The Algorithum Based On Global And Non-global Patch Matching For Image Denoisign

Images,as an important medium for humans to obtain visual information,play a vital role in medical,military and other fields.However,images are often polluted by noise in the process of image acquisition and transmission due to the influence of various factors such as equipment and humans.Noise will not only reduce the image quality,but also affect subsequent image processing such as segmentation.Therefore,image denoising has always been the focus of researchers at home and abroad.In recent years,researchers have proposed many low-rank denoising algorithms based on non-local similarities(NSS)patch priors.However,how to represent the local structure of the image and select the similar patch groups have always been difficult and key.Based on the existing low-rank approximation algorithm,this dissertation focuses on the above problems,the kernel wiener filtering algorithm based on low-rank approximation and the low-rank algorithm based on singular value decomposition(GMMLR)are proposed.The specific research work has the following aspects:(1)(External patch guided internal clustering for image denoising,PCLR)used the clean image priors to form non-local similar patch groups by guiding the clustering of noise patches,but confused the noise blocks and similar patches.To solve this problem,a kernel function is used to represent the local complex structure of images and by minimizing the mean square error between the noisy image and the denoised image to further remove noise in the paper.Experimental results show that the improved algorithm achieve better denoising performance.(2)Local block matching can avoid noise interference,but the selection range is small,global block matching can select an adequate number of similar blocks while it is greatly affected by noise.Therefore,this paper combines the advantages of global and local block matching,using the preprocessed image as an approximate solution of the original image by(Block matching 3D algorithm,BM3D)to guide the noise block to generate non-local similar block groups.In addition,to solve the nuclear norm minimization problems,the soft threshold operator is used.It only considers non-local singular values greater than the threshold,but ignores image information contained in the singular values less than the threshold.All non-zero singular will be re-estimated in this paper.Experimental results demonstrate that the algorithm can better recover edge and detailed information.In this paper,a large number of grayscale images are simulated and the best parameter values are selected.And the objective evaluation and visual effect are compared with other top denoising algorithms.The all simulation experiment results show that the proposed method is superior to other related advanced denoising algorithms in this paper,especially for high-level noise.