Performance Evaluation of Palm Vein Feature Extraction Techniques using Linear Discriminant Analysis and Principal Component Analysis

ABSTRACT Palm vein recognition is a biometric identification method that integrates automated scientific applications to provide a way or means to identify a person by scanning a person’s palm vein in order to gain access.  Palm vein recognition technology is secure because the authentication data exists inside the body, it is very difficult to forge and it is highly accurate.
The palm vein feature extraction task is another challenging problem in the infrared hand palm recognition tasks. In this research two set of extraction techniques have been suggested to represent the veins grid attributes: (i) the Linear Discriminant Analysis and (ii) the Principal Component Analysis. Each extracted feature list is assembled as a feature vector used to distinguish between palms belonging to different persons.
The training and testing were performed on static printed images of palm vein for the database with the palm vein samples taken with different angular prints and that of normal print. The experiments were performed for the recognition system at cropped resolutions of 20x20, 30x30, 40x40, 50x50, and 60x60 pixels.  The system developed had average recognition performance accuracy of 96.86% and 90.45% for the LDA and PCA respectively. LDA performs better in terms of accuracy.
The best recognition performance was obtained when the LDA algorithm was used and this however shows why it is commonly used in authentication and verification process. The implemented system serves as an extendable foundation for future research.
TABLE OF CONTENTS Page    Title page        i Certification   ii Dedication      iii Acknowledgements    iv Table of Content   v List of Figures            ix List of Tables x Abstract                      xi       
CHAPTER ONE  INTRODUCTION 1 1.1 Preamble   1          1.2 Statement of problem                   5 1.3 Aim and Objectives            6 1.4 Method of study   6
CHAPTER TWO LITERATURE REVIEW                    7 2.1       Biometrics7 2.1.1 Enrollment Module       10 2.1.2 Identification Module   10 2.2       Biometrics Performance        11 2.3       Historical Background of Feature Extraction               11 2.4       Overview of Palm Vein Feature Extraction     13 2.4.1    The Description of PalmVein               13 2.4.2       The Principal Line Detection and Datum Point Location     16 2.5       Steps Taken in Palm Vein Feature Extraction  18 2.5.1    Preprocessing             18 2.5.1.1 Binarizing the Palm Image       18 2.5.1.2 Boundary Tracking                    18 2.5.1.3 Feature Detection                      19 2.5.1.4 Establishing a Coordination System              19 2.5.1.5 Extracting the Central Part       20 2.5.2    ROI Extraction           20 2.5.3    Feature Extraction and Matching         20 2.5.3.1 Line Based Approach                20 2.5.3.2 Subspace Based Approach        22 2.5.3.3 Statistical Approach                  22 2.5.3.4 Coding Approaches                   22 2.5.4    Accept and Reject      23 2.6       Palm vein Features Extraction Process            23 2.6.1    Image Acquisition   24 2.6.2    Feature Extraction      24 2.6.2.1 Line- Based    25 2.6.2.2 Appearance-Based                     26 2.6.2.3 Texture-Based 26 2.6.3    Features Matching      28 2.6.3.1 Geometry Based Matching       28 2.6.3.2 Features Based Matching 29 2.7  Performance Measures           30 2.8       Feature Extraction based on Linear Discriminant Analysis 31 2.8.1    Algorithm for Linear Discriminant Analysis   31 2.9       Feature Extraction based on Principal Component Analysis            32 2.9.1    Algorithm for Principal Component Analysis          33
CHAPTER THREE RESEARCH METHODOLOGY        34 3.1       Background Theory                34 3.2       Stages in the Development of the Proposed System             36 3.2.1    Palmvein Image Acquisition                 36 3.2.2    Image Preprocessing  36 3.2.2.1 Image Enhancement                  36 3.2.2.2 Segmentation  37 3.2.3    Histogram Equalization37 3.2.4    Gray Scale Conversion  38 3.2.5    Cropping                     38 3.2.6    Noise Filtering           38 3.3       Recognizing images using Eigenspace             39 3.4       Training and Classification    40 3.5       Recognition and Testing        40
CHAPTER FOUR RESULTS AND DISCUSSION          42 4.1       Analysis of Result                  42 4.2       Discussion42 4.2.1    Computational Time  42 4.2.2    Accuracy                     43        4.2.3    False Rejection Rate  43 4.2.4    False Acceptance Rate  43
CHAPTER FIVE CONCLUSION AND RECOMMENDATIONS            60 5.1       Conclusion60 5.2       Recommendation                   60 REFERENCES     61 APPENDIX A 65
LIST OF FIGURES Figure             Page 2.1         The Line Pattern of Palm Vein 15 2.2       The Distribution of Palm Vein Features           17 2.3       ROI Extraction                       21 3.1       Palm Vein Recognition System      35 4.1       Graph Showing the Result of Computational Time              45 4.2       Graph Showing Accuracy Performance of PCA and LDA Feature
Extraction Comparison of PCA and LDA.       49 4.3       GUI Window Developed        51 4.4:      Information Window              52 4.5       Equalization Operation          53 4.6       Raw Histogram of a Palm Vein Image             54 4.7       Equilised Palm Vein             56 4.8       Coded-PCA Features              57 4.9       Coded-LDA Features             58
LIST OF TABLES Table               Page 4.1.1   Computation time                  44 4.1.2    Accuracy   46 4.1.3    False Rejection rate 47 4.1.4    False Acceptance rate  48