Research On High Speed Data Processing Systerm Of JPEG2000 Based On GPU

At present,the solution based on IBM blade cluster has the disadvantages of large power consumption and inefficient.What's more,it is unsuitable for the development of on-board and miniaturized applications.The development cycle of Tilera based multi-core platform is relatively long and the cost is high.The JPEG2000 standard has been developed from the original JPEG standard.This new generation of image compression standard is more difficult.In additional to a better quality than the original standard image,there are many other advantages.It not only can support the progressive data transfer type,but also can provide support for lossless data compression and lossy data compression.In practical application,the requirements of JPEG2000 standard image compr ession have strong real-time nature.The existing one is based on CPU,plus util ization of high-level language,which has difficulty in meeting with the real-time requirement.Even if it can barely be up to the standard,it will consume lots of labor and material resource,which is a less cost-effective technique.As time goes by,the development of science and technology is also very rapid particularly in computer science.And its development has changed with each passing day.In recent years,a programmable graphics processing unit(GPU)with powerful para llel computation hascome into being.Its features has attracted worldwide attentio n,especially with the birth of the CUDA,completely conquer the weakness of long-term cycle and inflexibility when develop software before.The core of this thesis is based on the principle of RS decoding and C language program to analyze RS implementation of HPC.And using the irrelevance between the RS code set to analyze the parallel implementation of RS decoding,which is one of the emphases of the thesis.JPEG2000 standard adopt entropy encoding EBCOT algorithm.EBCOT,based on code block and bit plane embedded encoding algorithm includes T1 and T2.This thesis studies the JPEG2000 encoding stream organization to realize the parallel processing of T2 part.In addition,T1 decoding contains MQ decoding and T1 decoding arithmetic embedded bit plane decoding.Encoding and decoding processing between blocks are independent.Research based on block parallel high-speed decoding technology is the key to JPEG2000 decoding,and it is one of the crucial techniques ofthe thesis.JPEG2000 decoding uses discrete wavelet transform(IDWT),which is a kind of complicated mathematical transformation,and can achieve multi resolution image points Classification and analysis of image quality.Two-dimensional image after wavelet transform,the pixel value of the original image is transformed into convenient coefficient matrix encoding.Analyzing discrete wavelet image parallel processing and implementing of parallel processing of discrete wavelet transform is one of the pivotal crucial techniques of the thesis.Plenty of research work,software parallel processing research work and hardware and software testing work have been in this thesis.After the two stages of software testing contrast and the experiment,it has proved that through the research of this thesis we can fully achieve the current CPU blade cluster processing function.