Face recognition is a biometric authentication system for human security and personal identification that has become a field of interest in pattern recognition and computer vision societies in recent years as it has become increasingly important and commonly used for legal and personal identification in various fields such as visa information system, access control and multimedia search engines. However, distinct illumination, pose and blurring of facial images have become a big challenge in finding important facial features and facial representation in these fields. Therefore, this thesis proposes a facial recognition framework based on multi-feature selection approach. The framework in this thesis consists of eight stages: face preprocessing, segmentation, detection, cropping, transformation, extraction, classification and verification. The experiments were performed on gray scale frontal facial image with 750 images applied from three different standard facial databases namely BioID, ORL and Yale. In face segmentation, detection and cropping stages, Voronoi Diagram and Delaunay Triangulation methods have been applied. Wavelet transform and moment invariants methods have been used to extract facial image features. All features were fed into Radial Basis Function neural network for classification and verification purposes. The results show that a recognition accuracy rate of more than 92% has been achieved as compared to other proposed methods. Therefore, the framework in this thesis would be beneficial for the field of face authentication or verification due to its robustness and invariance to pose, illumination, and expression.
展开▼
