Image Processing Using Patches Reordering And NL-means

As the most important source of information for people,the image plays an increasingly important role in our life.So how to get clear images from damaged images becomes a hot topic for scholars all over the world.More and more research results are applied to all aspects of life,greatly improving the quality of people's life.In recent years,the digital image processing methods proposed by a large number of papers recovery the corrupted image not from the whole image.The recovery algorithms based on image patches have been applied to all areas of image restoration,and achieved excellent recovery results,and some of them even reached state-of-the-art results.The image restoration algorithm based on patches reordering is a novel restoration algorithm.The algorithm reconstructs corrupted image by reordering the image patches and filtering or interpolating the reordering signal.This algorithm has achieved good results.The thesis is organized around the recovery algorithm based on image patches reordering.The main research contents include:1.Study the current algorithm in image denoising.Study the concept of image patches reordering,patches-reordering algorithm and smoothing filter training.2.In terms of the shortcomings of the algorithm based on patches reordering in image denoising:1)the smoothing filters are not adaptive,and they require a separate training set to be learned from.2)it did not take advantage of the distances between the noisy image patches,which were used in the reordering process.The thesis proposes to reconstruct the neighborhood for each pixel after reordering the image patches,and estimate the pixels using NL-means algorithm.At the same time,we use the K permutations,patches classification and second iteration to further improve the denoising result.3.The thesis first try to apply the current algorithm to image interpolation,using different image patches similarity measure for different corrupted images,and analyze and discuss the experiment results.From the experiment results,we can see that the proposed method avoids smoothing filter training and makes full use of the distance between the image patches in image denoising.The algorithm is simple and easy to implement.Also,the denoising result of the thesis is better than the current algorithm.And the thesis proposes analysis about the bad interpolation results in interpolation.