BIMODAL RECOGNITION BASED ON THUMBPRINT AND THUMB IMAGE USING BAYESIAN CLASSIFIER LOW ZHI WEI UNIVERSITI TEKNOLOGI MALAYSIA

BIMODAL RECOGNITION BASED ON THUMBPRINT AND THUMB IMAGE USING BAYESIAN CLASSIFIER LOW ZHI WEI UNIVERSITI TEKNOLOGI MALAYSIA

BIMODAL RECOGNITION BASED ON THUMBPRINT AND THUMB IMAGE USING BAYESIAN CLASSFIER LOW ZHI WEI A project report submitted in partial fulfillment of the requirement for the award of the degree of Master of Engineering in Computer & Microelectronic Systems Faculty of Electrical Engineering Universiti Teknologi Malaysia APRIL 2010

Dedicated to my family for their support and love iii

iv ACKNOWLEDGEMENT I am heartily thankful to my thesis supervisor, Assoc. Prof. Dr. Syed Abd. Rahman Al-Attas, whose encouragement, guidance and support from the initial to the final stage enabled me to develop my understanding on the subject. Without his continued support and interest, this thesis would not have been the same as presented here. I would like to show my gratitude to my family members especially my parents, brother and fiancée Mei Mei for the moral support and assistance given during my critical time. Lastly I would like to thank my fellow coursemate for their valuable views and tips for my thesis.

v ABSTRACT The purpose of this project is to develop a Thumb image classification module which able to predict the gender from the image input. This module can be integrated into the current thumb print recognition system to form a bi-modal biometric system. With this add in module, the performance of the recognition system will significantly increase since the database search time is reduce into almost half when only the gender matched is considered. The development of this module is based on the Bayesian Classifier method by having input of the textural analysis, thumb area consumption and the thumb width size. The textural analysis is using the GLCM (Gray Level Co-occurrence Matrix) with its properties of contrast, correlation, energy and homogeneity. The thumb area and size calculation is based on a cropped image which has the thumb over a certain boundary. Due to the usage model of searching the database, the training set and the verification set is coming from the same data sets. The Bayesian Classifier algorithm is implemented in the MATLAB code. Few GLCM pixel distance analysis was done to evaluate the module performance. With the distance pixel of 2, it had shown the best accuracy among the result of other pixel combination. Result for male matching 82.35% and female matching is 81.82%.

vi ABSTRAK Projek ini bertujuan untuk membina satu modul klasifikasi ibu jari yang mampu meramal jantina daripada imej ibu jari. Modul tersebut boleh diintegrasikan dengan sistem pengenalpastian imej ibu jari untuk membentuk satu sistem biometric dwi-modal. Dengan adanya modul ini, prestasi sistem pengenalpastian dapat ditingkatkan kerana masa untuk menaksir and membanding hanya perlu dilakukan kepada jantina yang berpadan sahaja. Pembinaan modul ini berdasarkan kepada klasifikasi Bayesian dengan ciri-ciri tekstur, kawasan ibu jari dan saiz ibu jari. Analisis ciri-ciri tekstur adalah berdasarkan matrix sama kejadian paras kelabu (GLCM) dengan sifat-sifat seperti kontras, kolerasi, tenaga dan kehomogenan. Pengiraan terhadap kawasan dan saiz ibu jari adalah berdasarkan imej yang dipotong dengan batasan mengandungi ibu jari. Kumpulan data latihan and kumpulan data pengesahan adalah daripada kumpulan data yang sama oleh sebab penggunaan modul ini adalah untuk memadan imej dalam database sahaja. Algoritma Bayesian klasifikasi ditulis dalam kod MATLAB. Beberapa analisis terhadap jarak GLCM piksel telah dijalankan untuk menilai prestasi modul ini. Dengan menggunakan jarak piksel 2, keputusan ketepatan adalah terbaik member ketepatan 82.35% bagi golongan lelaki dan 81.82% bagi golongan perempuan.

