Complexity Reduction of Multi-frame Motion Estimation in H.264

Abstract

H.264 video coding standard gains significant improvement in compression efficiency. However, the computational complexity also considerably increases. Many fast algorithms are introduced to speed up the encoding. Distinguishing from pervious multi-frame based fast motion estimation algorithms which only exploit temporal correlation, a novel algorithm both employing temporal and spatial correlations is proposed in this paper, which has much better prediction accuracy. To further speed up the encoding process, an adaptive mode reduction scheme which will discard small probability modes is also proposed. Combining the novel fast multi-frame based motion estimation algorithm and the adaptive mode reduction scheme, experimental results show that the proposed algorithms improve the encoding speed by an average of 9 times faster than fast full search and about 2.5 times faster than reference method which only employs temporal correlation, PSNR dropping and bitrate increasing are negligible.