Development of Blind Deblurring Based on Deep Learning

Abstract The rapid development of computer vision has greatly promoted the development of camera in the fields of target detection, remote sensing analysis, target recognition and so on. As the carrier of information exchange in contemporary society, image contains a large number of information elements. When taking an image, it will blur the image and destroy many photos due to various factors, resulting in the inability to obtain relevant information from the image. Restoring a clear image from the blurred image is a hot spot in the field of computer vision and image processing. This paper expounds the causes of fuzziness in detail, summarizes and combs the current blind deblurring methods and research status based on deep learning, makes a detailed overview of the development of deblurring network based on deep learning from three aspects: convolution neural network, cyclic neural network and generation countermeasure network, and summarizes the advantages and disadvantages of various methods, Some networks are compared, and the problems of feature extraction, evaluation index and data set construction in deblurring are analyzed. Finally, the development trend of image motion blur restoration technology is prospected.