融合尺度和仿射不变特征的倾斜立体影像配准  

Registration of oblique stereo images by integrating scale and affine invariant features

作　　者：于瑞鹏 金丽鸿 高宁[3] 赵星涛 

机构地区：[1]山东省第一地质矿产勘查院,济南250014 [2]北京建筑大学,北京102600 [3]国家海洋环境监测中心,辽宁大连116021 [4]北京数维翔图高新技术有限公司,北京100073

出　　处：《测绘科学》2016年第7期138-143,共6页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41371438)

摘　　要：针对倾斜立体影像通常存在较大的几何变形从而难以自动匹配的问题,该文提出了一种融合MSER特征和SIFT特征的自动配准算法:选取MSER特征获得初始匹配及其对应邻域的仿射几何关系,对MSER候选匹配邻域内SIFT的特征点进行局部几何约束的仿射不变归一化相关系数匹配,定义邻域内SIFT特征的匹配正确率为邻域支持度,并据此进行原始MSER误匹配进行剔除。对上述未能匹配的MSER和SIFT特征,采用基于局部单应约束进行迭代匹配传播以获得更多的配准控制点并用于精确估计投影变换模型,完成影像的配准。同现有方法相比,所提算法在匹配点数、匹配正确率和配准精度方面有较大优势。Aiming at the problem that it is difficult for the oblique stereo images to perform automatic registration due to the presence of large geometric deformation, the paper proposed an automatic method with integrating the characteristics of MSER and SIFT： the MSER detection method was performed on the reference image and the image to be registered, respectively. Original matches and the affine transforma- tion between their matches＇ neighbor area were obtained. The SIFT interest points located in each neigh- bor were matched by Affine Invariant Normalized Cross Correlation （AINCC） algorithm. Then Neighbor- hood Support Strength （NSS） was proposed to eliminate the original mismatching points. A robust matc- hing propagation under a local homography constraint was performed to obtain much more corresponding image points. With initial matches and propagating matches, accurate projective transformation of the im- age to be registration was achieved by using RANSAC algorithm, then registration of stereo images was finally finished. Compared with existing methods, the proposed algorithm would have better performance in terms of matching number, correcting ratio and registration accuracy.

关 键 词：影像配准 倾斜立体影像 最大稳定极值区域 尺度不变特征变换 单应映射 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]