Image Super-resolution Reconstruction Algorithm Based On Deep Learning

As a carrier of objective object information,image is widely used in human society.In the era of information explosion,human beings are increasingly desperate for high-resolution images.However,due to the high cost of hardware equipment and the adverse environmental factors of image shooting,the resolution of the collected images often fails to achieve the desired effect.Therefore,how to improve the image resolution through cheaper and more convenient software technology is particularly important.Image super-resolution reconstruction is a process in which a series of algorithms are used to reconstruct a high-resolution image from a given low-resolution image.In recent years,due to the continuous development of computer technology,deep learning algorithm has achieved good reconstruction effect in the field of image super-resolution reconstruction.This paper focuses on the research hotspot in recent years and studies the defects of Image Super-resolution Using Deep Convolutional Networks(SRCNN).In order to solve the problem of the loss of some important high frequency texture details during the preprocessing of low-resolution images by using Bicubic interpolation and the optimization of the network model,a super-resolution reconstruction method combining continuous fraction interpolation with convolution neural network is proposed.On the basis of the original lightweight SRCNN network model,the Newton-Thiele continuous fraction interpolation function is firstly used in this paper to interpolate the low-resolution image to the target size;then three convolution layers are used for image feature extraction,nonlinear mapping,reconstruction and optimization;at the same time,cosine attenuation method is used to gradually reduce the learning rate,and Radam algorithm is used to adjust the learning rate adaptively when the network converges.The experimental results show that the network model designed in this paper can obtain richer texture details and clearer image edges under the lightweight convolutional neural network.