Image Denoising Combining Nonlocal Bi-regularization And Nonlocal H^-1 Norm

Image denoising has always been an important part in the field of image process-ing.In the early stage,the local denoising method based PDE was the main method.In recent years,the nonlocal denoising methods by using the similarity between adjacent or overlapping blocks to restore the unknown pixels have become the focus of research We first study the nonlocal OSV model and its original dual solution,and the Split Bregman algorithm is used for the first time.The numerical algorithm of the method is given in theory,and the experimental results show that the new algorithm has better denoising effect,and can effectively remove texture image noise and reduce the stair-casing effect when denoising the smooth area.Moreover,the new algorithm improve the ability of maintaining image texture and details,and improve the shortcomings of previous algorithms.Then we introduce the bi-regularization nonlocal OSV model.A quadratic regularization term is introduced into the nonlocal OSV model,which fur-ther improves the ability of noise removal and detail preservation.Finally,a nonlocal CEP-H-1 model is proposed.In the new model,the nonlocal total variation and nonlo-cal laplacian regularization term form the new regularization term,and the data fidelity term is replaced by the nonlocal H-1 norm.The validity and properties of the new mod-el have been proved.In the process of solving the model,the Split Bregman method of the model is proposed by combining the minimization of alternating directions with the Split Bregman iteration.The experimental results show that the new model is superior to the nonlocal CEP-L2 model in noise removal,especially in images with rich texture.