Research On Single Image Blind Deblurring Reconstruction

Image deblurring aims to reconstruct a deblurred image using only one blurred input image.According to whether the blur kernel is known or not,image restoration can be classified into two categories:blind deblurring and non-blind deblurring.Blind deblurring is more useful in real applications,because it doesn't require known blur kernel beforehand.With the development of sparse representation technology,over-complete dictionary sparse representation,which can effectively recover image structures and details due to its redundancy,has been applied to blind deblurring.This paper focuses on single image blind restoration based on the framework of sparse representation technology.Main works are as follows:At the present time sparse representation based image blind restoration algorithms suffer from high computational complexity,limited ability of detailed recovering,high incidence of artifact and so on.In order to deal with problems above,this paper proposes a new blind image restoration algorithm,which combines both dictionary sparse representation and image gradient prior.The proposed algorithm introduces a novel way of non-overlapping image blocking,unites the novel image gradient prior and the blur kernel prior using efficient sparse coding,and finally constructs a new cost function.Experimental results show that the proposed algorithm can effectively overcome the common defects of the current algorithms,and improve the quality of image restoration.Currently,most blind image deblurring algorithms are proposed under the assumption that images are subjected to uniform blur,which is not in accordance with real facts,for example,the image blur caused by camera shake,one of the most typical non-uniform blur.In this paper,we research on the geometric model of camera shake,mapping the three-dimensional blur kernel into local regions of the image to generate non-uniform local PSFs and propose a non-uniform image deblurring method based on framework of sparse representation.The testing experimentation shows that the proposed method achieves ideal results on actual blurred images.