Texture Feature Extraction Using Hybrid Wavelet Type I & II for Finger Knuckle Prints for Multi-algorithmic Feature Fusion

(1)

Procedia Computer Science 79 ( 2016 ) 359 – 366

Peer-review under responsibility of the Organizing Committee of ICCCV 2016 doi: 10.1016/j.procs.2016.03.047

ScienceDirect

7th International Conference on Communication, Computing and Virtualization 2016

Texture Feature Extraction using Hybrid Wavelet

Type I & II for Finger Knuckle Prints for

Multi-Algorithmic Feature Fusion

Vandana Yadav, Dr.Vinayak Bharadi, Sunil K Yadav

MEIT Student Information Technology Thakur College of Engg & Technology,Mumbai, India

Associate Professor Information Technology Thakur College of Engg & Technology,Mumbai, India

Assistant Professor Information Technology Shree L.R Tiwari College of Engineering,Mumbai, India

Abstract

The finger knuckle print (FKP) of a particular person is found to be unique and can serve as a biometric feature has been revealed recently by the researchers. In this research Finger Knuckle Print has been used as a biometric feature. Hybrid Wavelet Type I and Hybrid Wavelet Type II were used for feature extraction from the images in order to process it further. The important role of hybrid wavelet transform is to combine the key features of two different orthogonal transforms so that the strengths of both the transform wavelets are used. In this research the different transforms like (Discrete Cosine Transform) DCT, Haar. Hartley, Walsh and Kekre are used in combination for generation of 20 different hybrid wavelets. These hybrid wavelets are applied on the database images to generate feature vector coefficients and they are then subjected to Intra Class testing And Inter Class Testing and their performance is evaluated and compared. Proposed system has given up to 77% of EER for TAR-TRR (PI) for multi-algorithmic implementation of HWI+HWII.

Peer-review under responsibility of the Organizing Committee of ICCCV 2016.

Keywords: Finger knuckle print (FKP), hybrid wavelet, region of interest (ROI), Transform, Hybrid Wavelet Type I (HWI), Hybrid Wavelet Type II (HWII).

(2)

1. INTRODUCTION

Biometrics deals with identifying humans by their characteristics or traits. Computer science makes use of biometrics for performing identification and access control operations. Individuals under surveillance and in groups can also be identified using their biometric features [1,3]. Biometric identifiers are distinct, measurable characteristics used to tag and describe individuals. Biometric identifiers of an individual are often categorized as their physiological and behavioral characteristics [1,2].

This system has used Finger knuckle print (FKP) a newly emerged biometric trait. The texture in the outer finger surface, also known as the dorsum of hand, is found to have the potential to do personal authentication. The texture design obtained by bending the finger knuckle of a person is distinct and thus can be used as a biometric trait. The skin pattern on the finger-knuckle is highly rich in texture due to skin folds and creases, and hence, can be considered as a biometric identifier. Texture rich features make Finger Knuckle Print (FKP) more advantageous and in addition to this it is easily accessible, contact-less image acquisition can be done, it is invariant to feelings and other behavioral aspects such as fatigue, stable features and tolerability in the society [4,5,23]. Typical Knuckle-Print image is shown in Figure 1, taken from the PolyU Hong Kong FKP database.

Fig 1: Finger Knuckle Print Database [7]

2. EXISTING SYSTEMS

The finger knuckle print identification system is a newly developed biometric identification system. It is based on features of the finger knuckle, texture and uniqueness of lines visible when the finger is folded. The research in [6] is based on combination of local-local information for an efficient finger-knuckle-print (FKP) based recognition system. Local information of the FKP are extracted using Scale Invariant Feature Transform (SIFT) and Speeded Up Robust features (SURF) and they are fused at matching score level. The system is evaluated for various scales and rotations of the query image. It is observed that the system performs with CRR of atleast 98:62% and EER of atmost 5:25% for query image down-scaled upto 60%and performs with CRR of 99:75% and EER of 0:925% for any orientation of query image. It only focuses on the local features.

In [12] DAISY descriptor is implemented to get the features for the finger knuckle print based authentication. The score level fusion technique is used to combine the scores of individual FKPs obtained using the DAISY features. It pays attention only towards the feature extraction methods not on the efficiency of the information retrieval [12].

