Face Image Recognition System

In this thesis the questions about computer face image recognition are investigated. The objective is to implement a general face image recognition system and it is expected, by doing this, to provide some indices for opening out the human being visual mechanism, reasoning mechanism and give some helps for artificial intelligence research. It is supposed that there exists each face image for each person in the data bases, however the face images to be recognized are diversified. They may suffer from lighting, scale, rotation, deformation etc.Firstly, the image segmentation is discussed. It is known the location of the face in a image can be gotten by detecting the face contour, then the face region can be segmented out of a image. To detect the face contour a general deformable model is put forward and is used to represent various face contour. At the same time the internal energy function and external energy function of the face contour model are constructed. The Hough transform is used to get the initial location of the face contour in the image. Followed by this a local matching means is used to determine the finial location of the face contour in the image.Principal component analysis (PCA) method can be used to extract the principal component of the sample sets and the eigenvectors gotten can compactly represent the each sample individual. In addition, PCA can be used to recognize face image and present good performance for same kind of face image, however for different kind of face image, e.g., images are changed by lighting, scale, rotation and deformation etc., then the performance is bad. In the thesis, the reason is given based on the deeply study of the principle of PCA method. According to the same reason various face image space models can be obtained and be used to determine the scale and rotation pose of face image to be recognized.Also, PCA method is used to analyze the local features of face images, and some local feature models can be constructed. A feature detection algorithm, locating step by step, is presented. It can accurately detect the local...