Motion Transitive Based Fast Multi-Frame Motion Estimation Algorithm For MPEG-4 AVC/H.264

The complexity of conducting motion estimation in multiple reference frames is much higher than that of a single reference frame. Generally, we know that temporally neighboring frames have strong correlations with each other. In this paper, the interframe motion correlations are used to estimate motion vectors in multiple reference frames mode efficiently. Our method reuses the pre-coded motion vectors without requiring additional information. The proposed method takes only 23.06% execution time as compared with the approach applying the full search in all the reference frames. Meanwhile, the coding efficiency and visual quality of the proposed method are still comparable to those of the full search.