Research On Image Restoration Algorithm Based On Sparse Representation Method

Modern society,the digital information has developed vigorously.Digital images as an important way of storaging and transmitting visual information have penetrated into all aspects of life.In the whole process of actual acquisiting to applicating,due to the impact of object displacement,imaging equipment and transmission environment,people often obtained the distorted images which can't meet the life and research needs.Therefore,it is still an important work to recover the original clear image from existing distorted image.In recent years,sparse representation theory has won favor and received a good effect in a variety of image processing problems.In this paper,based on the sparse representation theory,the group based sparse representaion model is selected for it has not only expressed the sparsity but also expressed the nonlocal similarity of the image.In addition the gradient histogram preserving regularization is added to the model stated above.Then the algorithm based on the conbination of the iterative histogram specification algorithm and inexact Lagrange multiplier method is used.Finally the improved model and algorithm are been used to serve the image denoising and image deblurring problem.The main work of this paper is as follows:(1)The prior models of image,the sparse representation theory and the related algorithms are introduced emphatically.Additionlly,This paper introduces the traditional algorithm of solving sparse representation coefficients and learning dictionary training in sparse representation model with image patch as sparse representation unit.and their advantages and disadvantages are analyzed.(2)Image denoising mehod based on the Group based Sparse Representation is proposed and different algorithms have been used to solve the model,finally the inexact Lagrange multiplier method is selected.Compared with traditional denoising method,the group of similar image patches is the unit of sparse representation and an adaptive dictionary for each group is designed,which has the advantages of high efficiency and low complexity.(3)The group based sparse representation model is researched and the conclusion is that although it can get higher peak signal-to-noise ratio and structural similarity index metric,there is still room for improvement in texture preserving.So the gradient histogram preserving regularization is added.And the iterative histogram specification algorithm is combined with the inexact Lagrangian multiplier method for solving the new model.The experimental results show that the proposed method can get high peak signal-to-noise ratio and similarity of structure,low computational complexity,and can keep the fine texture of image.Finally,the improved method is applied to image deblurring,which fully demonstrates the advantages of the method in preserving the texture effect in the process of image restoration and it is of practical value and practical significance.