Three Dimensional Face Recognition

Face recognition is one of the most popular research fields at present. Recently, most of the researches on face recognition are based on 2D face images. Because of the influence of illumination, pose variation and expression, the improvement of recognition accuracy of 2D face recognition is greatly impeded. This makes it still difficult to build a robust face recognition system. 3D model holds more rich information than 2D image, so implementing face recognition on 3D face model is one of the effective approaches to tackle the present problems.The thesis mainly focuses on 3D face recognition problem. After extracting face region from the raw scan and necessary preprocessing, feature points are located automatically, and registration is done to transform face models into the same pose and position. Finally precise and effective recognition algorithms are used to achieve high accuracy. The contribution of the thesis is summarized as follows:In the part of "3D face extraction and preprocessing", an approach is proposed to separate head from shoulders based on the shape of the raw scan. Face region is precisely cropped after locating feature points from head. Several noise smoothing methods are discussed in theory and analyzed with experimental results.In the part of "feature points location", a method based on continuous Shape Index and geometric restriction is proposed. An algorithm of "iterative neighboring vertices retrieve" is proposed to tackle the problem of retrieving neighboring vertices when computing continuous Shape Index. A region-labeling algorithm is proposed to solve the problem of labeling connected regions on 3D meshes. Finally the real feature regions are extracted based on a statistical distribution model. Experiment results show that the proposed method is robust against pose variation.In the part of "3D face registration", feature points based pie-registration method is implemented with the help of feature points location method. When there is too much noise, or the feature regions are lost, the feature points based method may fail. In order to tackle this problem, a pre-registration method based on principle axis analysis is proposed. Traditional ICP algorithm is improved by using bucket algorithm to accelerate closest points searching. Experiment results show that the proposed method is robust against noise.In the part of "3D face recognition", existing recognition algorithms are carefullyreviewed and discussed in theory. Methods ofPCA, LDA, ICP and Hausdorff distance are implemented, and their effects and accuracy are exhibited and analyzed with experiment results.