A fast Coding Tree Unit depth Prediction for HEVC

High Efficiency Video Coding (HEVC) standard employs Coding Tree Unit (CTU) quadtree partition technology to improve the compression efficiency significantly, but it also consumes an enormous encoding computational complexity. In this paper, a fast CTU depth prediction algorithm is proposed. First, the spatio-temporal correlation of CTU depths among the current CTU, the spatial neighbor CTUs, and the same located CTU in the reference frame are utilized to predict the depth range of current CTU. Second, the rate-distortion (RD) costs of the spatio-temporal neighbor Coding Units (CUs) and the parent CU are employed to further reduce CU partitions. Experiment results show that the proposed algorithm reduces the encoding time effectively with maintaining a good RD performance as the HEVC test Model (HM). Compared to the state-of-the-art depth prediction algorithm, the proposed algorithm also obtains about 10% more time saving for various test sequences, while has a similar RD performance.