top of page
gerard van weyenbergh

AI revolutionizes art authentications, with an 98% certainty.

AI finds inconsistency in historic Raphael masterpieceThis study, which focuses on the Virgin with the Rose, shows to what extent machine learning could become a formidable tool for art historians.

The Virgin and the Rose is generally attributed to Raphael, but at least one other artist probably contributed to the masterpiece. © Wikimedia Commons

In addition to generating images, machine learning-based systems are now capable of extracting patterns invisible to humans which can reveal tons of hidden information about a work. These features can, for example, be used to detect deepfakes… or to solve mysteries surrounding historical masterpieces, such as the Madonna della rosa (The Madonna of the Rose).

This famous painting on wood, painted at the beginning of the 16th century, is the subject of a great debate which has agitated the art community for more than 200 years. Originally, it was attributed without dispute to Raffaello Sanzio, better known as Raphael. But from the 19th century, doubts began to emerge.

Indeed, specialists remain convinced that certain features - notably Joseph's face - are not consistent with the rest of the scene. They therefore believe that they come from the hand of another artist. Today, there is still no real consensus.

AI takes on historical masterpieces

But this could finally change thanks to new tools powered by artificial intelligence. To find a definitive answer, British and American researchers have developed a new analysis algorithm. They began by training him by submitting other paintings by Raphael, those whose origin is considered certain. This allowed him to learn to detect the characteristic nuances that define the master's style.

“  We used other authenticated images of Raphaël to train the computer to recognize his style down to the smallest details, from the brushstrokes to the palette to the representation of light  ,” explains Hassan Ugail, researcher at the English University of Bradford and co-author of the study.

This technique has already produced great results. The team showed that this system was capable of authenticating a painting by Raphael with 98% accuracy. This notably made it possible to attribute another painting, known as Brécy Tondo, to the Italian master. Unfortunately, the algorithm hit a wall with the Madonna. The researchers failed to obtain sufficiently conclusive results to remove the last doubts.

So they changed strategy. Instead of making the algorithm work on the entire work, they forced it to focus on isolated portions, like faces . Well, they were lucky, because a clear trend has finally started to emerge. The program was completely convinced that Raphael had indeed painted all the characters himself... with one exception: Joseph, for whom the probability drops below 40%. In other words, it is very likely that it is an edit added by another artist.

But then, who painted Joseph?

There still remains a burning question: if it wasn't Raphael who painted this character, who was responsible for it? Unfortunately, this is outside the scope of this algorithm, and the question remains open.

Most experts already have a name in mind: his apprentice Giulio Romano, to whom some authors already partially attribute the painting. Others also cite Gianfrancesco Penni, another lesser-known Italian painter. But whatever it is, further studies will need to be conducted to prove it. And that might be difficult. Unfortunately, these candidates produced fewer famous paintings that could be used to train an algorithm.

A great tool for art specialists

In the press release from the University of Nottingham, the authors point out that such results do not constitute definitive proof . “ Attribution and authentication are among the most important challenges for art scholars  ,” says David G. Stork, co-author of the study. “  They must study the provenance, the materials, the condition of the painting, the iconography, and finally the work itself: brushstrokes, color, composition… However, studies in computer science applied to art — including ours — focus mainly on these latter criteria. The results of our work should therefore not be considered sufficient  .”

On the other hand, they insist that their work shows to what extent machine learning can be a formidable tool in art history . “  This is a step forward for authentication protocols,  ” Stork said.

“  As databases grow and algorithms improve, researchers will be able to refine their methodology. These computer methods will improve and can be widely used in art history and criticism  ,” he concludes. We can therefore expect that new puzzles that have tortured specialists for decades will soon be solved thanks to machine learning. Seen on France Info Culture - Antoine Gautherie

 

コメント

5つ星のうち0と評価されています。
まだ評価がありません

評価を追加
bottom of page