Con Artist: Scanning program can discern true art

Until now, discerning an artist’s style has been in the eye of the beholder. However, a new mathematical tool distills style into an array of statistics as a potential means to spot forgeries. In a recent study, the technique distinguished eight drawings by the 16th-century artist Pieter Brueghel the Elder from five imitations attributed to the master until a decade or so ago.

MANY HANDS. A statistical analysis of Perugino’s “Virgin and Child with Saints” (top) suggests that at least four artists contributed to the work. Only areas 1, 2, and 3 cluster in a representation of six areas (bottom). PNAS

Several digital-imaging researchers, including the study’s authors, agree that the work is only a first step toward a reliable fraud-detection technique. However, the preliminary findings are encouraging, comments David Donoho, a statistician at Stanford University.

The technique employs a process called wavelet decomposition to break down a digital image into a collection of more-basic images, called subbands. Just as a musical tone consists of a low fundamental frequency with higher-frequency overtones, an image’s low-frequency subbands show the broad strokes, while higher-frequency subbands depict details. Wavelets have been used in a wide range of image-processing applications, such as layering detail onto the animated creatures in the film A Bug’s Life.

Wavelet decomposition is good at analyzing textures. For instance, a smooth, untroubled surface such as a blue sky would show up mostly in the low-frequency subbands, while blades of grass would produce activity primarily in higher-frequency subbands.

Now, in an upcoming Proceedings of the National Academy of Sciences, researchers report capturing the texture of an artist’s strokes.

“A master might have smooth, consistent strokes, say, while an imitator is jerky,” says study coauthor Hany Farid, a computer scientist at Dartmouth College in Hanover, N.H.

In the new study, the researchers measured 72 statistical features, such as the percentage of dark portions in a given subband. The team found that the genuine Brueghels all had similar statistics, while the imitations were significantly different from the Brueghels and from each other.

The researchers also studied the painting “Virgin and Child with Saints,” created in the studio of the Italian artist Pietro Perugino around the turn of the 16th century. They conclude that it was probably painted by at least four artists, in keeping with historians’ opinion that Perugino painted only a portion of the work.

While the results are interesting, a study of only two artists isn’t enough to make the art world adopt the method, says Nadine Orenstein, a curator at the Metropolitan Museum of Art in New York. “I think they need to study a much larger sample of material,” she says.

“I think the broad idea is the right one,” says Jitendra Malik of the University of California, Berkeley. However, he adds, it’s not clear whether the 72 features the team examined are the most effective ones. “Testing the approach on many more artists will probably enable us to get a handle on what are the best features for art authentication,” he says.

More Stories from Science News on Math