Deep Learning Technology to Improve the Coding Efficiency of H.266/VVC

H.266/VVC modifies the quadtree structure of HEVC and adopts the Quadtree with nested multi-type tree (QT-MTT) encoding structure to search for the best encoding unit. Although the QT-MTT encoding structure has better encoding efficiency, it also increases the computational complexity and encoding time. This paper mainly focuses on the QT-MTT structure of H.266/VVC intra-frame coding and proposes the use of convolutional neural networks (CNNs) based on deep learning to prematurely terminate the decision of the horizontal binary tree, horizontal ternary tree, vertical binary tree, or vertical ternary tree of $32\times 32$ coding units, and skip the rate distortion optimization (RDO) step to save encoding time of H.266/VVC. Experiments show that this paper only approximately increases BDBR by 0.45 dB, but can reduce% of encoding time.