Image Adaptive Deblurring Algorithm Based On Blur Parameter Estimation

Image deblurring is a classic problem and research hot-spot in the field of image processing,especially in recent years,the image deblurring method based on deep learning has made a breakthrough,compared with the traditional method,it has made a significant improvement in image quality.In practical application,the blur degree of image obtained in different shooting distance,shooting device and shooting environment is also different.Therefore,how to realize a deblurring network to adaptively enhance images with different degrees of blur is an urgent problem to be solved.A unified image deblurring network model can be obtained by the mixed training of image samples with different blur degrees,but the performance of the model will decline with the increase of blur types,and it is lack of pertinence to specific blur degrees.The research shows that the input of accurate image blur parameters plays a key role in the enhancement of blur image,but in practical application,the blur parameters of blur image are usually unknown.In order to solve this problem,this paper studies the image self-adaptive deblurring method based on the blur parameter estimation.The blur degree information of the image is obtained through the blur estimation network,which can guide the subsequent image deblurring network learning and improve the adaptability of the image deblurring model to different blur degrees.Specifically,the work of this paper includes the following aspects:(1)The performance of blur parameter estimation of residual network and u-net network is analyzed experimentally.Based on the advantages of the two models,a u-net blur parameter estimation network model with joint residual learning is proposed.The model uses the "short connection" mode of residual network to strengthen the connection between the front and back layers of the network,and the "long connection" mode of u-net network to strengthen the connection between the shallow layer and the deep layer of the network,so as to improve the accuracy of blur parameter estimation of the network.(2)In order to solve the problem of too much parameters and difficult network training caused by the direct combination of u-net network and residual network structure,the convolution kernel multiplication structure in u-net network is improved to the same number of convolution cores in each coding block and decoding blockwithout affecting the accuracy of blur parameter estimation,so as to significantly reduce the network parameters and further improve the model efficiency.(3)On the basis of the above,an adaptive image deblurring algorithm based on blur parameter estimation is proposed,and an end-to-end network framework is constructed,which combines the u-net blur parameter estimation network of joint residual learning with the non-blind deblurring network based on SRMD.The image with different blur degrees is processed by the blur parameter guidance deblurring network.The experimental results show that compared with MSWNNM,the PSNR of image objective quality of this method is improved by 1.29 db,which shows that this method has better adaptive enhancement function for different degrees of blur image,which is of great significance for improving the image clarity in practical application.