卷积自编码融合网络的红外可见光图像融合  

Convolutional Auto-encoding Fusion Network for Infrared and Visible Image Fusion

作　　者：杨勇[1] 刘家祥 黄淑英[2] 张迎梅 吴嘉骅 李露奕 YANG Yong;LIU Jia-xiang;HUANG Shu-ying;ZHANG Ying-mei;WU Jia-hua;LI Lu-yi(School of Information Technology,Jiangxi University of Finance and Economics,Nanchang 330032,China;School of Software and Communication Engineering Jiangxi University of Finance and Economics,Nanchang 330032,China)

机构地区：[1]江西财经大学信息管理学院,南昌330032 [2]江西财经大学软件与通信工程学院,南昌330032

出　　处：《小型微型计算机系统》2019年第12期2673-2680,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61662026,61862030)资助;江西省自然科学基金项目(20182BCB22006,20181BAB202010,20192ACB20002,20192ACBL21008)资助;江西省教育厅科技项目(GJJ170312,GJJ170318)资助

摘　　要：可见光图像能够充分反映场景的细节信息,红外图像能够反映目标的热度信息,利用两者的互补信息进行融合,可以得到具有目标信息和场景细节的图像.本文提出一种基于卷积自编码融合网络的红外与可见光图像融合的方法.首先利用卷积编码网络从两类源图像中提取相应特征,产生在不同维度上包含源图像各项信息的特征图;接着对两类图像的特征图利用等权重相加的融合规则将其进行融合,得到包含两类源图像信息的特征图,然后利用卷积解码网络对其进行重建,最终得到一张包含两类源图像信息的融合图像.通过对多组红外与可见光图像进行实验,实验结果证明本文提出的方法可以得到较好的融合结果.该方法的融合结果无论从主观视觉效果还是客观定量评价,都优于一些主流的红外与可见光图像融合方法.The visible image can fully reflect the detailed information of the scene,and the infrared image can reflect the thermal radiation distribution of the target.The fusion information of those two images can be used to obtain the image with the target information and the scene detail simultaneously.This paper proposes a fusion method of infrared and visible images based on convolutional autoencoding fusion network.Firstly,the convolutional coding network is employed to extract the corresponding features from the two types of source images,and the feature maps containing the information of the source images in different dimensions are generated.Then,these feature maps are fused by an equal-additional fusion rule to achieve a feature map containing two types of source images information.Finally,a convolutional decoding network is performed on the feature map to obtain a fused image containing the source images information.Experimental results on multiple sets of infrared and visible images can show that the proposed method can obtain better fusion results.The fusion results of this method are superior to some mainstream infrared and visible image fusion methods in terms of both subjective visual effects and objective quantitative evaluation.

关 键 词：红外与可见光图像融合 卷积自编码 深度学习 卷积神经网络 纹理梯度信息提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]