vii TABLE OF CONTENT CHAPTER TITLE PAGE DECLARATION DEDICATION ACKNOWLEDGEMENT ABSTRACT ABSTRAK TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES LIST OF ABBREVIATIONS LIST OF APPENDICES ii iii iv v vi vii x xii xiii xiv 1 INTRODUCTION 1.1 Project Background 1.2 Objectives 1.3 Scope of Work 1 3 3 2 LITERATURE REVIEW 2.1 Previous Work 2.2 Grey Level Co-occurance Matrix (GLCM) 2.2.1 Textural Features extraction based on GLCM 2.3 Bayesian Classification 2.3.1 Naive Bayesian Classifier 4 5 9 11 12

viii 3 METHODOLOGY 3.1 Project Flow 3.2 Instrumentation 15 17 4 THUMB IMAGE FEATURE EXTRACTION AND ALGORITHM 4.1 Feature Extraction 4.1.1 Thumb Area and Size Extraction 4.1.2 GLCM Texture Extraction 4.2 Bayesian Classifier Algorithm 19 20 23 28 5 VERIFICATION AND RESULT ANALYSIS 5.1 Images for the Bayesian Classification system 5.2 Result of Bayesian Classifier using GLCM pixel distance of 1 pixel 5.3 Result of Bayesian Classifier using GLCM pixel distance of 2 pixels 5.4 Result of Bayesian Classifier using GLCM pixel distance of 4 pixels 32 33 37 40 6 CONCLUSION 6.1 Concluding Remarks 6.2 Recommendation for Future Work 6.2.1 Smaller group range for probability calculation instead of quartile 6.2.2 Better image capturing and processed using colored image 6.2.3 Larger sample size 44 45 46 46 46 REFERENCES 48

Appendices A-E 49 ix

x LIST OF TABLES TABLE NO. TITLE PAGE 4.1 Summary of parameter available for graycomatrix MATLAB 25 function 4.2 Mathematical form of GLCM properties 27 5.1 Summary table for the thumb image training set and 33 verification set 5.2 Summary of the Result using GLCM distance of 1 pixel 37 5.3 Summary of the Result using GLCM distance of 2 pixels 40 5.4 Summary of the Result using GLCM distance of 4 pixels 43 6.1 Summary of system accuracy on different GLCM pixel distance 45

xi LIST OF FIGURES FIGURE NO. TITLE PAGE 2.1 Diagram of GLCM reference pixels to neighboring pixel 6 in various angle 2.2 Analyzed data and GLCM table with matching operation 7 2.3 4x4 Analyzed image with four gray levels 8 2.4 General Form of the GLCM table 8 2.5 GLCM calculation of 0 degree 8 2.6 GLCM calculation of 45 degree 9 2.7 GLCM calculation of 90 degree 9 2.8 GLCM calculation of 135 degree 9 3.1 Project Flow Chart 17 3.2 Feature extraction of the cropped image 17 4.1 Flow chart of the gender classification MATLAB code 20 4.2 Thumb image and the cropped area for thumb area and 21 size calculation 4.3 Cropped thumb image and the binary form thumb image 22 4.4 MATLAB code and high level block description for 22 thumb area and size calculation 4.5 Example of MATLAB function on imread and rgb2gray 23 4.6 Original images and the cropped area for texture feature 24 extraction 4.7 Example of MATLAB function on graycomatrix 24 4.8 Diagram of offset illustrating angle and position 25 4.9 Example of MATLAB function on graycoprops 27 4.10 Output of GLCM properties function in MATLAB 28

xii 4.11 Output of thumb area and size calculation function in 29 MATLAB 4.12 MATLAB code for Bayesian Classifier threshold 30 calculation 4.13 Value used for Bayesian Classification threshold 30 4.14 Mathematical calculation for gender identification on 31 Bayesian Classifier 5.1 MATLAB function graycomatrix with 1 pixel distance 33 5.2 Contrast distribution for the GLCM distance of 1 pixel 34 5.3 Correlation distribution for the GLCM distance of 1 pixel 34 5.4 Energy distribution for the GLCM distance of 1 pixel 35 5.5 Energy distribution for the GLCM distance of 1 pixel 35 5.6 Thumb area data distribution 36 5.7 Thumb size data distribution 36 5.8 MATLAB function graycomatrix with 2 pixel distance 37 5.9 Contrast distribution for the GLCM distance of 2 pixels 38 5.10 Correlation distribution for the GLCM distance of 2 pixels 38 5.11 Energy distribution for the GLCM distance of 2 pixels 39 5.12 Homogeneity distribution for the GLCM distance of 2 39 pixels 5.13 MATLAB function graycomatrix with 4 pixels distance 40 5.14 Contrast distribution for the GLCM distance of 4 pixels 41 5.15 Correlation distribution for the GLCM distance of 4 pixels 41 5.16 Energy distribution for the GLCM distance of 4 pixels 42 5.17 Homogeneity distribution for the GLCM distance of 4 pixels 42

