3D Realistic Face Reconstruction Based On Orthogonal Images

Building realistic 3D face models has always been a persistent challenging task in the field of computer graphics, image processing and computer vision. It is widely used in virtue reality, movie making, computer games, remote communication, remote education, etc. In this paper, 3D face reconstruction is studied, and significance and worthiness is obvious in application.In the premise of analysis the applications and present situation of 3D face modeling research at home and abroad, the thesis presents the basic methods and key processes of 3D facial modeling. In this method, use improved CANDIDE model as generic 3D face model. Firstly, based on extracting feature information from facial photos, special face model is obtained by modifying generic model. Then texture picture is synthesized by the frontal and profile views of face. It is projected on special face model, thus it looks really. The main content is as follow:(1) Build a generic 3D face model. In this paper, we use CANDIDE-3 mesh model as generic face model, which is made by Image Coding Group of Sweden University. And improve the number of feature points and the size of coverage area. The improved model has simple structure and good reference for construction, which lays good foundation of personalized facial reconstruction.(2) The method of facial feature extraction with facial feature templates is presented. CANDIDE model which has been defined feature points is as templates, and feature information is extracted interactively. This method can identify feature points in special face conveniently and availably.(3) The paper studies the method of deriving the 3D coordinate of the feature point from the 2Dimages. With this feature points, we get a given face model by modifying the generic face model. The modification of generic face model includes two steps, one is modification of feature points, and the other is modification of non-feature points. (4) Analyzing the disadvantage of traditional view-independent facial texture generation method, we apply an approximate view- independent facial texture generation and mapping method. Realistic 3D face model is obtained by OpenGL rendering. Experiments show that this method is simple and the personalized facial model textured is realistic.(5) Based on the object-oriented methods, the 3D face model reconstruction system is built using the tool of Visual C++ and GpenGL with the proposed algorithm. And a nice experimental result is gotten.