基于SURF-HOG与显著性特征的红外可见光图像配准融合  

Infrared and visible light images registration and fusion based on SURF-HOG and saliency feature

作　　者：姜迈 郑岩[1,2] JIANG Mai;ZHENG Yan(Criminal Investigation and Counter-Terrorism College,Criminal Investigation Police University of China,Shenyang 110854,China;Ministry of Public Security Key Laboratory of Trace Inspection and Identification 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,辽宁沈阳110854 [2]中国刑事警察学院痕迹检验鉴定技术公安部重点实验室,辽宁沈阳110854

出　　处：《激光与红外》2023年第2期261-270,共10页Laser & Infrared

基　　金：公安部技术研究计划项目(No.2020JSYJC26,No.2019JSYJC23);公安理论及软科学研究计划项目(No.2019LLYJXJXY055,No.2019LLYJXJXY057);辽宁省自然基金指导计划项目(No.2020-MS-131);中国科学院海洋信息技术创新研究院前沿基础研究项目(No.QYJC201913)资助。

摘　　要：针对现有红外与可见光图像配准不精确,边缘及细节纹理缺失,融合时间较长,不能突出重点目标等不足,提出一种基于SURF-HOG描述符与红外显著性特征的红外与可见光图像融合方法。首先,在红外与可见光图像配准阶段,在SURF(Speed-Up Robust Features,SURF)框架内构建基于HOG(Histogram of Oriented Gradient,HOG)的特征点描述符,并通过NNDR(Nearest Neighbor Distance Ratio,NNDR)进行红外与可见光图像的特征点匹配;其次,在显著特征提取阶段,先通过四叉树算法对源红外图像分解,然后通过贝塞尔插值法重建红外图像背景,接着分别对红外图像中的背景及目标进行自适应抑制以提取目标红外显著性特征;最后,结合已配准的可见光图像与重建后的红外图像以获取最终融合结果。实验结果表明,所提方法对不同场景下的红外与可见光图像具有较高的配准精度,不同场景下的融合结果不但主观视觉上具有显著的目标特征,同时背景纹理和边缘细节清晰,整体对比度适宜,运行时间最短,并且在客观评价指标上也取得了较好的效果。In order to solve the problems of existing infrared and visible images, such as inaccurate alignment, loss of edges and detailed textures, long fusion time, and failure to highlight key targets, a fusion algorithm based on SURF-HOG descriptor and infrared saliency feature is proposed in this paper.Firstly, in the infrared and visible image alignment stage, HOG(Histogram of Oriented Gradient, HOG) based feature point descriptors are constructed within the framework of SURF(Speed-Up Robust Features, SURF),and feature points matching between infrared and visible image is performed by NNDR(Nearest Neighbor Distance Ratio, NNDR).Secondly, in the salient feature extraction stage, the source infrared image is decomposed by the quad-tree algorithm, and then the background of the infrared image is reconstructed through Bessel interpolation, followed by adaptive suppression of the background and target in the infrared image respectively to extract the target infrared saliency feature.Finally, the final fusion image is obtained by combining the reconstructed infrared image and aligned visible image.The experimental results show that the proposed algorithm has high alignment accuracy for infrared and visible images in different scenes, and the fusion results in different scenes not only have significant subjective visual characteristics of the target, but also have clear background texture and edge details, appropriate overall contrast, and the shortest running time, and also achieve better results in objective evaluation indexes.

关 键 词：红外与可见光图像 SURF-HOG 四叉树分解 红外显著性特征 配准融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]