Research On Image Denoising Method Based On Dictionary Learning

As a kind of effective image analysis tools,sparse representation can use as little description as possible to get as much information as possible,the more sparse the image,the higher the quality of reconstruction,it has been widely applied to visual tracking,super resolution analysis,target tracking,hyperspectral detection,computer vision and image processing field.The essence of image denoising is to extract useful information and suppress useless noise information,at the same time enhancet the ability of protecting the texture and edge details,sparse representation denoising is based on the sparsity of the image itself and the non-sparsity of the noise image.But in the condition of strong noise,can not get the effective training dictionary,the denoising effect is not obvious and easily lead to blurred edges.To solve these problems,based on the sparse representation and dictionary learning as the theoretical background,combined with the image denoising based on joint sparse representation and total variation denoising and the solution is calculated by using augmented Lagrange method,finally improve the image denoising ability,the main research work is as follows:(1)This paper briefly introduces the coefficient decomposition algorithm and dictionary construction method in sparse representation problem,and introduces how to apply the sparse representation model to the inverse problem of the actual image processing,and to verify the denoising ability of the common denoising algorithm.(2)In order to improve the traditional sparse denoising method denoising effect is not obvious and easy to cause the edge fuzzy phenomenon in the condition of strong noise,proposed a kind of based on improved K-SVD dictionary learning and total variation regularization denoising method based on dictionary learning and sparse representation of image denoising model.Firstly,considering the advantages of fast convergence properties of orthogonal matching pursuit algorithm and the augmented Lagrange method atomic strong recovery capability,combined with the K-SVD dictionary algorithm,proposed a kind of improved K-SVD dictionary learning.Secondly,making the total variational denoising theory into sparse denoising model,the total variation regularization term is introduced as a new constraint into the original model,the solution combined with improved K-SVD method.Experiments show that the improved method on the condition of strong noise and edge rich image has very good denoising effect.(3)In order to further improve the denoising ability and enhance the processing capacity of enriching the image of texture and detail information,the calculation principle is simple and fast in convergence properties with the augmented Lagrange method,based on dictionary learning and sparse representation of image denoising model,proposed a kind of method based on augmented Lagrange multiplier method to achieve denoising task,using the augmented Lagrange method of image reconstruction model and variational constraint model,from the structural similarity,the peak signal to noise ratio,the running time compared to other algorithms.