The Research And Implementation Of Face Detection And Recognition

Face recognition has been attended significantly with the improvement of intelligence and the enhancement of security in recent years. It is an active rearch in the field of imgae process and analysis, and can be applied to human ID management and security system and so on.In this thesis, ther are three key technologies dicussed, including face detection, feature extraction and face recognition. The rearch is focus on the human face detetction in this thesis, and the following aspects of work have been done.The methods of face detection based on the skin model are rerearched in this thesis. The difference method is used to get rid of the background image and get the regiones including human face at first. According to the skin color in YCbCr color space binary image is generated by binarization, and the face is be located by the geometrical fuatures. Then, the eyes and mouth are localized, and the face can be localized accurately according to the positons of the human eyes. After searching, reading and analyzing lots of papers, a new algorithm of eye localization based on the chrominance, geometrical features and grayness has been presented. The eye map of chrominance can be generated by this method based on the observation that there are high Cb and low Cr values around the eys. The regiones of eyes is segmented by the map. Non-eye regiones can be deleted according to the geometrical features and skin mask. The eye map of gray also can be established in accordance with low gray values and large changes in the level of gray. The accurate position of eyes can be detected by this map. Experiments show that the feasibility and effectiveness of the method.Additionly, the method of combination of the Independent Component Analysis (ICA) uesd in feature extraction is described in this thesis. Facial features space for a human face whose various mutual components are independent can be formed by FastICA algorithm. The coordinate values of face in the independent facial features space are to be facial characteres. In that case that the samples are less, SVM is selected to train samples. Face recognition is finished by SVM finally.