Noise Estimation And Noise Removing Based On Fast Non-local Means Algorithm

Digital image processing is a new field derived from mathematical technology, optics, graphics and computer technology. But images are inevitably contaminated by noise in the process of acquisition and transmission, which will seriously affect the visual effects and the results of subsequent processing. Therefore, image denoisng has been an important preprocessing step in various digital image related applications. It has been studied nearly half a century, but image denoising is still one of the most classical and challenging tasks in the field of image processing.Non-local means(NLM) algorithm which proposed by Buades et al is a totally different denoising method, it is another milestone in the field of image denoising. Its main idea is to make full use of redundant information of the image, build similar function between the denoising pixel neighborhood and the pixel neighborhood in the search window, then calculate the similarity of these pixels using the similar function, finally weight averagely of all the similar pixels in the search region to reduce the noise.Non-local means preserves textures and edges of the image while reducing noise, it is acknowledged one of the best denoising algorithms. However, its application has several limits because of uncertainty of filtering parameter and high complexity. According to denoising system, this paper did the following work under the non-local means.1. Aiming to solve the uncertainty problem, this paper proposed a new criterion to gain the sensitive parameter accurately, making sure that the filtering effects are optimal or near-optimal;2. By combing the filtering algorithm and block algorithm in the field of noise estimation, this paper proposed an improved block-based algorithm to estimate the noise variance which affects the filter performance. Meanwhile this paper pointed out as well as resolved the underestimation problem, which has been overlooked for a long time;3. This paper overviewed the existing fast non-local means algorithms which aim to reduce algorithm complexity, also introduced the characteristics of them in detail, and at last pointed out their deficiencies through theoretical and experimental analysis;4. Proposing a totally new and effective fast NLM algorithm, which reduced the algorithm complexity while maintained high denoising quality by quoting the idea of box filtering, hence enhanced the practicability of non-local means algorithm.