基于几何不变性和局部相似特征的异源遥感图像配准算法

Registration Algorithm for Heterogeneous Remote Sensing Images Based on Geometric Invariance and Local Similarity Features

针对异源遥感图像在图像配准中的几何形变问题,本文提出了一种基于几何不变性局部相似特征的异源遥感图像配准算法。GISS算法利用加速鲁棒特征算子先对存在几何差异的异源遥感图像进行预匹配,然后根据特征点的方向特征对图像进行旋转仿射校正,最后引用局部相似性描述符并集成相似性度量来考察预匹配点对的相关性,选取其中相似相关性最优的点对实行图像配准。实验结果表明,对于存在几何形变的异源遥感图像,具有较好的配准实现效果,可以有效的解决异源遥感图像之间的几何形变差异问题,具有较好的鲁棒性和配准精度。

This paper proposes a registration algorithm based on geometric invariance and local similarity features (also known as GISS (geometric invariant self-similarities)), to address the problem of geometric deformation of remote sensing images during image registration. The GISS algorithm first uses the SURF operator and the Euclidean distance to pre-match the heterogeneous remote sensing images with their geometric differences, thereafter rotates the images according to the directional characteristics of the feature points, and finally uses the local self-similarity descriptors and integrates similarity measures to examine the phase of the pre-matched point pairs. The experimental results show that, for remote sensing images with geometric deformations, it has a superior effect on registration, can effectively resolve the problem of geometric deformation between the images, and therefore, ensures better robustness and registration accuracy.