VVC Fast ME Algorithm Based on Spatial Texture Features and Time Correlation

• Published in: 2021 International Conference on Digital Society and Intelligent Systems (DSInS) pp. 371 - 377
• Main Authors: Guan, Xiaohan, Sun, Xiaoyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.12.2021
• Subjects: affine; AMVR; Decision making; Encoding; H.266; ITU; Machine learning; Machine learning algorithms; Motion compensation; MTT; Prediction algorithms; Prewitt
• DOI: 10.1109/DSInS54396.2021.9670606

Faced with the application of 5G technology, the demand for high-quality video transmission is more urgent. The H.26X series coding framework led by the ITU (International Telecommunication Union) launched a new generation of H.266 encoders in July 2020 in response to emerging scenarios. While bringing average bit rate savings, it also brings high computational costs. The study found that the affine motion mode produced huge computational complexity in the new framework. The fast ME algorithm studied in this subject terminated the division of some blocks ahead of time, corrected the best prediction mode of the sub-CU, and based on the time domain. Features use affine skip mode for blocks that do not require affine prediction, thereby reducing RDO calculations. In the C and D test sequences, the coding time was saved by an average of 15.50%, resulting in an average increase in BD-RATE of 0.502%.