Artful Software Spots Faked Masterpieces

Stefan Lovgren
for National Geographic News

November 23, 2004

As soon as people began paying for art, the lucrative business of art forgery was born. How widespread is the problem today? By the wildest estimates, 15 percent of the art sold at auction houses may not be authentic.

No wonder forgery detection has become an art in itself.

Authenticating works of art has long been the domain of art historians steeped in the works of a particular artist. In addition to their discerning eyes, these art sleuths have also been able to use surface-material examination and x-ray analysis to determine if a work of art is authentic.

Now they may have another tool in their arsenal.

Scientists have developed a new digital authentication technique that analyzes and classifies paintings based on a digital analysis of the artist's style.

The method could also be used to determine if more than one hand was involved in creating the work, something physical forensic science has never been able to do.

"This technique, in conjunction with traditional methods, could play an important role in art forensics," said Hany Farid, a computer scientist at Dartmouth College in Hanover, New Hampshire. He is the co-author of a report describing the technique published this week in the Proceedings of the National Academy of Sciences.

Smooth Compressions

The digital authentication technique builds a statistical model of an artist from scans of a set of authenticated works. Other works are then compared against this baseline data.

The process finds consistencies and inconsistencies in the works. The technique is similar to how a digital camera compresses an image by removing so-called statistical redundancies.

"Imagine comparing one painting by a very talented artist with a smooth, elegant brush stroke with another painting by an imitator whose strokes are more clunky," Farid said.

"When compressed, the image that is smooth and continuous is going to be easier to compress, because there's a lot of redundancy and consistency, while the image that's choppy and broken up is going to be harder to compress," he said.

In an experiment, researchers applied the technique to a set of 13 drawings. The computational model automatically grouped together the eight authenticated works by the 16th-century Flemish artist Pieter Bruegel the Elder, separating them from five imitations.

"I don't think this is a technology that will much influence the detection of fakes [created later to fool collectors]. But it could well revolutionize identification of many sincere imitations of various major artists that occurred during their own lifetimes," said Stuart Fleming, author of Authenticity in Art.

Complementing Scholars

While there may be few modern-day forgers, most master painters have had their works imitated or forged over the years. For example, the National Gallery of Scotland once reportedly held an exhibition of five paintings of the same subject, all attributed to Leonardo Da Vinci. At least three of the works were copies.

Still some art historians are skeptical about the new digital authentication technique.

"It's an interesting idea. But before curators and art historians will jump in and use this, I think a much larger sample of work has to be tested," said Nadine Oberstein, curator at the Department of Drawings and Prints at the Metropolitan Museum of Art in New York.

If the technology turns out to be accurate, she says, it could be used to complement the expertise of scholars. This is actually what Farid envisions, too.

"What the human eye can see is often very difficult for the computer to extract and vice versa," he said. "What computers are good at, humans are typically not good at."

Multiple Hands

Most important, Farid says, the digital technique may determine how many hands were present in a single painting, something forensic technology has never been able to address.

Many great Renaissance painters had large workshops and painted only portions of their works. They let apprentices paint other parts. The practice has caused scholars to argue over who painted what in many masterworks.

Using his computer model, Farid isolated the faces of six individual characters in the painting "Madonna With Child," by the Italian Renaissance painter Perugino. Three of the faces proved to be statistically similar, while the three other faces were all different, indicating that at least four hands contributed to the painting.

"If further studies prove conclusively that this computer program effectively discriminates between different artistic hands, it will lead to a whole new era in the study of the workshop practice of artists such as Perugino and Peter Paul Rubens," said Katherine Hart, the interim director of the Hood Museum of Art at Dartmouth College.

For now, scientists themselves are driving the technology rather than museum curators and auction houses. There may be an obvious reason for that.

"If you spent 80 million [U.S.] dollars to buy a painting, I don't think you'd be eager to know that it's a fake," Farid said. "It's a dirty little secret that a lot of paintings are not real."

Don't Miss a Discovery
Sign up for the free Inside National Geographic newsletter. Every two weeks we'll send you our top news stories by e-mail.