Several key problems in model-based image sequence compression by using interframe AUs correlation

Model-based image coding has received much attention for it could gain a big compression ratio and well form the basis of very low bit rates visual transmissions. Many authors have contributed a lot to this area. However, some key problems are not well solved in practice. In this paper, the human face model and its invariant representation, textural mapping for natural synthesizing of facial images, the occlusion problem in the case of face rotation, real time automatic extraction of face feature points, and a better performance criteria suitable for facial expression analysis/synthesis are discussed. The results obtained by computer simulation are given.<
>