Research On Motion Degradation Image Restoration Method

Image as an important carrier of information,it is particularly important to ensure its integrity and readability.Therefore,restoring high-quality images from low-quality images has become an urgent problem to be solved,which has extremely important research significance and application prospect.Aiming at the image restoration task,this thesis mainly studies the motion degradation image restoration algorithm.In view of the shortcoming of traditional motion degradation image restoration based on the minimum mean square error filtering to produce ringing effect,an improved method is proposed to add a window function with the same size as the motion degradation image to be restored.At the same time,the image restoration method based on U-Net is studied.Aiming at the problem of information loss caused by continuous pooling of this method,a multi-stage step restoration network algorithm based on smoothed dilated residual block is proposed.Specific research contents are as follows:(1)In view of the ringing phenomenon caused by the traditional minimum mean square error filtering method,an improved minimum mean square error filtering method is proposed.Firstly,the traditional minimum mean square error algorithm is analyzed and simulated in detail,then the restoration results are optimized,and the optimal window algorithm is added to solve the ringing effect caused by the mutation of pixel value in image boundary by the minimum mean square error filtering method.The simulation results show that the improved algorithm proposed in this thesis reduces the change of pixel values at the boundary of fuzzy image,reduces the ringing effect effectively,and improves the quality of image restoration.However,the image restoration method based on the minimum mean square error filtering can only obtain a globally optimal average value,but cannot obtain the optimal value of the restoration result of each pixel in the image.This situation is more prominent when there are local motion degradation regions or multiple motion blur types in an image.To this end,the restoration method of motion degraded image based on deep learning is studied.(2)Aiming at the deep learning method,a deblurring network algorithm for motion degraded image based on multi-stage stepwise restoration is proposed.Firstly,to solve the problem of information loss caused by continuous pooling in U-Net networks,smoothed dilated residual block is proposed to replace the down sampling pooling operation in traditional U-Net networks,which not only increases the receptive field,but also eliminates the grid artifacts caused by dilated convolution.Secondly,the channel attention block(CAB)is introduced to enhance the correlation between feature maps and screen features to retain important information to the maximum extent,so as to improve the effect of image restoration.Simulation results show the effectiveness of the algorithm.