J2K Image Decoding Technology Research Based On GPU

Focusing on image decoding based on GPU and its parallel processing technic, thispaper shows the research of technology of improved JPEG2000(J2K) image-Bayeriamge decoding. Firstly, Bayer image decoding algorithm would be analysised inparallel. Combined with the parallel computation theory, the algorithm was decomposedinto three sections, the data-parallel section, the task-parallel section and the serialsection. Then, the efficiency of the data-parallel section designed to improve by GPUoptimization technology. Based on GPU plaform, an optimization program wasachieved to advanced the data-parallel section under the CUDA framework. Some GPUoptimization technique were used in this program after studying the current GPUtechnology，which mainly include the methods based on parallel model and the methodsbased on the memory model. The parallel model optimization method is stream, whilethe memory model optimization methods are data bandwidth optimization, texturememory optimization, global memory coalesced accessed optimization and sharedmemory optimization. Next, this paper shows the analysis of the task-parallel sectionand the multi-core and multi-thread CPU parallel technique. After that, the task-parallelsection was optimized by the means of OpenMP technique. At the same time, this papershows the research on the optimization of OpenMP, which mainly include avoidingfalse sharing and the optimization of task decomposition and scheduling. In addition,the RAID disk technology and parallel cache model were researched. A RAID diks wasmade up of two SATA disk, whose read-write speed is speedup in2times. With thesetwo technique, a file-save system was set up and the file-save efficiency was advanced.At last, pipeline was got into uese in the system, which made the file-save time washidden. In a word, all these technologies were used to improve the efficiency of theBayer image decoding system. Comparing with the original algorithm, ours excutefaster and could achieve in engineering. For example, Bayer image decoding speed is4times improved than the origin for1Kx1Kx16bits resolution, which meets the demandsof the engineering.