A method for the registration of spectral images of paintings and its evaluation Athanasios Zacharopoulos 1 , Kostas Hatzigiannakis 1 , Polykarpos Karamaoynas 2 , Vassilis M. Papadakis 1,3 , Michalis Andrianakis 1 , Kristalia Melessanaki 1 , Xenophon Zabulis 2 August 2, 2017 Abstract This work concerns the automatic registration of spectral images of paint- ings upon planar, or approximately planar, surfaces. An approach that cap- italizes upon this planarity is proposed, which estimates homography trans- forms that register the spectral images into an aligned spectral cube. Ho- mography estimation methods are comparatively evaluated for this purpose. A non-linear, robust estimation method that is based on keypoint features is adopted, as the most accurate. A marker-based, quantitative evaluation method is proposed for the measurement of multispectral image registration accuracy and, in turn, utilized for the comparison of the proposed registration method to the state of the art. For the same purpose, characteristic for this application domain, benchmark datasets that are annotated with correctly corresponding points have been compiled and are publicly availed. 1 Introduction Multispectral Imaging (MSI) is a non-destructive diagnostic technique, combining digital imaging with spectroscopic analysis, to recover spatial and spectral informa- tion about a surface or an object. MSI is applied in medicine, agriculture, remote sensing, food industry, and cultural heritage sciences, including art conservation, archeology, and art history [1, 2]. In MSI, a monochromator, i.e. a series of bandpass filters, is placed in front of the illumination system or the imaging sensor. This sensor then acquires a sequence of quasi-monochromatic images at consecutive narrow spectral bands, mainly in the visible spectrum but often extending to the infrared (IR) and/or the ultraviolet (UV). The acquired grayscale images are called spectral images and their sequence a spectral cube. Analysis of the spectral cube provides information related to stratigraphy of a painting, that is the paint layers that comprise it. The intensity values along a column of the spectral cube form the extracted reflectance spectrum, at the column location. This spectrum is informative about the imaged materials and used in pigment studies [3, 4, 5]. Revealing the pigments used in artworks, assists the 1 Institute for Electronic Structure and Laser, Foundation for Research and Technology-Hellas, N. Plastira 100, Heraklion 700 13, Greece. 2 Institute of Computer Science, Foundation for Research and Technology-Hellas, N. Plastira 100, Heraklion 700 13, Greece. 3 Aerospace Non-Destructive Testing Laboratory, Faculty of Aerospace Engineering, Delft Uni- versity Of Technology (Aerondt), Kluyverweg 1, 2600 GB, Delft, The Netherlands. 1