Fast VVC Inter Coding Algorithm Based-On Frequency Domain and Deep Learning

Versatile Video Coding (VVC) has achieved significant improvements in coding efficiency over its predecessor, High Efficiency Video Coding (HEVC), but at the cost of a substantial increase in complexity. In VVC, inter coding is a crucial part of the encoder. Owing to the fact that the Rate-Distortion Optimization (RDO) entails an exhaustive search, the encoding complexity is relatively high. To accelerate this process, this paper proposes a method that combines frequency-domain analysis and deep learning. By constructing a large-scale dataset and designing a region-aware residual network, the proposed method is capable of predicting the coding mode of the Coding Unit (CU). The experimental results indicate that this approach can reduce the encoding time of VVC by 15.17%, while the Bjøntegaard Delta Bit-Rate (BD-BR) only increases by 1.71%.