diagram of proposed finger knuckle recognition FKP database FK images acquired from proposed device Extracti on of ROI of finger knuckle images Enhance ment of FK images Kekre Wavelet Transfor m domain represent ation Generate feature vector of claimer Matching classifier Decisio n Unit Feature vector of enrolled user

1.2. P The deform can be Finger Propo • • • T back C are d captu FK trans user. of fi comp finge imag back T devic acqui Solid beca Mod acqu and w at di has p roposed finger e proposed prot mation due to h e relieved from r knuckle imag

sed device desc Lighting con over exposur The acquired The backgro restrictions a The prototype F kground, finger Camera captures darker than wh uring device is images are wi slation or rotatio . These limitatio inger. Finger br pared to comme er supporter. N ges be acquired k side of the dev The distance bet ce (100mm). H isition of FK im d Works I3as sh ause of its roug dified FKP devi uire FK images w

we use white ba fferent time are potential for per

knuckle acquisi to type finger k hand orientation m careful planni es used in prop cribe the uncont

ndition and the re or under expo d image should ound contains are not tough an Finger knuckle supporter and d

Fig. 2. (a) s FK images w hite backgroun made with prop ith high resolu on take place. ons are overcom racket is design ercial system. F Notch size is op d. The supporte vice and its size tween the came Height of the de mages. Figure3 s hown in Figure3

h design. In m ice is compact with high resolu ack ground. By e similar to each rsonal identifica ition device knuckle acquisi n and perspectiv ing of acquisiti posed system a trolled acquisitio camera exposu osured

cover whole fin no color that nd are easily sat e capturing de digital camera (

) prototype device ( ith white and b nd. Hence ima per design. Mo ution. While ca

Such moveme me by using fold

ned for this pu For acquisition ptimum, so use er is rectangular e depends on dif era and support

evice is reduce shows detail str . In prototype odified device, in size (140 m ution (4300 x 3 using proposed h other while im ation.

ition device sur ve projection. on environmen are acquired by

on environment ure have no spe

nger, and finger is similar to t isfied for most p vice is compos (SonyDSC-W38

(b) frame work of d black backgroun

ges with whit odified device is

apturing Finger nt increases var ding bar as guid urpose. The to n, user places f er with any fin r in shape so it e fferent model of ter is 89.04 mm ed from 120 mm

ructural design device user fac finger notch is mm x 130 mm 200 ) .Here dist d image acquisit mages from dif

rvives under un Hence, users o nt and unnatura ordinary came t as follows. ecial restrictions r should occupy the skin color. portable device sed of light w 80)12as shown i device (c) Modified nd. The capture e background s compact in siz r Knuckle ima riation in finge ding structure fo ouching area of finger from the nger size can pl

easily fixes fing f digital camera m. This is less a m to 90 mm to of modified FK ce problem of p s flexible and m x90 mm), cost tance between t tion devices, im fferent fingers a ncontrolled light of biometric aut l hand orientati eras equipped o s. But there sh the image as m . According to s in practice. weighted acrylic in figure 2. FKP device ed images with

are used. The ze, user friendly ages, finger m er knuckle featu or finger tips and f the proposed

notch at front lace finger on s ger. Camera soc a. as compare to o o avoid outside K image acquis placing finger o mounted on the t effective and the camera and mages of the sam

are different; thi

ting conditions thentication sys ion during capt n portable devi hould not be sev

much as possible o experience, th c box with wh black backgrou e modification y and captures t movement such ure of authentic d part of back s system is smal side of device supporter and F cket is provided ld design of sa e reflection duri sition device us on finger suppor e finger support user friendly a knuckle is 89 m me finger collect is implies that F and stem ture. ices. vere e. hese hite und in the as cate ide ller on FK d at me ing sing rter ter. and mm ted FK

Fig.3 1.3. F As midd prese finger consis teach in tw from Figur

3. (a) base view, (b

Finger knuckle i s per hand geo dle knuckle of r nt only on the f r knuckle of rig sts of 50 users ( ing, students an o phases with a the proposed d re 4 (b) and Figu a User1_1righ

b

) font view for user

image database

metry, number right index and four fingers exc ght index and m (18 to 68 years) nd workers. To average interval evice. These im ure(c) respectiv

r to insert the finge

e

of finger is av right middle fi cept thumb. Sma middle knuckles

) from an educa consider varian

l of days and tim

mages are cropp ely.

