Research On Sparse Model Based On Clustering And Weighted Nonlocal For Image Denoising

Image is an important way for people to get information. In many fields, we need to make a decision or explain phenomena by images, such as remote sensing, medical diagnosis and so on. However, image will inevitably be severely influenced by noise during the collection and transmission process. Therefore, image denoising has an important practical significance.In recent years, sparse representation theory has been extensively studied, and also has been applied successfully in image denoising. The theory of the method is that a clean image can be sparse representation under the proper over complete dictionary, so we can reconstruct the original image to achieve the purpose of removing noise. In this paper, we carry on the research for image denoising based on sparse representation which include two parts as follows:First, this paper proposes an image denoising method based on clustering and K-SVD. The traditional K-SVD algorithm doesn’t take into account the differences of image’s structure. So we use clustering algorithm to divide similar structure into a class, then train these image blocks to get a dictionary. The dictionary can be better represent this type of image structure, so detail features of image can be well protected.Second, this paper proposes a weighted non-local sparse model for image denoising. We introduce the non-local means’ weight to non-local sparse model for making the difference between the image block in similar collection. So we can get more suitable dictionary atom to represent image block in similar collection.