Image Restoration Research Based On Sparse Representation

Image denoising, image deb luring and image super resolution restoration are the three basic problems of image restoration, which are affecting people's lives. Image restoration requires us to remove the noise, remove the blur, restore the high resolution, what's more, we should retain the edges and detail information at the same time, while a pair of contradictions was appeared in application. How to solve these two problems at the same time becomes the urgent matter of the moment. In order to solve these problems, this paper studies the theory of sparse representation which is very hot in image restoration, and discusses the two major problems which are called dictionary construction and sparse decomposition, and put forward some new ideas in two aspects with my own understanding base on the previous work.In the sparse representation theory, there are two ways to construct the dictionary, which are analytical dictionary based on fixed basis set and learning dictionary based on training sample, The latter tends to recover images better and can retain more detail information of texture. This paper researches the KPCA dictionary which is popular in the learning dictionary, it is formed by the combination of the K mean clustering method and the PCA principle, on the basis of this, a series of improvements are made in order to improve the performance of image restoration:1. Optimizing the K-means clustering method to select the initial cluster center, in order to enhance stability and enhance the accuracy of the dictionary.2. After improving dictionary using the first step, the noise elimination scheme is introduced to form a IKPPCA dictionary to reduce the impact of noise on the dictionary.3. Improving sparse coefficient estimation as the core part of the the non-local weighted sparse representation algorithm combined with the IKPPCA dictionary.The experiment shows that the improved algorithm can improve the smoothness of homogeneous regions while still preserving more texture and edge details in image denoising. As the sparse algorithm always has good performance in the field of image restoration, this pager further researches the algorithm application in image deblurring and image super resolution and also reach good results, at last makes a GUI interface and provides a more intuitive contrast display.