Abstract
The new video coding standard, H.264/AVC, has achieved significant compression gain by utilizing several advanced techniques. Motion estimation (ME) is one of the most important elements to reduce the motion residual. However it results in heavy computational burden. In this paper, a novel fast ME search algorithm, which exposits the correlation character between predicted and optimal motion vectors (MVs), is presented for H.264/AVC. According to motion vector prediction, search window is adaptively divided into several regions with calculation of its motion correlation probability. Specifically, it combines MVs' direction correlation and locality correlation. A novel Priority Region List, which indicates search process and selects the number of search points, is dynamically updated. The simulation results show that the proposed algorithm reduces 23% of motion estimation encoding time and 60% of average search points while maintaining the same RD performance compared with that of UMHexagonS algorithm, which is adopted in H.264 reference software. © 2012 IEEE.