ser1_5right_ind

er1_5right_inde

er (c) camera socket

vailable for acq inger. Among f all finger knuck of all classes t ational institute nt finger knuckl me. Figure 4 ( ped to obtain re

t in which the camer

quisition. Propo five fingers, low kle is too small i to cover the ent

of different cat le location and o a) shows some egion of interest right_middle... r1_1right_ midd ra is inserted and(d)

osed system int wer, middle and in area. We cre tire population. tegories such as

orientation, FK

of the sample F t (ROI) and are

User1_5right_m

dle... User1_5r

) shows the actual se

tent to capture d upper knuckle ated the databa

FK image data s VIP, teaching

images are acq

FK images acqu resized as show middle right_middle etup only es are ase of abase , non quired uired wn in

User1_1right_index... User1_5right_index User1_1right_ middle... User1_5right_middle

Fig. 4. (a) raw finger knuckle (b)cropped(c)resized finger knuckle.

Image captured from the modified FKP device are raw images. These images are enhanced to get better clarity of unique features such as ridges, creases around the phalageal joint of finger knuckle surface. This is very important step to improve the recognition. We use Wiener filter and reflection removal as pre-processing step to enhance the quality of FK images acquired from the proposed device. Images with many edges are handled by local W iener filter. Hence we de noise all FK images using Wiener filter is done using Wiener filter15 .Original images has curvature surface hence results in non uniform reflection. To obtain well distributed texture, we use the reflection removal technique9_{. The reflection filtered image is enhanced image from which feature are}

extracted. Features are extraction using Kekre Wavelet Transform. Proposed technique extracts more number of features with same iterations and improves the performance index and reduces the error equal rate (EER).

1.4. Kekre wavelet transform

Avoid hyphenation at the end of a line. Symbols denoting vectors and matrices should be indicated in bold type. Scalar variable names should normally be expressed using italics. Weights and measures should be expressed in SI units. All non-standard abbreviations or symbols must be defined when first mentioned, or a glossary provided.

Kekre transform matrix of size MxM defined as under 10

Elements of above matrix are calculated as,

ܭݔݕ ൌ ൝

ͳǢ ݔ ൑ ݕ െܰ ൅ ሺݔ െ ͳሻǢ ݔ ൌ ݕ ൅ ͳ

ͲǢ ݔ ൐ ݕ ൅ ͳ

(1)

The advantage of Kekre wavelet is that wavelet matrix is integer but need not the power of two. So we can form any size wavelet. We consider Kekre transform matrixes (M) of 5 x 5. We can form Kekre wavelet of 15 x 15 and up to maximum size of M2 x M2.

In this paper, FK images are pre-processed and resized to 128x128 and transformed to Kekre Wavelet Transform domain (KWT). Keke wavelet is formed from Kekre Transform of 64 x 64 with spreading factor 2 in the first iteration. In further iteration, matrix is reduced by 2 and so on 13.

1.5. Feature extraction using Kekre wavelet transform

Kekre Wavelet Transform (KWT) is used to extract localized spectral information from the region of interest (ROI) of FK image. We considered region of interest of size 256 × 128 pixels. The region of interest is divided into three blocks each of size 128 × 128 pixels such as left, centre and right block. These images are transformed using the 128x128 KWT matrices. Wavelet transform is applied up to three levels. At this level four set of coefficients i.e. Low Low (LL), Low High (LH), High Low (HL), and High High (HH) are available as

shown in figure 5. The feature vector is formed using LH HL, HH coefficients while LL is used for further transformation13_{. The LH, HL, and HH are divided into 2X2 parts of size matrix (N/4 x N/4). A set of four}

wavelet coefficients obtained from LH, is given by the following equation

ܹܧ ൌ σ௪ିଵ௜ୀ଴ σ௪ିଵ௝ୀ଴ܹܥሺ݅ǡ ݆ሻʹ (2)

Where W is size of wavelet component (64, 32, and 16). Similarly a set of 4 wavelet coefficients are obtained from HL and HH. The values of these 3 set of coefficients gives 12 features from the first transformation. For the second transformation the LL1 from the transformed matrix is taken and padded with zero to keep the size of matrix 128 x128. This matrix is inversed transformed and down sampled. The down sampling is done by selecting alternate rows and columns. The new matrix is of size 64 x 64. The KWT of size (64x64) is used for second transformation and the same procedure is carried out on LL2 and another set of 12 features extracted. The third level of transformation is done with the down sampled LL2 in 2nd stage and transformed using KWT of size 32 x 32. So we obtain total 36 features i.e. energy coefficients are obtained in three steps13. In FK images

database, among the 10 samples of each user, 7 samples are used as training and rest 3 samples as testing. Testing samples are 150 and training samples are 350. Features are extracted for all samples and feature vector is formed.

