Cross-Modal Image Registration

Cross-modal image registration estimates how images from different sources—such as MRI and CT, infrared and visible light, or cameras and point clouds—overlap in space. Methods range from mutual-information and anatomical-feature matching to learned shared representations, coarse-to-fine correspondence networks, and geometry- or uncertainty-aware optimization; they may estimate rigid poses, local deformations, or both. The goal is to transfer information reliably across modalities for tasks such as image fusion, navigation, and measurement, including when appearance, viewpoint, or tissue shape differs. A central challenge is that stronger overlap scores do not always mean anatomically or geometrically correct alignment, so generalization and task-specific evaluation remain important.