Research Of Individual Face Modeling Based On Single Image

The model of 3D human face has been wildly applied in film and cartoon, advertising publicity, security certificate, medicine science and other fields. How to obtain 3D face models is particularly important. There are two main ways for reconstructing 3D face models, one is by using some advanced equipment and the other is reconstructing face modes with knowledge of computer graphics.Particular face information can be obtained by using dimensional scanner and some other equipment. The particular information is useful for constructing 3D face models. However, these equipment can not be used widely. Because they are expensive in most case. Reconstructing 3D face models from 2D images is a effective and lower cost way. Face models reconstructed in this way can have certain realistic, although there are some problems needed to be solved. This paper focus on a method for constructing personalized 3D human face models based on a single image.The paper's contribution are summarized as follows:Some related technology, such as face detection and feature points extraction, are used in 3D face reconstruction method based on image(s). These technology will be introduced one by one in this paper. In this paper, the present research of face detection and feature points extraction are described. A method for face detection based on Adaboost and a method for feature points extraction based on ASM are introduced particularly.After the introduction of progress of 3D face reconstruction based on image(s), two bottleneck problems in reconstruction of 3D face from only one image are proposed. Not only depth information of feature points but also enough texture information can not be got from only a frontal image. To deal with lack of texture information, a texture mapping algorithm based on one ring neighborhood of vertexs'and classification of triangular patches is proposed in this paper. Triangular patches are classified according to the number of how many their vertexs can get validate texture coordinators directly. If validate texture coordinators of a vertex can not be obtained directly, the vertex will be processed in different priority depended on the class of its patch. Feasibility and effectiveness of the algorithm is proved by experiment.A 3D face modeling system base on a single image is designed and implemented with Microsoft Visual C++6.0, OpenGL and OpenCV. Some simple expression and animation also be simulated. Presently, human vision is still a primary way to evaluate the effect of a algorithm for 3D face modeling. This paper try to propose a quantized evaluation criterion.