Fast Intra Coding Algorithm Based on Image Texture Analysis for HEVC

Abstract

The newest High Efficiency Video Coding (HEVC) standard adopts a quadtree-based coding unit (CU) partitioning structure with intra prediction modes up to 35, which achieves much better improvement in coding performance, but causes a dramatic increase in computational complexity. Hence, the fast coding algorithm is in great demand for HEVC optimization. In this paper, we propose a fast intra coding algorithm, including a CU size skipping and early splitting termination algorithm, as well as fast modes decision, based on video image texture analysis for HEVC. We first execute edge detection and gradient processing of video images to estimate texture complexity and direction features, which are closely correlated with CU splitting and mode decisions. Subsequently, specific thresholds adaptive to each depth level are selected based on theory analysis and considerable experimentation. Depth checking of CU with probable improper size is skipped. Moreover, we propose a gradient based rough mode selection instead of the original rough mode decision. Therefore, computational complexity of intra prediction is dramatically decreased. Experimental results show that the proposed algorithm achieves a coding time saving of 55% with little Bjontegaard delta rate increase.