Abstract: With the analysis of the current methods for image registration, an improved method based on feature point is proposed. Firstly, segmentation are conducted to the reference image and a number of feature points are extracted from each sub-picture to ensure the feature points are evenly distributed over the image. Subsequently, the known corresponding points are fitted to the transformation equation to obtain the rough matched points with the pending registration points. Then a small area search is carried out for precision registration with the rough matched point as the center according to the similarity measurement. Finally, the correct corresponding points are determined and the overall image registration is fulfilled. To validate the efficiency of the method, simulation experiments are applied for the optical images with large deformation and with large displacement of the sensor. The results show that the improved method proposed is a good way to solve the registration of optical remote sensing images. It can also reduce time complexity of extracting feature points and finding corresponding points.

Key words: remotesensing image, image mosaic, image registration, feature point, corresponding point

摘要： 在分析当前主要的图像配准方法的基础上, 提出一种改进的基于点特征的遥感图像配准方法: 首先对参考图像进行分块, 每块提取一定数量的特征点, 以确保各块图像特征点分布均匀; 根据已知的同名点对, 拟合变换方程, 将待配准点代入变换方程得到粗匹配点坐标, 再以粗匹配点为中心在一个较小的范围内搜索, 根据相似性测进行精配准, 确定正确的同名点位置, 在此基础上实现整体遥感图像配准。为了验证该方法的有效性, 针对变大和传感器位移较大的光学遥感图像设计了仿真实验。实验结果表明, 改进方法能很好地解决光学遥感图像之间的配准问题, 降低特征点提取和同名点匹配的时间复杂度。

关键词: 遥感图像, 图像镶嵌, 图像配准, 特征点, 同名点