Remote Sensing Image Registration Based on Feature Points of Global Edge

In this paper, in view of the grayscale information and morphological changes of remote sensing images, a registration algorithm based on the distribution information of the global edge and the local information of the feature points is proposed. Lifting wavelet transform is used for multi-scale edge extraction of stable global edge, such as river and coastline. On the basis of SIFT feature point detection, the Equivalent DoG (EDoG) pyramid is constructed and the scale-invariant feature points on edge are extracted. The SIFT descriptor is combined with the global descriptor to construct Global-SIFT (G-SIFT) descriptor. The algorithm is tested on the remote sensing images before and after severe natural disasters such as earthquakes, floods and storms. Compared with the performance of SIFT algorithm, the algorithm has advantages in terms of algorithm stability, matching accuracy and registration precision. It can meet the registration accuracy requirement in remote sensing image with large changes.