A Novel Approach to Improve Rate-Distortion-Complexity in Versatile Video Coding Standard

Versatile Video Coding (VVC) achieves up to 30% bitrate reduction at the same quality level compared to its predecessor, High Efficiency Video Coding (HEVC).  It could support resolutions from 4K to 16K as well as 360° videos. Some new coding tools, such as AFFINE, Integer Motion Vector (IMV), Decoder-side Motion Vector Refinement (DMVR), and Triangle are proposed for VVC to improve the encoder efficiency. But, these new coding tools usually impose high computational complexity on the encoder side. In this paper, we provide a new approach to reduce the computational complexity of the Rate-Distortion Optimization (RDO) process in the encoder side of VVC. In the proposed approach, first, the effectiveness of each coding tool at various parts of the scene is estimated. The results of the experiments show that some of the coding tools--,i.e., AFFINE and IMV, have much better performance in borderline CTUs. So, the proposed approach suggests considering these coding tools in the RDO process, just for the borderline CTUs. This way the computational complexity is decreased considerably without affecting the coding performance. Simulation results show that by disabling the AFFINE and IMV coding tools in the rate-distortion optimization process of non-borderline CTUs, the encoding gain is reduced by only 0.88% and 0.72% BD-rate, but the processing time is reduced by 11.70% and 63.91%, respectively. As the second approach, the correlation between the various coding tools is investigated. Our simulation results show that the AFFINE and Triangle coding tools are highly correlated to each other. So, in the rate-distortion process, if the encoder decided to disable the AFFINE coding tool, the Triangle coding tool is also can be considered disabled without examining the rate-distortion process for this coding tool. This way, the computational complexity is reduced, by 4.96%, on average, without affecting the encoding gain considerably.