Research On Image Denoising Based On Sparse Representation Theory

In recent years, sparse representation theory has attracted much attention ofscholars and it has been successfully applied in the field of image denoising. Viaselecting or designing a proper over-complete dictionary, sparse representation caneffectively obtain essential features of an image, representing the image in a way asconcise as possible. Moreover, compressive sensing is a newly developed theorybased on sparse representation. It further illustrates the important value and greatpotential of the sparse representation theory.This article focuses on image denoising methods based on sparse representationtheory. Firstly, this paper comes up with an image denoising method, which canefficiently reduce Gaussian noise while exploiting much more image textureinformation. This proposed method treats image as a combination of two differentattributes, texture and cartoon, making use of morphological component analysistechnique to extract texture from a noised image. According to the different propertiesof image texture part and cartoon part, the method separately appliesover-completeddictionaries of different training strategies to deal with the noise in each part of theimage. Experiment results have shown excellent denoising ability of the proposeddenoising method, which is based on image decomposition and sparse representation,with denoised results of better visual quality.Secondly, a denoising method based on adaptive compressive sensing isproposed. This method introduces the generalized principal component analysis(GPCA) into the framework of compressive sensing, making full use of priorknowledge of an image such as image texture information. Experiment results showthat the image denoised by the proposed adaptive compressive sensing method is ofhigh visual quality, remaining more detailed information by projecting the noisedimage into GPCA domain to preserve the most significant principal components.