Research On 3D Face Reconstruction And The Key Technology

Face is the most important expressed part of human body, because of its characters: individuation and diversification, so we can distinguish identities and express our feelings by it, and it plays an important role in our daily life. Generally speaking, one face has a complicated geometry and surface, although a lot of problems can't be solved by the 2D method the face recognition. The varieties of facial expression always hinder the recognition rate as a key factor. The rate and robustness of recognition will be improved if we can artificially synthesize the facial expressions and actions by a 3D model, making full use of the information of 3D face. Face reconstruction is used in the area of computer animation as well. In this article, an improved framework of morphable model is proposed based on the study of 3D reconstruction; the preliminary system of face reconstruction is implied, then we pay more attention on the technology of image registration, as a key technology in the face reconstruction. My primary work as following:1. The set of 3d front face and the index informations are built after the operations of origal 3-d face, such as texture projection, texture insinuation and so on. The new front face images need to be preprocessed, eg. The light normalization. Then, they are classified by the age and sex.2. An improved framework of morphable model is proposed as a result of of huge data of the original 3D face model. The main work is: (1) the original data of face model is sampled by right size; (2) the process of reconstruction is optimized by the new framework of 3D face reconstruction.3. We have a analysis on the features of face data, and describe them by the style of computer data structure. Windows API is called to take out the system of 3D face reconstruction as well as some drawing functions in the OpenGL, and the face model can be fast rendered.4. We emphasis on the study of image registration as the core technology of 3D face reconstruction. A fusion algorithm of AdaBoost and Optical flow is proposed to detect and track faces, which uses the algorithm of AdaBoost to detect faces followed by the algorithm of Optical flow for estimating the continuity of video frames and obtaining the position of faces in the frames. Taking advantages of the each algorithm, the effects of the rotation, partial occlusion, expression variation and illumination to face detecting and face tracking are partly solved.5. A new algorithm is proposed for image registration based on AAM and Warp, then the virtual face image set is built by it, which is used as a part of training samples in the process of face recognition. The result of face recognition is improved by the virtual face image set.