Automatic Extraction Of Local Features In Face Recognition Applications

The research purpose of this thesis is to develop a system for face automatic recognition. The research is primarily involved three model blocks: face detection and localization, the pick-up of feature, matching and identification.This three model blocks are the foundations and cores of face automatic recognition system.This paper gives a comprehensive survey of the theory, literature, face library and relative products in the domain of face recognition combined withexpatiation of the perception laws. Then we established a small scaled face library------theOF_L.Based on the Support Vector Machine (SVM) technique, we developed a practical real-time face detection system. By introducing the idea and classical theory of Local Feature Analysis (LFA), we presented the improvement LFA (IFLA) method. According to the local feature points aggregates which are picked-up by the ILFA method, we experiment on several matching methods: computing degree of matching directly by block matching method, Singular Value Decomposition (SVD) method base on LFA, eight neighborhood vector method, the matching method of the local feature points accumulated inside block. It has been proved by practice that the above methods are proper and feasible. Finally, by consulting the OpenCV library, we established our face automatic recognition system programming with Matlab mixed with C++.