Efficient AVS3 Intra Prediction Hardware Design for Real-time Applications

Abstract

a hardware-efficient hybrid greedy CU (coding unit) partition algorithm for AVS3 intra prediction, which has advantages over the traditional regression algorithm on both scheduling complexity and resource consumption, is presented. Compared with the NVidia hardware acceleration of HEVC, the proposed algorithm achieves 21% performance improvement on AI (all-intra) configuration for UHD 4K video encoding.