摘要
针对红外与可见光图像配准的准确性与鲁棒性问题,提出一种基于R-MI-rényi测度的由粗到精红外与可见光图像配准方法。首先通过分析红外与可见光传感器的成像原理,确定红外与可见光传感器在相对位置固定时的配准变换参数;然后采用Hough变换检测模板图像的直线特征,并利用待配准图像对应直线段的长度和斜率确定粗配准参数;根据粗配准参数确定搜索区域与匹配窗口尺寸,利用rényi互信息与Harris角点函数相结合的R-MI-rényi匹配测度在粗配准对应区域内搜索匹配点对;最后使用RANSAC方法完成图像的精确配准并求解图像转换参数矩阵。分别选取标准测试图像集和真实测试图像集对本文方法和现有代表方法进行综合对比,实验结果表明,方法在像素误差、标准差以及时间消耗等方面均优于其他对比方法,说明方法具有较高的配准精度和效率、较好的鲁棒性,综合性能最优。
Aiming at the accuracy and robustness of infrared and visible images registration,a coarse-to-fine image registration method for infrared and visible images based on R-MI-rényi measurement is proposed in this paper. Firstly,the transformation parameters are determined when the infrared and visible sensors located at the fixed locations by analyzing the different principles of infrared and visible imaging. Secondly,the line features of different images are detected through the Hough transformation,and the parameters of coarse registration are computed with the length and slope of the detected straight-line segments. Thirdly,the sizes of searching regions and matching windows are defined with the parameters of coarse registration,and the matching couples of feature points in the corresponding areas are presented by using the R-MI-rényi measurement which combined the rényi mutual information and Harris function. Finally,the fine images registration is finished and the transformation matrix could be solved by RANSAC. The standard and real test image sets are employed to evaluate the proposed method and the typical existing methods. And the experimental results show that the proposed method has the better performances of the pixel error,standard error and time consumption compared to the other methods. It indicates that the proposed method has higher accuracy,rapider computing efficiency and better robustness.
出处
《电子测量与仪器学报》
CSCD
北大核心
2018年第1期1-8,共8页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(61772255,61462062,61401190)
江西省创新驱动“5511”工程优势科技创新团队(20165BCB19007)
江西省优势科技创新团队计划项目(20152BCB24004)
航空科学基金(2015ZC56009)
江西省青年科学基金(20171BAB212012)
江西省重点研发计划项目(20161BBE50080)资助