(CN) – Scientists are using artificial intelligence to analyze X-ray images of paintings, sorting through the massive amounts of data available and offering potential new insights that could aid investigations into unlocking the secrets of ancient works of art, according to new research published Friday.
In a research paper published in the journal Science Advances, scientists have discovered a way to improve X-ray scans of paintings. Hyperspectral imaging, macro X-ray fluorescence scanning and other new techniques can provide a wealth of information about the history and composition of historic paintings. But the imaging techniques have limitations and the sheer volume of data generated can be daunting to analyze.
For example, polyptychs – paintings divided into multiple sections – often contain side panels that are painted on both sides and set on hinges, so that they may be arranged to depict different scenes. Analysis of such double-sided paintings using X-rays produces overlaid X-ray images that are difficult to decipher.
To improve on previous approaches to separate blended content into two X-ray images, Zahra Sabetsarvestani from the University College London and colleagues at Duke University and in the U.K. proposed a “self-supervised framework,” whereby a deep neural network learned from standard color images of each of a panel’s sides.
To test this approach, the new method was applied to portions of the depictions of Adam and Eve on the outer panels of the Ghent Altarpiece, for which high-resolution photographs and overlaid X-ray images were available in an online repository. The team produced near-perfect separations of the mixed X-ray images.
The Adoration of the Mystic Lamb, painted in 1432 by the brothers Hubert and Jan Van Eyck and more commonly known as the Ghent Altarpiece, is arranged in two vertical registers, each with double sets of foldable wings containing inner and outer panel paintings. Sometimes called the “most stolen artwork of all time”, the polyptych is currently undergoing a conservation and restoration campaign begun by the Belgian Royal Institute for Cultural Heritage in 2012.
Applying this method to double- and single-sided paintings could reveal superimposed or altered compositions, as well as damages that are not apparent on the surface.
The researchers said that as AI technology advances, “the development of new algorithms capable of ingesting such complex datasets will not only have far-reaching implications for art investigation but can open entirely new vistas both in computer and heritage science.”