Research On Image Blind Deconvolution Method Based On Sparsity And Self-similarity

Due to factors such as environment,weather,equipment,operating,the image often have a certain degree of blur and noise in the process of image acquisition.Those blur and noise caused great inconvenience for practical applications,especially caused greatly challenge in the image identification,analysis and processing.So restoring the noisy blurred image using the image processing method has been applied in more and more fields.Blind image deblurring aims to recover the latent sharp image from a noisy blurred image when the kernel is unknown.Image deblurring is a typical ill-posed inverse problem.In this paper,we will using the sparse theory and the self-similarity theory to analysis and research the image deconvolution algorithm.The main content and innovation of this thesis are reflected in the following two aspects:1.We propose an image blind deconvolution method based on the group sparse representation and self-similarity.This algorithm sets the self-similarity prior and the group sparsity prior of the image as a regularization constraint conditions,and uses the non-local group of similar image block as the basic unit of the sparse representation.Furthermore,this algorithm combined with the inherent local sparsity of natural images and non-local self-similarity,based on the similar structure of the image block for group sparse representation to better restored image.At the same time,by using the self-similarity of the image in the different scales,we will search multiple similar image block in the different scales of image to linear said each image block.This method can better estimate the point spread function and the latent sharp images.2.We propose an image blind deconvolution method based on self-similarity and low-rank representation.This algorithm can solve the problem of the noisy blurred image recovery.In our method,we use the low-rank prior and the self-similarity prior of the image as a regularization constraint conditions.On the one hand,using the low-rank and similarity in the structure of the image,this algorithm can do the low-rank representation for the group of the image block with the similar structure.This is a good way to eliminate the noise in the image and reduce the effects of noise on the recovery operation.At the same time,by using the self-similarity of the image in the different scales,we will search multiple similar image block in the different scales of image to linear said each image block.This method can better estimate the point spread function and clear images.