用Harris-Laplace特征进行遥感图像配准

为克服图像在旋转及分辨率不一致时自动配准的困难,提出了一种新的自动配准方法:包括新的尺度空间投影算法和新的特征匹配算法。基于Harris-Laplace(H-L)特征的尺度不变性,新的尺度空间投影法运用H-L方法提取图像特征,然后将特征空间建立在依据特征点主方向的图像投影信息上,使得特征空间具有对图像旋转和分辨率大小不变的特性。新的特征匹配法则采用特征空间k-d树欧氏距离匹配和RANSAC一致性位置检验相结合的方法,实现了高效率无差错的特征匹配。通过比较分析与实验证明,该自动配准方法能够对不同分辨率、不同旋转角度的图像精确地实现自动配准。

Remote Sensing Image Registration by Using Harris-Laplace Features

This paper presents a new method for automatic image registration, in which. Harris-Laplace detector is applied to extract image features, the scale space projection based on scale-invariance of Harris-Laplace is proposed, then, a feature space on the projection information in the direction of feature main orientation is constructed, making feature space rotation-invariant and resolution-invariant. The feature matching adopts feature space k-d tree distance matching to combine with RANSAC, which implements feature matching effectively and satisfactorily. The experiments and comparison analyses show that the method in this study can performe automatic image registration accurately in the different resolutions and rotations.results proved that the studied method was effective and robust.