Adaptive Motion Vector Prediction Based on Spatiotemporal Correlation

Motion vector prediction (MVP) is an important issue in the hybrid block-based video coding. The more accurate MVP comes with the better coding efficiency in terms of bits reduction for motion vector (MV) and residual. This paper proposes an adaptive spatiotemporal MVP scheme for H.264/AVC. In the proposed scheme, a novel criterion is put forward and the predicted motion vector (PMV) of current block is calculated by either the available MVs of neighboring blocks or the MV of co-located block in the previous frame adaptively. Due to the proposed criterion, the more accurate PMV can be obtained by effectively exploiting the spatial and temporal MV correlation. Experimental results show that the proposed scheme can improve the coding efficiency for various video sequences