Research On Optimization Of Intra Coding Algorithm For Screen Content Video Based On VVC

The new generation of video coding standard VVC(Versatile Video Coding)includes a series of screen content video coding tools,such as IBC(Intra Block Copy),PLT(Palette)and ACT(Adaptive Color Transform),etc.,to improve the encoding performance of screen content video.However,the introduction of multiple new tools significantly increases the coding complexity.In this paper,the fast intra-frame coding algorithm for screen content video is studied,and two new algorithms are proposed to reduce the computational complexity by making fast decisions on coding tools and adaptive selection of the reference region fo IBC tools.The main contents and innovation points are as follows:(1)The fast coding mode decision technology based on machine learning is studied,and a fast intra mode decision algorithm is proposed based on feature crossing for screen content video.By introducing the idea of feature crossover,the algorithm improves the accuracy of screen content CU(Coding Unit)and natural content CU classification,and then accelerates the mode decision-making process by assigning different coding modes to different types of CUs.According to the characteristics of screen content video,the algorithm designs numerical features and category features for classification.The Adaptive Factorization Network(AFN)is used to cross-feature the class features and construct features with stronger discriminating ability.The algorithm improves the structure and loss function of AFN network,and further improves the CU classification ability.Experimental results show that under the All Intra(AI)configuration,compared with the VTM10.0reference model,the proposed algorithm reduces the encoding time by an average of29.32%,and the BD-rate increases by only 1.76%.(2)The paper studies the expansion problem of the IBC reference region,and finds that the performance of the extended reference region is related to the data characteristics of the region: for the area with similar graphic structure concentration,the extended reference region can effectively improve the IBC coding performance,and the increase in coding time is not obvious;For areas with similar graph structures scattered,the performance gain from expanding the reference region is not significant and increases encoding time.Based on this discovery,an adaptive IBC reference region selection algorithm is proposed.Firstly,the algorithm uses the improved AFN network in(1)to determine the coding mode of the current CU.Then,for CUs that select IBC mode,the perception hash operator is used to detect the similarity between the 32×32 region where the CU is located and the 32×32 region outside the original reference region of IBC.Finally,the reference area of IBC is judged according to the similarity.Experimental results show that under the AI configuration,compared with VTM10.0 with expanded IBC reference region,the proposed algorithm saves an average of 11% of coding time,and the BD-rate increases by 0.8%.