Research And Application Of Image Super-Resolution Based On Convolutional Neural Networks

Image resolution is an important indexes to measure image information and the higher the resolution is,the richer the information the image contains.Image super-resolution is a technology to improve the resolution,which can up-sample the given low-resolution image to the high-resolution one.Compared with physical methods,image super-resolution technology has the advantages of low cost and high flexibility,and can be widely used in many scenes.In recent years,the super-resolution convolutional neural network has achieved remarkable success,but most of the networks have the disadvantages of high computational complexity and slow reconstruction speed,so their practicability is limited.To solve this problem,a lightweight super-resolution convolutional neural network is proposed and the algorithm is deployed to practical applications.The main contents and innovations of this paper are as follows:(1)A dual-branch convolutional neural networks for super-resolution is proposed.The feature extraction part of the network consists of two branches with different receptive field,this structure can extract both global and local features to improve the multi-scale ability.The network uses deconvolution to up-sample the feature map,which avoids the large-scale operation and reduces the computational complexity.By introducing the residual learning structure,the training difficulty of the network is greatly alleviated,and the reconstruction result is also significantly improved.Experiments on several benchmark datasets show that our algorithm is practical compared to several state-of-the-art methods.Taking B100 data set as an example,the PSNR index of the algorithm is 0.06 db higher and 32 fps faster than that of VDSR algorithm.(2)A super-resolution system is designed and implemented.The system is equipped with the trained model which can improve the image resolution.This paper combines the system with the cloud disk developed by Communication and Computer Network Laboratory of Guangdong Province to enable remote image processing.(3)A Universal Windows Platform app is developed.The app not only supports the cloud disk file management function,but also provides an entry for super-resolution of cloud disk image,which saves the cost of local deployment.