Research And Implementation Of Image Super-resolution Reconstruction Based On GAN

Image super-resolution reconstruction technology refers to the process of using relevant knowledge and methods to recover corresponding high-resolution images from low-resolution images,and is widely used in medical image processing,video surveillance,biometric identification and other scenes.In the field of computer vision,super-resolution reconstruction techniques based on deep learning have undergone considerable advancement;however,certain limitations remain,such as network learning difficulties,insufficient feature extraction,blurred image generation and lack of realism.The main work of this paper is as follows:(1)To address these problems,this paper proposes a new image super-resolution reconstruction algorithm based on the improved generative adversarial network.The algorithm is improved on the basis of the original generative adversarial network,and the overall architecture is composed of a deep convolutional neural network and residual blocks.1)In the design of generator network,in order to solve the problem of insufficient feature extraction,this paper proposes a dual network structure.The dual network structure is divided into an upsample subnetwork and a refinement subnetwork,which upsample and optimize a low-resolution image,respectively.In a scene with large upscaling factors,this structure can enhance the utilization of high-frequency details,thereby generating highquality reconstruction results.Besides,in order to solve the problem of generating image blur,this paper uses the perceptual loss,which exhibits a faster convergence speed and excellent visual effect,to guide the generator network training.In this part,this paper applies the ResNeXt-50-32×4d network,which has few parameters and a large depth,to calculate the loss to obtain a reconstructed super-resolution image that is much sharper and highly realistic.Finally,in order to further enhance the learning ability of the network,this paper improves the components of the residual block and applies it to the research of the algorithm model to assist the model to better extract features.2)In the design of the discriminator network,in order to solve the problems of the generative adversarial network exhibits limitations such as unstable training and the occurrence of the vanishing gradient and mode collapse,this paper introduces the Wasserstein distance in the WGAN and the gradient penalty in WGAN-GP to measure the distance between the generated super-resolution image and the real high-resolution image,thereby optimizing the network,enhancing the discrimination ability of the network and the stability of the model.Specifically,this paper employs this distance to eliminate the activation function in the last layer of the discriminator network and avoid the use of the logarithm in calculating the loss function.(2)Experiment with the algorithm model proposed in this paper to verify its effectiveness.This paper uses the Image Net dataset to pre-train the proposed model to improve the overall performance of the model.In this paper,extensive experiments are carried out on the DIV2 K,Set5,Set14,and BSD100 datasets,and the experimental results are compared and analyzed in detail.The experimental results demonstrate that the proposed algorithm in this paper can effectively increase the image resolution and enhance the image quality,and the PSNR and SSIM evaluation indexes of the reconstructed image are improved.Overall,the experimental results fully demonstrate the effectiveness of the proposed algorithm.(3)Design and implementation of image super-resolution reconstruction system.Based on the above algorithm research,this paper designs and implements a visual image super-resolution reconstruction system with a Web interface.This system builds a graphic website platform,which integrates image loading module,model processing module and system interaction module to provide users with online image super-resolution reconstruction function.The user sends the image super-resolution reconstruction request through the Web front-end page,the server receives the request and makes the correct response,and displays the finally result to the user.This system greatly reduces the network bandwidth required for image transmission and saves the cost.