The authors in [15] has proposed method that is a hybrid feature selection method of Lempel-Ziv Feature Selection and Principle Component Analysis is used for feature extraction and an artificial Neural Network based on Scaled Conjugate Gradient is used for the recognition. This process is very rigorous and time consuming. As it makes use of artificial neural science it requires lot of training samples and time. All these above methods only provide implementation of finger knuckle print recognition system using different methodologies but no comparison of these are done with respect to other methods. Due to this the efficiency of these different methods are not known [15].

The authors in [4] had extracted the FKP features using using Kekre’s Wavelet and Haar Wavelet transform. TAR-TRR analysis is performed for both Kekre‟s Wavelets and Haar Wavelet based feature vector on the database. Euclidian distance based classification was used, it was found that Haar Wavelets and Kekre‟s wavelets give same

(3)

Equal Error Rate (EER) of 80% for True Acceptance Rate (TAR) Vs. True Rejection Rate (TRR) analysis. The EER for False Acceptance Rate vs. False Rejection Rate (FRR) was found to be 20% [4].

3. PROPOSED ALGORITHM

The proposed system for this research can be referred from paper [26]. It gives the detail of the system.

Proposed Algorithm for feature vector generation is as follows:

Step 1. The Finger Knuckle Print image present in .jpeg format is read from the database and converted into grey image array of size 256 x 256.

Step 2. These images are then pre-processed to obtain the Region of Interest (ROI) consisting of the wrinkled part of the phalangeal joint. This extracted image is of size 256 x 128.

Step 3. Next the ROI is divided into three parts left, centre and right. This gives image of size 128 x 128 each and it becomes convenient to apply wavelets.

Step 4. The Finger Knuckle Print image present in .jpeg format is read from the database and converted into grey image array of size 256 x 256.

Step 5. These images are then pre-processed to obtain the Region of Interest (ROI) consisting of the wrinkled part of the phalangeal joint. This extracted image is of size 256 x 128.

Step 6. Next the ROI is divided into three parts left, centre and right. This gives image of size 128 x 128 each and it becomes convenient to apply wavelets.

Table 1. 5x4 = 20 Combination of Hybrid Wavelets

Transforms Wavelets

Kekre Walsh DCT Hartley Haar

Kekre x Kekre- Walsh Kekre- DCT Kekre- Hartley

Kekre- Haar Kekrelets

Walsh Walsh- Kekre x Walsh- DCT Walsh- Hartley

Walsh- Haar Walshlets

DCT DCT- Kekre DCT- Walsh x DCT- Hartley DCT- Haar DCTlets Hartley Hartley- Kekre Hartley- Walsh Hartley- DCT x Hartley- Haar Hartleylet s

Haar Haar- Kekre Haar-

Walsh

Haar- DCT Haar-

Hartley

x Haarlets

Step 7. Then Hybrid Wavelet Type I and Hybrid Wavelet Type II are applied on each block of the image to extract feature vector. One block gives feature vector of size 192 and such three blocks are there for one finger knuckle image hence total feature vector of size 576 is obtained for one finger knuckle sample of one user. Step 8. To reduce the complexity of calculation composite feature vector is calculated for every user sample data.

Composite feature vector comprises of all the four image samples of one user present in database as left index, left middle, right index and right middle. Multi-Resolution analysis is performed to obtain the composite feature vector using Hybrid Wavelet Type I and Hybrid Wavelet Type II upto 4 levels.

(4)

applied upon each sample as shown in table 1. And the 5 x 4 resulting wavelets are (1) Kekrelets having combinations such as Kekre-Walsh, Kekre-DCT, Kekre-Hartley, Kekre-Haar. (2) Walshlets having combinations such as Walsh-Kekre, Walsh -DCT, Walsh -Hartley, Walsh –Haar. (3) DCTlets having combinations such as DCT-Kekre, DCT-Walsh, DCT-Hartley, DCT-Haar. (4) Hartleylets having combinations such as Hartley-Kekre, Hartley -Walsh, Hartley -DCT, Hartley -Haar. (5) Haarlets having combinations such as Haar-Kekre, Haar -Walsh, Haar -DCT, Haar-Hartley. All these gives 20 hybrid wavelets. The resulting feature vector for every wavelet is then stored in the database.

