Studies Of Automatic Human Face Recognition Based On Modified Feature Location Algorithm

With the development of society and technology, the new demand for computer vision and artificial vision is increasingly urgent. As the most active theme in the computer vision and pattern recognition, auto human-face recognition technology is the research focus with an extensive application foreground.Auto face recognition system is composed of face detection, facial feature location, facial representation and face matching recognition. The research work of this paper focuses on the feature detection and face representation to improve the robustness and efficiency of face representation algorithms. In this thesis, the improvement of the traditional template matching algorithm, based on the face pictures detected by Adboost algorithm, for eye detection is carried out firstly. The improvement on accuracy and rapidity of eyes locating are revealed by three experiments. The next mission of this thesis is the research on the Gabor Bunch Facial Graph for auto face recognition.The main initiative works of this paper are as follows:①As the limitation in the precision of traditional template matching, thesis is proposes new templates definition by extending eye templates to eye-brow model. The modified templates avoid the miss location caused by the high similarity between some eyes and some brows.②The thesis also proposes template matching algorithm for eye locating with template selected by correlation coefficient. The algorithm reduced the redundancy of the template set by taking the correlation coefficient to select the efficient templates, and then, the man-made error in template selection is avoided. Thus the location precision and speed are improved obviously.③This paper take the bunch face graph as the train template for train set, and for test set at the same time. Save the BFG in the system memory. When the system get a new face, the system utilize the BFG to pick up Gabor value of the new face and add it in to total system train set for recognize this new face. Then, the system does not need update the BFG for new face recognition.