Deep Learned Fast H.266/VVC Inter-Coding with Background Analysis

VVC significantly improves coding efficiency through advanced encoding techniques, achieving 30 % to 50 % bitrate savings compared to the previous standard, HEVC. Among all the techniques, the Quad-Tree and Multi-Type Tree (QTMTT) partition structure plays a crucial role. In addition to QT partitioning, QTMT adopts Binary Partitioning and Ternary Partitioning, enhancing the appropriateness of block partitioning for scene information. VVC also provides new techniques in inter-frame coding to further reduce temporal redundancy, with affine motion estimation (AME) providing better predictive capabilities for rotational motion. While these techniques improve coding efficiency, they also bring a significant amount of computational time. To address this issue, we proposed a fast inter-frame partition algorithm to improve the following issues: (1) Design CNN models to predict the suitable partition mode for the current Coding Unit (CU) to by-pass low-probability modes (vertical and horizontal) during the RDO test. (2) Use statistical analysis to find the correlation between motion compensation and static areas in a frame, determining whether the current region should undergo a No-partition test. These proposed techniques aim to achieve an overall acceleration of the encoding process.