An Intra Fast Coding Algorithm For HEVC Screen Content Coding Based On Decision Tree

The wide application of screen content video in computers,mobile displays,etc.has made the research of Screen Content Coding(SCC)attract much attention.As an extended version of the new High Efficiency Video Coding(HEVC),the new technology of SCC for screen content characteristics improves the video coding performance.Meanwhile,The new tools of SCC makes the entire coding complexity higher,which affects the real-time application of SCC seriously.Therefore,it is a hot topic that how to reduce the coding complexity of HEVC SCC in current research.This thesis mainly optimizes the HEVC SCC intra prediction coding to reduce the coding complexity.The main work done specifically is as follows:Aiming at the CU partitioning process of intra prediction coding in screen content,this thesis proposed a fast CU partitioning algorithm based on decision tree to reduce the coding complexity by reducing the CU depth traversal layer by fast CU decision.Firstly,the thesis expounds the complexity of CU partitioning and some influencing factors,and analyzes the feasibility of the algorithm.Secondly,the thesis summarizes the video sequence characteristics,and selects some frames from the standard test sequences for the training set of decision tree training,and then analyzes the proposed feature values,and extracts the original data,and generates decision tree model for offline training by calling function.Finally,different eigenvalues are calculated according to different sizes of CUs at the current layer,and different decision tree models are called for fast CU partitioning and termination.By the method of reducing the CU calculation,the intra prediction coding process of the screen content coding is accelerated,and the coding complexity is reduced.Experiments show that,compared with the standard algorithm of SCM7.0,the algorithm can reduce the coding time by 23.33% when the average bit rate is increased by 0.81% and the PSNR is reduced by 0.044 dB.Aiming at the high complexity of PU modes selection for intra prediction of screen content coding,This thesis focuses on reducing the coding complexity from two aspects: reducing candidate lists and fast termination and skipping or specific modes.Firstly,the distribution characteristics of PU modes in different video classes is analyzed.Then,a fast decision tree algorithm for PU modes selection is proposed based on the CU partition of decision tree.On the one hand,by improving the original eigenvalues,the offline model is trained to make decisions for different sizes of PUs.The way to classify candidate patterns reduces the number of candidate sets,thereby reducing the coding complexity.On the other hand,the fast termination and skipping of characteristic mode for PU which fails to produce accurate results can accelerate the prediction coding process.Compared with the SCM7.0 standard algorithm,the algorithm can save 17.51% coding time while ensuring that the video quality is basically unchanged.Finally,for the two fast algorithms proposed in this thesis,the joint simulation is performed,and the experimental result is analyzed to ensure the effectiveness of the integrated algorithm.