Fusion of Reflectance and X-ray Fluorescence Imaging Spectroscopy Data for the Improved Identification of Artists’ Materials Open Access
Downloadable ContentDownload PDF View PDF in Browser Report an accessibility issue with this item
Works of art often are complex objects, and to fully understand how they were constructed requires a variety of tools. The tools are used to provide information about the object, including its age, origin, construction methods, and materials. Art conservators and historians use this information to better preserve and restore the object and to learn about the artist’s working methods and motivations. While those working in the cultural heritage field routinely use multiple point-based tools to draw conclusions, because a single tool may be insufficient, a full analysis of an object will require an automated approach for merging the results from multiple image-based tools. In this dissertation, algorithms for automatically registering, classifying, and fusing data from multiple sources are described and demonstrated to identify the pigments used in the creation of works of art. The registration algorithm provides a means for registering and producing mosaics from three types of datasets that are commonly used in conservation science: 1) sets of overlapping sub-images, registered to a common reference image, from which a mosaic is generated (x-radiographs, IR reflectograms), 2) sets of images captured with different bandpass filters registered with one another (multispectral image sets), and 3) hyperspectral image cubes (reflectance imaging spectroscopy (RIS) and x-ray fluorescence (XRF) imaging spectroscopy, including data collected with an imaging system utilizing a scan-mirror), registered to a common reference image. An XRF scanner was constructed at the National Gallery of Art (Washington, D.C.) to obtain XRF imaging spectroscopy data, and a sum-of-Gaussians fitting algorithm was produced for generating element maps from the XRF imaging spectroscopy data, as well as maps indicating the confidences in those element maps. A feature-based algorithm was produced for generating pigment maps from the RIS data, as well as maps indicating the confidences in those pigment maps, and a fusion algorithm was produced for merging information from the element and pigment maps. Finally, validation steps were performed to assess the accuracy of the registration, classification, and fusion algorithms, and the fusion was judged to provide improvements in the pigment maps and their associated confidences.
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.