A Fast Affine Motion Estimation Algorithm Based on Motion Vector Difference

作者：Mengmeng Zhang, Mingkang Wu, Fei Gao, Qing Li, Zhi Liu · 年份：2023 · DOI：10.1145/3604078.3604139 · 被引用次数：1 · 研究领域：Video Coding and Compression Technologies、Advanced Vision and Imaging、Advanced Image Processing Techniques

In Versatile Video Coding (VVC), Affine Motion Estimation (AME) mode has been introduced for inter frame motion estimation, to improve the accuracy in predicting complex motion. Meanwhile, it brings high computational complexity in deciding AME mode. In this paper, a fast affine motion estimation algorithm based on Motion Vector Difference (MVD) is proposed to solve this problem. Through the statistical analysis, it is observed that there exists notable correlation between the MVD value of the current CU and the skipping affine mode. Therefore, one metric based on MVD of TME (Translational Motion Estimation) is defined to guide the decision of skipping affine mode, which is also used to help SKIP type coding block to skip affine mode, where SKIP type coding block means that its parent CU or neighboring CU is Skip mode in this algorithm. The other metric based on CPMVD (Control Point Motion Vector Difference) of four parameter affine model is defined to skip only six parameter affine mode. The experimental results show that compared with VTM10.0, the total time complexity of the proposed algorithm in Random Access (RA) mode is reduced by 9.77%, the total affine motion estimation time complexity is reduced by 40.65%, and the BDBR coding loss is only 0.07%. In Low Delay B (LDB) mode, the total coding time is reduced by 18.64%, the total affine motion estimation time is reduced by 51.54%, and the BD-BR coding loss is only 0.22%.