一种基于虚拟三角形的图像自动配准方法  

An Automated Image Registration Method Based On Virtual Triangle

作　　者：吕金建[1] 文贡坚[1] 李德仁[2] 

机构地区：[1]国防科技大学电子科学与工程学院ATR重点实验室,湖南长沙410073 [2]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《信号处理》2008年第5期737-741,共5页Journal of Signal Processing

基　　金：国家自然科学基金项目(40571103);国家863项目(2006AA12Z140)

摘　　要：图像配准是图像处理和分析的关键技术之一。本文提出了一种基于虚拟三角形的图像自动配准方法来处理具有全局刚体变换的图像配准问题。该方法主要分三步:首先采用改进的Harris算子从参考图像和待配准图像中分别提取角点特征,将每三个角点顺次连接起来构成一个虚拟三角形;然后运用刚体变换模型下匹配虚拟三角形对全等的准则找到全等性最好的一对虚拟三角形,利用它们的对应顶点求解刚体变换模型参数的初始值;最后根据刚体变换模型参数的初始值和一个预设的门限得到所有的匹配角点特征,通过它们求得最终的刚体变换模型参数。实际图像实验结果表明:本文提出的图像自动配准方法是正确和有效的,并具有较高的配准精度。Image registration is one of crucial technologies of image processing and analysis. An automated image registration method based on virtual triangle is proposed in this paper to deal with registration of images with rigid geometric distortion. The method mainly consists of three steps：firstly, an improved technique based on Harris is used to extract corners from the reference image and sensing image respectively, and then every three Corners are connected to compose a virtual triangle;Secondly, get the most congruent two virtual triangles by the criterion that the matched virtual triangles are congruent under rigid geometric distortion and compute the initial parameters of rigid transformation model by their corresponding points ; Finally,obtain all matched corners according to the initial parameters of rigid transformation model and a preset threshold and then estimate the optimal parameters of rigid transformation model by them. The experiment results of real images prove that the proposed method is accurate, effective and has comparatively high registration accuracy.

关 键 词：特征提取 虚拟三角形 特征匹配 刚体变换 图像自动配准 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP751[自动化与计算机技术—计算机科学与技术]