Fig. 5. Feature extraction using KWT

1.6. Matching Module

Matching scores between query FK images and enrolled FKP in database are obtained by using Euclidean distance method based on feature vector10. The matching is done for genuine and impostor user. Verification is done for genuine matching. The features of the selected test sample (test1) a r e matched with the seven t r a i n i n g samples of same user. Seven Euclidean distances (ED1 to ED7) are calculated by the following equation.

ܧܦ ൌ ඥσ ሺݐ݁ݏݐሺͳǡ ݉ሻ െ ݐݎܽ݅݊ሺͳǡ ݉ሻʹெ

ଵ (3)

The average of these seven Euclidean distances gives EDavg1. Similarly for second test sample EDavg2 and

third test EDavg3 is calculated. In identification the features of the selected test sample (test1) are matched with

the seven training (train1-train7) samples of each user present in the database except for the same user. The

Euclidean distance matching is performed for recognition. Using the Euclidean distances threshold is calculated for verification and identification process. User recognition depends upon the threshold. Threshold value is compared with the EDavg and if it less than EDavg, user is recognized otherwise not recognised 13.

Table 1 shows different threshold value for genuine test and imposter test. The true acceptance starts to recognize at threshold 220 while true rejection at 340. Such 150 and 7350 tests are taken for genuine and imposters respectively and calculate the recognition accuracy in terms of true acceptance rate (TAR) and true rejection rate (TRR).

2. Fusion of Local and Global Features

We compute the phase congruency along with local orientation and local phase from a local image patch. These three local features are independent of each other and reflect different aspects of image local information.

These local features are combined with Fourier transform coefficients, which are global features. This feature integration is known as local feature integration (LFI). The matching distance between training and test images is determined. These distances are calculated by matching three local features PCD, ORID, and PHAD. These

three distances are fused together to get the final matching distance14. Furthermore these local features are fused w i t h global features. The features are integrated using Matcher-Weighting (MW) rule, where weights are assigned according to equal error rate (EER) obtained on a training dataset by different matchers.

3. Result and Discussion

Finger knuckle samples of right hand are selected as most people use it frequently and easy to get different types of finger knuckle samples. Middle knuckle of index and middle finger are taken for each user. Five samples per user are taken so total 10 samples per user. So we have total 50 *10= 500 finger knuckle samples data base.

Table 1. True acceptance and false acceptance at different threshold (Own FKP database)

Threshold Genuine Imposter

TA FA TR FA 0 20 40 60 80 100 120 140 160 180 200 0 150 0 150 0 150 0 150 0 150 0 150 0 150 0 150 0 150 0 150 0 150 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 7350 0 220 2 148 7350 0 240 260 280 300 320 340 360 3 147 4 146 5 145 7 143 9 141 13 137 14 136 7350 0 7350 0 7350 0 7350 0 7350 0 7349 1 7348 2

We consider 3 test samples for verification and 7 training samples for identification. All samples of FK in both modes are tested using MATLAB (R2009a). The program runs in MATLAB with Intel 3 processor, 4GB RAM and windows 7(32bit) operating system.

Figure 6 shows graph of True Acceptance Rate (TAR) Vs. True Rejection Rate (TRR). The X-axis represents a scaled down Euclidean distance (threshold). The Y–axis represents a percentage of TAR and TRR. Genuine test shows 100% false rejection till threshold is 200. After this false rejection decreases and true acceptance increases. At end it shows only 5 to 6 % false acceptances to performed tests. For impostors test, it shows 100% true rejection up to threshold value of 320. When threshold increases it shows some false acceptance. From the Table 1, genuine test shows false rejection from the threshold 220 and impostors test give false acceptance from the threshold 340.These ranges give the range of the FRR and FAR. Table 2 presents the recognition accuracy of finger knuckle using KWT.

In accur thresh featur Fi local feature e racy of 92.5%. hold. Table 3 sh res and wavelet

Fig.7.

ig.6. (a) TAR -TRR Table 2. Perfor M Kekre wave features tes [Hong-Kon Proposed Ke tested on Ow Proposed Ke features teste [Hong-Kong] extraction algori Figure 7 shows hows the recog based features.