Step 10. These feature vectors of the finger knuckle print are then compared using Euclidian distance based K-NN classifier which then are put to Intra Class Testing and Inter Class testing, which generates Genuine and Forgery results.Finally these distances are considered to build a Multi-Algorithmic Finger Knuckle Print System by combining results of HWI and HWII.

4. RESULTS.

TABLE 1 Comparison of Performance Index (PI) by Multi Instance Analysis

Shown below is the graphical plots of highest Performance Index (PI) of Hybrid Wavelet Type II fig 6. Multi-Algorithmic Fusion Wavelets PI : HWI + HWII Left Middle (LM) % Increase in PI Right Middle (RM) % Increase in PI KW 75.8772 5.6 77.63158 8.04 KDCT 76.90058 5.52 72.73392 -0.2 KH 71.34503 -2.89 72.80702 -0.9 KHaar 70.39474 -5.5 72.36842 -2.85 WK 73.02631 -3.1 76.97369 2.13 WDCT 75.36549 4.25 76.0965 5.26 WH 75.0731 4.16 70.39474 -2.33 WHaar 74.1959 2.94 75.2193 4.36 DCTK 74.34211 0.89 76.16959 3.37 DCTW 73.39182 2.45 71.05263 -0.82 DCTH 74.04971 -1.75 73.02631 3.52 DCTHaar 72.95322 -3.39 70.97953 -2.22 HK 73.61111 -2.33 77.04678 2.23 HW 71.34503 -5.61 75.73099 0.19 HDCT 73.02631 1.01 72.44152 0.2 HHaar 71.92982 -0.2 71.41813 -0.91 HaarK 69.73685 -3.54 73.02631 1.01 HaarW 73.31871 5.03 71.34503 2.2 HaarDCT 74.1228 4.75 73.24561 3.51 HaarH 74.1959 3.89 73.24561 2.56

(5)

Fig 5. SPI Plot for Kekre Walsh Hybrid Wavelet I + Hybrid Wavelet Type II

Then the performance metrics like Security Performance Index (SPI) and Performance Index (PI) are calculated for comparison of results generated by different 20 hybrid wavelets.

In table 1 and 2, the abbreviations K-Kekre , W-Walsh, DCT-Discrete Cosine Transform, H- Haartley and Haar-Haar transform. Both the tables show the results generated for 20 different combinations of these hybrid wavelets. The columns are highlighted by using three colors, green-highest three values, orange-lowest three values and red-most highest value.

Table 1 shows the Comparison of Performance Index (PI) by Multi Algorithmic Analysis of the different 20 hybrid wavelets for Hybrid Wavelet type I and Hybrid Wavelet type II. Here multi-algorithm is formed using the Finger Knuckle Print samples of Hybrid Wavelet Type I and Hybrid Wavelet Type II of user’s left middle finger and right middle finger. 1st_{column i.e Wavelets shows which orthogonal transform is used to form the Hybrid Wavelet.} Then followed by Performance Index given by feature vector fusion of Hybrid Wavelet type I and Hybrid Wavelet type II. By going for multi- algorithmic method of result generation we got a considerable % increase in PI of 4 to 5%. From the table below it is clearly visible that the PI of Hybrid Wavelet type I +Hybrid Wavelet type II formed by Kekre and Walsh transform indicated by KW gives the highest PI of 77.63%.

Table 2 shows the Comparison of Security Performance Index (SPI) by Multi Algorithmic Analysis of the different 20 hybrid wavelets for Hybrid Wavelet type I and Hybrid Wavelet type II. Here multi- algorithm is formed using the Finger Knuckle Print samples of Hybrid Wavelet Type I and Hybrid Wavelet Type II user’s left middle finger and right middle finger. 1st_{column i.e Wavelets shows which orthogonal transform is used to form the Hybrid} Wavelet. Then followed by Security Performance Index given by feature vector fusion of Hybrid Wavelet type I and Hybrid Wavelet type II. By going for multi- algorithmic method of result generation we got a considerable % increase in SPI of 80 to 95%. From the table below it is clearly visible that that the SPI of Hybrid Wavelet type I +Hybrid Wavelet type II formed by Kekre and Hartley transform indicated by KH gives the highest SPI of 2.99%.

