Research On Image Denoising Based On Nonlocal Information

An image is always affected by noise in its collection, transmission and recording. Noise not only degrades image quality, but also affects the results of following image processing modules. Therefore, image denoising becomes a fundamental and crucial step in the image processing works. On the basis of the preprocessing, improving image quality can improve reliable basis for the subsequent image processing. The goal of image denoising is to removing noise and also preserving the important edges ortextures. Thus, how to preserving the edges or textures becomes the important research in image denoising work.After researching NLM algorithm and nonlocal structure similarity, this paper mainly focuses on preserving the edges and textures and thus proposing some improved denoising algorithm or model. The main achievements and innovations are follows:(1) To overcome the shortcomins of the fixed filter parameter, a n adaptive non-local means(ANLM) algorithm for image denoising is proposed. This method extracts edge contents of the noisy image and determines the structure features of the image area which the current block is located. And two thresholds are set to select filtering parameter adaptively. ANLM has solved the problem of over-smoothing and preserved more edges and textures.(2) In the original NLM, similarity between pixels is computed soly based on pixel gray values, thus more errors are introduced in the step of computing weights. To solve this problem, a new similarity measure method based on LBP feature is proposed. Furthermore, a mixed similarity measure method combined LBP features and pixel intensity is also proposed. The experiments results show that these modifications significantly improve the robustness of similarity measure a nd therefore boost the quality of denoised images.(3) To make full use of the nonlocal structure similarity, a new regularized imge denoising model is proposed. This model has introduced the nonlocal structure similarity regularization into TV model, thus more structural features are preserved. Split Bregman algorithm is used to solve the model iteratively and thus achieved the purpose of image denoising effectively.