The field of art history is steeped in scientific analysis, from analyzing an artist's technique to studying the materials used to create the painting. One issue that concerns everyone in the art world--from auction houses to museums to the average art collector--is the problem of forgery. The authentication and detection of forged or duplicated works is perhaps the area where the stakes are highest, especially financially.
In the past, it has taken a highly trained eye to tell the difference, but determining forgeries can be very subjective, costly, and time-consuming. Now, optical engineers are making headway into the authentication of paintings, as well as determining artists' individual styles.
Daniel Keren, chairman of the Department of Computer Science at the University of Haifa, Israel, has been working on software that enables machine vision equipment to identify an artist's exact style, currently with 86% accuracy. Keren has trained the program to recognize the work of Salvador Dali, Vincent van Gogh, Rembrandt Harmenszoon van Rijn, Rene Magritte, and Wassily Kandinsky. The program breaks down the paintings as a series of mathematical symbols, and then uses that coding to break down other paintings and determine whether they were painted by the same artist.
The van Gogh painting Cornfield and Cypress Trees as analyzed by Daniel Keren's algorithm. The areas colored red are identified as "van Gogh style," the blue areas are identified as "Dali style." The gray areas are "Undecided."
"I was just curious to see whether a computer can do what people do quite well -- recognize the 'style' of a painter," Keren says, "thus being able to recognize a painting by him/her even if all other paintings they've seen depict totally different things." The abilities of his software program are nowhere near those of humans--it still has a hard time determining faces, for example--but Keren is hopeful. "I believe this [program] can be improved."
Jim Coddington, Agnes Gund chief conservator at the New York Museum of Modern Art, and co-chair of the IS&T/SPIE conference "Computer Image Analysis in the Study of Art" had seen fractal analysis applied to Jackson Pollock's drip paintings but wanted to see if the results were reproducible. Working with Daniel Rockmore of Dartmouth College (Hanover, NH), they created a team to help try to reproduce some of the previous fractal analysis done on Pollock's paintings. While their tests found that fractal analysis was accurate overall, another team presenting at the same conference found that fractal analysis was not an effective analysis tool. "I think that's why it's pretty important to reproduce [previous] work to see if there were any assumptions wrong or data missing that might help further clarify the situation," says Coddington.
Fractal analysis might prove to be more effective if it is one tool among many, he said. "I think that some of the tools we already have in the conservation science tool kit for identification of materials are more robust and have better predictive quality than fractal analysis," Coddington says. "That said, whether it's fractal analysis or some other statistical analysis of brushwork, there are all kinds of possibilities out there."
David Stork of Ricoh Innovations co-chaired the conference with Coddington, and is a firm believer in the usefulness of image analysis. Stork is part of an international consortium studying van Gogh's brush strokes. "Computer wavelet analysis of the brushstrokes of Vincent van Gogh are helping to identify copies and forgeries and identifying who painted different portions in a single painting," says Stork.
Computers have already outperformed some art connoisseurs on some types of problems, according to Stork. "It is almost as if these new methods are a new kind of microscope or telescope, to extend and expand the understanding of paintings beyond what a connoisseur can do just with his or her eye."
Of course image analysis and authentication is not a perfect science. Using computer analysis for authentication of art still has a ways to go before it can, if ever, replace art history experts and connoisseurs.
"I think that for authentication there are still a lot of big questions there," Coddington says. "The interesting questions are going to be when image analysis can tell us more about how the artist painted, how they constructed the work, helping us get inside the creative mind, and that's when image analysis is going to serve us in really big ways."
To read about other applications of optical engineering to the art world, read the article Optic Art in the April 2008 SPIE Professional.