Optimization Of Affine Prediction Algorithm For VVC

In order to predict complex motion,affine model was introduced by the general video coding standard VVC(Versatile Video Coding).However,only the 4-parameter and 6-parameter affine models are introduced in VVC,which limits the ability to describe complex motion.Simultaneously,the introduction of affine model brings high computational complexity to VVC.In this thesis,the problem of improving the efficiency and reducing the complexity of the affine model is studied,and two optimization algorithms are proposed.The main contents and innovations are as follows:In order to improve the performance of affine model,an optimization algorithm for affine model based on Control Point Motion Vector(CPMV)is proposed.On the one hand,the affine candidate list constructed by VVC is not comprehensive enough,which can not provide potentially optimal CPMV;On the other hand,only4-parameter and 6-parameter affine models are introduced by VVC,which limits the ability to describe complex motion.Information from partial non-adjacent blocks of the current coding block is introduced to improve the candidate list.Moreover,by extending the CPMV of the affine model,an 8-parameter affine model is introduced,and an affine motion estimation algorithm with low complexity is designed,which includes an 8-parameter affine model,which is designed to enhance the representation of complex motion: The Control Point Motion Vector Difference(CPMVD)of the4-parameter affine model is used to decide whether to skip the calculation of the6-parameter affine model.The 8-parameter affine model is only considered when the cost of the 6-parameter affine model is less than the 4-parameter affine model.Experimental results show that compared with the VTM10.0 reference model,the BD-rate of Y,U and V components in the proposed algorithm can be saved by 0.04%,0.21% and 0.26% respectively in the LDP configuration mode,and the increase in encoding time is basically negligible.In order to reduce the computational complexity of affine motion estimation,a fast affine motion mode decision algorithm based on Motion Vector Difference(MVD)is proposed.According to the MVD value of the current block under the translation model,the algorithm decides whether to skip the subsequent affine motion mode decision process,and the decision threshold value of the coding block is discussed in detail;For coding blocks that preliminary require a decision-making process for executing a 6-parameter affine mode,the CPMVD of 4-parameter affine model is used for a secondary decision on whether to skip the decision-making process of6-parameter affine motion mode.The experimental results show that compared with the existing VTM10.0,the proposed algorithm reduces the total encoding time by14.57% in LDP mode,the total affine encoding time by 55.98%,and the BD-BR only increases by 0.3%,which can be ignored.