Researchers Use Facial Recognition to Identify Raphael Work

Using facial recognition technology, researchers in the United Kingdom concluded that a tondo once attributed to an unknown artist was “highly likely” made by Renaissance master Raphael.

The “de Brécy Tondo,” acquired by the late British businessman and art collector George Lester Winward in 1981, depicts a classical Madonna and Child scene. The Madonna wears a dress with a white gathered collar, a dusty pink bodice, and what appears to be a loose teal skirt. A beige cloth covers the mother’s hair as she holds her baby close to her chest. The description applies, with uncanny accuracy, to Raphael’s “Sistine Madonna,” commissioned by Pope Julius II in 1512. 

Gemäldegalerie Alte Meister in Dresden, Germany, which holds “Sistine Madonna” in its collection, has long thought Winward’s Tondo to be a copy. The collector, however, was convinced that the work was an original Raphael and founded the de Brécy Trust in 1995 to sponsor research projects that could support his claim. 

Winward was posthumously vindicated, so to speak, when a team of researchers from the University of Nottingham and the University of Bradford in the UK announced their findings on Monday, January 23.

“This study demonstrates the capabilities of machine learning to give a probability of the same artist between different ‘Old Master’ paintings,” said Christopher Brook, an honorary research fellow at the University of Nottingham and expert in digital image analysis. He added that the research, which he co-authored, “promises much for the future examination of works of art.”

Using facial recognition systems developed by Hassan Ugail, a professor of visual computing at the University of Bradford, the team compared the face in the “de Brécy Tondo” with Raphael’s “Sistine Madonna.” They found a 95% similarity between the Madonnas in the two paintings and an 86% similarity in the Child. In research terms, a similarity of 75% or higher is enough to label two works “identical.”

The technology was originally created to identify international criminals from grainy CCTV security footage. After feeding the software an image, the machine learning algorithm closely reads the face, collecting information on features such as skin color, face shape, and expression. That data is compared to a second known image to determine how similar the two faces are.

“We’ve never applied this algorithm to paintings,” Ugail told Hyperallergic in an interview. “At the end of the day, the software is analyzing two images. From a technical point of view, we didn’t have to do anything extra; it’s exactly the same software.”

This finding confirms previous research by Howell Edwards, a scientific adviser to the de Brécy Trust which argued that the painting was not a Victorian-era copy. Pigments in the Tondo were studied and found to be typical of pre-1700 work. The team’s findings will be published soon in a peer-reviewed paper.