Research On Image Denoising Based On Nonlocal Means And Regularization Model

After entering the information age,image has become an important source of infor-mation acquisition and transmission,and plays an important role in human life.How-ever,in real life,images will inevitably be disturbed by various noises in the process of acquisition and transmission,resulting in a significant reduction in image quality.The processing to solve this problem is called image denoising.Image denoising is a basic problem in the field of computer vision and image processing.It is very important for subsequent image processing and image application and can guarantee people to get more accurate and effective information of images.The key of image denoising is how to keep the structure and details of the image while removing the noise.Over the years,a lot of research efforts have been invested toward the development of efficient methods for this problem.Among the existing image denoising methods,nonlocal means(NLM)shows great potential in image denoising.At present,the research on NLM mainly focuses on the following two aspects:1)designing better similarity measures;2)establishing a regu-larization framework related to NLM methods.On the one hand,this thesis profoundly analyzes some drawbacks of NLM and proposes an improved NLM algorithm.The proposed algorithm obtains better image denoising results than the original method and several existing NLM methods.On the other hand,from a mathematical point of view,image denoising is an ill-posed problem,and one of the effective ways to solve the problem is a method that bases on regularization.That is,according to the hypothesis which is based on the images,via a large number of prior information to construct the regularization term to generate a function model,which has its unique advantages.However there are also some shortcomings in this method,such as step effect and tex-ture protection.Because the NLM method solves the staircasing effect to a certain extent,this thesis fully excavates the prior information of the image,such as piecewise smoothness(gradient-based prior),nonlocal self-similarity and low-rank characteristic-s,and makes a thorough research on regularization model for image denoising.In addi-tion,since the most advanced low-rank algorithms rely on an iteration step to boost the denoising performance,this thesis also proposes a low-rank image denoising method based on adaptive boosting technique.This technology can ensure that more details are retained while removing noise.The main research contents in this thesis are outlined asfollows:1.The drawbacks of NLM method in similarity measure and parameter selec-tion are analyzed intensively,and an image denoising method based on edge similarity measure and adaptive parameter selection is proposed.The main contributions of the proposed method are as follows:1)An anti-noise difference operator is designed to improve the accuracy of edge detection;2)A new similarity measure method,which combines edge similarity and neighborhood similarity,is proposed;3)The filter param-eters used in weight function are adaptively determined by using the image structure as constraints;4)A two-stage denoising scheme is adopted to optimize the results.The experimental results show that compared with the original NLM method and other ex-isting NLM algorithms,the proposed algorithm can effectively preserve the fine texture structure and achieve the state-of-the-art results in terms of both PSNR and SSIM.2.This thesis profoundly analyze the regularization denoising model based on im-age priors.Aiming at the problem that some existing methods tend to smooth the fine image textures while suppressing noise,degrading the image visual quality,an adap-tive texture-preserving denoising method is proposed.Mainly,our contributions are as follows:1)Two types of priors(gradient histogram matching priors and nonlocal self-similarity priors)are proposed,and their combination is used for image denoising;2)We introduce a hyper-Laplacian distribution of the gradient histogram matching prior,which enforces the gradient histogram of the denoised image to be as close as possible to the estimated reference histogram from the original image.Meanwhile,the pro-posed model obtained by introducing the NSS priors effectively solves the staircasing effect;3)To improve the accuracy of the method,a content-adaptive parameter selec-tion scheme based on edge detection filters is proposed;4)Because the optimization problem of objective function is a non-convex optimization problem,a new numerical solution based on augmented Lagrangian multiplier and alternating direction multipli-er method(ADMM)is proposed.Experimental results demonstrate that the proposed method effectively preserves the texture features of the denoised images and outper-forms several variational methods and other state-of-the-art methods in terms of various evaluation indices and visual quality,especially at medium and high noise levels.3.In this thesis,we make in-depth research on low-rank approximation denoising model.It is found that there is a specific functional relationship about singular values between the original image and a series of noisy images,which can be used to construct the singular values of a noise-free image.Based on the above theory,a novel denoising method based on low-rank approximation theory is proposed.The major contributions of the proposed method are as follows:1)Estimating the noise energy distribution of the group matrix in the singular value decomposition(SVD)domain by using the energy characteristics of the image with different noise levels.The energy distribution of the noise is shrunk to obtain the energy distribution of the true signal;2)Based on the optimal energy compaction property of SVD,the low-rank property of matrix is constrained in the SVD domain to obtain the low-rank approximation of the matrix;3)A new noise standard deviation estimation approach is proposed for the iterative boosting process,which effectively optimizes the denoising effect during the iteration.Experimental results show that the proposed method efficiently decreases the noise and achieves comparable denoising performance to the state-of-the-art methods regarding both quantitative measurement and visual effect.4.In order to further improve the performance of low-rank based denoising method,a low-rank image denoising method based on adaptive boosting technique is proposed on the basis of improving the most advanced weighted nuclear norm minimization(WNNM)method.The contributions of this method are summarized as follows:1)To further improve low-rank denoising effect,an adaptive boosting technique is proposed.Considering the availability of previous denoised image to strengthen the signal,we prove the advantage of adaptive boosting approach from the statistical analysis of boost-ing techniques.Then,in each iteration of the boosting scheme,the dynamic boosting parameters are motivated by optimal solution analysis,which ensures the convergence of the algorithm;2)In the process of image denoising,an adaptive patch search scheme is put forward to gain an effective similar patches index;3)Aiming at the number of iterations,a stopping criterion with correlation coefficient is applied to adaptively deter-mine the optimal number of iterations.Experimental results demonstrate the proposed approach can preserve more detail information while removing noise and outperform various state-of-the-art denoising algorithms in terms of both PSNR and SSIM.