Hany Farid and his colleagues at Dartmouth College adapted a technique they had developed to detect tampering in digital images. They first analyzed a high-resolution image of artwork, computing the strength of light and dark stripes of various orientations and spacings. They then generated a mathematical "fingerprint," which summarizes the statistical prevalence of these textures and also how easily the properties of one region could be predicted from its neighbors. Farid says that this predictability may correspond to a property known from handwriting analysis: authentic signatures tend to be smoother than forgeries. So far, however, the digital fingerprint is a purely mathematical construct, and the researchers have not determined what features in the image cause differences between fingerprints.
The researchers found that fingerprints for eight authenticated drawings by the 16th-century artist Pieter Bruegel the Elder could be mathematically distinguished from five contemporary imitations. They also examined six faces on the painting ¿Madonna with Child" (see image), attributed to the Renaissance master Perugino. The three faces on the left had a similar statistical signature, whereas the other three were each distinct, suggesting that four different artists painted the faces. Ordinary forensic analysis couldn't answer this ¿how many hands¿ question, Farid says, because all the artists were probably working in the same studio.
Nadine Orenstein, an expert on Bruegel at the Metropolitan Museum of Art in New York City, provided the examples of his work to the Dartmouth researchers. She says that automated analysis of imitations is an interesting possibility, and could add to the battery of technical tests that art analysts already use to complement their evaluation. But she cautions that a lot more work needs to be done to find out how useful this new tool will be, especially for evaluating the works of artists whose style changed over time.