Image Denoising Based Sparse Representation And Dictionary Learned

Image denoising is one important problem in the field of digital image processing, because the image can be polluted by some specific noise in the process of acquiring, storage and transport, leading to the quality degradation. The purpose of image denoising is to eliminate the noise interferes with variety treatments, so as to improve the quality of image.In recent years, the sparse representation theory has drawn wide attention, and also has been applied successfully in image denoising. Its theoretic proof is that a clear image with smooth character has its sparse decomposition under the over-complete dictionary. Therefore, we can accomplish the image denoising through computing the sparse representation after choosing or designing an appropriate dictionary. There are two kinds of dictionaries in the application for image denoising. One is fixed analytic dictionary, the other is adaptive dictionary, which can be trained by appropriate model using the given image information data.Due to its adaptivity, the learned dictionary can get better denoising results than the fixed dictionary. Therefore, in this paper, we improve a dictionary learning algorithm introduced by predecessors, whose aim is to learn a structured dictionary as unions of orthonormal bases. The improved algorithm can update the coefficients simultaneously, increasing the learning efficiency. Moreover, the structured adaptive dictionary (L-ONB dictionary) is combined with the denoising algorithm at last. The simulation demonstrates that the new algorithms has a better denoised effect comparing with the fixed dictionary.