xiii LIST OF ABBREVIATIONS ASM - Angular Second-Moment GLCM - Gray Level Co-Occurrence Matrix IPT - Image Processing Toolbox MATLAB - Matrix Laboratory

xiv LIST OF APPENDICES APPENDIX TITLE PAGE A MATLAB Code for Feature Extraction 49 B MATLAB Code for Threshold Calculation 52 C MATLAB Code for Probability Calculation 55 D MATLAB Code on Verification Model 62 E MATLAB Code on Thumb Area and Size Calculation 70

CHAPTER 1 INTRODUCTION This project implement a thumb top view classification module using the Bayesian classifier based on the texture, thumb area and size analysis. The Bayesian classification is developed and implemented using the MATLAB software. With the addition of thumb area and size analysis from previous research, it is target to hit higher accuracy of the classification system. In the first chapter, overview of the project background is presented followed by the discussion on the problem statement, project objectives and the scope of work. The organization of this report is presented at the end of the chapter. 1.1 Project Background Bio recognition system is getting popular nowadays especially on the thumb print recognition since most of the high end laptop is integrated with the thumb print recognition device. With the cheaper cost of implementing the thumb print recognition system compared to the 90 s, the government agency such as National Registration Department (Jabatan Pendaftaran Negara) started to record the resident s

2 thumbprint for their record. Besides that, the bank as well implemented the thumbprint record upon the registration of a new account. With this implementation, the huge database will takes time to retrieve the data when bank or the government wants to retrieve the customer information upon verification process. Thus to improve the verification process throughput time, bi-modal thumbprint recognition is suggested where the thumbprint recognition is integrated with the gender recognition using the Bayesian classification method. This can be done with a camera capturing the top-view of the thumb image while the user is providing the thumbprint through the thumb print scanning device. From the topview image captured, it is feed into the MATLAB code to differentiate the male and female while the thumbprint captured is used to get the actual identity recognition. It will help to reduce the database query time since the query can just limit to the male or female category only. In this project, a thumb classification module based on the Bayesian classification technique will be developed to classify the top-view image captured to male and female category. A few parameters will be extracted from the image to feed into the Bayesian classifier such as the space occupied of the thumb in a pre-defined area, texture feature like contrast, hue, homogeneity and the size of the thumb skin. The algorithmic detail will for these implementations will be discussed in Chapter 2. The main focus on this project would be to increase the accuracy of differentiating the image to be male of female compared to previous project and integration with the thumb print recognition.

3 1.2 Objective The main objective of this project is to develop a thumb classification module that can integrate with thumb print recognition system to differentiate between male and female in order to reduce the database searching time. The target achievements are as below: 1) To develop a thumb classification module based on the Bayesian classifier technique using the data from top-view captured image. 2) To able to integrate the thumb classification module with the thumb print recognition system. 1.3 Scope of Work The scope of work can be divided into the following: I. Understand the Bayesian classifier based on the texture, shape, and size analysis of the thumb. II. Analysis will be done on the thumb in grayscale and on a cropped region of the thumb images. III. Develop the MATLAB code that will differentiate male and female based on the thumb texture, shape and size based on the offline data. IV. All programming coding will be done in MATLAB.

48 REFERENCE 1. Tom M. Mitchell, Machine Learning, 1 st edition, McGraw Hill (1997) 2. Robert M. Haralick et al., Textural Features for Image Classification, IEEE Transaction on System, Man and Cybernetic 1973 3. Sajad Shirali-Shahreza & M. E. Mousavi, A New Bayesian Classifier for Skin Detection, IEEE The 3rd Intetnational Conference on Innovative Computing Information 2008 4. Robin Hanson et al, Bayesian Classification Theory, Nasa Artificial Intelligence Research Branch 5. Alberto Tellaeche et al, A vision-based method for weeds identification through the Bayesian decision theory, Science Direct 2007 6. Tom M. Mitchell. Generative and Discriminative Classifier: Naïve Bayes and Logistic Regression. McGraw-Hill 2006. 7. Thomas P. Weldon. Improved Image Segmentation with a Modified Bayesian Classifier. International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2006) 8. Daniel P. Stormont. An Online Bayesian Classifier for Object Identification. International Workshop on Safety, Security and Rescue Robotics 2007