Parameter analysis and synthesis for MPEG4 facial animation

Abstract

In MPEG-4, 68 facial animation parameters (FAP) are defined to animate a facial object. Most of the previous facial animation reconstruction systems were focused on synthesizing natural-looking facial animation from manually or automatically generated FAP, while parameter analysis systems were focused on analysis of 3D motion from a 2D image sequence of a natural video scene. But these are still two separate research fields. A real visual communication system can be established only when these two fields are merged seamlessly. Selection of the subset of FAP and concealment of parameter distortion are the main issues in such communication systems. The extraction of a full set of FAP from image analysis is nearly impossible and unnecessary. Thus a practical selection of a sufficient subset of FAP is essential for an analysis-synthesis MPEG-4 visual communication system. On the other hand, the distortion of parameters extracted from a natural video scene by video analysis always exists. A concealment algorithm for such parameter distortions in facial animation synthesis is proposed in this paper.