Research And Implementation Of Motion Blurred Image Restoration Based On Deep Learning

During image capturing,the captured image is blurred due to factors such as movement of the subject,out of focus,and the like.These degraded images usually cannot meet people's needs,so removing motion blur and improving image quality have gradually become the focus of people's research on image restoration.The motion blurred image restoration problem can be divided into two categories according to whether the kernel is known or not.One is to determine the non-blind deblurring problem of the image of the kernel,and the other is to blindly deblur the image with unknown kernel.But in real life,the blur kernel of most motion blurred images is unknown.In recent years,with the rise of artificial intelligence,it has important scientific research value and practical significance.Therefore,this paper mainly uses the knowledge of deep learning to study blind blur motion blur,reduce noise impact and recover clear images.This paper focuses on blind motion blur removal for a single image.The main work includes several aspects:(1)Aiming at the problems of traditional image de-motion blur method,such as noise amplification and ringing effect,an image motion blur blind removal algorithm based on gradient domain and depth learning is proposed.Based on the image deblurring by using filtering,the method of using the filtering to suppress the inconsistency details,using L0 filtering to enhance the edge of the image,preprocessing the image,and using the image of the gradient domain to train is proposed.The trained model parameters are extracted to realize kernel estimation and image restoration.While removing the image motion blur,the noise amplification is suppressed,and the image restoration effect is good.(2)The traditional method to achieve image restoration,we must estimate the exact kernel,but the method of estimating the kernel is very sensitive to image noise,so the process of estimating the kernel needs to be carefully designed.In addition,the erroneous kernel will directly affect the quality of the restored image,resulting in ringing phenomenon;using CNN to achieve image deblurring is only effective under a specific model,and the scope of use is limited.Therefore,the convolutional neural network model based on generating the anti-network is proposed,and the end-to-end method is adopted to avoid the problem ofinaccurate kernel estimation.At the same time,CNN deblurring is no longer limited to a specific model.The method generates a confrontation network based on the depth residuals of the U-net and the residual network,without estimating the kernel,which reduces the complexity of the algorithm and reduces the image ringing effect.(3)In order to extract the detailed information in the image more deeply,use coarse to fine methods,combine the multi-scale processing of the image with the generated adversarial networks,and use the end-to-end structure to avoid the inaccuracy of the kernel estimation to affect the image restoration and reduce the ringing effect in the restored image.Better recovery of details in the image.Through a series of experiments to verify and analyze the effectiveness of this method,and compared with other existing methods.The experimental results show that this paper can effectively remove the image motion blur,reduce the ringing effect and improve the image restoration quality.