Study On Nonlocal Image Denoising Method Based On Markov Chain Monte Carlo Sampling

Image is one of the most important information form. At present, the image processing technology has been widely used in military, aerospace, medical and health, digital entertainment, security monitoring and automation of industrial and agricultural production management, etc. In all application areas, the images with high quality is of great significance for subsequent analysis and treatment. But due to the limitation of resolution of the acquisition equipment, acquisition environment, The noises of equipments, Communication conditions, Images are polluted by noise in the process of acquisition, transmission, and record, which result in radiometric distortion and greatly reduces the interpretability of image target information. For the purpose of analysis and using of the original image, it is necessary for us to denoise the original image first of all.In order to overcome the high time complexity of non-local denonising method in choosing similar blocks, and make use of structures between similar patches collected to have better performance in noise removing, image details protecting and visual quality, in this thesis, we do some researching on some basic but important problems in nonlocal image denosing. Some novel method as well as corresponding efficient algorithms are proposed. The main work is summarized as follows.1.Combining nonlocal Markov chain Monte Carlo sampling and low-rank approximation of matrix method, an approach for image noise removal is presented in this paper. The cluster of similar patches is searched by using Markov chain Monte Carlo sampling. The cluster matrix of similar patches is decomposed by singular value decomposition method, and the image noise is suppressed by applying the low rank structure from decomposing. The simulation results show that the proposed method outperforms the Block Method of 3-D dimension(BM3D)and the nonlocal means (NLM)method in computational complexity. The proposed method has a better performance in protecting image details compared with the NLM method, and has some advantages over the BM3D method in terms of visual quality.2. A novel image denoising method is proposed by using non-local approximation of low-rank based on random projection. The cluster of similar patch for each pixel point is found by using methods of non-local searching, and then compute low-rank approximation of matrix corresponding to the cluster of similar patches using two-side random projection. Finally, the image noise is suppressed by using the Low rank structure. Results show that the proposed method have the low computation cost. Comparing with one-side random projection method, the proposed algorithm ensure lower reconstruction error,and comparing with BM3D method, proposed method have appealing visual quality of images.3. A new stochastic nonlocal denoising method based on adaptive patch-size is presented. The quality of restored image is improved by choosing the optimal nonlocal similar patch-size for each site of image individually. The method contains two phase. The first phase is to search the similar patches based adaptive patch-size. The second phase is to design the denoising algorithm by making use of similar image patches obtained in the first step. The multiple clusters of similar patches for each pixel point are searched by using Markov-chain Monte Carlo sampling many times. Following, we adjust the patch-size according to the consistency of multiple clusters.This processing is repeated until we obtain the optimal patch-size and corresponding optimal patch cluster. We get the estimation of noise-free patch cluster by employing modified two-directional non-local (TDNL) method. Furthermore, the denoised image is obtained by using the method of superposition approach. The theoretical analysis and simulation results show that the method is feasible and effective.4. A divide-and-conque image denoising method based on stochastic technique is proposed in the fifth chapter of thesis. The procedure is divided into two phases:the appropriate random sampling strategy is adopted to search for similar patches, then the original image is estimated by these patches. Specifically, in order to reduce the sampling rejection rate, the observed image is decomposed into different frequency bands by 2D wavelet transform, then the similar patches are collected by alterable direction Markov-Chain Monte Carlo (MCMC) sampling with a properly chosen rejection criterion. Rather than taking the weighted average of similar patches, we use two-directional non-local (TDNL) method in order to take full use of the similarity between similar patches collected. The simulation results show that the proposed method improves the efficiency of searching similar patches. Compared with the NLM and BM3D method, our approach has lower computational complexity, better performance in protecting image details and higher visual quality, respectively.