The Research And Realization Of Face Recognition In Video Retrival

The thesis makes a series of research on face detection and face recognition, and realizes a system of automatic face recognition. This system first detects faces in the video frame. After preprocess of these faces, feature extraction and class recognition is implemented with key points Gabor-Fisher method. The main research includes the following:1. In face detection, we study the method of face detection based on Adaboost. This method first constructs a series of basic classifiers according to Haar-Like features, then forms a stage classifier by training these basic classifiers with Adaboost, finally cascade a large number of stage classifiers into classfier cascade. The result shows that the method realizes the real-time face detection and has better veracity.2. The thesis finds a new method of eye location. This method first fixes the region of eyebrow and eye with rectangle method, and then locates the accurate eye location with integral projection. We implement geometry normalization and grey level normalization on the basis of eye coordinate.3. In face recognition, We study EigenFace method, Fisher face method, direct sample Gabor-Fisher method and key points Gabor-Fisher method. Compare with EigenFace method and Fisher face method, the Gabor-Fisher method has better tolerance and more accurate recognition with respect to the change of gesture, expression and illumination.At last, the thesis realizes the system of automatic face recognition. Experiment with the Adaboost method, eye location method and four recognition methods, the result indicates that key points Gabor-Fisher method is the best.