Robust Video Coding Based on Tree Structure Motion Compensation

One inherent problem of network video is the sensitivity of the compressed bit-stream to data loss. Error propagation is one of the most important factors that affect the quality of network video. This paper discusses a new scheme for robust video coding to cope with this problem. First, error propagation in Block Motion Compensation (BMC) in H.263 is analyzed using a Markov chain model. Several observations are made from a numerical evaluation of the Macro-Block (MB) error probability. Since a high degree of correlation exists across multiple frames in typical videoconferencing sequences, a tree-structure motion estimation and compensation (TMEC) scheme is proposed based on the above observations. This structure reduces the length of error propagation and implicitly increases the MB intra-refresh rate. Simulations show that the new scheme has better performance than that of traditional H.263 encoder in dealing with error propagation.