High resolution multimodal images registration of Théo Van Rysselberghe’s La dame en blanc (1904) using mutual information

Multimodal imaging techniques—including high-resolution photography under visible and Ultraviolet light, infrared reflectography (IRR), X-ray radiography (XRR), hyperspectral imaging (HIS), and macro X-ray fluorescence (MA-XRF)—are essential for studying a painting's material composition, hidden features, and conservation needs. However, comparing these multimodal images poses challenges due to resolution mismatches, reconstruction-induced deformations, and gigapixel-scale scale. A two-step registration using mutual information and the Insight Segmentation and Registration Toolkit (ITK) was developed to address these issues. Initially, an affine transformation achieves coarse alignment, followed by a refined B-spline transformation. Both steps use a pyramidal approach at multiple scales with an L-BFGS-B optimizer maximizing mutual information between images. Images are rescaled and aligned to a common high-resolution visible-light photograph reference. Registered images are tiled with Photoshop's Zoomable plugin and displayed in a custom viewer, facilitating smooth multimodal navigation. This method was applied successfully to Théo Van Rysselberghe's La dame en blanc (1904), belonging to the collection of Liège's Museum of Fine Arts.