Multimodal Remote Sensing Image Registration Method Based on Multi-Registration Model Fitting

Remote sensing satellites generate vast amounts of multimodal image data, and the synergistic utilization of these images enables information complementarity. Image registration serves as the fundamental prerequisite for this process. However, establishing a reliable registration relationship remains a complex and challenging task due to significant nonlinear radiance differences (NRD) and geometric distortions (GD) resulting from different imaging mechanisms. To address these challenges, this paper proposes a multimodal remote sensing image registration method based on multi-registration model fitting. First, a feature detection map is constructed using a weighted moment equation of image phase information, which outperforms conventional gradient-based methods in terms of robustness to NRD. Next, the maximum index map (MIM) is employed to describe the extracted features. Finally, through fitting two types of registration transformation models, the influence of outliers caused by GD is effectively mitigated, thereby establishing a robust registration relationship. Experimental results demonstrate that the proposed method achieves excellent performance in multimodal remote sensing image registration, outperforming existing approaches in both accuracy and robustness.