一种结合局部与半全局几何保持的影像匹配算法

An image matching algorithm combining local and semi-global geometry preservation

遥感影像匹配是众多遥感应用中数据处理的关键前置步骤,但高程差导致的影像局部畸变和影像匹配的复杂性严重限制了高分辨率影像的匹配精度。提出一种适用于局部畸变和高外点比例的鲁棒匹配算法,首先利用Delaunay剖分算法在假定匹配点集上施加几何约束,得到特征点局部邻接关系;然后基于邻接信息进行预过滤;采用多尺度的策略建立局部邻接关系一致约束模型;最后定义三角形相似度函数实现匹配恢复。利用3组高分辨率影像开展对比实验,实验结果表明该算法的平均精度比RANSAC提高7.69%,在外点率高于90%时仍旧稳健。

Remote sensing image matching is an essential preprocessing step for many remote sensing applications.However,the distortions caused by elevation differences and the complexity of remote sensing image matching severely limit the matching precision of high-resolution remote sensing images.This paper proposes a robust feature matching algorithm suitable for local distortion and high outlier ratio.First,the Delaunay triangulation algorithm is used to impose geometric constraints on the initial matching point set,and the local adjacency relationship of the feature points is obtained.Second,pre-filter is conducted based on the adjacency information.Third,a multi-scale strategy is used to establish the local adjacency consistent model.Finally,a triangle similarity function is defined to achieve matching recovery.The experimental results on high-resolution images show that the average accuracy of our algorithm is 7.69%higher than that of RANSAC,and it is still robust when the outlier ratio is higher than 90%.