摘要
使用数学形态学的"膨胀算子"对影像进行预处理,提出了一种改进的基于高斯拉普拉斯算子的面状特征提取和细化方法,并利用边界代数快速标注边界封闭的面状特征。在提取面状特征的基础上,利用奇异值分解算法,实现了基于面状质心的遥感影像匹配,进而完成精确配准。实验结果表明,与传统方法相比,此方法在速度与准确度上具有明显优势。
Extraction and matching of conjugate image features is prerequisite for registration of multi-sensors images.Feature of image includes points,lines and polygons.We focus on area feature-based image registration for the reason that area features improves registration accuracy.More importantly,area features are often the sole basis for image registration.We propose using ＇dilation＇ operator in mathematical morphology as a pre-processing procedure to prevent boundaries extracted using conventional Laplacian of Gaussian（LoG） operator from becoming discontinuous.We use a boundary algebra algorithm to mark area features with closed boundaries rapidly.We explory singular value decomposition（SVD） to match images based on centroids of area features extracted beforehand.Experiments confirmed that the proposed methods are superior over conventional methods in terms of speed and accuracy.
出处
《武汉大学学报（信息科学版）》
EI
CSCD
北大核心
2011年第6期678-682,共5页
Geomatics and Information Science of Wuhan University
基金
国家自然科学基金资助项目(41071286)
湖北省自然科学基金资助项目(2007ABA276)
关键词
面状特征
高斯拉普拉斯算子
边界代数
奇异值分解
影像配准
area features
Laplacian of Gaussian（LoG）
boundary algebra
singular value decomposition（SVD）
image registration
