Research On Fast Intra Mode And Coding Block Partitioning Decision Algorithms For VVC

In recent years,video has gradually become an important medium for people to socialize and receive information due to its intuitive perception effects.A variety of video applications have participated in our lives,such as high dynamic range(HDR)video,ultra-high definition(UHD)video and screen content video(SCV).Video is developing towards high resolution,high frame rate and high dynamic range,which not only puts pressure on storage and transmission,but also makes efficient video coding methods become a research hotspot.To achieve higher coding efficiency,the Versatile Video Coding(VVC)standard adopts a series of new coding techniques on the basis of the High Efficiency Video Coding(HEVC),such as the more flexible quadtree plus multi-type tree(QTMT)structure,intra block copy(IBC),intra sub-partitions(ISP)and multiple-reference line(MRL)prediction,etc.However,this makes the intra coding more complicated,as it needs to traverse all the partition types and coding tools to find the optimal combination.This dissertation has carried out exploration and research on the new features brought by the new coding tools in VVC and proposed several low-complexity coding algorithms from two aspects of mode selection and partition terminating.The major contributions of the dissertation are summarized as follows:1)We propose an adaptive mode pruning algorithm and design corresponding decision models for the newly introduced coding tools(IBC and ISP).IBC and ISP modes are not applicable to all coding units(CU).To adaptively remove redundant prediction modes based on the CU attributes,this dissertation explores the distribution characteristics of IBC and ISP modes,and designs mode decision models and corresponding decision mechanism based on massive data.To improve the classification accuracy,novel coding features are designed.Experimental results show that our algorithm achieves nearly 20% time savings under the All-Intra configuration,while the coding loss is only 0.23%.2)An ensemble decision strategy is proposed to sort the candidate modes after the rough mode selection,and a rate-distortion terminating model is constructed to terminate the prediction of redundant candidates.Experiments prove that the distribution of the optimal mode in the candidate list is unreasonable,and the probability that candidates selected as the optimal mode in the candidate list is not strictly decreasing in the order of the modes.To solve this problem,the ensemble decision strategy is proposed to calculate the probability of each candidate being selected as the best mode.It integrates several factors related to the optimal mode distribution,including the Hadamard cost,MRL prediction and spatial information.Experimental results demonstrate that this method achieves nearly 12% coding time savings with negligible performance loss.3)A mode-dependent QTMT termination algorithm is proposed to solve the problem of low prediction accuracy of CU partition types.It is found through experiments that different optimal modes will lead to different termination probabilities of QTMT partition.This dissertation pre-classifies CU based on the relationship between the optimal mode,CU texture and partition termination mechanism.Then different early termination models are designed for different CU types to evaluate whether the partition can be terminated at the current CU level.Experimental results show that this method achieves 33% complexity reduction with negligible coding efficiency loss.