Artificial Intelligence Identifies Long-Overlooked Raphael Masterpiece

When he bought the painting in 1981, art collector George Lester Winward was sure he’d stumbled upon a genuine Raphael.

The mysterious, 37-inch-long work had no known artist. Looking into the faces of the Virgin Mary and Jesus, Winward noticed an uncanny resemblance to the two figures at the center of another painting: the Sistine Madonna, a 77-inch-long Renaissance masterpiece completed by Raphael around 1513.

Winward became convinced that the same artist was behind both works. But his view was in the minority. Experts thought the painting, now known as the de Brécy Tondo, was more likely a Victorian copy of the Sistine Madonna.

Shortly before his death in 1997, Winward set up the de Brécy Trust, donating his paintings to the organization to allow researchers to continue rooting out their secrets. Scholars have been studying the de Brécy Tondo ever since.

Now, after 40 years of speculation, researchers from the University of Nottingham and the University of Bradford, both in the United Kingdom, have made a breakthrough. Using facial recognition technology and artificial intelligence (A.I.), they have concluded that the painting is “highly likely to be a Raphael masterpiece.”

“Looking at the faces with the human eye shows an obvious similarity, but the computer can see far more deeply than we can, in thousands of dimensions, to pixel level,” says Hassan Ugail, an expert on visual computing at Bradford who developed the A.I. facial recognition system used in the study, in a statement.

The tool is an algorithm trained on millions of images, and Ugail has been working on it since 2002, reports Artnet’s Sarah Cascone. It can quantify how similar one face is to another by comparing features and recognizing patterns.

“These features may be the physical attributes (e.g., shapes, colors and textures of the face) but also include a lot (potentially thousands of features) which cannot be described visually or physically,” Ugail tells CBS News’ Caitlin O’Kane. “In this sense, the analysis carried out through these facial recognition systems can compare two facial images in much greater detail and can outperform humans.”

If two faces are more than 75 percent similar, Ugail considers them identical. In this case, the tool found that the two Madonnas were 97 percent similar, while the two versions of baby Jesus were 86 percent similar.

The de Brécy Tondo is, as the London Times’ Russell Jenkins wrote in 2013, “one of the most talked-about and scientifically tested Renaissance masterpieces.” The A.I. analysis is the latest in a series of discoveries made in the decades researchers have been studying the painting.

One of these finds came about when Howell Edwards, then an expert on molecular spectroscopy at Bradford, concluded that the work was likely a Renaissance painting, not a later copy. By analyzing paint pigments, he found evidence that the work contained certain colors and chemicals that were characteristic of Renaissance art—including massicot, a yellow pigment that wasn’t used after 1700.

Now, says Edwards in the statement, those findings have been “further vindicated by the facial recognition analysis.”

The researchers presented their analysis at the Conference on Software, Knowledge, Information Management and Applications in Phnom Penh, Cambodia, in December. Their work will be published in an academic journal later this month.

Ellen Wexler | | READ MORE

Ellen Wexler is Smithsonian magazine’s assistant digital editor, humanities.