FACE RECOGNITION IN CRIMINAL INVESTIGATIONS (UNCONTROLLED ENVIRONMENT)

Presented To

Electrical/Electronic Engineering Department

📄 Pages: 115 🧠 Words: 13111 📚 Chapters: 5 🗂️️ For: PROJECT

👁️‍🗨️️️ Views: 604

⬇️ Download Now!

ABSTRACT

The application of face recognition through public surveillance cameras in criminal

investigations is non-intrusive, inconspicuous and faster compared with fingerprint

or DNA-sampling biometric evaluation methods. It assists the law enforcement

agency during the preliminary investigation to narrow down the candidate list of

suspects or person-of-interest. In this project, the fusion between Gabor filters and

Maximum Response (MR) filters with Random Forests classifier is proposed for face

recognition in criminal investigation. The proposed method differs from the

algorithms such as deep neural networks in which the deep neural networks require

larger training datasets. However, it is difficult to obtain large volume of the face

images of the same individuals in reality. Deep neural networks are also more

computation exhausting. The Gabor and MR filters are the facial features extractor.

The Gabor filters are the hybrid of the Gabor magnitude filters and oriented Gabor

phase congruency (OGPC) filters. Gabor magnitude filters produce the magnitude

response while the OGPC filters produce the phase response of Gabor filters. The

MR filters produce the edge- and bar-anisotropic filter responses and isotropic filter

responses.

The variable selection using Monte Carlo Uninformative Variable

Elimination Partial Least Squares Regression (MC-UVE-PLSR) is used to pick the

most useful features in order to save computation costs and time which is crucial in

criminal investigations. Random Forests is used in the classification of the generated

feature vectors. The algorithm is applied in uncontrolled environment where the

unconstrained parameters of facial images such as uncontrolled illumination, pose

and expression variations commonly present in the tape recorded by surveillance

cameras during criminal investigation. The algorithm performance is evaluated

using two unconstrained facial image databases: Labelled Faces in the Wild and

Unconstrained Facial Images (UFI). The images of the databases include different

illumination, face expressions and pose variations. The implemented method

achieved 81.28% and 67.33% of recognition rates and 97.07% and 93.06% of

Receiver Operating Characteristics (ROC) curve on LFW and UFI databases.

TABLE OF CONTENTS

DECLARATION iii

APPROVAL FOR SUBMISSION iv

ACKNOWLEDGEMENTS vi

ABSTRACT vii

TABLE OF CONTENTS viii

LIST OF TABLES xii

LIST OF FIGURES xiii

LIST OF SYMBOLS / ABBREVIATIONS xvi

LIST OF APPENDICES xvii

CHAPTER

1 INTRODUCTION 1

1.1 General Introduction 1

1.2 Importance of the Study 1

1.3 Problem Statement 2

1.4 Aims and Objectives 3

1.5 Scope and Limitation of the Study 4

1.6 Contribution of the Study 4

1.7 Outline of the Report 4

2 LITERATURE REVIEW 6

2.1 Introduction 6

2.1.1 Face Recognition System 6

2.1.2 Face Detection Method 10

2.1.3 Feature Extraction 12

2.1.4 Feature Selection 19

2.2 Face Recognition 21

2.2.1 Subspace-based Method 21

2.2.2 Learning-based Method 25

2.3 Summary 29

3 METHODOLOGY AND WORK PLAN 30

3.1 Introduction 30

3.2 Databases 32

3.2.1 Labelled Faces in the Wild (LFW) 32

3.2.2 Unconstrained Facial Images (UFI) 34

3.3 Image Pre-Processing 35

3.4 Feature Extraction 36

3.4.1 Gabor Filters 36

3.4.2 Oriented Gabor Phase Congruency Filters 39

3.4.3 Maximum Response Filters 40

3.5 Feature Selection 42

3.5.1 Monte Carlo Uninformative Variable Elimination

Partial Least Squares Regression (MC-UVE-PLSR) 42

3.6 Classification and Prediction Using Learning Framework 43

3.6.1 Random Forests 44

3.7 Fusion of Gabor Filters and Maximum Response Filters 47

3.8 Software 49

4 RESULTS AND DISCUSSIONS 50

4.1 Introduction 50

4.2 The Pre-Processing Phase 50

4.2.1 Face Detection 50

4.2.2 Conversion into Grayscaled Format 51

4.2.3 Tan-Triggs Illumination Normalization 51

4.3 Feature Extraction 53

4.3.1 Gabor Filters 53

4.3.2 Maximum Response (MR) Filters 57

4.4 Feature Selection 58

4.5 Classification using Random Forests 61

4.6 Fusion of the Random Forest Prediction Scores of Gabor

Magnitude, OGPC and MR filters 64

4.7 Finalization of Parameters 65

4.8 Comparison with Other Existing Algorithms 68

4.8.1 The Use of Cross Validation in the Evaluation of

Algorithms 68

4.8.2 Labelled Faces in the Wild 70

4.8.3 Unconstrained Facial Images (UFI) 78

4.8.4 Summary of the Comparison of Algorithms using LFW

and UFI Databases 85

5 CONCLUSIONS AND RECOMMENDATIONS 86

5.1 Conclusions 86

5.2 Challenges and Recommendations for future work 86

REFERENCES

📄 Pages: 115 🧠 Words: 13111 📚 Chapters: 5 🗂️️ For: PROJECT

👁️‍🗨️️️ Views: 604

⬇️ Download Now!