Research On Global Motion Deblurring Of Single Image

Image deblurring is an indispensable technical aspect in the field of image processing. In recent years, image restoration technology has been rapid developed by academics’assiduous efforts. Of course, the practicability for some of them is to be improved. Thus, the significance of the image deblurring technology not only stays in the theoretical value in the field of digital image restoration, but also is in the application of the reality scene.In order to solve the problem of motion blur image restoration, it is important to deeply study blur kernel estimation techniques in deblurring algorithm of single global motion picture, and prior knowledge in the image deconvolution algorithm and natural images for us. We propose image deblurring algorithm based on outlier and image blind deblurring algorithm based on rational polynomial priori model.Experiment results demonstrate the effectiveness of the proposed algorithm. For objective evaluation criteria, the proposed algorithm has been well parametric. For the subjective evaluation, the proposed algorithm has good visual effects, and significantly inhibits the ringing effects and saves cost calculation. The main work of this dissertation includes the following aspects:(1) The purpose and significance of image restoration technology are briefly introduced, and blurred image restoration techniques for blurred image generated will be explored and summarized by the relative motion, and the major research work will be stated.(2) From two aspects of the blur kernel estimation and deconvolution, the global motion deblurring technology of single image and get clear image is researched and improved by the method of the edge estimation combined with the rapid deconvolution algorithm. Considering the effect of image outlier on restoration results, we propose image deblurring algorithm based on outlier to reduce the ringing effects in this paper.(3) Considering the issue caused by Gaussian or mixture Gaussian function fitting image gradient distribution of natural scenes, we propose to use rational polynomial function to substitute a Gaussian or mixture Gaussian function by images and prior knowledge. It improves the goodness of the fitting gradient distribution of the natural scene image and the binding of boundary conditions, so that the estimation of blur kernel is better precision and effectively inhibits the ringing effects, and improve the quality of image recovery.(4) The main contents of this paper will be summarized, and the ideas for the future research direction are proposed in the section.