LGRI: A Multimodal Image Registration Method Integrating Adaptive Log-Gabor Filtering With Rotation-Invariant Descriptors

Multimodal image registration aims to accurately align images with differences in spatial structure or optical characteristics, serving as a key technology in remote sensing image processing and analysis. Currently, most registration methods are designed only for single-modal or limited cross-modal scenarios, often leading to distortions, artifacts, and registration errors when processing multimodal data. To address this issue, we propose a multimodal remote sensing image registration method called LGRI (Log-Gabor Rotation-Invariant), which integrates adaptive Log-Gabor feature extraction with rotation-invariant local binary descriptors. Specifically, we first design an adaptive Log-Gabor Structural-Frequency Detector (LGSFD) to extract highly repeatable and discriminative cross-modal feature points by leveraging spatial structure and frequency response. Then, we propose a Rotation-Invariant Noise-Robust Local Binary Descriptor (RINR-LBD), which improves robustness via arc-segment mean smoothing and ensures descriptor consistency through principal direction normalization. Finally, we refine the matches using Fast Sample Consensus (FSC) to effectively eliminate mismatches and improve registration accuracy. Experimental results show that our method outperforms state-of-the-art approaches in both the average number of correct matches and the matching success rate.