Research On Image Super-resolution Method Based On Generative Adversarial Network

In the process of obtaining images in various fields,high-resolution images cannot be obtained due to factors such as shooting distance,camera resolution,blurring,noise,down-sampling,and deformation.However,high-resolution images are widely used in various fields,such as medicine,military,and transportation.Therefore,how to transform the obtained low-resolution image into a high-resolution image and improve the image quality has always been an urgent problem in the field of imaging technology.The traditional method of using hardware is expensive and the quality of the obtained image is poor.The popularity of deep learning has brought new research directions to image over scoring.This thesis will continue to develop in the direction of image super-resolution technology based on generative adversarial networks in the field of deep learning.On the research of this issue,this thesis mainly carries on the following aspects:First,the theoretical knowledge of deep learning and image super-resolution technology is studied in detail.First of all,we systematically analyze and study the field of deep learning.Then we understand the classification and evaluation indicators of super-resolution reconstruction technology,and introduce the general process and classic models of image super-resolution reconstruction based on deep learning.Afterwards,we systematically analyze and research the generative adversarial network model and the SRGAN based on the image supermodel of the adversarial network.A series of experiments were conducted to explore the impact of standardization on the SRGAN model.Then,aiming at the shortcomings of the SRGAN model,combined with the newer network models that have appeared recently,a new model structure SRCWGAN is constructed.Finally,use the new network model for experiments,and use 4 classic datasets in the field of image super-segmentation: Set5,Set14,BSD100,Urban100 for testing,and existing model structures such as Bicubic,SRCNN,VDSR,DRCN and SRGAN We compared the numerical indicators of PSNR and SSIM to verify that the network model constructed in this paper has better performance in image super-reconstruction.