Research On Technology Of JPEG Image And AVS Intra Frame Coding

With the development of computer technology and communication technology,digital images and videos as information carriers have penetrated into all aspects of society and have become an indispensable part of our lives.Image and video coding technology has been highly valued by scholars because it can greatly improve the storage and transmission efficiency.As one of the most important academic research directions at present,research on image and video coding is in the ascendant.At the same time,with the progress of the Internet and mobile devices,especially the development of image acquisition equipment,image and video coding technology has become one of the core technologies of the multimedia industry.It can be said that who has achieved the leadership of image and video coding technology,they can also get the lead in the future multimedia application and marketing.So at present,the research of image and video coding technology has important academic and industrial significance.Intra coding technology was first used in the field of image compression.It mainly uses the spatial correlation of images to process ordinary images such as natural images.As one of the most important parts of coding technology,it has been widely used i n image and video compression.On the one hand,with the emergence of new coding technologies represented by deep learning and new coding scenarios represented by VR(Virtual Reality)and Screen Content Coding(SCC),there is a huge space for further improvement of intra coding technology;On the other hand,as the most important image coding standard,the space for JPEG file storaging is large,and the image quality of decoding image of JPEG is low.If the latest intra coding technology can be introduced into JPEG,the above two problems of JPEG can be expected to be solved.In view of this,this paper has carried out a series of research on the application of intra coding technology in JPEG and AVS3.The main contributions of this paper are as follows:Firstly,for the application of intra coding technology in JPEG file recompression,this paper proposes a lossless recompression hybrid coding framework for JPEG images based on transform domain intra prediction,called Lossless Recompressor of JPEG(LLJPEG).LLJPEG attempts to guide the existing intra prediction technology from the pixel domain to the transform domain to meet the requirements of JPEG lossless recompression.What’s more,LLJPEG’s intra prediction,transform,quantization and entropy coding have been redesigned.In LLJPEG,intra prediction is first used to obtain the predicted block,and then DCT transform and quantize the predicted block to obtain the predicted coefficients.Subtract the predicted coefficients from the original coefficients to obtain the DCT coefficient residuals,and finally entropy encode the DCT coeffici ent residuals.The experiment shows that LLJPEG can reduce 29.43% and 26.40% of storage space on Kodak and DIV2 K datasets respectively,while maintaining low decoding complexity.LLJPEG is completely based on the traditional intra coding technology,without any additional hardware requirements(such as GPU),so it can be easily applied to various scenarios.Secondly,for the application of intra coding technology in image enhancement,this paper proposes a JPEG decoding method using nonlinear inverse transform network and progressive recursive residual network,which is called JPEG Decoding Network(JDNet).Due to the low compression ratio of JPEG,its decoded image often has strong compression effect.In view of this,this paper proposes a JPEG decoding fra mework based on CNN(Convolutional Neural Network)-JDNet,which includes a CNN-based inverse transform network(i TNet)and a CNN-based post-processing network(Post-processing Network Utilization Local and Nonlocal Similarities in Multi-scale Space: LNLMS).Specifically,i TNet learns the nonlinear mapping from DCT coefficients to original pixels,so it can reduce the error propagation of i DCT process and obtain more accurate reconstruction.LNLMS uses local and nonlocal correlation in multi-scale space to further improve the quality of decoded images.Experimental results show that JDNet provides better JPEG decoding image quality than the most advanced post-processing methods.Thirdly,for the application of intra coding technology in AVS3,two improved schemes are proposed in this paper.First,AVS3’s Angle based Intra Prediction(AIP)and Intra Block Copy(IBC)are still designed based on the linear copy of surrounding reconstructed pixels,which inevitably limits the performance of intra coding.In view of this,this paper designs a Multi-scale Intra-Prediction Filter(MIF)based on CNN,which is used to extract the nonlinear correlation between the surrounding reconstructed pixels and the current prediction block,and filter the AIP and IBC mode predict ion results of AVS3.The test result shows that the performance of AIP and IBC increased by 1.37% and 2.42% respectively after using MIF m ethod.Second,as an important coding scene in AVS3,screen content coding has large room for improvement in the early version of AVS3 reference software HPM.This paper improves the search candidate generation and motion estimation process of AVS3 SCC,and proposes three improved techniques to further improve the coding performance.The experimental results show that the SCC coding performance of AVS3 has been significantly improved by using the methods of this paper.