Remote Sensing Image Registration Using Equivariance Features

We propose a deep learning-based framework for remote sensing image registration using equivariance features. Unlike conventional methods, the networks in the framework are trained on not invariance features but equivariance features since keeping invariancy in the areas of registration of remote sensing images can reduce the accuracy of matching results. Our framework is tested on four sets of KOMPSAT-3 remote sensing images and compared with the conventional machine learning based and the invariant feature based deep learning methods. The experimental results show that the proposed approach outperforms all the comparing methods.