The Application Of The Non-local Algorithm In Image Denoising

As an important preprocessing step in various digital image related applications,image denoising is still one of the most classical and challenging problems, although ithas been studied for nearly half a century. So far, Many methods have been proposed forimage denoising. However, the most of these methods are just underlying the regularityassumptions on the original image, and the details and the fine structure are smoothedout. Non-local means (NL-means) method which proposed by Buades et al Effectivelyovercomes those weaknesses.Digital images have a lot of redundant information,Nonlocal Means method uti-lizes those information redundancy to compute the similarity between the noised pointand those points in the searching region, then weights averagely the similar parts of animage to reduce the noise. This method produces state-of-art texture preserving, edgepreserving and denoising performance, however, the computation cost is high for itspixel-to-pixel neighborhoods matching.In this paper, according to the geometric features of the image, we did the follow-ing work under the framework of non-local mean.1. In this paper, we introduce a new method to provide more precise noise varianceestimation based on the discrete wavelet transform, the smoothing spline estimation andgeneralized cross-validation criterion,which provides a more precise parameter for thefollow-up works;2. In this paper, we introduce the corresponding neighborhoods'Gaussian cur-vature based on the original neighborhoods'gray value to compute the wights, whichconstructs the nonlocal bilateral filter algorithm.3. In this paper, we propose a new adaptive non-local means algorithm for im-age denoising based on the geometrical structure region via pixel region growing andmerging algorithm, which Gives a more reasonable search region;4. In this paper, we introduce and perform the nonlocal total variation algorithmbased on the nonlocal operator.Experiment results demonstrate that the proposed method is robust and efficientfor image denoising.