Development of feature extraction techniques for face recognition

Abstract

As one of the most successful application in image analysis and understanding, face recognition has recently received significant attention, especially during last few years. The first reason behind using face recognition is its wide range of commercial and law enforcement applications. Second reason is availability of feasible technologies after few years of research. In research of face recognition, we always want to achieve the correct classification rate according to the characteristics required. Feature extraction greatly affects the design and performance of the classifier, and it is one of the core issues of face recognition research. As an important component of pattern recognition, feature extraction has been paid close attention by many scholars, and currently has become one of the research hot spots in the field of pattern recognition. This thesis presents a discussion of feature extraction. This Ph. D. Thesis presents new approaches to Automatic Face Recognition using concepts of Information theory and Neural networks. This research work has been focused in new approaches to improve the robustness of Automatic Face Recognition Systems taking into account different variations of face images of each individual. We divide the proposed methods into statistical methods, information theory based methods and spatial-frequency based methods viz Chi-Square test, Principal Component Analysis, Entropy, Mutual Information and Discrete Wavelet transform. In order to evaluate proposed techniques, database consisting of high number of individuals and rich in variations (in the facial expression and pose) among images of each individual are used. MATLAB software is used to implement the different feature extraction techniques. From MATLAB, the toolboxes used are Image processing, Statistics, Wavelet and Neural network. The advantages and limitations of every method are studied. A combination of information theory and statistical technique is proposed viz.