Research On Optimization Algorithm For Motion Vector Prediction And Multi-type Tree Partition In VVC

Compared with High Efficiency Video Coding(HEVC),the compression performance of Versatile Video Coding(VVC)has been further improved.The bit rate is reduced by 30% ? 50%,and has significant compression efficiency improvement on4 K and 8K ultra-high definition videos.Due to the introduction of a large number of new coding tools,the time complexity of VVC encoder increases greatly.To reduce the time complexity of inter coding process in VVC,based on spatio-temporal correlation of Coding Unit(CU),the fast algorithm of extended Merge prediction and Multi-Type Tree partition decision is studied in this thesis.The main contents and innovations are as follows:(1)With the optimization of Merge tools in VVC,the prediction efficiency of Motion Vector(MV)has been improved.However,the computational complexity of Merge tools is increased,either.Based on the statistical analysis of Skip mode,an early Skip mode decision algorithm based on CU motion features is proposed.It is found in our work that in the inter coding mode,the probability of choosing Skip mode as the best prediction mode is much high,and this decision has a great correlation with the intensity of movement of the coding block,and the relatively static block has a high probability to use Skip mode.A three-frame difference scheme is designed to judge the motion status of the current CU.The block with mild motion status is determined to use Skip mode in advance,and the specific Skip mode(Affine?Skip or Merge?Skip)of such block is determined using the Sum of Absolute Transformed Difference(SATD)cost in rough selection process.For CUs that is not determined to use Skip mode in the previous step,the proposed algorithm uses spatial correlation to determine whether it is Affine Merge mode or regular Merge mode.For other CUs,the standard decision process is applied.The experimental results show that compared with VTM-11.2,the coding time of the proposed algorithm decreases by 22.36% in Random Access configuration,with BD-rate increasing by 1.08% and PSNR decreasing by 0.04 d B.(2)The correlation of the partition decision of Multi-Type Tree with the motion intensity and texture characteristics of coding blocks is studied,and a fast Multi-Type Tree partition algorithm based on spatio-temporal correlation for inter prediction is proposed.In the inter prediction process,the coding blocks with intense motion or complex texture are usually further partitioned,and the relatively static background areas with simple texture are often represented by large blocks.The proposed algorithm uses Average Sum of Square Difference(ASSD)and gradient amplitude to describe the motion intensity and texture complexity of the coding block,respectively.For the CU with relatively static motion status and simple texture,the Multi-Type Tree partition process is terminated in advance.For the CU that needs to be further divided,the gradient direction is used to predict the division direction and to reduce unnecessary partition modes.The experimental results show that compared with VTM-11.2,the coding time of the proposed algorithm decreases by 25.06% in Random Access configuration,with BD-rate increasing by 1.23% and PSNR decreasing by 0.03 dB.