ABSTRACT
Research in computer vision and machine learning is a significant part of research in computer science departments of many leading institutions resulting in ideas and products that have direct applications in different industries such as medical image segmentation in the medical industry, and face recognition and tracking in the entertainment and security industry.
Face recognition is a significant part of research in computer vision and machine learning and has a wide range of applications in security, human computer interaction and artificial intelligence in general. The main goal of this thesis was to build a code repository to facilitate research in computer vision and machine learning at The African University of Science and Technology, Abuja. Our work concentrated on implementing some statistical shape and appearance algorithms used in face recognition research.
We trained an appearance model and active shape models for an experiment in face verification. We evaluated the use of parameters from the appearance model for face verification using four very common metrics: Mahalanobis distance, Euclidean distance, normalized correlation and Manhattan distance.
Our results showed that normalized correlation performed least while there was very little difference in the performance of the others.
TABLE OF CONTENTS
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1 Introduction 1
1.1 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Background and Literature Review 4
2.1 Statistical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Active Shape Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Active Appearance Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Differences between AAM and ASM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6 Linear Discriminant Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.7 Distance and Similarity measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7.1 Mahalanobis Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7.2 Euclidean Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7.3 Manhattan Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7.4 Normalized Correlation . . . . . . . . . . . . . . . . . . . . . . . 9
2.8 Face Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
v
3 Active Shape Models 11
3.1 Landmarked Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Aligning the Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Modeling Shape Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Modeling Local Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.5 Multi-resolution ASM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Appearance Model 19
4.1 Modeling the Texture variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Combining the shape and texture model . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Fitting the Appearance Models Using Active Shape Model . . . . . . . . . . . 23
5 Face Recognition 26
5.1 The Face Verification Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 Evaluation of the Face Verification System . . . . . . . . . . . . . . . . . . . . . . . 27
5.3 Decision Threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
6 Experiments 31
6.1 Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.1.1 MUCT Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.1.2 Model Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.1.3 Partitioning Data for Verification Experiment . . . . . 32
6.2 Face verification experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.3 Experiment: Quantitative comparison of the results of the use of four metric (Mahalanobis, Euclidean, Manhattan, Normalized Correlation). . . . . . 34
6.3.1 Learning the Global thresholds (Mahalanobis, Euclidean, Manhattan, Normalized Correlation . . . . . . . . . . . . . . . . . 34
6.3.2 Error Rate Comparison (Mahalanobis, Euclidean, Manhattan, Normalized Correlation) . . . . . . . . . . . . . . . . . . . . . . . 37
6.4 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
CHAPTER ONE
INTRODUCTION
Research in computer vision and machine learning is a significant part of research in leading computer science departments worldwide. It has led to many breakthroughs in both academic research and commercial applications.
A currently thriving area of research in computer vision and machine learning is face recognition. It was described in [30] as one of the most successful applications of image analysis and understanding, stating two reasons for the strong research efforts in this area as the wide range of applications it provides and the availability of the technology to support the research.
A study of some of the leading institutions in this field and in other fields in computer science reveals each has a thriving code repository which has been built over the years by researchers and is available to new researchers to build upon thus speeding up research work.
Examples include VisionX of Cornell University Vision and Image Analysis Group1. FSL, of the Analysis Group, FMRIB, Oxford, UK2 and STAIR Vision Library (SVL)3 developed by a Stanford PhD student for research initially to support the Stanford AI robot project.
One goal of the computer vision and machine learning (CVML) group at The Africa University of Science and Technology, Abuja is to build its own code repository from ground up to facilitate research within the group. Many of the algorithms implemented for the repository are not freely available elsewhere or at least not in the organized form implemented in the repository.
Our main goal for this thesis was to build a code repository for the computer vision and machine learning group focusing on research in face recognition and computer animation, and to utilize the code base to perform some experiments in face verification.
The experiments performed evaluated the use of Mahalanobis distance, Euclidean distance, Manhattan distance and normalized correlation as metrics for face verification using parameters
2
obtained from an appearance model. These measures were chosen because they are the most commonly used.
This chapter briefly describes the initial design of the code repository, the scope covered for this thesis, our contributions and the layout of the report.
1.1 Scope
The figure below shows the proposed framework for the code repository for development during this thesis work.
Figure 1.1: CVML Code repository initial framework
3
Our tasks in this thesis can be summarized in these steps.
1. Develop the following models
a. Texture Model
b. Appearance Model (APM)
c. Active Shape Model (ASM)
2. Experiment with the ASM and the APM for face verification.
1.2 Contributions
The following are the contributions made towards the goal of this thesis
1. Wrote code for building Texture model
2. Wrote code for building Appearance model
3. Contributed code for building Active Shape Model
4. Used the repository code for doing verification experiments with ASMs and APMs
1.3 Layout
The remaining chapters of this thesis are organized in the following way:
Chapter 2 discusses the literature on some background concepts needed for proper understanding of the work done in the thesis, as well as some of the mathematical techniques used and different approaches to face detection and face recognition.
Chapter 3 describes the process involved in building the active shape model.
Chapter 4 describes building appearance models and how to use an ASM to interpret an image.
Chapter 5 explains our face verification experiment framework.
Chapter 6 describes the experiments performed and discusses the results obtained.
Chapter 7 summarizes the achievements of the thesis, the challenges faced and future work.
IF YOU CAN'T FIND YOUR TOPIC, CLICK HERE TO HIRE A WRITER»
