Abstract: Aiming to the high accuracy registration between heterogenous images with differences in angles and scales, the spatially assistant circles with invariant angle and same scale are introduced, and the normalized mutual information robust to spectrum differences is adopted to compute the statistical features in the assistant circles. It confines the searching range to the assistant circle, which increases the registration precision while the searching efficiency is guaranteed. Optical and SAR images with considerable differences in scale and angle are tested, and the registration method based on spatial assistant plane is used for comparison. The experimental results indicate that the registration method based on spatially assistant circle and normalized mutual information is robust to differences in angle and spectrum. Moreover, the registration model’s error is lower than one pixel.

Key words: heterogenous images registration with multi-angle and multi-scale, spatially assistant circle, invariant angle, normalized mutual information

摘要： 针对具有显著尺度和角度差异的异源遥感影像配准中存在的问题，引入具有角度不变性和同一尺度的空间辅助圆，克服角度和尺度偏差；选取对异源影像的光谱差异具有高鲁棒性的归一化互信息测度，基于空间辅助圆提取统计特征进行区域配准。该方法还将同名特征点的搜索范围限制在辅助圆内，在保证算法搜索效率的同时提高了配准的精度。选取具有较大尺度和角度偏差的异源遥感影像进行实验，通过与基于空间辅助面的配准方法相比较，证明基于空间辅助圆的归一化互信息配准方法对角度和光谱偏差具有较高的鲁棒性，配准模型的精度小于一个像素。

关键词: 多尺度多角度的异源遥感影像配准, 空间辅助圆, 角度不变性, 归一化互信息