Dictionary Analysis And Image Denoising Based Non-local Sparse Models

Image is an important tool for people getting information. However, Due to the restriction of the imaging equipment and imaging condition, Image is inevitably polluted by noise in the process of collection, conversion and transmission. So, Image denoising occupies an important position in image processing area. Image denoising is the premise work of image edge detection, feature extraction, segmentation and pattern recognition. The noise in the image can be approximate to Gaussian White Noise. Therefore, it is an important researching area to remove Gaussian White Noise from noisy image.It is confirmed that there are similarity between image patches, by the performance of non-local mean filter denosing additive white Gaussian Noise. Recently, image sparse representation has been the hot topic of image representation. This method is used in image denosing field and has a good performance. Therefore, for dealing with additive Gaussian white noise in image, Non-local Sparse Models for Image Restoration was present, and also had a good performance. This paper proposes dictionary research of image restoration based on non-local sparse model. It includes three aspects:Firstly, this paper proposes an image denoising method based on non-local sparse model with PCA, using PND dictionary initializes sparse dictionary of similar signals and updating dictionary with similar signals. The sparse representation efficiency is improved by using low redundant dictionary.Secondly, this paper proposes an improved image denoising method based on non-local sparse model. A new dictionary initialization method is proposed, using similar signals in similar set. The sparse representation of similar signals with updating initialized dictionary uses to denoising noisy image, overcoming the linear said fitting and owed fitting problems of sparse representation and outperforming the state of art.Lastly, this paper proposes an image denoising method based on non-local joint sparse model. The similar signals are decomposed into public components and characteristic components by sparse decomposing. Then, the characteristic components include image noise and we chose applicable dictionary for the characteristic components sparse representation. This method has kept the advantage of the second method. And it also makes up the shortage of the second method’s worse performance to suppress large noise. Comprehensive denoising ability has been improved.This work has received the support of the National Natural Science Foundation of China (Grant No.61072106), the Program for Cheung Kong Scholars and Innovative Research Team in University (No. IRT0645) and the Fundamental Research Funds for the Central Universities (No. JY10000902032).