Research On Super-resolution Image Reconstruction Method Based On Deep Learning

Image resolution is an important evaluation indicator for image quality and definition.In the process of image acquisition,many objective factors will affect the resolution of the image,such as the limited imaging capabilities of hardware instruments and environmental factors,etc.The function of the super-resolution image reconstruction algorithm is to use the existing low-resolution images to obtain high-resolution images.How to improve the information that can be transmitted by the image and then obtain high-resolution images to meet the practical needs will be of great research value and significance.In this paper,the image reconstruction algorithm based on generative adversarial network is improved,which makes the content of reconstructed image substantial and the visual effect is good.The algorithm model training strategy is optimized to improve the convergence speed of the model and reduce the time complexity of model training.Finally,a super-resolution image reconstruction system is designed and implemented based on the proposed algorithm.The specific work of this paper is as follows:(1)Aiming at the problem that the high-resolution image generated by the super-resolution image reconstruction algorithm based on generative adversarial network contains too many artifacts and the potential information of the image is lost,a super-resolution image reconstruction algorithm based on generative adversarial network and edge detection(ESRGAN)is proposed.In this algorithm,an edge detection model is designed,and a new network loss function is proposed by combining with the generative adversarial network.At the same time,the depth of the network is deepened,and the residual network structure with better effect is adopted to further improve the effect of image detail restoration.Experimental results show that the algorithm has a good performance in super-resolution image reconstruction and is superior to other contrast algorithms.(2)Aiming at the problem of slow convergence speed and high complexity of model training time in super-resolution image reconstruction algorithm model based on generative adversarial network and edge detection,a model training strategy is proposed.This strategy optimizes the process of image extraction during image preprocessing and the network structure during model training,introduces pixel offsets to realize the maximum extraction of feature information in the image,and adaptively adjusts the model structure during model training,and changes the introduce time of edge detection algorithm model.The experimental results prove that the algorithm model convergence speed is accelerated and the model training time complexity is reduced after the optimization strategy is used.(3)According to the proposed algorithm,a super-resolution image reconstruction system is designed and implemented.The system adopts the form of browser page access to provide user registration,image reconstruction,image reconstruction history query,model training and other functions,effectively solve the user demand for high-resolution image also provides important support for the subsequent upgrade and iteration of the algorithm model.