An efficient long-term memory motion compensated prediction scheme and application to error-resilient video transmission

This paper presents a long-term memory motion compensated prediction (MCP) scheme for video sequences based on reusing previously computed motion vectors. The proposed method is similar to the long-term memory MCP scheme introduced by Wiegand et al. (1999) in that multiple frames are involved in the prediction; however, instead of having to store all these frames, only the motion vectors associated with these frames are stored and reused for the prediction of the current frame. As the proposed MCP scheme requires less computations and less bits to represent the motion information compared to existing schemes, it is very suitable to provide error resilience to video coders. Simulation results are presented to illustrate the performance of the proposed scheme as well as its application to provide error-resilience in video transmission.