Research On Nonlocal Regularization Method For Image Denoising

Images are easily disturbed by noise during the processes of receiving, transmitting and storing and this will deteriorate the quality of images as well. Noisy images shall dramatically keep people from obtaining useful information from the images. In this case, it is necessary to do denoising before analyzing and employing images. Image denoising is an ill-posed problem, and one of the effective ways to solve the problem is a method that bases on regularization. That is, according to the hypothesis which is based on the images, via a number of prior information to construct regular terms to generate function model. Compared to the traditional denoising model, classical ROF Model can achieve better denoising result, however, which is more likely to arouse staircase effect. Inspired by the Non-local Average Smoothing Ideology, scholars put forward the Total Variation Regularization of Model NL-ROF under the non-local framework, which not only effectively reduce the staircase effect caused by ROF but also achieve better result in the aspect of maintaining image borders and other detail information. On the basis of deep analysis of Non-local Average Smoothing, this thesis have introduced the leading NL-ROF Denoising Model, and also made an intensified research on the regularization of function of this model. Main researches including:1. This thesis has analyzed the researching background and significance, and summarized the current situation at home and aboard. It also has introduce the theoretical basis of image denoising, including the presentation of digital image, degraded image model, noisy model and its classification, as well as the evaluation criterion of image denoising performance.2. This thesis regards the Non-local Average Smoothing as a kind of development of Airspace Average Smoothing. At first, it introduced the principle and developing process of Airspace Average Smoothing, and also puts forward several classical denoising methods of Airspace Smoothing. Then it emphasized on the introduction of the principle and features of NLM Denoising Algorithm. In the end, it made a comparative research on every denoising algorithm through simulation experiments.3. This thesis has introduced the principle of Regularization Denoising Algorithm, briefly analyzed its development, emphasized on the research of Regularization Denoising Model under the non-local framework (NL-ROF), and also made a comparative research on the classical NLM Algorithm, ROF Algorithm.4. The improved algorithm of NL-ROF centered on the Non-local Regularization of Function. While calculating the weight number, it puts forward to using SSIM instead of Euclidean Distance to measure the similarities between two images. Specific to the problem that noise may influence the SSIM estimation value of original image, it puts forward a two-step calculating method. According to results of Matlab simulation experiment, the NL-ROF Denoising Model which blends SSIM can better maintain the image borders and other detail information.