AI reveals which Renaissance master is likely responsible for this mystery painting

The de Brécy Tondo is seen here next to Raphael's Sistine Madonna. (University of Nottingham)

A painting with a long-unknown origin was almost certainly a work of the master painter Raphael, according to new analysis using artificial intelligence.

Researchers from the University of Nottingham and the University of Bradford found that facial recognition technology puts the painting in question at a 97 per cent match with one of Raphael’s most famous works, suggesting that the exact same models were used for the two paintings.

The de Brécy Tondo, a 95-centimetre circular painting, has long been theorized to be a work of Raphael, with researchers and historians carrying out numerous investigations of the painting over the past 40 years.

But this is the most definitive connection to Raphael yet, researchers say.

"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,” Hassan Ugail, a professor and Director of the Centre of Visual Computing at the University of Bradford, said in a January press release.

"Based on the high evaluation of this analysis, together with previous research, my fellow co-authors and I have concluded identical models were used for both paintings and they are undoubtedly by the same artist."

The peer-reviewed research was published in February after being presented at an international conference at the Cambodia University of Technology and Science in December.

But the journey to this moment started in 1981, when the painting was acquired by collector George Lester Winward at an auction in North Wales.

The de Brécy Tondo shows a woman holding a baby, referred to as a Madonna and Child painting, and although there was no confirmed artist, it was even then immediately clear that there was some connection to Raphael.

The expression, pose, clothing and composition all looked strikingly similar to Raphael’s Sistine Madonna, a painting commissioned in 1512 by Pope Julius II.

Although the similarity is obvious even to the untrained eye, historians knew there was still the possibility that the de Brécy Tondo was a deliberate copy, or painted by one of Raphael’s assistants. The art gallery which Winward acquired the painting from had judged it to be a later copy.

Extensive research between 1987 and 1991 determined that while the de Brécy Tondo almost certainly pre-dated the Sistine Madonna, it wasn’t clear if Raphael himself had painted it first as the template for the Sistine Madonna.

Prior to his death, Winward set up the de Brécy Trust Collection so that his collection of paintings would continue to be displayed and studied by art scholars. The trust contains 30 works of art.

The theory that the de Brécy Tondo could be a Victorian copy was later dispelled by research in 2007 by Howell Edwards, a professor of molecular spectroscopy at the University of Bradford and an honorary scientific advisor to the de Brécy Trust. His work showed that the pigments contained in the Tondo could be dated to early pre-1700 Renaissance work, placing it in the time period of Raphael.

But this latest study took the work one step further by involving artificial intelligence.

The facial recognition technology that Ugail utilized was trained on millions of faces in order to recognize and compare facial features.

The Madonnas in both paintings were found to have a 97 per cent facial similarity, while the Child had an 86 per cent similarity between the two works. In the field of facial recognition, a similarity above 75 per cent is considered identical, according to the researchers.

“This study demonstrates the capabilities of machine learning to give a probability of the same artist between different ‘Old Master’ paintings,” Christopher Brooke, an honorary research fellow at the University of Nottingham and co-author of the study, said in a separate press release. “In this case study, direct facial comparison comes out at a match of 97 per cent – a very high statistical probability that the artworks are by identical creators.”

Ugail said that this highlights how artificial intelligence can be used for a positive goal in art — narrowing down the authors of long-lost masterpieces.

“Facial recognition technology can be applied for a variety of purposes, including analysis of art and even to health care,” Ugail said. “Using it in this way, to determine the similarity of portraits in paintings, is yet another example of its wide-ranging potential of artificial intelligence-assisted computer vision.”