Research On JPEG2000 Image Compression Encoding Technology Based On GPU

Digital image compression technology,as an important mean of information technology is widely used in many fields,which involves such as aerospace,multimedia communications,digital photography,medical imaging and so on.As a new generation image compression standard,comparing with the traditional JPEG standard,the JPEG2000 standard not only can obtain excellent quality restoration image at high compression rate,but also has the advantages of support multi-resolution progressive,region of interest coding(ROI),flexible file format and good fault tolerance.The coding process of JPEG2000 standard is high complex.With the image size and the number of images increasing,the processing images time-consuming coding will become very long.However,we need very high real-time requirement for processing image data in some special applications.But it is difficult to meet real-time requirements by using the traditional image compression algorithm based on CPU.In recent years,with rising and development of Graphics Processing Unit(GPU)general computing technology,GPU is no longer limited to complete graphics rendering computing task.When the CUDA programming language of NVIDIA was released,the programming performance of GPU is improved greatly.We can achieve to accelerate the JPEG2000 image compression algorithm by using powerful computing resources of GPU based on CUDA programming platform,which provide new idea for improving coding efficiency.This thesis firstly introduces the current development and research situation of digital image compression technology,JPEG2000 image compression standard,GPU and CUDA programming language.Subsequently it studies and analyzes the basic theory of JPEG2000 image compression algorithm and the steps of each module operation in detail.According to the characteristics of JPEG2000 algorithm we determined to parallel realize three coding modules for Preprocessing,Discrete Wavelet Transform(DWT)and Tier1.Then the thesis introduces GPU architecture and CUDA programming platform in detail and we determined to adopt the hardware system scheme of CPU+GPU heterogeneous.The CPU is responsible to deal with logic complex work,while GPU complete image data parallel computing work.We accomplished the program development and design of JPEG2000 image compression coding based on GPU by using CUDA programming language.Finally,we tested time-consuming coding of our CUDA program by comparing with Kakadu,Jasper and Openjpeg.Experimental results show that coding speed increased significantly and the maximum speedup achieve 10 approximately after GPU parallel optimization.From the visual and PSNR test results can be seen,the restoration image quality is excellent and the loss image details is very small.