Fast CU patition based on image similarity using neural network

作者：Cao, Yinglie; Wu, Wenjin; Zhou, Zhiheng^*; Xu, Haoqi; Yue, Wanlin; Zhuge, Shang

来源：MULTIMEDIA TOOLS AND APPLICATIONS, 2023.

DOI：10.1007/s11042-023-16962-x

• HEVC
• CU partition
• Image similarity
• Neural network

High Efficiency Video Coding (HEVC) has introduced a quad-tree (QT) based coding unit (CU) partition structure, which has significantly improved the compression performance compared with Advanced Video Coding (AVC). However, the use of rate-distortion optimization (RDO) techniques in the search for the optimal CU partition has increased the encoding complexity of the video. In this paper, we propose a fast CU partitioning algorithm based on image similarity, which makes decisions on the partitioning of the parent CU by comparing the similarity of the image content of four sub-CUs. We propose four different neural networks based on this algorithm, and experimental results demonstrate that our proposed network structure reduces encoding time by 59.8%, 58.7%, 58.5%, and 59.3% respectively, while increasing the Bjontegaard delta bit-rate (BDBR) by 2.32%, 1.99%, 1.82%, and 1.91%, respectively.