A Fast CU Depth Estimation Algorithm for HEVC Inter Coding

Abstract

This paper proposes a fast coding unit (CU) depth estimation algorithm based on genetic algorithm (GA). Finding the optimal splitting pattern of each coding tree unit (CTU) from many possibilities is represented as an optimization problem and this splitting pattern is rapidly searched by a low-complexity optimizer. A suitable chromosome structure and an efficient fitness function are firstly proposed. The experimental results show 69.2% computational time on average can reduced by the proposed method compared with HM16.5. Compared with start-of-the-art fast encoding method, the proposed one can achieve 5.2% time saving on average under a comparable BD-PSNR.