Palmprint Identification Using Gabor and Wide ... - Science Direct

Available online at www.sciencedirect.com

ScienceDirect Procedia Computer Science 93 (2016) 706 – 712

6th International Conference On Advances In Computing & Communications, ICACC 2016, 6-8 September 2016, Cochin, India

Palmprint Identification using Gabor and Wide Principal Line Features Hemantha Kumar Kalluria,∗, Munaga V. N. K. Prasadb a CSE

Department, Vignan’s Foundation for Science Technology and Research University, Guntur, Andhra Pradesh, India for Development and Research in Banking Technology (IDRBT), Castle Hills, Masabtank, Hyderabad, India

b Institute

Abstract In this paper proposed palmprint identification using Gabor features, Gabor and Wide Principal Line Image (WPLI) features. Extracted a fixed size ROI from palmprint images. Resize the extracted ROI into 64 x 64. Apply the Gabor filters to extract the features from the resized ROI. Dissimilarity distance is used to measure the dissimilarity between the query palmprint and database palmprint images. Experiments were conducted on Polyu Palmprint Database using Gabor features, Gabor and WPLI features. Experimental results shows that the proposed approach using Gabor and WPLI features obtains better results compared with the existing methods. © by by Elsevier B.V.B.V. This is an open access article under the CC BY-NC-ND license c 2016  2016The TheAuthors. Authors.Published Published Elsevier (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of ICACC 2016. Peer-review under responsibility of the Organizing Committee of ICACC 2016

Keywords: Palmprint; Biometrics; Region of Interest; Feature Extraction; Palmprint Recognition.

1. Introduction Automatic identification of humans is a very essential for law enforcement, public places such as airports and shopping complexes etc.,. Traditionally documents used for human recognition are passports, ration cards, driving licenses etc.,. But these are very susceptible for forgery. Biometric based systems are possibly the best solution for human recognition 1 . Kung et al. 2 demonstrates palmprints can also be used for human identification. Zhang et al. 3 stated that palmprint contains more information compared to fingerprint. The central region of the palmprint is referred as ROI 4 . ROI extraction is the first and important step in palmprint registration, identification and verification process. In the literature 5,6,7,8,9 , a majority of researchers used ROI size of 128 x 128 on PolyUPalmprint Database 10 . Zhang et al. 11 used the circular Gabor filters to perform palmprint identification. Dale et al. 12 extracted 128 x 128 ROI from palmprints. The ROI is resized to 64 x 64 and the resized image was divided into four non-overlapping sub-images. For each sub-image applied 2-D DCT and obtained transformed coefficients were grouped into nine groups. For each group the standard deviation was calculated, i.e. for each ROI obtained 36 standard deviations were formed as the feature vector. Guo 13 et al. partitioned the ROI into several smaller sub-images, then the feature vectors ∗

Corresponding author. Tel.: +091-9490776374 . E-mail address: khk [email protected]

1877-0509 © 2016 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 ICACC 2016 doi:10.1016/j.procs.2016.07.272

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

were extracted using: Fourier transform, DCT transform, Gabor transform and Local Binary Pattern (LBP). Pan et al. 14 applied the Gabor filter and divided the resultant image into two level non-overlapping partitions and calculated the mean and standard deviation as features. Mu et al. 15 used the Gabor filter to generate Gabor magnitude (GM) features. Logarithm transform was applied to generate LogGM features. The LogGM feature image was divided into subblocks. Mean and standard deviation of LogGM feature of all the blocks was considered as the feature vector. Literature survey 12,13,14,15 shows that Statistical Features (SF) extracted after applying transformation such as Gabor filter, Fourier transform etc. Hence, palmprint identification using Gabor features without using statistical functions has to be explored. The remaining part of the paper is organized as follows: Section 2 provides an overview of ROI extraction. Section 3 provides Gabor feature vector and WPLI generation. Section 4 provides details of palmprint registration and palmprint recognition approaches using Gabor Features, Gabor and WPLI features. Experimental results using Gabor features, Gabor and WPLI features are discussed in section 5. Section 6 describes the conclusions. 2. ROI Extraction To identify the key points, adopted the key point identification approach proposed by Kalluri et al. 16 . The obtained key points were used to align the palmprint image. After align the palmprint image extracted 128 x 128 size ROI. The sample palmprints and extracted ROIs are shown in Figure 1.

Fig. 1. Palmprints and its corresponding ROIs.

3. Feature Extraction 3.1. Gabor Feature Vector Generation ROI image is resized into 64 x 64 by using bi-cubic interpolation. The circular Gabor filter is an effective tool for texture analysis 17 , and has the following general form: x 2 + y2 1 exp(− )exp{2πi(μxcosθ + μysinθ)} (1) 2πσ2 2σ2 Apply the Gabor filter on the resized image to obtain the magnitude, real and imaginary features. After obtaining these values to perform binarization the obtained values are greater than zero are considered as 1 otherwise the vales are considered as 0. G(x, y, θ, μ, σ) =

707

708

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

To extract Wide Principal Line Image (WPLI) adopted the WPLI extraction approach proposed by Kalluri et al 18 . Sample ROIs and it’s related WPLIs are shown in Figure 2.

a

b

c

d

Fig. 2. Sample ROIs (a) & (b) and its corresponding WPLIs (c) & (d).

