Block information strategy for multi-modal remote sensing image registration

Registration of multi-modal remote sensing image pairs (MRSI) is challenging given the distinct imaging mechanisms of multi-modal data sources, which lead to substantial geometric and radiometric distortions and inaccuracies in correspondences. To tackle this issue, we propose a novel approach that integrates local image information into feature representations through the design of local regions and the extraction of local information. The latter comprises of two key components: rank-based feature redistribution and residual information extraction utilizing a pyramid-like structure of local patches. This enhanced feature representation technique, termed Reinforced Local Information of LSS (RLILSS), embeds local information to improve the performance of the Local Self-Similarity (LSS)-based framework for MRSI registration. RLILSS strengthens feature characterization across various regions and addresses the limitations of supplementary information. This enables more reliable correspondences between images. Experimental results show that the proposed method achieves higher accuracy and better registration across diverse multi-modal datasets. Detailed analyses confirm its superiority over state-of-the-art methods in both accuracy and robustness. This approach holds significant potential for applications in automatic geographic registration and disaster area reconstruction.