FINGER VEIN RECOGNITION USING LOCAL MEAN BASED K-NEAREST CENTROID NEIGHBOR AS CLASSIFIER By Saba nazari Thesis submitted in fulfillment of the requirements for the degree of bachelor of Science 2012 ACKNOWLEDGEMENT i
First of all I would like thank God for conceding me many affluence and protecting me. I would like to take this opportunity to express my profound sense of gratitude and respect to all who helped me through the duration of this thesis. I would like thank my supervisor, mr arash mahyari, who guiding me, teaching me how to improve my ideas supporting me, encourage me and helping me for everything that I needed for this research. Special thanks to my lovely husband ali whom important reason of mine to expressed my goals in my life for always encouraging me and also standing the detachment from each other. I am grateful for my father and my mother who confidence me in each steps of my life and helping my dreams come true. Lastly, I would like thank of all my friends, special Sepehr Monfared,Mehran Mirsafai,Sina Ashoori and Pegah Moradi, for helping me during the process of writing and implementing the programs. ii
TABLE OF CONTENTS FINGER VEIN RECOGNITION USING LOCAL MEAN BASED K-NEAREST CENTROID NEIGHBOR AS CLASSIFIER...i saba nazari... Error! Bookmark not defined. ACKNOWLEDGEMENT...i TABLE OF CONTENTS...iii LIST OF TABLES...vi LIST OF FIGURES...vii LIST OF ABBREVIATIONS...x Abstract...xi Abstrak...xiii CHAPTER 1...1 INTRODUCTION...1 2 1.1 Overview...1 1.2 Biometric...2 1.3 Finger vein recognition and PCA...3 1.3 Biometric Identification and Verification...5 1.3.1 Biometric Identification System...5 1.3.2 Biometric Verification System...6 1.4 Classification methods...6 1.5 Problem statement...8 1.6 Research objectives...9 1.7 Scope of Research...9 1.8 Thesis outlines...10 CHAPTER 2...11 LITERATURE REVIEW...11 2.1 biometric systems...11 2.1.1 Identification...12 2.1.2 Verification...13 2.2 Finger vein recognition...14 iii
2.2.1 Finger vein feature...17 2.2.2 General model of finger vein recognition...18 2.3 Image acquisition...20 2.3.1 2.4 Image processing...22 Feature extraction...27 Principal component analysis (PCA)...27 2.4.1 Background mathematics...28 2.4.2 Advantages of PCA...30 2.4.3 Implementation of PCA...31 2.4.4 Mathematics of PCA...35 2.5 2.5.1 The K-nearest neighbor classifier (KNN)...37 2.5.2 The LMKNN classifier...39 2.5.3 The Nearest centroid Neighbor (NCN)...40 2.5.4 The KNCN classifier...42 2.5.5 THE LMKNCN classifier...44 2.5.6 Comparison between the LMKNCN and the LMKNN...45 2.6 3 Classification...37 Summary...48 CHAPTER 3...49 METHODOLOGY...49 Introduction...49 3.1 Proposed method...49 3.2 Principal Component Analysis (PCA) algorithm...52 3.3 KNN classifier...53 3.4 LMKNCN classifier...54 3.4.1 3.5 4 Mathematics of LMKNCN...56 Summary...60 CHAPTER 4...62 RESULTS AND DISCUSSION...62 4.1 Introduction...62 4.2 MATLAB Software...62 iv
5 4.3 Database information...63 4.4 How to get the optimum Size of the images...64 4.5 How to find the Accuracy...66 4.6 Experimental results and analysis...66 4.6.1 The results of KNN...66 4.6.2 The results of proposed method (using LMKNCN)...72 4.7 Comparisons...77 4.8 Summary...80 CHAPTER 5...82 CONCLUSION...82 5.1 Summary...82 5.2 Future work...83 REFERENCES...84 APPENDICES...87 Comparison between image size of 10 30 and 20 60 in different number of training and testing images...87 v
LIST OF TABLES Table 2.1 Comparison of Various Biometric Methods at Seven Factors 16 Table 2-1 Comparison of Various Biometric Methods 16 Table 4.1 The obtained accuracy of KNN and LMKNCN for the different number of training and testing finger vein images 78 vi
