A new method of image analysis developed by three Dartmouth researchers may help art historians distinguish more easily between authentic artwork and forged copies, according to Daniel Graham, a post-doctoral researcher in mathematics and one of the tool's developers.
The model, which uses a technique known as "sparse coding" to quantify artistic style, was also co-developed by Daniel Rockmore, a mathematics and computer science professor, and James M. Hughes, a third-year Ph.D. student in computer science.
The technology models how people interpret art by capturing features that are particularly relevant to the human visual-processing system, Graham said.
It breaks down an image into a simpler set of components, called basis functions, Hughes said. Graham described the process as similar to separating a work of literature into the letters of the alphabet.
In their research, the team compared eight authentic drawings by Flemish artist Pieter Bruegel the Elder to five well-known imitation artworks, which were "long-thought to be authentic," Hughes said.
The team first fragmented the authenticated Bruegel works into an "alphabet." To test authenticity, they sought to recreate the work in question using as few letters of this "alphabet" as possible.
While authentic works can be identified using relatively few basis functions, imitations require significantly more "letters," rendering them dissimilar enough to be identified as fake, Hughes said.
Although art historians have previously used other scientific methods, including chemical analysis of pigments and canvas thread counts, to determine authenticity, there has been a recent trend of using computers to detect forgeries, Graham said.
"Increasingly, we're finding ways to turn phenomena into data," Rockmore said, "which means that they are potentially comprehensible through tools of mathematics and computer science."
The group's method is different from current models because it is "adaptive" and "data-driven" and because it uses information derived from authentic works as the basis for analysis, Hughes said.
The technology currently uses grayscale images and can only be used for drawings, but Graham said the researchers hope to incorporate color and medium into future versions and to make the analysis applicable to paintings.
The team hopes to work with art historians to help answer questions that historians cannot answer on their own, Graham said.
Hughes emphasized that the tool is not intended to replace traditional methods of authentication but can add an extra degree of confirmation, which could have monetary implications as well, he said.
"This approach is certainly going to be used by both parties by museums for conservation and art history purposes as well as by sellers of artwork," Hughes said.
Although this method can be used to spot fakes, there may be better ways in which the tool could be used, Rockmore said.
"The hope is that it provides an interesting way to study the evolution of a particular style of art or to distinguish between different styles of art," he said.
The team's research was published in the online edition of the Proceedings of the National Academy of Sciences on Jan. 5.
