Researches On Face Recognition Method Based On Weber Local Descriptors And Sparse Representation

Face recognition is one of biometrics identification technologies based on facial features. Compared to others, it develops faster and faster and attracts more and more attention, and its market share is constantly growing. Most existing face recognition methods can achieve good performance in a controlled environment. However, their accuracies degrade significantly with pool face image acquisition conditions and without user’s complicity. Hence, there still exist challenges in practical applications.The work in this thesis focus on two aspects:facial feature extraction and classification, aiming at the variations in facial expressions, poses, illuminations, and partial occlusions etc. On one hand, a method for face representation based on the Weber Local Descriptor (WLD) is proposed. On the other hand, a Kernel Group Spare Representation Classification (KGSRC) is proposed for face recognition. The main work is described as follows:1. Facial feature extraction based on WLD:The WLD is a2D histogram feature, concatenating its two components:differential excitation and orientation, which can describe details in digital images. In this thesis, the WLD is used to represent facial images and effects of the parameters to recognition performance are discussed. Then, the multiscale analysis is employed to improve its robustness and discriminant power.2. Recognition based on SRC:The SRC has good robustness in face recognition to variations in illumination conditions and partial occlusions. Based on the basic SRC, we combine Group SRC (GSRC) and Kernel SRC (KSRC) to propose a new algorithm, called Kernel Group SRC (KGSRC) in this thesis. In the KGSRC, Block Orthogonal Match Pursuit (BOMP) algorithm is employed to compute the sparse coefficients.A number of experiments over Yale, AR, FERET databases are conducted to testify the effectiveness of the proposed method. Experimental results show that the proposed algorithm is of good robustness to variations in facial pose, expressions, illumination conditions, especially in partial occlusions.