Research On Image Denoising Based On The Sparse Representation And Dictionary Learning

Image denoising is an important part of image processing, people pay more andmore attention on it, the technology plays an important role in medicine, aerospace,military, agriculture, etc.. The noise image denoising, not only to facilitate a betterunderstanding of the image information, but also easy for people to carry outsubsequent image processing. In recent years, the sparse representation for imagedenoising is a hot spot in the field. This paper is based on the theory of sparserepresentation of image denoising method, were used to study the sparserepresentation denoising method and flow using fixed dictionary and learningdictionary. In the sparse representation based on the understanding of the principleof the noise reduction algorithm, we propose a new denoising method, it canimprove the denoising effect and speed. The specific contents of this paper are asfollows:（1）Analysis of the basic theory and method of traditional image denoising,and experiment the traditional denoising methods, such as median filtering, Wienerfiltering, wavelet denoising. Depth study of the sparse representation theory,analyzes the basic model of sparse denoising, and conducted experiments based on afixed DCT dictionary and Gabor dictionary. Compare the performance of eachalgorithm in image denoising.(2) Analysis of the denoising model based on K-SVD learning dictionary, basedon this framework, expounds the realization process of dictionary learning andsparse decomposition of the two key links. Compared to the traditional fixeddictionary, because learning dictionary is obtained by machine learning, the packagecontaining the image has its own features, so under normal circumstances the betterdenoising effect. In the process of denoising experiments on learning dictionary,classify the image according to the amount contained in texture, and uses theK-SVD global dictionary and adaptive dictionary conducted experiments, comparedthe denoising effect of these two kinds dictionary: the structure image with lesstexture using the global dictionary for has a better denoising effect; the more textureimage using the adaptive dictionary has a better denoising effect. （3）Analysis the learning dictionary denoising effect, we put forward the ideaof MCA image denoising based on sparse decomposition: Decompose the imageinto structure and texture, and then use the global dictionary denoising on structure,use adaptive dictionary denoising on texture. This can be more focused on differentparts of the image denoising, the experimental results show that this method hasimproved the denoising effect.（4）To improve the image decomposition based on sparse representation MCAdenoising algorithm speed, introduced the dictionary structure of a pair of sparse, tospeed up the adaptive dictionary constructing, in the case of de-noising effect is notlost, improve the speed of image denoising, and achieved satisfactory experimentalresults.