LIST OF FIGURES Page Figure 1.1 Finger vein authentication devices 3 Figure 2.1 Biometric Industry Reveries and Percentage of Biometric Market by Application 2005-2010(USD $M) 12 Figure 2.2 The block diagram of identification and verification system consists of all components 14 Figure 2.3 The general model of finger vein recognition 19 Figure 2.4 Image acquisition devices 20 Figure 2.5 The basic structure of finger vein capturing devises and one sample of the original finger vein image 21 Figure 2.6 (a,b,c) Original captured images, binarized images, cropped images 24 Figure 2.7 (a,b) using high pass filter to retain high frequency components 26 Figure 2.8 The resized finger vein images (top) and also their enhanced images 26 Figure 2.9 Exapmle of PCA data 31 Figure 2.10 Plot of normalised data 33 Figure 2.11 The plot of new data pont after applying the PCA 34 Figure 2.12 The reconstruction from the data that was derived using only a single eigenvector 34 Figure 2.13 The comparison between the KNN and the KNCN when k=5 43 Figure 2.14 A comparison between LMKNCN and LMKNN in two-class classification problem and k=5 in two dimensional feature space. 46 Figure 3.1 The overall view of proposed method 50 Figure 3.2 Different types of implementations by LMKNCN and KNN classifiers from each class 51 Figure 3.3 The PCA implementation flow 53 vii
Figure 3.4 The basic algorithms to propose LMKNCN classifier 55 Figure 3.5 LMKNCN algorithm implementation flow 60 Figure 4.1 The captured images from one person 63 Figure 4.2 Excremental testing results in two different sizes of images 65 Figure 4.3 database, 9images to train, 1 image to test 67 Figure 4.4 database, 8images to train, 2images to test 67 Figure 4.5 database, 7images to train, 3 images to test 68 Figure 4.6 database, 6images to train, 4images to test 68 Figure 4.7 database, 5images to train, 5 images to test 69 Figure 4.8 database, 4 images to train, 6 images to test 69 Figure 4.9 database, 3images to train, 7images to test 70 Figure 4.10 database, 2images to train, 8 images to test 70 Figure 4.11 database, 1image to train, 9images to test 71 Figure 4.12 database, 9images to train, 1image to test 72 Figure 4.13 database, 8images to train, 2images to test 73 Figure 4.14 database, 7images to train, 3images to test 73 Figure 4.15 database, 6images to train, 4images to test 74 Figure 4.16 database, 5images to train, 5images to test 74 Figure 4.17 75 viii
database, 4images to train, 6images to test Figure 4.18 database, 3images to train, 7images to test 75 Figure 4.19 database, 2images to train, 8images to test 76 Figure 4.20 database, 1image to 9images to test 76 Figure 4.21 Comparison between the percentage of having highest accuracy between LMKNCN and KNN 79 Figure 4.22 The differences between the percentage of accuracies of KNN and LMKNCN 80 ix
LIST OF ABBREVIATIONS PCA: Principal Component Analysis ROI: Region of Interest NIR: Near-Infrared LED: Light Emitting Diode KNN: K Nearest Neighbor LMKNN: KNCN: Local Mean Based K Nearest Neighbor K Nearest Centroid Neighbor LMKNCN: Local Mean-Based K-Nearest Centroid Neighbor x
FINGER VEIN RECOGNITION USING LOCAL MEAN BASED K-NEAREST CENTROID NEIGHBOR AS CLASSIFIER Abstract Nowadays, the security requirement has been rapidly increased. Bank robbery, financial losses and other security systems weakness due to identity theft are the evidences to introduce the high importance of the identification systems. Biometrics is one of the best technologies which link the identity of people to behavioral or physical characteristics of them in order to provide security and safety.one of the newest method of biometric systems is finger vein recognition which is a unique and successful way to identify human based on the physical characteristics of the human finger vein patterns. As the database used in finger vein is image, Principal Component Analysis (PCA) is used in this thesis to extract the valuable features from the database. After feature extraction, the purpose of classification and determining which dataset belongs to which one is the next step which is done by taking advantage of a newly proposed method called Local Mean-based K-Nearest Centroid Neighbor classifier(lmkncn) which is believed as a big improvement on traditionally used methods such as K-nearest neighbor classifier(knn). All in all, in this research a new method is proposed in which PCA is used as a feature extraction method and LMKNCN as a classifier. Finally, the significance of the proposed method is proven by comparing the results of LMKNCN classifier with those of the KNN classifier. Experimental results demonstrate that LMKNCN classifier performs the best by not only employing the nearest and geometrical distribution of neighbors around query pattern, but also takes into account the local mean vector of K-neighbors from each class. xi