Automatic feature extraction and face synthesis in facial image coding

A real-time method that automatically extracts facial feature points is proposed. Based upon this technique, the authors also present a procedure to encode facial images efficiently. For clarity the proposed method is divided into an analysis part and synthesis part. The analysis part includes face tracking, feature extraction, and pose estimation. All of these procedures are applicable for live video applications. In the synthesis part, a new face image is synthesized using one base image and a small number of feature images. The current performance of feature extraction is about 8 frames per second on a Pentinum II 233 MHz PC with an image size 320/spl times/240.