Research On Intra Prediction Of H.266/VVC

The development of video compression technology has always been accompanied by a contradiction:on the one hand,the explosive growth of video data requires video standards to further compress information redundancy and improve compression efficiency by continuously improving coding tools and adding prediction modes;On the other hand,the increase of computing complexity with the increase of video compression efficiency also needs to introduce corresponding fast algorithms to further alleviate the pressure of computing resources.Aiming at the above two problems,this paper proposes two optimization algorithms to optimize the coding performance of intra prediction:1.A fast mode decision algorithm aiming at optimizing the intra prediction complexity.In order to improve the accuracy of angle prediction,the intra prediction modes in H.266/VVC are extended to 67 based on H.265/HEVC,resulting in a sharp increase in the computational complexity in the mode decision-making process.According to the algorithm flow characteristics and complexity distribution of intra prediction mode decision-making process,an improved mode fast decision-making algorithm based on image texture features is proposed to reduce the number of candidate modes in rough mode decision process(RMD)and rate distortion optimization process(RDO)and reduce the complexity.Experimental results show that the algorithm can save 18.321%computing time on VTM 10.0.2.Cross component intra prediction model aiming at improving intra prediction accuracy.In order to eliminate the component redundancy in YCbCr color space and improve the performance of chrominance component intra prediction coding,an improved cross component intra prediction model is proposed as a new chrominance prediction mode to participate in mode decision-making.Firstly,two CNN branches are used to obtain the spatial information of luminance component and the information of luminance chrominance cross component respectively.Then,the attention module senet is introduced to record the contribution of each adjacent reference sample when calculating the predicted value of each chroma pixel,selectively enhance the beneficial features and suppress the useless features;Finally,the full connected network is used to generate the final chroma prediction value.Experimental results show that the algorithm reduces the BD-rate of YUV component by 0.668%,4.356%and 4.768%respectively.