| dc.description.abstract | In this ever-changing era, with the advancement of the Internet and the development of technology, people will never stop pursuing higher-quality things, and the same is true for high-resolution images. In order to compress these huge videos more efficiently data volume, VVC adopts some more novel technologies, such as rectangular coding tree unit, rate-distortion optimization, etc., but at the same time, it also causes an increase in the complexity of coding calculations. This paper combines the very popular in recent years Deep learning and machine learning, namely convolutional neural networks and random forest classifiers, are applied to VVC coding unit depth decisions. Different from the original VVC recursive operation coding unit rate distortion cost, this paper first uses support vector machine and convolutional neural network to divide the square coding unit blocks at the beginning of coding, and then uses random forest classifier to Subdividing the rectangular coding unit block, the classified block will only be coded once, thereby greatly saving the time required for coding, and then using random forest decision-making to assist the original VVC to filter the prediction mode, reducing the overall calculation to Less than 20%. Subsequently, a three-channel residual neural network architecture is introduced at the decoding end to compensate our distortion at the encoding end with different information. In this way, the concept of distributed video coding is realized, and the fast prediction mode is combined with post-processing at the decoding end to compensate for image quality. Experimental results Compared with VVC, when the overall average BDBR is reduced by 1.63%, the overall side decoding time can be saved by about 51.48%. | en_US |