(6)

TABLE 2 Comparison of Security Performance Index (SPI) by Multi Instance Analysis. Multi-Algorithmic Fusion

Wavelets

SPI : HWI + HWII Left Middle

(LM) % Increase in SPI Right Middle (RM) % Increase in SPI

KW 1.22 -13.48 2.75 95.04 KDCT 1.3 -7.8 2.97 110.64 KH 1.57 12.95 2.99 115.11 KHaar 1.27 14.41 2.33 109.91 WK 1.35 42.11 2.57 170.53 WDCT 0.92 4.55 2.46 179.55 WH 0.82 0 2.38 190.24 WHaar 0.56 -18.84 2.4 200 DCTK 1.6 11.89 2.83 97.9 DCTW 1.18 -32.95 2.42 37.5 DCTH 1.21 -36.65 2.71 41.88 DCTHaar 1.71 -22.97 2.63 18.47 HK 1.33 40 2.53 166.32 HW 0.82 13.89 2.3 219.44 HDCT 0.97 10.23 2.25 155.68 HHaar 0.59 -14.49 2.16 170 HaarK 1.26 5.88 2.42 103.36 HaarW 0.92 -7.07 2.25 127.27 HaarDCT 0.91 -6.19 2.17 123.71 HaarH 0.95 5.56 2.32 157.78 5. CONCLUSION

This research has performed Finger Knuckle Print recognition using 20 different hybrid wavelets for type I and type II for multi-instance combination for its left middle and right middle finger knuckle print images. The results clearly indicates that Mult- Algorithmic system gives better performance when compared to Unimodal Systems.

REFERENCES

1. Anil K. Jain, Arun Ross, and Salil Prabhakar, “An Introduction to Biometric Recognition”, IEEE transactions on circuits and systems for video technology, Vol. 14, no. 1, january 2004.

2. Alfred C. Weaver, “Biometric Authentication”, IEEE Computer Society, Feb. 2006, Volume 39, No. 2, pp. 96-97.

3. Debnath Bhattacharyya, Rahul Ranjan, Farkhod Alisherov A. and Minkyu Choi, “Biometric Authentication: A Review

4. H B Kekre and V A Bharadi, “Finger-Knuckle-Print Verification using Kekre’s Wavelet Transform”, in ICWET’11, February 25–26, 2011, Mumbai, Maharashtra, India, ACM 978-1-4503-0449-8/11/02.

5. Chetana Hegde, P. Deepa Shenoy, K. R. Venugopal and L. M. Patnaik, “Authentication using Finger Knuckle Prints”,Received: 26 June 2012 / Revised: 15 October 2012 / Accepted: 1 March 2013 / Published online: 19

(7)

April 2013, in Springer-Verlag London 2013.

6. G S Badrinath, Aditya Nigam and Phalguni Gupta, “An Efficient Finger-knuckle-print based Recognition System Fusing SIFT and SURF Matching Scores”. Department of Computer Science and Engineering, Indian Institute of Technology, Kanpur, 208016, India.

7. http://www.comp.polyu.edu.hk/~biometrics/FKP.htm

8. V A Bharadi and Pallavi Vartak, “Performance Improvement of Hyperspectral Face Recognition by Multimodal and Multi Algorithmic Feature Fusion of Hybrid and Kekre Wavelets based Feature Vectors”,ICCUBEA,PCCOE Pune 2015.

9. V A Bharadi and Pallavi Vartak, “ Hyperspectral Face Recognition by Texture Feature Extraction using Hybrid Wavelets Type I & II and Kekre Wavelet Transform”,ICCUBEA,PCCOE Pune 2015.

10. Loris Nanni, Alessandra Lumini, “A hybrid wavelet-based fingerprint matcher”, Received 19 June 2006; received in revised form 3 January 2007; accepted 27 February 2007, in Elsevier.

11. Abdallah Meraoumia, Salim Chitroub and Ahmed Bouridane, “Fusion of Finger-Knuckle-Print and Palmprint for an Efficient Multi-biometric System of Person Recognition”, in IEEE ICC 2011 proceedings, 978-1-61284-231-8/11.

12. Neha Mittal, Madasu Hanmandlu and Ritu Vijay, “A Finger-Knuckle-Print Authentication System Based on DAISY Descriptor”, IEEE 2012 12th International Conference on Intelligent Systems Design and Applications (ISDA), 978-1-4673-5119-5/12.

