Research On Blind Image Restoration Based On Dictionary Sparse Representation

With the rapid development of multimedia and computer technology, imageprocessing technology becomes more and more widely used. Blind image restoration is ofgreat part of image processing. When the PSF (Point Spread Function) is unknown, usethe prior knowledge of the blurred image, the degradation system and the original image,blind image restoration technology could build the blind image restoration model toreconstruct the original image. Since blind image restoration doesn’t depend on PSF, it hasa wide range of practical applications, and has become an active topic in recent years.This thesis mainly studies the blind image deblurring algorithm based on dictionarysparse representation. The major content and innovatation of this thesis is mainlyincluding the following three aspects:Firstly, briefly describe the principle and the application prospect of blind imagerestoration technology, and some typical blind image restoration algorithm. It analysis thecontinuous and discrete image degradation model in detail, and the ill-posed problems inblind restoration process as well as two kinds of solutions are discussed.Secondly, when make use of the image restoration based on the dictionary sparserepresentation, the artificial effects among image blocks cannot be avoided. In order tosolve this problem, the statistical property of image gradient, as a priori knowledge, isadded into the deblurring model, and a blind image deblurring algorithm based ondictionary sparse representation and sparse gradient is presented. Experimental analysisand results indicate that the proposed algorithm can eliminate the artificial effects to someextent, remain the structure characteristics and gliding property.Finally, propose a blind image deblurring algorithm, which is based on double sparsedictionary learning. By analyzing the dictionary structure characteristics, research to trainthe double sparse dictionary by the framework of K-SVD dictionary learning method.Then we add the double sparse dictionary learning method into the deblurring model forupdating the clear dictionary, it contributes to reconstruct a more excellent image.Experimental analysis and results indicate that the proposed algorithm can acquire a betterrestoration effect.