Fast Algorithm of Video Coding Based on AVS3

Abstract

This study proposes a LCU (Large Coding Unit) level fast algorithm of video coding based on AVS3. Nowadays, AVS3 (Audio Video coding Standard 3) has implemented traversal algorithm on CU (Coding Unit) prediction modes. It first computes the current blocks RD cost of inter-frames mode and then compute the RD (Rate Distortion) cost of intra-frames mode. By comparing RD cost, it chooses the best prediction mode. Inter-frames modes include three modes: skip, direct and inter. While intra-frames modes include two: intra and IBC (Intra Block Copy). This method will run different prediction modes algorithm when it encodes one CU and can choose the optimal prediction mode, though with large computation complexity. Based on previous researches, we do more surveys and expand the LCU of AVS3 from 64*64 to 128*128 and add more prediction tools. Therefore, the prediction mode in space express stronger correlation and we present a fast prediction mode algorithm based on apposition LCU. According to apposition LCU image content of adjacent frames in the same video and correlation with optimal prediction mode, we can provide prediction mode reference for the LCU not encoded and reduce the number of traversals. The time complexity is decreased significantly. In our experiment, it shows that under the LDP (Low delay P-frame) model, twelve general test cases reduce 16.67% in time complexity and the performance loss is only 0.61%. With fewer performance loss, we dramatically reduce time complexity and the experiments results is overall satisfying.