Parallel Technology Based On K-SVD Image Denoising Algorithm In General Computing Platform

The digital image is inevitable contaminated by noise due to the influence of the performance of the imaging equipment,the environmental conditions and the human factors in the process of acquisition and transmission.Therefore,image denoising is an important research direction in image processing.In recent years,the high-speed increase of digital image data not only promote the change of people's lifestyle,but also makes great challenges to the digital image processing technology.There is an increasing demand for high resolution image.The real-time requirement of image processing algorithm is getting higher and higher.Ordinary serial algorithm is difficult to give consideration to the processing speed and processing results.With the rapid development of multi-core computing platform,this problem can be solved by designing parallel algorithm based on multi-core platform.In recent years,sparse representation theory has been widely studied.The sparse representation of the image is a good way to extract the essential features of the image and it can express the image in a more concise way.In this paper,the sparse representation denoising algorithm and parallel algorithm optimization are studied.First of all,the knowledge of multi-core processor platform is introduced in this paper.Secondly,the sparse representation denoising theory is studied,and the basic theory of sparse representation and the image denoising model based on the theory are analyzed in detail.The sparse decomposition algorithm OMP and dictionary update algorithm K-SVD are mainly introduced.And according to parallel algorithm theory the serial sparse denoising algorithm is transformed to the parallel version.Not one atom but multiple atoms updated simultaneously is the biggest change in the dictionary update part.After that the convergence of the algorithm is analyzed.In the sparse reconstruction part,data parallelism model is used to transform the algorithm,the multi block images are updated at the same time.Then the results are averaged to obtain the denoised image.Finally,four kinds of international standard test images and multi band solar images are used to test the effect of denoising and acceleration performance based on Intel(R)E5 2640v2 16 cores computer and Linux operating system.The test results show that the parallel algorithm can effectively remove the noise and greatly improve the signal to noise ratio.Speedup of the program is 13.563 and the parallel efficiency is 0.848 for a 2160x2560 resolution sun image.Comparing with the parallel denoising algorithm based on GPU at home and abroad,the algorithm in this paper has better acceleration performance.The parallel K-SVD denoising algorithm based on general multi-core CPU in this paper has been successfully applied to the adaptive optical image preprocessing system.This parallel algorithm can fully utilize the multi-core CPU hardware resources,greatly reduce running time and easily to transplant in multi-core platform.