基于增强MSER和Harris-Laplace互补不变特征的遥感图像配准  

Remote Sensing Image Registration Based on Strengthened MSER and Harris-Laplace Local Invariant Features

作　　者：王晓华[1] 李克[1] 邓喀中[2] 杨化超[2] 

机构地区：[1]河南理工大学,河南焦作454000 [2]中国矿业大学江苏省资源环境信息工程重点实验室,江苏徐州221116

出　　处：《红外技术》2015年第1期20-24,共5页Infrared Technology

基　　金：河南理工大学2014年度博士基金资助项目;编号:B2014-07;测绘地理信息公益性行业科研专项项目;编号:201412020;校内青年基金;编号:Q2014-02A

摘　　要：针对具有倾斜的遥感图像的自动配准问题,提出一种增强自动配准方法。该方法首先应用最大极值稳定区域(Maximally Stable Extremal Regions,MESR)特征的仿射不变性结合匹配能力较强的SIFT(Scale Invariant Feature Transformation,SIFT)描述子进行粗匹配,初步校正倾斜图像的空间变换;然后利用Harris-Laplace(H-L)在图像旋转、光照变化条件下能最稳定的提取2维平面特征点和在3维尺度空间中能最稳定高效地提取特征点的特性结合随机一致性检验(Random Sample Consensus,RANSAC)方法进行精匹配。通过实验分析证明,与SIFT配准方法相比该方法能够对倾斜的遥感图像实现更精确的自动配准。A strengthened local invariant feature automatic registration method is proposed for registration of remote sensing images with tilt. The image space transform is corrected by coarse matching with the affine invariant of the MSER features and SIFT descriptor. Then the images are precisely matched by using H-L that can be stable and efficient to extract feature points in 2D and 3D scale space on image rotation and illumination changes condition. Experiment shows that this method can achieve more accurate registration and correct matching rate than SIFT matching method with respect to tilt remote sensing images.

关 键 词：特征配准 最大极值稳定区域 尺度不变特征变换 Harris-Laplace特征 

分 类 号：TN911.73[电子电信—通信与信息系统]