Fast Intra CU Size Decision for HEVC Based on Machine Learning

High Efficiency Video Coding (HEVC) is the new generation of video coding standard. A quad-tree based Coding Unit (CTU) partitioning scheme is used to adapt to different video contents. However, it brings the dramatically increasing of coding complexity because there are a large amount of CU partition structure to traverse. In this paper, we proposed a fast CU size decision method based on machine learning. CU features is extracted and Support Vector Machine (SVM) model is trained to classify CU splitting or non-splitting. Experimental results show that our proposed method can achieve 40.23% encoding time saving on average and the BD-rate loss is only 0.83% under All Intra (AI) configuration.