Research On Fast CU Partition Algorithm For HEVC Intra-frame Based On Ensemble Learning

With the development of digital information technology,high-definition and ultra-high-definition video applications have begun to spread all over the society.High-definition video and diversified application scenarios have imposed higher requirements on video encoding compression.In order to meet people's demand for ultra-high definition video,Joint Collaborative Team on Video Coding(JCT-VC)has developed a new generation of video coding standard—H.265/HEVC(High Efficiency Video Coding).Compared with the previous coding standard H.264/AVC,HEVC can save 50% of bitrate under the same coding quality,providing an effective solution for the efficient compression of video.However,HEVC improves coding performance and brings higher coding complexity,such as a flexible quadtree block coding structure,rate-distortion decision technology,etc.,which introduce high computational complexity,which results in the inability to implement real-time encoding and affects the application of HEVC in high-definition video services such as video conferencing and webcasting.Therefore,this dissertation regards the optimization of HEVC intraframe coding as a research topic,analyzes in detail HEVC's characteristic coding technology and the reason of coding complexity.Based on the analysis of the current coding optimization algorithms,two kinds of fast HEVC intraframe coding units based on ensemble learning by studying the excellent algorithmic process of machine learning.For the optimization of the partitioning of the intra-frame coding unit,The main contents and innovations of this paper are as follows:(1)Aiming at the high complexity caused by the recursive partitioning of HEVC intra-frame coding units,a fast partitioning algorithm based on Decision Tree and Support Vector Machine heterogeneous ensemble(DT and SVM ensemble,DT-SVM)is proposed.The partitioning of the coding unit is modeled as a two-class problem.The texture feature statistics are used as feature attributes.The decision tree and the support vector machine are used to simply and efficiently predict whether the current coding unit is divided.If and only if the prediction results are inconsistent,use HEVC's rate distortion technique to make decisions,which is avoiding unnecessary Rate Distortion Cost(RDCost)calculations and reducing the computational complexity of the encoder to some extent.In order to explore the relationship between the texture features and partitioning of coding units,the wrapper(WrapperSubsetEval + Bidirectional Search)strategy was used to evaluate the best feature subsets.(2)Aiming at the disadvantages of the DT-SVM model training and the complexity of HEVC intra-frame coding units,in order to further improve the HEVC encoding speed and coding unit partition prediction accuracy,a fast partition algorithm of HEVC coding units based on Random Forest Classification(RFC)is proposed.The random forest uses filter method(GainRatioAttributeEval + ranking search strategy Ranker)to select the optimal attribute feature,and use it as the basis for determining whether the coding unit is divided.The algorithm process is no longer performed by the RDCost calculation of HEVC.In order to further improve the prediction accuracy of the classifier,a GridSearch method was used to optimize the parameters of the random forest.Experimental results show that the HEVC encoding fast algorithm based on excellent random forest can reduce the coding complexity more effectively.