Research Of Super-Resolution Image Reconstruction Based On Generative Adversarial Network

Super-resolution is a technology that uses software or hardware to obtain the corresponding high-resolution based on existing low-resolution pictures.Nowadays,given a set of images observed at low resolution,trying to reconstruct the original scene with high resolution is also a research hotspot in the field of images.Image super-resolution technology is widely used in video compression and recovery,medical image quality improvement,auxiliary diagnosis,video surveillance and security.It is particularly important to improve the image restoration capabilities of image super-resolution technology,especially to optimize the structure of the neural network,while reducing the consumption of computing resources while improving the effect.This article is based on generative adversarial network.Based on the existing research,it will optimize the adversarial network,modify the network structure and optimize the loss function to improve its super-resolution capability.In particular,the generative adversarial network has a better ability to recover image details to improve the texture quality of image generation.This paper also uses this algorithm as the core to design and complete a visual image enhancement system based on generative adversarial networks.The main work of this article is as follows1.Researched and analyzed the traditional image super-resolution methods,and found that the traditional methods have the problems of recovering the texture loss of the image and smoothing the texture changes where the image texture details are rich.The solution uses an improved residual network in the generator,which is better than SRResNet in SSIM and PSNR indicators,and the details of the obtained picture texture are clearer and the changes are more abundant2.By optimizing the complex neural network structure,removing the batch normalization layer,adding a bicubic interpolation information compensation layer,modifying the neural network structure and modifying the loss function,effectively improving the model's super-resolution capability.The multi-resolution enhanced processing time is compressed to the order of seconds3.Based on the image super-resolution optimization algorithm model,a visual image super-resolution enhancement system is designed and implemented.Pictures are input on the front-end web interface and uploaded to the online server through the network.The server returns the super-resolution enhanced pictures.The system saves computing resource consumption,makes more calculations focus on the server,and greatly improves the availability of the algorithm model;...