基于生成对抗网络的单帧红外图像超分辨算法
投稿时间:2017-12-12  修订日期:2018-05-21  点此下载全文
引用本文:邵保泰,汤心溢,金璐,李争.基于生成对抗网络的单帧红外图像超分辨算法[J].红外与毫米波学报,2018,37(4):427~432].SHAO Bao-Tai,TANG Xin-Yi,JIN Lu,LI Zheng.Single frame infrared image super-resolution algorithm based on generative adversarial nets[J].J.Infrared Millim.Waves,2018,37(4):427~432.]
摘要点击次数: 380
全文下载次数: 254
作者单位E-mail
邵保泰 中国科学院上海技术物理研究所 shaobaotai@sina.com 
汤心溢   
金璐   
李争   
基金项目:国家十三五国防预研项目(Jzx2016-0404/Y72-2),上海市现场物证重点实验室基金资助项目(2017xcwzk08)
中文摘要:高分辨率红外图像的获取受到了硬件性能的限制, 利用信号处理的方法实现红外图像的超分辨率重建可以有效地提高红外图像的分辨率.将基于深度学习的超分辨方法应用于红外图像, 实现了单帧红外图像的超分辨率重建, 获得了更好的评价结果.通过引入对抗训练的思想, 以及添加基于判别网络的损失函数分量, 提高了放大倍数的同时, 获得更好的高频细节恢复, 图像边缘锐化, 避免了超分辨率红外图像过于模糊.
中文关键词:红外图像  超分辨率重建  深度学习  生成对抗网络
 
Single frame infrared image super-resolution algorithm based on generative adversarial nets
Abstract:Image processing makes super-resolution infrared image reconstruction effectively improve infrared images resolution, w hich breaks through hardw are performance limits. Based on deep learning, super-resolution method is applied to infrared image, w hich enables the super-resolution reconstruction of single-frame infrared image. Thus, better evaluation results are acquired. Derived from adversarial thoughts, adding a loss function based on discriminant netw ork can improve magnification, w hich can access to better high-frequency details of the restoration and can sharpen image edge and avoid blurred super-resolution infrared images.
keywords:infrared image, super resolution, deep learning, GAN
查看全文  查看/发表评论  下载PDF阅读器

版权所有:《红外与毫米波学报》编辑部