Research On Image Super-resolution Based On Deep Learning

Image is a very popular information carrier,which has a wide range of applications.In terms of resolution,the image has high-resolution image and low-resolution image.High-resolution image means that the image contains more pixels and carries more information.High resolution image is often more valuable than low resolution image.So scholars study from many aspects to improve the resolution of the image.Image super-resolution technology as a software research results,has an important research significance.With the development of technology,super-resolution technology has a higher development prospect.In this paper,the super-resolution technology of single image is studied,and the corresponding high-resolution image of single low-resolution image is generated by building network model structure using the relevant knowledge of deep learning.In the research of image super-resolution,in order to improve the quality of the generated image and show a clearer detail texture,the method in this paper is based on the generative adversarial networks in deep learning.The main research contents include three aspects: one is the construction of the generation network structure;In order to improve the generation quality of the generation network,improve the generated network model structure,use different large and small convolution check features to extract,modify the residual network model structure,delete the batch normalization layer,and improve the feature extraction ability of network structure by deepening the network hierarchy.The second is to establish a double identification network structure,which is used for the identification of real or false images and the identification of content loss.Third,in order to optimize the network structure,reduce the error loss of data and improve the ability of network generation and identification,this paper defines a number of loss functions to judge the loss in the network,Then the optimization experiment of the model is carried out by the optimization method;finally,through a variety of methods to analyze the experimental results.The method is compared with bicubic and srgan methods,and the experimental results are evaluated by PSNR and SSIM.The experimental results show that the high-resolution image generated by this method has a good display on the clarity and texture.