A study on the motion vector prediction schemes for AVS

The motion vector prediction (MVP) is an important part of video coding. There have been numerous workings on the topic done by researchers before. In this paper, a continue study on MVP of video coding based on the workings of predecessors is made. The video sequences with various motion characteristics are further investigated. The characteristics of motion vectors of objects in video scenes are discussed briefly. Then, summarizing these characteristics, two MVP schemes for a new coding standard, Audio and Video Standard (AVS), are proposed. In these schemes, current block's MV can be predicted based on statistical correlation of MVs of spatial contiguous neighbor blocks. A correlation criterion is employed to measure how correlated between two MVs. With the correlation criterion, the correlated MVs of neighbor blocks are determined. Then, the predicted MV of current block can be obtained with some simple algebraic operations on determined MVs. The two proposed schemes, as the alternative ones of median predictor, are suitable for different video sequences with different motion characteristics, respectively. The experimental results show that the bit rate savings are achieved with these schemes in most of typical video sequences, compared with the median predictor implemented in AVS.