Single Image Super-Resolution Based On Interleaved Group Convolution

With the continuous development of image processing technology,super-resolution reconstruction technology plays an increasingly important role in the field of image restoration and reconstruction due to its advantages such as no hardware involved,fast and flexible algorithm,easy to be combined with other theories,and close connection with practical requirements.The so-called super-resolution reconstruction,simply speaking,is the use of a single or multiple low-resolution images,after a certain algorithm processing of the relevant scenes of high resolution images.Super-resolution reconstruction technology has been widely used in remote sensing,military,surveillance,biomedicine and other fields.With the continuous development of artificial intelligence,deep learning algorithm has been studied in more and more fields.Based on its high data fitting ability,many scholars have successfully applied deep learning to the super-resolution reconstruction of images.However,the increasing depth and complexity of the network brings information loss,data and computing redundancy,which are not suitable for small devices.To solve these problems,a new lightweight network model based on interlaced group convolution is proposed for the super-resolution reconstruction of single frame images.This paper first introduces the research background,basic theory and research status of super resolution reconstruction technology.The development,research significance and common applications of super-resolution reconstruction algorithm are described in detail.In addition,we study the super-resolution reconstruction algorithm of single frame image based on deep learning.Based on the analysis of the traditional classical algorithm,this paper proposes an algorithm to introduce the residual elements and convolution into the classical neural network structure.The core idea of this algorithm is to broaden the network structure and enhance the sparsity of the convolution kernel by means of convolution,so as to reduce the amount of calculation and parameters and optimize the algorithm.After a lot of experimental evaluation,the super-resolution reconstruction algorithm based on single frame image proposed in this paper improves the shortcomings of the classical algorithm,and its performance is better than that of the classical algorithm based on learning.Under the same experimental conditions,and through the peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)these two kinds of image evaluation standard of performance comparison experiment shows that this algorithm can give attention to both scale and objective evaluation index,neural network model with a smaller number of ginseng reached a relatively better results,and has a better effect on the vision.