基于改进ORB算法的近红外和可见光图像特征匹配方法  

Near infrared and visible image feature matching method based on improved ORB algorithm

作　　者：杨金玲 曹先革 王霞迎 万伟 YANG Jinling;CAO Xiange;WANG Xiaying;WAN Wei(School of Surveying and Mapping Engineering,East China University of Technology,Nanchang 330013,China;Center for Higher Education Research and Evaluation,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学测绘工程学院,南昌330013 [2]东华理工大学高等教育研究与评估中心,南昌330013

出　　处：《测绘工程》2022年第6期7-12,共6页Engineering of Surveying and Mapping

基　　金：东华理工大学2019年度博士科研启动基金项目(DHBK2019194,DHBK2019193)。

摘　　要：提出一种针对近红外和可见光图像匹配的改进ORB算法,首先通过引入BRISK算法构建金字塔的思想进行图像金字塔的构建,然后对探测出的特征点进行子像素和尺度上的进一步插值,得到修正后的特征点,最后将特征点按匹配质量得分大小进行排序,采取逐渐增大采样样本数的策略进行误匹配的剔除,并根据非随机性和极大值策略停止“采样-验证”循环。实验结果表明,提出的改进算法与ORB算法相比,在近红外图像上提取特征点数增加了17.21倍、在可见光图像上提取特征点数增加了12.34倍、正确点对数增加了23.11倍;基于渐进采样一致性算法进行误匹配点对的剔除,其用时与经典RANSAC算法相比减少约14%。In this paper, an improved ORB algorithm for near infrared and visible image matching is proposed. First, the idea of constructing pyramid by BRISK algorithm is introduced to construct the image pyramid. Second, the feature points are further interpolated on the sub-pixel and scale in order to get the refined feature points. Finally, the feature points are sorted according to the quality score, the strategy of gradually increasing the number of samples is adopted to eliminate the false matches, the "sampling-validation" cycle is stopped according to the non randomness and maximum strategy. The experimental results show that compared with ORB algorithm, the number of feature points detected in near infrared image is increased by 17.21 times and in visible image is increased by 12.34 times, the number of correct matches increased by 23.11 times. In this paper, we use progressive sample consensus algorithm to eliminate false matches, compared with the classical RANSAC algorithm, the time consumption is reduced by about 14%.

关 键 词：改进ORB算法 近红外图像 可见光图像 BRISK算法 特征匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]