Creating Animatable MPEG4 Face

Realistic face modeling and animation is a challenging topic in computer graphics. In this paper, we develop an approach to create realistic human face with minimum user interaction. The face created is animatable and MPEG-4 compatible. The system takes cylinder mapped texture image and sparse or dense range data, which can be obtained from any available resources, as input. After locating 11 major feature points in the texture image, an automatic feature locating method is applied to get all feature positions. Then the multi-step compactly supported radial basis function (RBF) is used to automatically adapt from a generic face model to a specific face model. If dense range data are available, the face model is refined by Triangular B-spline which takes the RBF adapted face model as its initial value. After that automatic texture mapping is used to obtain a realistic face.