A Fast CU Decision Algorithm in Inter Coding Based on Residual Analysis

Inter coding in HEVC can greatly improve video compression efficiency. However it also brings huge computational cost due to adaptive partition of Coding Tree Unit (CTU) with the quadtree technique. In this paper, a fast CU (Coding Unit) decision algorithm is proposed to alleviate the computational burden. This algorithm is fulfilled by analyzing the residual using mean and dispersion. And the optimal RQT (Residual Quad-tree Transform) depth is used innovatively to measure the residual dispersion. First, the optimal RQT depth and avgdis (defined in 3.3) is obtained after inter coding in 2N x 2N PU mode. Then the decision of CU partition is determined through comparing avgdis with the corresponding threshold. Thresholds are predicted based on the distribution of CU partition in the encoded pictures and they can be adaptively changed as the video content changes. Compared to HM13.0 (HEVC test model), the improved algorithm could save about 56% of encoding time on average, with 0.2034% increase of bitrate and the influence on the quality of reconstructed videos is negligible.