红外图像与可见光图像记录着地物的不同属性信息, 两者融合能够优势互补, 弥补单一数据源信息的不足。然而由于两者成像原理不同, 热红外传感器与可见光传感器对同一场景获取的图像灰度差异较大, 二者图像误匹配多, 融合难度大。本文在分析红外与可见光图像共有特征的基础上, 提出了一种基于 SIFT与 ORB特征检测的匹配方法, 利用 SIFT算子与 ORB算子同时进行特征点检测, 先基于 RANSAC对 SIFT匹配得到的同名点进行筛选, 同时结合最近邻比次近邻算法获取 ORB匹配点, 再利用 SIFT匹配点对 ORB匹配点进行距离和角度的几何约束进一步剔除误匹配, 最终得到特征点分布均匀、可靠度更高的匹配结果, 解决因灰度差异较大产生的匹配效果不佳的问题。利用 4组红外与可见光图像进行实验, 结果表明, 本文算法特征点正确匹配数量相较于 SIFT分别提高了约 3.7倍、 3.2倍、3.6倍、3倍, 大幅地提高了红外与可见光图像的匹配数量, 为两者间的匹配提供了一种有效的方法。

Infrared images and visible light images record different aspects of the nature of a ground object, such that the fusion of two such images of the same object can compensate for a lack of information from a single data source. However, due to the distinct imaging principles involved, the difference between the same-scene images produced by a gray image sensor and a visible light sensor is large, resulting in mismatched images that are difficult to fuse. In this paper, a matching method based on the analysis of the common features of infrared and visible light images using SIFT and ORB feature detection is proposed. The SIFT operator and the ORB operator are used to simultaneously perform feature point detection. First, the same name is obtained, using RANSAC, for SIFT matching. The points are filtered, and the nearest neighbor neighboring nearest neighbor algorithm is used to obtain the ORB matching points. Then the SIFT matching points are used to geometrically constrain the distance and angle of the ORB matching points to further reduce the mismatch. Ultimately, the feature points are evenly distributed and the reliability is higher, solving the poor-matching-effect problem. The performance of the proposed method was compared with that of SIFT using four sets of infrared and visible images, with the proposed method achieving a number of correct matching feature points approximately 3.7 times, 3.2 times, 3.6 times, and 3 times higher than those achieved with SIFT. This significant performance improvement indicates the effectiveness of the proposed method.