Image Deblurring Research Based On Dictionary Learning

Image and video technology have been in rapid development in recent years, and become the main carrier of information transmission. But in the process of acquisition, transmission, storage and calculation, images can be affected by many factors, resulting in blur images, which largely influenced the visual effect and the research value. Sparse representation theory is a popular direction in the field of signal processing in recent years, and has been successfully applied to image denoising, image classification, and other fields. In this paper, adaptive dictionary learning algorithm based on sparse representation would be applied to the study of image deblurring, and through experiments proved the feasibility of the algorithm.The main research contents and results are as follows:1) For the traditional image deblurring method need to know the prior information of the point spread function (PSF), the dictionary learning algorithm is applied to the research of image deblurring, and achieves the blind deblurring for a single blur image. The optimization deblurring model is constructed by the principle of sparse restrictions, convergence and local minimization. We solve the equation by the method of assuming the initial fuzzy kernel, dictionary training and iterative update, and get the estimation of the fuzzy kernel finally. Experiments show that through the dictionary learning algorithm, better deblurring effect can be got.2) Our method is compared with traditional deblurring algorithms in the paper. Through the experiment of motion blur, gaussian random blur and disk blur, from the aspects of subjective visual and objective of mean square error ratio and gradient value, it is proved that more image texture and detail information and better deblurring effect can be got through this method.Deblurring method based on dictionary learning algorithm is a kind of blind image deblurring method, it is not necessary to know the image blur reasons in advance or the statistical analysis on multiple images. Through dictionary learning method based on sparse representation, we had better access to get the single image characteristic information, and avoided to solve the ill-posed problem. At the same time, the method of capturing part of the image block in this paper, on the one hand, greatly reduced the computation burden of the dictionary training, on the other hand can be reference to the study of global image blurred by different blur kernels for different parts.