基于角点和Hu矩不变量的可见光和红外图像自动配准方法  

Automatic Registration of Visible and Infrared Images Based on Corners and Hu Invariant

作　　者：窦建方[1] 李建勋[1] 

机构地区：[1]上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海200240

出　　处：《红外》2011年第7期23-27,共5页Infrared

基　　金：国家自然科学基金(60874104)

摘　　要：针对红外图像与可见光图像的自动配准问题,提出了一种基于图像角点特征和透视变换模型的方法。首先采用自适应阈值对红外与可见光图像进行分割,然后利用Harris因子分别在分割后的红外和可见光图像上检测角点。通过分析角点邻域在原始图像上的相关性实现角点的粗匹配。接着通过RANSAC算法对角点进行细匹配,删除outliers,再由Hu矩不变量进一步提纯角点。最后利用最终匹配的角点作为透视变换的控制点,得到透视变换模型。用该模型对待配准图像进行透视变换,可实现图像配准。实验结果表明,该方法的配准精度高,可以很好地完成红外与可见光图像的自动配准。According to the automatic registration of visible and infrared images,a new method based on Harris Corner and Hu invariant is proposed.In this method,the adaptive thresholds are used to segment visible and infrared images and the Harris factors are used to detect the corners in the segmented visible and infrared images respectively.After analyzing the correlation of the corner neighborhoods in the original images,the coarse corner matching is realized.Then,the RANSAC algorithm is used to remove the outliers in the coarse matched corners and the Hu invariant is used to further filter the wrong matching. Finally,by using the matched corner as the control point of projection transformation,a projection transformation model is obtained.The model can be used to conduct the projection transformation of the images to be matched and make the images registered.The experimental result shows that the method has a high accuracy in registration and can automatically register the visible and infrared images.

关 键 词：图像配准 角点 Harris因子 透视变换 Hu矩不变量 RANSAC算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]