Research And Application Of Image Processing In Parallel Programming Method

Our topic background is derived from the lower level processing of satellite ima ge. Suchas template convolution needs much more cycle processing, but at the same time needs toachieve the goal of real-time or near real-time processing speed. This is a real difficultproblem to the huge satellite image pixel size. In this paper, our solution is using of thecurrent mature parallel computing model, especially using of the processing ability of GUP, tohelp the accelerate.In this paper, we research the implementation method of three types parallelprogramming models. They are separately based on multi-core parallel processing model onCPU parallel library-OpenMP, based on multi-threaded processing on GPU parallelmodel-CUDA, and based on one which can do parallel processing across CPU and GPUplarform standard library-OpenCL. Comparing their similarities and differences, anaylisisingtheir own characteristics and the key point of implementation process. And give aspecification of the image processing in three parallel platforms with the flowchart way toachieve the implementation.The experimental in this paper is using different algotithm of image enhancement,imagerestoration,image segmentation applied to three different parallel platforms. With thesummary of experimental data, get the following conclusions: OpenMP model which basedon multi-core can get50%~90%accelerated effects. CUDA model and OpenCL model whichbase on GPU can get several times accelerated effect, some extreme case can get hundreds oftimes accelerated effect.With the real using background requirement, this paper designed a complete verificationprocess. Using a original satellite image which size is about2.5GB to block down, then putthem in the three parallel computing frame to do the midian filtering algorithm. With thecurrent low-end graphics cards, we got the best acclerated effect of19.07times in OpenCLmodel. Fully reflects the significance and pratical value of image parallel processing. Due tothe versatility of the experimental method, it can widely appied to other various imageprocessing areas, so that can contributing to the development of more industry.