4. Registration and Recognition Registration is one of the important steps in biometrics in general pattern recognition. The schematic diagram is depicted in Figure 3. For every palmprint in Registration Images, extract the ROI, resized the extracted ROI into 64 x 64 and apply the Gabor filter on the resized ROI. Perform the binarization on obtained features and store these features in the Feature Data Base. Apply WPLEs on extracted ROI (128 x 128) to generate WPLI and store the obtained WPLI on the Feature Data Base.

Fig. 3. Schematic diagram for registration.

4.1. Similarity Measures In Gabor features, the dissimilarity between the Database (testing) and the Query (training) image has been calculated by Dissimilarity Distance (DD) (2). Where D M , DR and DI are the magnitude, real and imaginary feature values of the database image. Similarly Q M , QR and QI are the magnitude, real and imaginary feature values of the query image 64 64 i=1 j=1 (D M (i, j)&Q M (i, j))&(DR (i, j) ⊕ QR (i, j)) + (D M (i, j)&Q M (i, j))&(DI (i, j) ⊕ QI (i, j)) DD = (2)  64 2 ∗ 64 i=1 j=1 (D M (i, j)&Q M (i, j)) For WPLIs, Let L1 and L2 are the training WPLI and testing WPLI respectively and their size is 128 x 128, then the matching score (MS) between L1 and L2 is defined as

709

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

128 128 MS (L1 , L2 ) =

j=1 L1 (i, j)&L2 (i, j) 128 128 i=1 j=1 L1 (i, j)

i=1

(3)

Similarity score (SS) is calculated using the following equation S S (L1 , L2 ) = max(MS (L1 , L2 ), MS (L2 , L1 ))

(4)

4.2. Palmprint Recognition using Gabor Features The Schematic diagram for palmprint recognition using Gabor Features is depicted in Figure 4.

Fig. 4. Schematic diagram for palmprint recognition using Gabor Features.

The proposed approach using Gabor features takes testing palmprint, Feature Data Base as input and returns the ID of the testing palmprint. Extract the ROI from the testing palmprint, resize the extracted ROI into 64 x 64 and apply the Gabor filter to generate Gabor features. Calculate the Dissimilarity Distances using equation (2) between query image feature vector and feature vectors of training images and palmprint wise identify the minimum Dissimilarity Distances. Declare the smallest Dissimilarity Distance palmprint is the identified palmprint and stop the process. 4.3. Palmprint Recognition using Gabor and WPLI Features The Schematic diagram for palmprint recognition using Gabor and WPLI features is depicted in Figure 5. The proposed approach using Gabor and WPLI features takes testing palmprint, Feature Data Base as input and returns the ID of the testing palmprint. Extract the ROI from the testing palmprint, resize the extracted ROI into 64 x 64 and apply the Gabor filter to generate Gabor features.

710

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

Calculate the Dissimilarity Distances using equation (2) between query image feature vector and feature vectors of training images and palmprint wise identify the minimum Dissimilarity Distances. If the smallest Dissimilarity Distance less than thresholda then declare the smallest distance palmprint is the identified palmprint otherwise the following steps are used for identification: Step 1. Apply the WPLEs on the extracted ROI(128 x 128) to generate 128 x 128 WPLI. Step 2. Retrieve the smallest dissimilarity distance palmprint Database WPLIs from Feature Data Base. Step 3. Compute the Similarity score between testing WPLI and retrieved WPLIs using equation(5). Step 4. If the maximum of SS ≥ T hresholdb then declare the current palmprint is the identified palmprint otherwise retrieve the next smallest dissimilarity distance palmprint Data Base WPLIs from Feature Data Base and goto step 3.

5. Experimental Results Experiments were conducted on the PolyUPalmprint Database 10 , to verify the efficiency of the proposed approach. 5.1. Data Sets for Experiments The data sets used by the previous researchers are considered here to compare the performance of the proposed approach. These data sets are given below: Data Set I: The first two palmprint images in the first session and first palmprint image in the second session of every palm were considered as the training palmprint images, all the palmprint images of all the palm were considered as the testing palmprint images. Mansoor et al. 7 obtained CRR was 90.17. Data Set II: The first 5 palmprint images of every palm in the first session were considered as the training palmprint images and the left over images of the first session were considered as the testing palmprint images. The number of training palmprint images are 1930 and the number of testing palmprint images are 1959. Xuan et al. 8,19 obtained CRR was 97.377 and 99.02 respectively.

