Research On Fast Algorithm For H.266/VVC Screen Content Coding

With the development of distance education,video conference,screen sharing and other applications,screen content coding has become a research hotspot.Considering that the screen content contains repeated local patterns,sharp edges and few overall colors,the new generation video coding standard H.266/VVC(Versatile Video Coding)introduces the screen content coding tools to improve the coding quality of screen content.However,the application of these coding tools leads to the high complexity of H.266/VVC screen content coding.Starting from the screen content coding structure and coding tools of H.266/VVC,this thesis proposes two fast algorithms for screen content coding based on H.266/VVC,which greatly reduces the coding time.In order to reduce the coding complexity of screen content in intra prediction,this thesis proposes a fast intra prediction algorithm combining fast Coding Unit(CU)partition,improved Intra Block Copy(IBC)mode and improved Intra mode.Firstly,CUs are divided into natural content CUs and screen content CUs according to the gradient information and luminance information of CU.Then,unnecessary CU partition modes are skipped according to the texture characteristics of coding units.Next,the hash search scope of IBC mode is extended moderately to improve the probability of finding matching blocks in hash search.Finally,the prediction range of Intra mode is reduced according to the gradient information of CU.Experimental results show that,compared with the original algorithm of H.266/VVC,the proposed algorithm can save52.14% coding time and 2.66% coding bit rate under all intra coding configuration,while keeping the encoded video quality unaffected.In order to save the coding time of screen content coding in inter prediction,this thesis proposes a fast inter prediction algorithm of screen content coding.Firstly,by using the temporal correlation between the frames,a fast decision algorithm of CU partition modes based on motion vectors is proposed,and a fast CU partition method based on texture features is used as a supplement.Then,according to the relationship between the distribution of residual coefficients and the performance of prediction modes,this thesis proposes a fast selection method of prediction modes based on normal distribution test.At the same time,in the process of traversing each prediction mode,the hash search range of the IBC mode is still moderately expanded,and the probability of local search being adopted is reduced.Experimental results show that compared with the original algorithm,the proposed algorithm saves 19.96% and 19.78%of the coding time under low delay and random access configurations respectively,while keeping the video quality basically lossless and the bit rate slightly reduced.