A fast CU partition method based on CU depth spatial correlation and RD cost characteristics for HEVC intra coding

摘要

The High Efficiency Video Coding (HEVC) could save half of bitrates than the previous standard AVC/H.264. However, the traverse of quad-tree partition for finding the optimal CU partition structure brings high computational complexity, which greatly restricts its development in real-time video application. In this paper, a fast Coding Unit (CU) partition algorithm is proposed, which exploits early-split and early-termination strategies for CU partition based on spatial depth correlation and Rate-Distortion (RD) Cost Characteristics. The CU's candidate depth range can be estimated by utilizing spatial neighboring CUs' correlation in partition depth. So if the current depth is not in the candidate range, Rough Mode Decision (RMD) and Rate-Distortion Optimization (RDO) could be directly skipped. Moreover, RD cost would be set to an adaptive threshold based on training frames. When a CU's RD cost is smaller than the threshold, quad-tree partition would be early terminated. Numerous experimental results demonstrate that our algorithm can save up to 67.41% and 43.37% on average in encoding time for all intra sequences with 0.0276 decrease in PSNR and 0.6161% increase in encoding bitrates compared with the reference test model (HM).