Generic, scalable and efficient shape coding for visual texture objects in MPEG-4

This paper presents a generic, scalable, efficient shape coding scheme for scalable object-oriented visual texture. The base-layer coding scheme used is similar to the binary CAE coding scheme adopted in MPEG-1 video. The proposed scheme introduces a new set of generic contexts to efficiently encode (predict) enhanced shape layer based on the lower spatial layer using a context-based arithmetic coder. It is not dependent on any specific sub-sampling filters, thus it can generate the exact shape matching the texture decomposed using any wavelet filters. The proposed shape coding scheme was adopted in MPEG-4 Version 2 Standard to enable the visual texture wavelet coding to be a true spatially-scalable object-based texture coding technique. In addition, when operated in macro-block mode, the proposed scheme provides full backward compatibility with the MPEG-3 scalable shape coding for video objects. A simple solution to solve the chroma shape mismatch is also presented and was adopted in MPEG-4 Version 1 visual texture coding part. The results show that the proposed scalable shape coding scheme also achieves significant better coding efficiency than the non-scalable shape coding and the other competing shape coding scheme