Efficient multi-frame motion estimation algorithms for MPEG-4 AVC/JVT/H.264

Abstract

In H.264 standard, a lot of computational complexity will be consumed in the encoder for motion estimation. It allows seven block sizes to perform the motion/ compensation, and refers to previous five frames for searching the best motion vector to achieve lower bitrate and higher quality. Since the H.264 reference software uses the full search scheme to obtain the best performance, it will spend a lot of searching time. In this paper, we propose efficient searching algorithms by reuse of the motion vector information from the last reference frame. The proposed algorithms use the stored motion vectors to compose the current motion vector without performing the full search in each reference frame. Therefore, our proposed algorithms can obtain the speed up ratio 4 in average for encoding, which benefits from the prediction of the motion vector for reference frames in advance and maintain the good performance. Any fast search algorithm can be utilized to further largely reduce the computational load for the motion estimation from previous one frame.