面向双模态夜视图像的混合尺度融合算法  

Research on mixed-scale image fusion algorithm for dual-modal night images

作　　者：刘文强[1] 姜迈 乔顺利[1] 李宏达[2] LIU Wenqiang;JIANG Mai;QIAO Shunli;LI Hongda(Criminal Investigation and Counter-Terrorism College,Criminal Investigation Police University of China,Shenyang 110854,China;School of Forensic Science and Technology,Criminal Investigation Police University of China,Shenyang 110854,China)

机构地区：[1]中国刑事警察学院侦查与反恐怖学院,沈阳110854 [2]中国刑事警察学院刑事科学技术学院,沈阳110854

出　　处：《兵器装备工程学报》2024年第5期291-298,共8页Journal of Ordnance Equipment Engineering

基　　金：公安部科技强警基础工作专项项目(2019GABJC06);公安部技术研究计划(2020JSYJC26,No.2019JSYJC23);辽宁省自然基金指导计划(2020-MS-131)。

摘　　要：针对传统红外与可见光图像融合算法存在的细节模糊、对比度降低、背景信息缺失等不足,提出了一种基于混合尺度的红外与可见光融合方法。通过潜在低秩表示变换将源图像分解低秩子带和显著子带;利用非下采样轮廓波变换将低秩子带继续分解为低频分量与高频分量;针对显著子带采用基于卷积稀疏表示的方法进行融合;并结合全局均值、区域均值与能量的优势融合低频分量;利用权重决策图融合高频分量。基于自建库及公开库的实验结果表明,与其他5种图像融合算法相比,所提算法在充分继承源图像有效信息的同时,融合图像整体对比度更均衡,有效提升了融合图像的清晰度,包含更丰富的图像细节信息,在主客观评价上均取得了更好的效果。Aiming at the problems of detail blur,low contrast and loss background information in the traditional infrared and visible image fusion algorithm,this paper proposes a new infrared and visible image fusion algorithm based on mixed-scale.Firstly,the source images are decomposed by latent low rank representation decomposition to obtain the low rank sub-band saliency sub-band,respectively.Secondly,through non-subsampled contourlet transform,the infrared and visible image low rank part are decomposed into low frequency and high frequency sub-bands,separately.Thirdly,the saliency sub-bands are fused by using convolutional sparse representation rules.Then,combining the global and reginal mean value,region energy to fuse the low frequency sub-band.Finally,the weighted decision map diagram is adopted to merge the low frequency sub-band.Experimental of self-built datasets and public datasets results show that the proposed algorithm can fully reserve the source image effective information,with more image balance contrast and sharpness,richness of detail information,superior to other five efficient image fusion algorithms in terms of subjective and objective evaluations.

关 键 词：图像融合 混合尺度 卷积稀疏表示 红外图像 可见光图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]