Skip to Main Content
In this paper, we extract the facial feature points and their corresponding feature vertices on the face model by hand according to the FDP of MPEG-4, and then improve the traditional radial basis functions (RBF) interpolation algorithm to fit the face model more smoothly and more realistically. In face model texture mapping, a new texture mapping method is proposed, which needs not create full view panoramic image mosaics in advance and avoids sampling them several times. Therefore, it preserves the original information of input face images, which makes the resulting 3D face more realistic and facial details clearer. In addition, when rendering the face model, we partition it into three blocks and make texture mapping for each one with its associated texture, so we render the whole face model only once, which decreases the complexity of computation and huge RAM, and improves face synthesis quality. The experimental results indicate the whole face synthesis process is fast and stable, and the quality of the resulting face is equivalent to the original photographs basically.
Date of Conference: 20-22 Oct. 2004