Blind Image Restoration Based On Michelson Channel And Recursive Convolutional Neural Network

As a main medium for information acquisition,expression and dissemination,digital images provide a convenient way of communication for the life and production.However,due to various factors in the process of image formation,the quality of the image formed by the shooting will be reduced.When the low-quality image cannot meet the needs of the production and life,it may cause certain economic losses.This requires researchers to be able to effectively recover images of low-quality images.With the continuous improvement of the digital image processing and the rise of deep learning,image deblurring has become a study hotpot in the digital image processing as an effective way to improve image quality.This paper studies the image deblurring based on the basic theory of digital image and semi-supervised learning.In order to effectively restore low-quality images,this paper proposes two effective image blind image deblurring methods,as following:Firstly,this paper introduces the dark channel,bright channel,the deblurring algorithms based on dark channel prior(DCP)and bright channel prior(BCP).However,DCP and BCP will cause some obvious ringing artifacts,and this observation has been confirmed theoretically by mathematical analysis and comparison.To reduce or eliminate the effect,a new image channel is proposed in this paper,which is called Michelson channel.Michelson channel can elevate the performance of existing motion deblurring algorithms by enhancing the Michelson pixels of the latent image.The proposed method mainly exploits the Michelson channel and its gradient information to recover the blurred image,which takes advantage of both Bright and Dark Channel Prior.In the process of image deblurring,by enhancing and retaining the dark and bright channels of the image,it captures sharper image detail and further eliminates the ringing artifacts of the recovered image.Massive experimental results demonstrate that the proposed method is more robust and outperforms the existing art-of-the-state of image deblurring methods on both synthesized and natural images.Secondly,based on the advantages and disadvantages of convolutional neural networks oriented to image deblurring,an image deblurring neural network based on wavelet transform and recursive convolutional neural network(R-DbCNN)is constructed.In the model training of R-DbCNN,blind image deconvolution is an ill-posed problem,which is mainly addressed by the regularization methods.Wavelet transform is an effective denoising method related to regularized inversion.In this paper,wavelet transform is utilized to decompose and extract the low and high frequency information of the blurred image,which is taken as the first step of the presented deblurring methods in this paper.However,when the process highlights the approximate portion of the image feature,the blurry image will be smoothed so that the image is distorted.Simultaneously,the overall wavelet transform will result in excessive data redundancy for the image.Thus in the second step,based upon the recursive convolution neural network,a deep recursive convolutional neural network(R-DbCNN)is designed,which can eliminate or weaken the characteristics of high data redundancy and image smoothness caused by wavelet transform to remove the blur of corrupted image.Comparing with the traditional convolution neural network in image restoration,R-DbCNN has better performance in deblurring with fast training speed.Thereafter a novel loss function is built to attain the best deblurring effect of the proposed method,that is,to balance the deblurring performance of wavelet transform and recursive convolutional neural network.The experiment results demonstrate that our method has practical applicability for image restoration with different blurs,such as the motion blur,Gaussian blur and out-of-focus by camera.