新一代通用视频编码标准 H.266/VVC: 现状与发展

Compared with the previous-generation standard, the new-generation versatile video coding standard H.266/VVC saves about 50% of the bit rate given the same quality and applies to a wide range of video application scenarios. The status quo, implementation, and application development of H.266/VVC are discussed in this paper from the perspective of its key technologies. H.266/VVC retains the dual-layer bitstream structure and hybrid coding framework of the previous standard while introducing technological innovations to all the major coding modules such as intra-frame prediction, inter-frame prediction, transformation, quantization, and loop filtering. Moreover, it provides efficient specialized coding tools for applications such as screen content videos. Currently, the H.266/VVC standard is in a practical stage, with the official reference software VTM and the open-source codecs VVenC/VVdeC serving as the most prominent software codec implementations. An analysis of the performance of H.266/VVC reveals that it achieves more notable coding gains for high-resolution videos. While some of the main coding tools contribute to improved performance, they may also increase complexity. Nevertheless, certain coding tools manage to enhance coding performance while reducing overall coding complexity. The hardware implementation of H.266/VVC faces many challenges, and its development lags behind software implementation. Following the release of H.266/VVC, the development of a next-generation video coding standard still focuses on the hybrid coding framework. There are two main directions: Enhanced compression in Beyond VVC concentrates on more advanced, non-neural-network-based coding tools, while neural-network-based video coding explores the use of neural-network-based coding tools. Furthermore, there is rapid progress in the development of end-to-end video coding that partially or completely deviates from the existing hybrid coding framework. In the future, the combination of video coding standards with neural networks will become a trend while facing the challenges of computational resource dependence and keeping a stable structure.