A fast CU depth decision algorithm based on learning how to use machines

摘要

High Efficiency Video Coding (HEVC) is the latest video coding standard, which has been proposed and developed by the Joint Collaborative Team on Video Coding (JVT-VC). With the evolution of the HEVC test model (HM), plenty of efficient coding tools are integrated into HM, and the HEVC outperforms H.264/AVC by almost 50% bitrate reduction for an equal perceptual video performance. However, it imposes a great deal of computational complexity on the encoder because of the optimization, especially with regard to the Rate Distortion Optimization (RDO) process. In this paper, we propose a fast Coding Unit (CU) depth decision algorithm to alleviate the heavy computational burden of the encoder, which attains a better trade-off between the Rate-Distortion (RD) performance and coding complexity. Then, a Support Vector Machine (SVM) is applied to training the data points, which has a better prediction of the CU depth decision. Experimental results indicate that under the "low delay" configuration, our proposed algorithm can achieve about 49.15% complexity reduction on an average with only a 2.35% BDBR increase when compared to the original HM12.0.