13. G S Badrinath, Aditya Nigam and Phalguni Gupta, “An Efficient Finger-knuckle-print based Recognition System Fusing SIFT and SURF Matching Scores”, Department of Computer Science and Engineering, Indian Institute of Technology, Kanpur, 208016, India.

14. Rui Zhao, Kunlun Li and Ming Liu, Xue Sun, “A Novel Approach of Personal Identification Based on Single Knuckle print Image”, IEEE 2009 Asia-Pacific Conference on Information Processing, 978-0-7695-3699-6/09.

15. Mobarakol Islam, Md. Mehedi Hasan, M. M. Farhad and Tanzina Rahman Tanni, “Human Authentication Process Using Finger Knuckle Surface with Artificial Neural Networks Based on a Hybrid Feature Selection Method”, 978-1-4673-4836-2/2012 IEEE.

16. H .B. Kekre, Tanuja Sarode and Rachana Dhannawat, “Image Fusion Using Kekre’s Hybrid Wavelet Transform”, IEEE 2012 International Conference on Communication, Information & Computing Technology (ICCICT), Oct. 19-20, Mumbai, India, 978-1-4577-2078-9/12.

17. H. B. Kekre, Archana Athawale and Dipali Sadavarti, “Algorithm to Generate Kekre’s Wavelet Transform from Kekre’s Transform”, International Journal of Engineering Science and Technology, Vol. 2(5), 2010, 756-767.

18. H. B. Kekre, Tanuja K. Sarode and Sudeep D. Thepade, “. Inception of Hybrid Wavelet Transform using Two Orthogonal Transforms and It’s use for Image Compression”, (IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, 2011.

19. H B Kekre, V A Bharadi, P P Janrao and V I Singh, “Face Recognition using Kekre’s Wavelets Energy & Performance Analysis of Feature Vector Variants”, ICWET’11, February 25–26, 2011, Mumbai, Maharashtra, India ACM.

20. H B Kekre, V A Bharadi, V I Singh and A A Ambardekar, “Palmprint Recognition Using Kekre’s Wavelet’s Energy Entropy Based Feature Vector”, ICWET’11, February 25–26, 2011 ACM., Mumbai, Maharashtra, India.

21. H B Kekre, V A Bharadi, P Roongta, S Khandelwal, P Gupta, B Nemade, V I Singh, S Gupta and P P Janrao, “Performance Comparison of DCT, FFT, WHT, Kekre’s Transform & Gabor Filter Based Feature Vectors for On-Line Signature Recognition”, , International Journal of Computer Application (IJCA), Special Issue for ACM International Conference ICWET 2011 extended papers, February 2011.

22. Vinayak A Bharadi, Vikas Singh and R R Sedamkar, “Hybrid Wavelets based On-line Handwritten Signature Recognition”, International Journal of Applied Information Systems (IJAIS) – ISSN : 2249-0868 Foundation of Computer Science FCS, New York, USA, Volume 2– No.2, February 2012.

23. H B Kekre and V A Bharadi, “Finger-Knuckle-Print Region of Interest Segmentation using Gradient Field Orientation & Coherence”, Third International Conference on Emerging Trends in Engineering and Technology (ICETET 2010), Paper Published on IEEE Xplore, 19-21 November, 2010, Goa, India.

(8)

from Multidimensional Data set for On-line Handwritten Signature Recognition”, 4th International Conference & Workshop on Advanced Computing, ICWAC 2013, TCET, Mumbai, 22nd & 23rd February 2013.

25. Bhavesh Pandya and Vinayak Bharadi ,“Multimodal Fusion of Fingerprint & Iris using Hybrid wavelet based feature vector”, 4th International Conference & Workshop on Advanced Computing, ICWAC 2013, TCET, Mumbai, 22nd & 23rd February 2013.

26. Vandana Yadav, Dr. Vinayak A Bharadi and Sunil K Yadav. “ Feature Vector Extraction based Texture Feature using Hybrid Wavelet Type I & II for Finger Knuckle Prints for Multi-Instance Feature Fusion”, 7th International Conference on Communication, Computing and Virtualization 2016 (Under Consideration)