Fast CU partition strategy for HEVC intra-frame coding using learning approach via random forests

Abstract

HEVC (High Efficiency Video Coding) achieves cutting edge encoding efficiency and outperforms previous standards, such as the H.264/AVC. One of the key contributions to the improvement is the intra-frame coding that employs abundant coding unit (CU) sizes. However finding the optimal CU size is computationally expensive. To alleviate the intra encoding complexity and facilitate the real-time implementation, we use a machine learning technique: the random forests, for training. Based on off-line training, we propose using the forest classifier to skip or terminate the current CU depth level. In addition, neighboring CU size decisions are utilized to determine the current depth range. Experimental results show that our proposed algorithm can achieve 48.31% time reduction, with 0.80% increase in the Bjantegaard delta bitrate (BD-rate), which are state-of-the-art results compared with all algorithms in the literature.