摘要
随着深度学习算法首次被应用于图像超分辨率重构,基于深度学习的重构方法取得了比传统图像超分辨率重构方法更好的重构效果.随后,一系列改进的深度学习算法相继提出,重构效果也不断提升.本文系统地总结了基于深度学习的图像超分辨率重构方法,主要可以分为:基于直连的浅层网络重构方法,基于深层特征的深层网络重构方法和基于生成式对抗网络重构方法.并且对比分析了不同网络模型的特点和不足.在主流数据集上对各种深度学习网络模型进行了比较,并根据当前基于深度学习模型的图像超分辨率重构方法的发展趋势,对基于深度学习模型的图像超分辨率重构方法未来的研究方向做了展望.
With the first application of deep learning algorithm to image super-resolution reconstruction,The reconstruction method based on deep learning achieves better reconstruction results than the traditional super-resolution reconstruction method. Subsequently,a series of improved deep learning algorithms have been proposed,and the effect of reconstruction has been improved continuously.This paper systematically summarizes the image super-resolution reconstruction methods based on deep learning,It can be mainly divided into: reconstruction method based on direct connection of shallownetwork,reconstruction method based on deep network and reconstruction method based on generative adversarial network. The characteristics and shortcomings of different network models are compared and analyzed.. various deep learning network models are compared on the main datasets. Then according to the development trend of the current image super-resolution reconstruction method based on deep learning model,the future research direction of the image super-resolution reconstruction method based on deep learning model is reasonably prospected.
作者
王威
张彤
王新
WANG Wei;ZHANG Tong;WANG Xin(School of Computer and Communication Engineering,Changsha University of Science & Technology,Changsha 410114,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2019年第9期1891-1896,共6页
Journal of Chinese Computer Systems
基金
国防预研项目(7301506)资助
国家自然科学基金项目(61070040)资助
湖南省教育厅科研项目(17C0043)资助
关键词
图像超分辨率重构
卷积神经网络
残差学习
密集连接网络
生成式对抗网络
image super-resolution reconstruction
convolutional neural network
residual learning
dense network
generative adversarial network
