Remote Sensing Image Registration Based on Spatial Transform Network and Phase Correlation Method

Abstract

Remote sensing image registration technology can weaken or even eliminate image position, size difference and deformation caused by imaging equipment, viewing angle and other influencing factors. It has been widely used in urban planning, ecological monitoring and land management. In this paper, the spatial transformation network(STN) model is used to extract the image features and train to obtain the affine transformation coefficients, so that the image to be registered can be adaptively transformed with reference to the coefficients to achieve the initial registration purpose. In order to obtain a more accurate registration effect, we use the phase correlation algorithm to calculate the phase shift of the frequency domain in the frequency domain caused by the translation of the two image spaces in the initial correction result. Then the relative motion vectors of the two images are obtained, and the entire registration phase is finally completed.