Research Of Blind Image Deblurring Based On Convolution And Depth Networks

Computer vision,as one of the three fields of artificial intelligence research,has received extensive attention.Image deblurring is one of the basic problems in the field of computer vision,which lays a foundation for subsequent image recognition,target detection and tracking.Especially,it has important application value for the recognition of moving target identity on traffic roads.Therefore,the study of image deblurring has important theoretical and application value.In this paper,aiming at the problems existing in the blind deblurring of single image motion blur,an end-to-end deblurring network is established,and the corresponding deblurring algorithm is proposed.The main work is as follows :(1)Aiming at the problem that the current convolutional neural network model is easy to cause the loss of image detail information in the process of image deblurring,resulting in poor deblurring effect,a deblurring method based on enhanced group convolution and dual task reconstruction network is proposed.By superimposing enhanced group convolution blocks,an enhanced group convolution neural network with shallow structure is constructed to extract the features of input blurred images.At the same time,in the dual-task reconstruction network,the conventional convolution network is combined with the wavelet transform to form a multi-layer convolution-wavelet model.The discrete wavelet transform is used to capture all the frequency and position information of the feature map,which expands the image receptive field and improves the sparsity of the feature information.Finally,the inverse discrete wavelet transform is used to output the information-enhanced high-resolution feature map.(2)Aiming at the problem that the current convolutional neural network model is easy to cause the loss of image detail information in the process of image deblurring,resulting in poor deblurring effect,a deblurring method based on enhanced group convolution and dual task reconstruction network is proposed.By superimposing enhanced group convolution blocks,an enhanced group convolution neural network with shallow structure is constructed to extract the features of input blurred images.At the same time,in the dual-task reconstruction network,the conventional convolution network is combined with the wavelet transform to form a multi-layer convolution-wavelet model.The discrete wavelet transform is used to capture all the frequency and position information of the feature map,which expands the image receptive field and improves the sparsity of the feature information.Then,the inverse discrete wavelet transform is used to output the information-enhanced high-resolution feature map.The experimental results show that the proposed two deblurring network models have strong generalization ability and can recover vehicle motion blurred images under different illumination environments.At the same time,compared with the classical deblurring algorithms,the proposed two models have higher performance indicators on the public data set,the restored image is closer to the real clear image,and the texture details are clearer.Therefore,the proposed two algorithms have certain application value,especially in the supervision of traffic road moving targets.