Research On Optimization Algorithm For Super-Resolution In VVC Coding In-Loop Filtering

The new generation of multifunctional video coding H.266/VVC technology is mainly for high-definition or ultra-high-definition video.The encoding process is still carried out on a block-based framework,which is the same as H.265/HEVC,and various distortions will be generated during the encoding process.In H.266,three loop filters are used to reduce these distortions,but the processing of HD and UHD videos is extremely complicated,because the default loop post-processing module in VVC lacks optimization for this type of video,so the paper Aiming at the problem of loop filtering,a new filtering structure and parallel optimization algorithm are designed.The main innovations and work content are as follows:First,the loop filter in VVC encoding takes up a lot of storage space when processing high-definition or ultra-high-definition video,and the processing effect is not good at low bitrates.The paper studies the information that directly or indirectly affects the distortion in the video.Combined with the super-resolution reconstruction network that has a good effect in the image restoration direction,a brand-new filter is designed,and the rate distortion optimization at the CTU level is used to determine whether to replace the first filter deblocking filter and the second filter in the VVC.Filter samples are adaptively compensated,and a loop filter optimization algorithm based on super-resolution reconstruction network is proposed.Use the CU and QP in the video sequence as the auxiliary input of the network,combined with the improved attention mechanism module and the residual module.Through continuous network training and structural optimization and adjustment,as well as combined use scenarios,the network model parameters and framework structure used are finally determined.Experimental results show that compared with the latest VTM-12.0 reference model,the BD-rates of Y,U,and V are reduced by 3.30%,12.11%,and 12.93% on average,and PSNR is improved by 0.6?1d B.Second,in view of the relatively complex calculation of loop filter structure in VVC encoding,the paper studies the characteristics of GPU parallel computing combined with the parallel encoding framework in VVC,and proposes a GPU-based parallel optimization algorithm for loop filter.According to the order of the three filter structures in the loop filter and the encoding characteristics,the algorithm selects the steps that require a large number of parallel operations and puts them on the GPU side for processing,and the rest is still processed on the CPU side,and the encoding is performed in CTU units.Parallel computing acceleration and computing performance are significantly improved.Experimental results show that,compared with the latest VTM-12.0 reference model,the proposed algorithm can reduce coding time by 36.8%and save BD-rate by 0.19%.