Research For Image Restoration Algorithm Based On Sparse Representation

Image restoration is referring to process images after degradation, in order to getclear, high quality images. It is one of the most popular topics in many fields, such asimages processing, computer vision and pattern recognition, etc. Image restoration ispossessed of important position in the fields of studying astronomy, remote sensingimaging, military and medical treatment, which mainly includes image denoising anddeblurring. The requirement is removing the noise, fuzzy and kept the image edges, details.But a pair of contradictions was apperared. For above issues, this paper had comparativelydone systematic study to the application of filtering restoration technique and sparserepresentation of image in image restoration. Some of the existing algorithm wereimproved, the following studies were done in this thesis.Firstly, for the linear wiener filter recovery to contain the edge of recovery richnon-stationary image easy deblur edge to improve its shortcomings, using a combinationof multiscale geometric analysis of image restoration wiener filter algorithm, through thewiener filtering on degraded image to blur preliminary results recovery are wavelettransform and contourlet transform, and the threshold processing respectively to achievethe purpose of further optimization.Secondly, we introduced natural images of wavelet coefficients statistical model,inorder to overcome not to use the decomposition coefficient of correlation betweenneighborhood characteristics of the threshold disposing of wavelet packet transform.Combined with wavelet gaussian mixture of scale model wiener, filtering imagerestoration is presented. First of all, we degraded image wiener filtering to fuzzy, thencombined with wavelet transform, the gaussian mixture model building scale waveletcoefficient neighborhood model, the application of the bayesian estimation of imagefurther optimized. The experimental results showed that the method was better than thoseof a single wavelet gaussian mixture model scale recovery or frequency domain wienerfiltering recover in visual characteristics and performance index of the algorithm.Finally, image sparse representation in image restoration application was studied. An improved algorithm based on structure clustering is proposed for the drawbacks of the bigdictionary of dimension and basic element number which are caused by traditional supercomplete dictionary of the learning process. Combined with the non-local similarity ofimage，this algorithm studied a child dictionary respectively of every kind of localcharacteridtics of image. But this study does not consider the effect of degradation andnoise when image block select matching dictionary. This paper is weighted with minimumeuropean distance formula. The dictionary is used in sparse regularization rehabilitationmodel. The experimental results proved by this algorithm can not only keep thereconstruction of clear image edge information, but also has better robustness.