Low Complexity Decision Algorithm for CU Partition Based on Image Texture Information and LGBM

In order to solve the problem of insufficient compression efficiency of High-Efficiency Video coding (HEVC) in the current video market, a new generation of Versatile Video Coding (VVC) has been proposed. However, the newly added Multi-Type Tree (MTT) in VVC leads to an increase in coding complexity, so this paper proposes a fast decision algorithm for Coding Unit (CU) based on image information and using Light Gradient Boosting Machine (LGBM) as the classifier for CU fast decision algorithm. This algorithm utilizes the pixel information and gradient information of the encoding unit to make decision and skip unnecessary partitioning methods. Meanwhile, efficient classifiers are utilized for partitioning decisions. And relevant features are extracted and trained for different partitioning decision problems, so as to obtain higher accuracy partitioning methods. The method balances the issues of coding efficiency and coding time. Compared to the VVC reference software (VTM), the method saves an average of 52.99% of coding time and increases BDBR by only 1.50%.