Research On Moving Image Deblurring And Super-resolution Reconstruction Based On Deep Learning

Fuzzy images are a highly pathological problem in the field of image processing,and image deblurring technology is a key research direction in this field.During the normal image shooting process,clear images cannot be captured due to factors such as camera shake and target object movement.Traditional image deblurring uses the method of estimating the blur kernel of blurred images,but the effect is generally not ideal.Similarly,for low resolution image super-resolution reconstruction,if a clear image super-resolution method is used to solve the problem of low resolution blurry image super-resolution reconstruction,the reconstructed image often contains obvious artifacts and unclear image texture.To address the above issues,this article adopts deep learning based technology to achieve image deblurring and image super-resolution reconstruction.In response to the existing problems mentioned above,the main research work of this article is divided into the following two aspects:(1)A method for generating adversarial networks based on multi-level jump residual groups was proposed.Firstly,by improving the residual block to construct a multi-level jump residual group module,the generative adversarial network can better combine the image features of each layer;Secondly,the method of multi loss fusion was used to optimize the network and enhance the true texture of the reconstructed image;Finally,an end-to-end mode is adopted to perform blind deblurring on motion blurred images and output clear images.The experimental results on the CelebA dataset show that compared to CNN based methods such as DeblurGAN and SRN DeblurNet(Scale current Network),the proposed method has improved peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)by at least 0.47 dB and 0.05,respectively.The proposed method has fewer model parameters,faster repair speed,and more texture details in the restored image.(2)An image super-resolution reconstruction network based on adaptive pyramid was proposed.Firstly,by building an image pyramid to achieve superresolution reconstruction of images with different resolutions,the resulting superresolution images have richer details and hierarchical structures;Secondly,using adaptive dense connected residual blocks as the core module for super-resolution reconstruction makes the network pay more attention to the high-frequency information of the image;Finally,in the discriminator,use the adaptive average pooling layer and 1 × The convolutional layer of 1 achieves fixed size output of feature maps and reduces the number of model parameters.Finally,the experimental comparison results on GoPro and other datasets demonstrate the superiority of our method,as the reconstructed image details,texture,and contour features are more abundant.