5.2. Experiments using Gabor Features On Data Set I, obtained CRR is 97.18 (i.e. 7534 palmprint images are correctly identified). On Data Set II, obtained CRR is 99.285 (i.e. 1945 palmprint images are correctly identified). The experimental results are shown in Table. 1. Using the proposed approach extract better results compared to existing methods 7,8,19 . 5.3. Experiments using Gabor and WPLI Features On Data Set I, obtained CRR is 99.15 (i.e. 7686 palmprint images are correctly identified). On Data Set II, obtained CRR is 99.846 (i.e. 1956 palmprint images are correctly identified). The experimental results are shown in Table. 1 . Using the proposed approach extract better results compared to existing methods 7,8,19 . Table 1. Performance (CRR) comparison of the proposed approach using Gabor Features Data Set

Author with reference Number used the Data Set

CRR of Existing Methods

CRR of the Proposed Approach using Gabor Features

CRR of the Proposed Approach using Gabor and WPLI Features

Data Set I

Atif Bin Mansoor et al. 7

90.17

97.18

99.15

97.377 99.02

99.285

99.86

Data Set II

al. 8

Xuan et Xuan et al. 19

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

Fig. 5. Schematic diagram for palmprint recognition using Gabor and WPLI Features.

6. Conclusion In this paper, palmprint recognition approach is presented based on Gabor and WPLI features. Experiments were conducted using Gabor Features, Gabor and WPLI features on PoluPalmprint Database. From these experimental

711

712

Hemantha Kumar Kalluri and Munaga V. N. K. Prasad / Procedia Computer Science 93 (2016) 706 – 712

outcomes, it can be concluded that the proposed approach using Gabor Features, Gabor and WPLI features obtains better performance in terms of the CRR compared with existing methods. It is also observed that CRR is improved when both Gabor and WPLI features are applied in combination.

References 1. D. Zhang, Automated Biometrics: Technnologies and Systems, Kluwer Academic Publishers, 2000. 2. S. Y. Kung, S. H. Lin and M. Fang, ”A Neural Network Approach to Face Palm Recognition,” in Neural Networks for Signal Processing , Cambridge, MA, 1995. 3. D. Zhang, Palmprint authentication, Kluwer Academic Publishers, 2004. 4. W. K. Kong and Z. David, ”Palmprint texture analysis based on low-resolution images for personal authentication,” in 16th International Conference on Pattern Recognition, pp.807-810, 2002. 5. C.L. Lin, T. C. Chuang and K. C. Fan, ”Palmprint verification using hierarchical decomposition,” Pattern Recognition, vol. 38, no. 12, pp. 2639-2652, 2005. 6. D. S. Huang, W. Jia and D. Zhang, ”Palmprint verification based on principal lines,” Pattern Recognition, vol. 41, no. 4, pp. 1316-1328, 2008. 7. A. B. Mansoor, H. Masood, M. Mumtaz and S. A. Khan, ”A feature level multimodal approach for palmprint identification using directional subband energies,” Journal of Network and Computer Applications, vol. 34, no. 1, pp. 159-171, 2011. 8. W. Xuan , L. Li and W. Mingzhe, ”Palmprint verification based on 2D-gabor wavelet and pulse-coupled neural network,” Knowledge-Based Systems, vol. 27, pp. 451-455, 2012. 9. M. Mu, Q. Ruan and S. Guo, ”Shift and gray scale invariant features for palmprint identification using complex directional wavelet and local binary pattern,” Neurocomputing, vol. 74, no. 17, pp. 3351-3360, 2011. 10. ”PolyUPalmprint Database,” Honkong Polytechnique University, [Online]. Available: http://www.comp.polyu.edu.hk/ biometrics. 11. D. Zhang, W. K. Kong, J. You and M. Wong, ”Online palmprint identification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 9, pp. 1041-1050, 2003. 12. M. P. Dale , M. A. Joshi and N. Gilda, ”Texture based palmprint identification using DCT features,” in Seventh International Conference on Advances in Pattern Recognition, pp. 221-224, 2009. 13. J. Guo, Y. Liu and W. Yuan, ”Palmprint recognition using local information from a single image per person,” Journal of Computational Information Systems, vol. 8, no. 8, pp. 3199-3206, 2012. 14. X. Pan and Q. Q. Ruan, ”Palmprint recognition using Gabor-based local invariant features,” Neurocomputing, vol. 72, no. 7-9, pp. 2040-2045, 2009. 15. M. Mu and Q. Ruan, ”Mean and standard deviation as features for palmprint recognition based on Gabor filters,” International Journal of Pattern Recognition and Artificial Intelligence, vol. 25, pp. 491-512, 2011. 16. H. K. Kalluri, M. V. N. K. Prasad, A. Agarwal, ”Dynamic ROI Extraction Algorithm for Palmprints,” in Third International Conference on Advances in Swarm Intelligence (ICSI 2012), Part II, LNCS 7332, pp. 217-227, 2012. 17. J.G. Daugman, High Confidence Visual Recognition of Persons by a Test of Statistical Independence, IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 15, no. 11, pp. 1148-1161, Nov. 1993. 18. H. K. Kalluri, M. V. N. K. Prasad, A. Agarwal, ”Palmprint Identification Based on Wide Principal Lines,” International Conference on Advances in Computing, Communications and Informatics (ICACCI-2012), pp. 918-924, 2012. 19. W. Xuan, L. Junhua and W. Minghe, ”On-line fast palmprint identification based on adaptive lifting wavelet scheme,” Knowledge-based systems, vol. 42, pp. 68-73, 2013.