The Improvement Research Of Non-local Means Image Denoising And Image Inpainting Algorithm

Digital image will be inevitably interfered by noise in the process of formation and transmission. The introduction of the noise will reduce visual quality of image, and cause serious influence on required follow-up processing and recognition. Digital image inpainting is also occupy an important position in the field of image processing, and has a broad application prospects. Image denoising and image inpainting algorithms are still have lots to improve constantly, especially the process for the texture and detail areas are often not enough satisfactory. To this end, this paper studies deeply all kinds of classic image denoising and image inpainting algorithm, and proposed improvement scheme on the basis of Non-local means(NLM).The core idea of NLM is to use a lot of redundant information present in natural images, compare gradation of the entire region around each pixel distribution, according to the distribution of the surrounding area similarity between pixel and another pixel to determine the value of the similar weight, finally compute of the Gaussian weighted average of these sampled pixels to get estimate gray value of pixels to be processed. Original NLM effect in smooth areas of the image denoising is good, but weak performance in the region of the edge area and texture structure that more complex performance. The main work and innovation are as follows:(1) In the aspect of image denoising, this paper improves the non-local means filtering algorithm. Firstly, for the smooth region of noisy image, using a relatively large-scale search window and similarity window pre-generated roughly similar set of images to obtain once denoised image. Then take it as only contains texture regions image, using relatively small-scale search window and similarity window pre-generated more similar set, discharge dissimilar sets, do the second denoising. Finally, gain the effect image. The simulation results show that the new algorithm in this paper improve better performance than NLM, the structure and texture region information of the image maintained better.(2) In the aspect of image inpainting, this paper presented a novel exemplar-based image inpainting algorithm which based on the NLM algorithm and Criminisi algorithm. In order to decrease error match rate, the new algorithm adopts sum of squared differences(SSD) searching for several exemplars to generate a candidate exemplar set, then make use of Hausdorff distance as measurement criterion to find match block by computing the weighted average of exemplar set so that image edge structure be preferentially propagated. Experimental results demonstrate that the improved algorithm keeps the image structure better and obtains better visual appearance.