First of all I thanks ALLAH for guiding me and taking care of me all of the time, my life is so blessed because of ALLAH.
I would also like to take this opportunity to thank my thesis supervisor, Assoc. Prof. Dr.
Rahib Abiyev for giving me the opportunity to work with him and guided and helping me throughout my thesis and other courses.
It is my pleasure to acknowledge and thank people who helped me accomplish my goal to pursue graduate studies. I would like to thank my lovely parents for their constant support and encouragement. They have made lots of sacrifices to help me with my education, for which I will always be grateful.
Lastly, I would wish to thank all the faculty members of the Computer Engineering Department at Near East University for their generous help and tremendous support through out the course of my M.S. program
ABSTRACT
Signature recognition plays an important role in the modern world, as it can solve complex problems and make the human’s job easier. Signature recognition has many applications. It has been used for decades in civilian applications while other methods (e.g., fingerprints) still have the stigma of being associated with criminal investigation.
In other words, signature verification is already accepted by the general public.
There are different techniques that can be used to recognize signatures. In this thesis the well known methods for recognition of signatures are discussed. It is known that one of the efficient methods for Signature recognition is artificial Neural Network (ANN).
Neural Network has such characteristics as: vitality, parallelism of computations, learning and generalization abilities, analytic description of linear and non-linear problems…etc. Due to these characteristics neural network have great of importance in the application areas such as artificial intelligence, pattern recognition, theory of control and decision making, identification, optimal control, robotics…etc.
In this thesis signature recognition using neural network is discussed. The feed-forward neural network is applied for signature recognition. The signature database is constructed and these signatures are used to train the neural network (NN). The simulation of the system is carried out using the MATLAB package. Neural network is used to train and identify signatures. The training is carried out using the Adaptive learning Algorithm. After training, a test was done for the noise and non noise cases and the simulation result satisfies the efficiency of signature recognition system.
AKNOWLEDGEMENTS i
ABSTRACT ii
CONTENTS iii
LIST OF FIGURES vi
LIST OF TABLES viii
INTRODUCTION 1
CHAPTER ONE STATE OF APPLICATION PROBLEMS OF NEURAL NETWORK FOR SIGNATURE RECOGNITION
3
1.1 Overview 3
1.2 Pattern Recognition 3
1.3 Pattern recognition by Human 4
1.4 Pattern Recognition in Machine 5
1.5 The review on methodologies used in Pattern Recognition 5
1.5.1 Linear Discriminant 5
1.5.2 Flexible Discriminants 6
1.5.3 Non- parametric Methods 8
1.5.4 Tree-structured Classifiers 9
1.6 Stat of Application Problems of Neural Network in Signature Recognition 10
1.7 Statement of problems 15
1.8 Summary 15
CHAPTER TWO SIGNATURE RECOGNITION 16
2.1 Overview 16
2.2 Introduction 16
2.3 Elements of Image Analysis 17
2.4 Patterns and Pattern Classes 19
2.5 Error Matrices 20
2.6 The Inverse DWT of an Image 20
2.6.1 Bit Allocation 21
2.6.2 Quantization 22
2.7 Object Recognition 24
2.7.1 Signature Recognition 24
2.8 Color Image Representations 25
2.8.1 RGB Images 25
3.3 Model of a Neuron 31
3.4 Activation Functions 32
3.5 Back-Propagation 33
3.5.1 Back-Propagation Learning 34
3.6 Learning Processes 36
3.6.1 Artificial Neural Network 36
3.6.2 Unsupervised learning 38
3.6.3 Supervised learning 38
3.7 Learning Tasks 38
3.7.1 Compression Networks 39
3.7.2 Architecture of Proposed Signature Verification System 40
3.8 Summary 41
CHAPTER FOUR DEVELOPMENT OF SIGNATURE RECOGNITION SYSTEM
42
4.1 Overview 42
4.2 Introduction 42
4.3 Quantifying Gradient Change in Images 44
4.4 SDM Analysis 46
4.4.1 Image Database (Source Image) 47
4.4.2 Grayscale Factor 48
4.5 Skeletonize 49
4.6 Clip Images 50
4.7 Problem Statement 51
4.8 Neural Network 51
4.9 Architecture 52
4.10 Initialization 53
4.10.1 Training without Noise 53
4.11 System Performance 54
4.12 Noisy Pattern Generalization 58
4.13 Results Comparison 59
4.14 Summary 61
CONCLUSION 62
REFERENCES 63
APPENDIX A 66
APPENDIX B 87
APPENDIX C 123
Figure 2.1 Elements of Image Analysis 18 Figure 2.2 Benefit’ of a Bit is the Decrease in Distortion Due to receiving that Bit. 22
Figure 2.3 Uniform Quantizer 23
Figure 2.4Schematic showing how pixels of an RGB color image are formed from the corresponding pixels of the three component images.
26
Figure 2.5 line detector masks. 27
Figure 3.1 Neural Network Aliases 31
Figure 3.2 Diagram of Abstract Neuron Model. 32
Figure 3.3 Hard Activation Functions 32
Figure 3.4 Sigmoid Functions 33
Figure 3.5 A two layer feed forward network for the restaurant problem. 34
Figure 3.6 Diagram of Synapse Layer Model 37
Figure 3.7. Activity UML diagram of stages in the proposed signature verification method.
41
Figure 4.1 SD Calculations. (a) (Negative Values); and (b) (Negative Values) 45 Figure 4.2 SD Computations (Zero Values) for (a) and (b) 45
Figure 4.3 Flowcharts for Methodology 47
Figure 4.4 DATABASE Fragment 48
Figure 4.5 Grayscale of (RAED) Signature 49
Figure 4.5 Skeleton Image for (a) “a”; (b) “d”. 49
Figure 4.6 Signatures after Segmentation 50
Figure 4.7 Signatures Clipped (Resized) 51
Figure 4.8 Neural Network Architecture 52
Figure 4.9 Network Initialization 53
Figure 4.10 Training without Noise 53
Figure 4.11 Digitize Image to Grid of Input. 54
Figure 4.12 Input-Hidden Layer Feed-Forward Connections 55 Figure 4.13 Hidden-Output Layer Feed-Forward Connections 55
LIST OF TABLES
Table 4.1 Testing results of the signatures 59
Table 4.2 Signatures recognition accuracy 60
Table 4.3 Signatures Recognition Accuracy of Hidden Markov Models & NN 61