Detector-Free Multimodal Image Matching

Abstract

Multimodal image matching is essential in image stitching, image fusion, change detection, and land cover mapping. However, the severe nonlinear radiometric distortion and geometric distortion of multimodal images severely limit the accuracy of multimodal image matching. To solve these problems, we propose a detector-free multimodal image matching approach to establish pixel-level dense correspondences. We mitigate the impact of modality differences on feature point extraction by establishing robust reference points. Specifically, we design a phase congruency module to keep the location of the reference point centered on the image edge structures. Simultaneously, a guiding correction module exploits the geometric relationships between pixels and reference points to establish accurate pixel correspondences. Finally, refined correspondences are obtained by finely positioning highly correlated pixel matches. Experiments show that our method can obtain sufficient and robust correspondences on multimodal images.