基于深度学习的红外和可见光图像融合  

Infrared and Visible Image Fusion Using a Deep Learning

作　　者：李宏伟 LI Hong-wei(School of Information Engineering,Hebei GEO University,Shijiazhuang 050030,China;Intelligent Sensor Network Engineering Research Center of Hebei Province,Shijiazhuang 050030,China)

机构地区：[1]河北地质大学信息工程学院,河北石家庄050030 [2]河北省智能传感物联网技术工程研究中心,河北石家庄050030

出　　处：《新一代信息技术》2021年第15期9-15,共7页New Generation of Information Technology

基　　金：国家自然科学基金(项目编号:61806069)。

摘　　要：近年来,深度学习已经成为一种非常流行的研究工具,在许多图像处理领域得到了应用。本文提出了一种有效的图像融合方法。首先,将红外图像运用中值滤波去除噪声,然后将源图像分解为基础部分和细节部分。然后进行融合。对于细节部分,使用深度学习网络提取多层特征。利用这些特征,使用l1范数和加权平均策略来生成融合细节部分的多个候选。得到这些候选后,用最大选择策略来得到最终融合的细节部分。最后,融合后的图像将通过融合基础部分和细节部分进行重构。实验结果表明,本文的方法在客观指标评价和视觉感官两方面都达到了较好性能。In recent years,deep learning has become a very popular research tool and has been applied in many image processing fields.An effective image fusion method is proposed in this paper.Firstly,the infrared image is denoised by median filter,and then the source image is decomposed into basic part and detail part.After that,it is fused by.For the detail part,deep learning network is used to extract multi-layer features.Using these features,norm and weighted average strategy are used to generate multiple candidates for fusion details.After these candidates are obtained,the maximum selection strategy is used to obtain the details of the final fusion.Finally,the fused image will be reconstructed by fusing the basic part and the detail part.The experimental results show that the method in this paper achieves good performance in both objective index evaluation and visual sense.

关 键 词：计算机软件与理论 红外与可见光图像 深度学习 图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]