结合tetrolet与红外显著性特征提取的红外与可见光融合  

Infrared and Visible Light Images Fusion Based on Tetrolet Transform and Infrared Saliency Feature Extraction

作　　者：姜迈 沙贵君[1] 李宁[2] JIANG Mai;SHA Gui-jun;LI Ning(Criminal Investigation Police University of China,Criminal Investigation and Counter-terrorism College,Shenyang 110854,China;Chinese Academy of Sciences,Shenyang Institute of Automation,Marine Information Technology Equipment Centre,Shenyang 110169,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]中国科学院沈阳自动化研究所海洋信息技术装备中心,沈阳110169

出　　处：《科学技术与工程》2022年第30期13398-13405,共8页Science Technology and Engineering

基　　金：公安部技术研究计划(2020JSYJC26);公安部科技强警基础工作专项(2018GABJC08);中国科学院海洋信息技术创新研究院前沿基础研究项目(QYJC201913);公安理论及软科学研究计划(2019LLYJXJXY055,2019LLYJXJXY057)。

摘　　要：针对红外与可见光图像融合过程中红外热目标不突出、纹理及边缘细节易缺失等问题,提出一种结合tetrolet变换域与红外显著目标特征提取的融合方法。首先,在鲁棒加速特征(speed-up robust features,SURF)框架内构建基于梯度直方图(histogram of oriented gradient,HOG)的特征点描述符实现红外与可见光图像的精确匹配;其次,基于贝塞尔面结合背景及目标进行自适应抑制完成红外目标显著性特征提取;接着,将处理后的红外与可见光图像通过tetrolet多尺度变换分解为低频和高频分量;然后,利用基于局部能量和相对亮度自适应规则对低频分量进行融合,对高频分量采用基于局部空间频率自适应融合规则;最后,将融合的低频分量与高频分量通过tetrolet逆变换,以获得最终的融合结果。实验结果表明,本文算法对不同场景下的红外与可见光图像的融合效果不但主观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,运行时间较其他算法得到了明显提升,并且在客观评价指标上也取得了较好的效果。An algorithm for infrared and visible image fusion based on tetrolet transform and infrared saliency feature extraction is proposed to resolve the traditional image fusion problems like dim red target,loss of edge and texture details.Firstly,the histogram of oriented gradient(HOG) feature point descriptor based on speed-up robust features(SURF) was constructed for infrared and visible image registration.Secondly,the infrared saliency feature extraction method through Bezier interpolation and background and target suppression was offered.Subsequently,the infrared and visible images were divided into low frequency components and high frequency components based on tetrolet transform,respectively.Then,for the low frequency components fusion rule,the adaptive local energy and relative intensity was used.In the high frequency components fusion rule,the adaptive local space frequency was adopted.The final fusion result was obtained by tetrolet inverse transform.Experimental results show that the algorithm with prominent target characteristics,clear background texture and edge details,suitable contrast can achieved in subjective visual,with the quickest running speed,and with advantages in the objective indicator evaluations as well.

关 键 词：图像融合 图像匹配 红外显著性特征 tetrolet变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]