Research On Face Detection And Recognition Algorithm

Face recognition technology is a kind of technology using computer to analyze face images and extract effective identifying information to verify identity or face status. It involves pattern recognition, image processing, computer vision, biology, psychology and many other disciplines of knowledge, and is one of the research hotspot nowadays.Face recognition technology consists of two parts:face detection and face recognition. In this paper, we first introduce the research background, significance and commonly used algorithms for face recognition, of which several kinds of face detection and face recognition algorithms are particularly discussed.The paper studies methods which directed against color images based on skin color segmentation and template matching of face detection, which can be divided into two parts:the color segmentation and template matching. First the skin regions are segmented by the model of skin tone, then, face candidate area could be obtained by mathematical morphology method. Finally we use this face candidate area to accomplish template matching.The paper studies the face recognition methods based on linear subspace. We discussed Eigenface approach, LDA approach, Fisherface approach, null space approach as well as the Small Sample Size problem.Face recognition method based on the spectral regression linear discrimination analysis (SR-LDA), SR-LDA solves the dense matrices eigen-decomposition problem encountered in sub-space method, and converts solving of the LDA Optimal projection vectors problem into solving of the linear equations problem. LDA Optimal projection vector can be evaluated by finding the regression of the linear equations, which avoids the large number of calculations brought by matrices eigen-decomposition. Directed against the problem of sample centralization in SR-LDA method, we propose a modified spectral regression linear discrimination analysis method which can increase the recognition rate of SR-LDA method and reduce the time of feature extraction.