Multimodal remote sensing image registration method and system

Abstract

The present invention is applicable to the technical field of image processing, providing a multimodal remote sensing image registration method and system. The method comprises the following steps: constructing a nonlinear scale space by respectively applying anisotropic diffusion filtering to remote sensing images of each modality, thereby obtaining a multi-scale image sequence for each set of images; based on the multi-scale image sequence, performing multi-directional co-occurrence filtering on images at each scale to generate a corresponding co-occurrence filter image pyramid; calculating a phase consistency map for each scale image according to the co-occurrence filter image pyramid and extracting feature points; constructing a feature descriptor based on weighted phase consistency according to the feature points; and performing bidirectional matching on the feature descriptors corresponding to the remote sensing images of each modality, followed by eliminating mismatches to obtain a registration result. This method can effectively increase the number of features, enhance descriptor robustness, and improve final registration accuracy, providing reliable technical support for the collaborative processing and application of multi-source remote sensing information.