动态人脸地获取、检测及识别地研究.pdfVIP

Abstract Research on the acquisition、detection and recognition of dynamic faces Abstract Face recognition is one of the most attractive and challenging problems in fields of pattern recognition and computer vision, which has great potential for research and practical application of biometrics. The dissertation reviews the existing theories and algorithms of face detection and recognition. The main contribution of this dissertation is summarized as follows: Firstly, an image acquisition module for Windows is designed, which has the functions of acquisition, capturing, and restoration of dynamic image. Secondly，the algorithms of face detection are presented in details. The application of optical flow estimation of color images and gray images is developed to locate face. Then a template matching algorithm of face detection is suggested based on optical flow. Experimental results indicate the effectiveness of the algorithm. Thirdly, some researches on feature extraction of face recognition have been conducted. Fisher linear discriminate analysis is used to reduce the feature dimensionality. Then, wavelet transform decomposition 、Fourier transform and wavelet packet is implemented to extract the face features respectively. Face recognition is implemented by Nearest Neighbor classifier. The results of experiments show that features extracted on frequency domain can improve the recognition accuracy. Finally, an approach to face verification based on the integration of individual eigenface subspace and SVD is presented in this dissertation, which first use wavelet transform to r