Fast Intra Candidate Selection and CU Split in Intra Prediction for Future Video Coding

Abstract

The latest video compression reference software Joint Exploration Model (JEM) achieved outperforming performance in intra prediction. It mainly benefits from the increase of intra direction prediction modes, which changed from 33 to 65 and more flexible CU partition structure quadtree plus binary tree (QTBT). However, these technologies caused very high computational complexity at the same time. This paper, proposed a fast intra candidate selection algorithm based on the Sum of Absolute Hadamard Transformed Difference (SATD) and an early quadtree split termination algorithm to reduce the computational complexity in JEM-7.1. Experimental results show that our first proposed algorithm reduces 10% encoding time on average with only 0.5% loss in terms of Bjontegaard delta bit rate (BDBR), and the second algorithm shows up to 21% time saving with 0.6% coding performance loss. These experimental results show the efficiency of our proposed algorithms.