关键词: 遥感图像, 图像配准, 控制点, 分布质量, 尺度不变特征变换, 随机抽样一致性

Abstract:

In order to improve the registration accuracy of remote sensing image,a method for selecting control points is proposed based on uniform distribution quality.In this method,first,control points are extracted via the scale-invariant feature transform,and wrong matched control points are deleted by means of random sample consensus（RANSAC）.Then,the common area of the image pair is divided into several sub-regions and suitable control points are selected according to the distribution quality.Finally,the parameters of the affine transformation model are estimated through the least squares approximation and the registration is conducted via the coordinate transformation of the input image.Experimental results of various kinds of remote sensing images prove that the proposed method effectively removes the wrong matched control points,guarantees the distribution uniformity of control points and reduces the registration error.

Key words: remote sensing image, image registration, control point, distribution quality, scale-invariant feature transform, random sample consensus