The Fast Super Block Partitioning Algorithm Based on Depth Residual Network
Wenmin Wang, Senke Yang, Ming Cheng · 2023
The superblock partitioning process based on rate distortion optimization in VP9 is analyzed in depth, and a fast algorithm for superblock partitioning based on deep residual network is proposed. Firstly, we perform statistical analysis on the division depth distribution of superblocks, and conclude that superblock partitioning has a high correlation with image content. Then, the superblock partitioning process is modeled as a classification problem, and a hierarchical deep residual network is designed to predict superblock partitioning. An effective neural network model is obtained through offline training. Experimental results show that under the premise of quality loss within an acceptable range, the method in this chapter can effectively reduce the complexity of VP9 coding.