Research On Light-compressed Image Coding For Low Delay Video-broadcast Technology

With the development of Internet and multimedia application technology,the demand for low delay video broadcasting technology is more and more urgent.Traditional image compression algorithms,such as H.26 x series and AVS series,belong to deep compression algorithms,which are suitable for the scenes with limited network transmission or storage space and low image quality requirements.Therefore,the standard of image shallow compression is coming out.In 2019,ISO/IEC officially released the JPEG XS codec standard,the compression rate is between 2:1 and 8:1,because JPEG XS is intraframe encoding,without considering the time dimension,so the video transmission delay is greatly reduced,and has become the main shallow compression standard used at present.For JPEG XS,a lightweight,almost lossless visual codec standard,the following research is carried out.Firstly,the implementation principle of JPEG XS standard is studied and the main attention is focused on the image data structure and key step algorithms in the standard,including the image plane division,component de-correlation,transformation,quantization,rate calculation and packaging steps.Then C language is used to complete the implementation of software codec,and finally PSNR(peak signal-to-noise ratio)is used to evaluate the image quality before and after codec.The evaluation result is that the PSNR is above40 d B,indicating that the image quality after decoding is almost lossless.Additionaly,as there is no domestic manufacturer to provide hardware JPEG XS IP at present,in order to speed up the localization process of JPEG XS IP core,we studies the IP implementation of some key algorithms in JPEG XS.The design scheme based on HLS can greatly simplify the design process and improve the design efficiency compared with the traditional pure RTL design.In this paper,the IP design of the four key algorithms in JPEG XS is mainly completed.The reversible component transform(RCT)IP core can convert the input image from RGB space to YUV space.The IP kernel of wavelet transform is divided into three types,which can obtain different sub-band widths respectively,complete the functions of Hal wavelet transform and inverse transform,LEGALL5/3 wavelet transform and inverse transform,etc.