Research On Image Deblurring Method Based On Generative Adversarial Networks

As the main carrier of information,image contains a wealth of intuitive and effective information.Compared with language and text,it is more conducive to people's record of beauty,sharing mood and transferring emotion,so it is widely used in daily production and life.However,in the process of image acquisition,blur often occurs due to various reasons,which seriously reduces the image quality and visual effect.As a main research direction in the field of image processing,image deblurring technology has important research significance and practical value.In this context,this paper mainly focuses on the problem of motion blur image de blurring.The previous research is based on the traditional methods,and proposes the method of deblur based on feature point matching,which has achieved good recovery results.However,this method has the limitation of most traditional methods,that is,it has a good processing effect on blurred images with globally unified blur kernels,and a poor processing effect on images with multiple blur kernels or local blur.Based on this,the paper extends the research on the method of blur elimination,and puts forward the method of blur elimination based on the generated countermeasure network.The main research work is summarized as follows:(1)In this paper,a method of de blur based on feature point matching is proposed.The feature point matching algorithm is used to match the target under the continuous frame image,and the accurate motion blur angle and distance are obtained through the matching information,so as to obtain the accurate blur kernel information.Finally,the improved Wiener filter restoration algorithm is used to remove the ambiguity.The experimental results show that the proposed method is better than the existing traditional methods.(2)A deblurring method based on generative adversarial networks is implemented.Adopting the condition based on Res Net to generate adversarial network,and introducing gradient penalty and perceptual loss to improve the accuracy of the network,directly recover the clear image from the blurred image end-to-end,thereby improving the efficiency of the algorithm and deblurring effect.Experiments show that the processing effect of this network structure is good,and it has universality compared with traditional methods.(3)A blury image data set was created and a blury image processing software system was developed.During the research on the deblurring method,the collected images were sorted,and a new blury image data was created in combination with the blury generation algorithm,which provided favorable conditions for network training.After the scientific goals of the supported scientific research projects were achieved,the research results were summarized and sorted out,and an image deblurring software platform system was developed.The software mainly includes two functions of image file selection and image deblurring,which has the characteristics of clear interface,simple and efficient operation.The experimental results show that the research of the method can effectively process the blur image and get a better clear image,at the same time,it has a significant advantage of timeliness.