Research On Fast Coding Algorithm For H.266/VVC

With the rapid development of the Internet,the emergence of a large number of different types of high-resolution video applications has put forward higher requirements for video coding standards.Therefore,researchers began to develop a new video coding standard named H.266 or Versatile Video Coding(VVC).At present,VVC has added new technologies such as Quad-Tree and Multi-type Tree(QTMT)and 67 intra prediction modes to improve coding efficiency of the encoder,but it also greatly increases the coding complexity.Therefore,reducing VVC coding complexity and ensuring its coding performance has very important practical significance.We conduct an in-depth analysis of the decision-making process and block partitioning process of the VVC intra prediction mode,and proposes a fast encoding algorithm for these two processes.The main contributions of this article are as follows:(1)A fast intra-frame coding algorithm based on support vector machine is proposed to quickly distinguish Planar and non-Planar prediction modes,avoid the rate-distortion optimization calculation of multiple intra-frame prediction modes,thereby reducing coding time.Its contributions are:First,a statistical feature based on directional gradient,named Statistics of Oriented Gradient(SOG),is proposed to extract feature information of image.By comparing the F-Score value with various common image features,it is confirmed that the SOG feature is highly distinguishable for intra prediction mode classification and can better represent the feature information of the image.Second,based on the SOG feature,a support vector machine model for decision-making between Planar and non-Planar prediction modes in H.266/VVC is proposed.Specifically,in the offline stage,the SOG feature value of each CU block is extracted as the sample input value,and the Planar mode is selected as the label,and 17 block size SVM models are trained respectively.In the online phase,the SOG feature value of the current CU block to be divided is extracted,and the corresponding trained SVM model is used for rapid prediction to determine whether the current CU block selects the Planar mode.Experimental results show that,compared with VTM5.0,the encoding time of the fast intra-encoding algorithm based on the support vector machine is reduced by 18.01%on average,while the Bj(?)ntegaard Delta Bit Rate(BD-RATE)only increases by 1.32%.(2)A fast block partitioning algorithm based on spatial features is proposed to terminate unnecessary partition mode predictions in advance,which greatly saves coding time.Its contributions are as follows:First,a new spatial feature is proposed,which is proved by experiments to be effective in terminating block division early.Second,a new adaptive threshold determination method is proposed.Specifically,when the intra-frame prediction CU block is divided,it is first detected whether the current frame is the first frame image of the GOP,and if it is,the conventional encoding process is performed,the feature value is recorded,and the threshold is updated.Conversely,if the current mode is the MT division mode,the spatial feature value of the CU block is calculated,and if the threshold condition is met,the current division mode can be terminated early.If the current division mode is not skipped,perform the regular division operation,recurse the above process,and then calculate the Rate-distortion Cost of the current CU division mode,compare it with the current best division mode,and update the current best Partition mode.Experimental results show that the fast block partitioning algorithm based on spatial features reduces the coding time by 43.84%compared with the VTM5.0 standard encoder,while the coding performance is almost no loss.