LAUSR

Coding tree unit (CTU) partition technique is one of the most advanced techniques, which devotes to the excellent performance of High Efficiency Video Coding (HEVC). However, the enhancement of coding performance is at the expense of increased coding complexity. To reduce the complexity of HEVC intracoding, a fast CTU depth decision algorithm based on texture features and convolutional neural network (CNN) classification technique is proposed herein. First, the relationship between texture complexity and coding unit depth is explored. Based on this, CTUs are divided into simple CTUs and complex CTUs in line with their texture complexity, which are limited to different depth ranges. Then, the CNN for HEVC intradepth range (HIDR-CNN) decision-making is proposed, which is used for CTU classification and depth range restriction. Finally, the optimal CTU partition is achieved by recursive rate-distortion cost calculation in the depth range. Experimental results show that the proposed algorithm can yield average 27.54% encoding time reduction with 0.99% BDBR gain or 0.05 dB BDPSNR loss compared with HM 16.9. The proposed algorithm contributes to promote HEVC coding efficiency under real-time environments.