HEVC coding-unit decision algorithm using tree-block classification and statistical data analysis

We propose a fast coding unit (CU) depth decision algorithm in the High Efficiency Video Coding (HEVC) procedure based on statistical analysis. First, we derive a set of optimized weights of surrounding CU decisions to predict the current CU decision for 3 different Largest Coding Unit (LCU) classes. Second, for a given predicted current CU decision, we analyze the possible true current CU decisions, aiming to find the correspondence. A corresponding table is found and can be used to achieve target prediction accuracy. Third, for early termination of the encoding processes, the 3 early termination methods in a state-of-the-art work, as well as their different combinations, are evaluated. We show that using one of them is sufficient for saving time while encoding to keep the implementation complexity low. Compared with full CU search in HEVC standards, the proposed method reduces the encoding time by 57 and 49 % on average with Low Delay and Random Access profiles, respectively, with acceptable bitrate and PSNR performances. Compared with two state-of-the-art methods, the encoding time reduction is up to 23 and 13 % with Low Delay profile, 7 and 3 % with Random Access profile, on average, whereas the performances of bitrate and PSNR are similar.