(a) TAR vs. TRR f

R plot on own FKP d rmance of finger kn Method

elet transform with sted on standard FK ng]

ekre wavelet with 10 wn FKP database

ekre wavelet with t d on standard FKP ]

ithm, by integra s the graphical r gnition accuracy

for fusion of local a

database (b) TAR -T nuckle recognition 36 K database 08 features tested 108 P database

ating local featu representation o y for own finger

and global features (

TRR plot on Hong-using KWT TAR-TRR 80 % 90% 90%

ures with globa of True Accepta r knuckle syste (b) TAR vs TRR fo -Kong Poly U FKP FAR-FRR 20% 10% 10% al features, it sh ance Rate, True m using integra

or wavelet based fea database

hows the recogn e Rejection Rat ting local and g

atures

nition te vs. global

Table 3. Performance of finger knuckle recognition on own database

Matching Distance

TAR-TRR

FAR- FRR Local and Global features

Wavelet based features(KWT)

92.5 %

92%

7.5%

10%

4. Conclusion

Finger knuckle an uncommon and upcoming physiological biometric trait has become popular for authentication. Edges, curves, creases present on middle knuckle are unique and capable to identify person. FK images are acquired from prototype compact in size, low cost and user friendly acquisition device. The proposed FKP technique is evaluated on own FK database shows that images acquired from proposed FK capturing device efficiently recognize the users. It is observed that pre-processing of FK images improve quality of raw samples individually and subsequently recognition accuracy. It shows that if more features are used recognition accuracy increases. The proposed technique has Error equal rate (EER) 10% on o wn FK image database which is almost 50% less as compare to conventional technique. A l g o r i t h m u s i n g i n t e g r a t i o n of local and global features system is t e s t e d on own database. The experimental result shows that the fusion of local and global features together performs better than using any of them.

References

1. S. Prabhakar, S. Pankanti, and A. K. Jain, Biometric Recognition: Security and Privacy Concerns IEEE : Security and Privacy Magazine;, 2003. pp. 33-42.

2. D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, book of Fingerprint Recognition, Springer NY; 2003. 3. J. Daugman, Recognizing Persons by Their Iris Patterns Cambridge University.

4. A. Eriksson and P. Wretling, How Flexible is the Human Voice? A Case Study of Mimicry Proc. Of the European Conference on Speech Technology;1997. pp. 1043-1046.

5. A. Kumar, C. Ravikanth, Personal Authentication using Finger knuckle Surface IEEE Transactions on Information Forensics and Security ;2009. pp 98–109.

6. Lin Zhanga, Lei Zhanga, David Zhanga and Hailong Zhub, Online Finger-Knuckle-Print Verification for Personal Authentication Biometrics.

7. Lin Zhang, Lei Zhang, and David Zhang, Finger-Knuckle-Print Verification Based on Band-Limited Phase-Only Correlation LNCS 5702, Springer-Verlag Berlin Heidelberg; 2009 pp. 141–148.

8. Loris Nanni, Sheryl Brahman, Alessandra Lumini, A User Dependent Multi-Resolution Approach for Biometric Data International Journal of Information Technology and Management; 2012. pp 112- 121.

9. G S Badrinath, Aditya Nigam and Phalguni Gupta,An Efficient Finger-knuckle-print based Recognition System Fusing SIFT and SURF Matching Scores Indian Institute of Technology, Kanpur, 208016, India.

10. Dr. H. B. Kekre, Archana Athawale, Dipali Sadavarti, Algorithm to Generate Wavelet Transform from an Orthogonal Transform International Journal of Image Processing; 2011.

11. Neha Gharat, Sujata S.Kulkarni,and .Trushita Chaware “Biometric Authentication using Fusion ,Technique” International conference and workshop on Advance Computing 2013 pp 214-18.

12. S.S. Kulkarni, R.D.Raut, Identification System using Finger Knuckle Features in UACEE International Journal of Advances in Electronics E n g i n e e r i n g ;2012

13. R.D.Raut, Sujata kulkarni, Neha N. Gharat, Biometric Authentication using Kekre’s Wavelet Transform International Conference of Electronic Systems, Signal Processing and Computing Technology, IEEE;2014.

14. Lin Zhang , Lei Zhang, David Zhang, Zhenhua Guo, Phase congruency induced local features for finger-knuckle-print recognition Pattern Recognition, Elsevier;2012.pp 2522–2531