MSTERDAM

WHO can say for sure that a great artwork is the real deal? It depends on whom you ask, and when: attributions of authorship often flip-flop from generation to generation, and a painting's journey from real to not real and back again can mean windfall profits for the lucky and despair for those who sell — or buy — too soon.

Considering the sums at stake, you might think the business of art authentication would have long ago given way to modern technology. But so far, sophisticated techniques like spectrometry, DNA testing and pigment analysis have failed to supplant expert opinion.

Now a team of researchers at the University of Maastricht, here in the Netherlands, are taking a stab at rationalizing connoisseurship, a word that in its art-historical context refers to the formal process of determining who created a work of art. They have developed a computer system that quickly examines hundreds of paintings for telltale patterns. The results, they say, can lend credence to existing attributions or help dismiss them.

Members of the team make modest claims for their system. "The computer will come up with data that show some patterns, but we cannot decide whether these patterns are meaningful or not," said Dr. Eric Postma, the leader of the project, known as Authentic, which is currently analyzing all paintings attributed to Vincent van Gogh. "For that purpose we need experts. We can provide them with numbers, and they can interpret the numbers. And this interaction is where the real value of the project is."

Though still in its early stages, Authentic and other projects like it may someday revolutionize a murky field. It uses the law of large numbers — the idea that the largest number of independent measurements produce the clearest view of a subject — to analyze digital images of van Gogh's accepted oeuvre for characteristics like color contrast, canvas structure and brush strokes. In the process, it highlights patterns that emerge throughout that group as a whole. The most consistent objects cluster together, while aberrations are flagged as suspect. "The computer can make an educated guess about whether it's genuine or fake," Dr. Postma said. "But more importantly, it can also indicate why it makes this decision."

Of the practice of defining authenticity through a body of paintings that may itself contain illegitimate works, Igor Berezhnoy — a Ph.D. candidate on the team, who is also exploring how to commercialize the technology — says: "Even if there'll be some fakes, statistically they will be pushed out from our set, because they would show completely different characteristics."

Dr. Postma compares this pattern-seeking technique to chess. "If you're a skilled player in chess," he said, "you can recognize the configuration on the board immediately. And the same is true for an expert in paintings." But the computer can process more data, churn through more potential patterns, and do it faster than a human being can. "That's why a computer can beat the grand master." And by discovering patterns that experts never thought to look for, Authentic can also serve as a tool for more general art-historical research.

This is not the first time artificial intelligence has been used in authentication. In Germany in 1998, a team at the University of Bremen's Center for Computing Technologies trained their computer to identify the drawings of Delacroix, which it managed to do with 87 percent accuracy. Ignoring previous techniques commonly used for the manual investigation of drawings, like analyzing the length, thickness and curvature of lines, the computer scanned the images, assigning each pixel to the category of either black or white, and then analyzed the ratio of one to the other — across a whole drawing, and then across the entire set. This method was cheap and required only a 300-dot-per-inch scan. But 87 percent accuracy, while high for a new technology, is low for people contemplating, say, a million-dollar purchase.

In a more recent project at the Catholic University of Rio de Janeiro, a computer distinguished between 23 paintings made by the popular Brazilian painter Candido Portinari and five by his contemporary Enrico Bianco. After detecting and analyzing thousands of brush strokes by the two artists, the computer was able to determine their maker 77 percent of the time. But by dividing the paintings into smaller segments, and then attributing each segment to either artist based on the majority of brush strokes in that segment, the accuracy jumped to 100 percent in terms of determining the authorship of a whole painting. Though lack of money has currently interrupted the project, Prof. Sidnei Paciornik, one of the authors of the study, said, the next step would be "training" the system further with "recognizedly fake paintings of Portinari and checking what the system would do with them."

Authentic started out in the same way a few years ago, with a test project that managed to discriminate between paintings by van Gogh, Cιzanne and Gauguin with 95 percent accuracy. Other tests have also sought to confirm what human experts already know, in order to prove that the system is up to the task. Its second test, for example, involved picking from a group of six van Gogh canvases the one that was known to have been painted on a different kind of material. By analyzing the canvas patterns that appear when the painting is X-rayed at high resolution, the Authentic system correctly picked out the misfit.

The team's latest study focuses on van Gogh's use of complementary colors. By analyzing several hundred low-resolution images of van Gogh paintings downloaded from the Web, the Authentic team still managed to pick out, in hours, patterns that would have taken much longer to detect using manual research — for example, the fact that van Gogh's use of complementary colors was greater when outlining human figures than it was for other objects in his paintings.

"As data, these Web images are terrible," said Mr. Berezhnoy. "Nevertheless, our method confirmed an increase in van Gogh's use of opponent colors. I could tell you that van Gogh started to use blue-yellow opponency in such and such year, and red-green later on, or that he used opponent color to highlight portrait silhouettes and human figures. Do you think I had time to draw those conclusions by studying all 854 paintings?"

Is this valuable information? Did van Gogh consciously highlight his human figures this way? That's for experts to say. But it does form one more tool in Authentic's kit.

Yet as these techniques gradually increase in sophistication, museums and auction houses aren't rushing to reinvent the process of attribution. The trained judgment of the art expert remains the gold standard in determining authorship, a situation perhaps best epitomized by the art critic Bernard Berenson, who said in 1933, "When I see a picture, in most cases I recognize it at once as being or not being by the master it is ascribed to; the rest is merely a question of how to try to fish out the evidence that will make the conviction as plain to others as it is to me."

Simon de Pury, chairman of the Phillips de Pury & Company auction house, confirms the continued emphasis on the human touch. "With many artists, you have an institution or an individual who becomes the final word, or the word that the market accepts as being the final word at any given time. And if the author or the expert that the market respects is not accepting the work, you cannot sell such a work."

It's not that the field of connoisseurship is allergic to technical change. "I think we're very open to any new technology," said Gregory Rubinstein, senior director in the department of old master paintings and drawings at Sotheby's auction house. "It's terribly sad that some art historians are so suspicious of it."

He mentions the example of a Vermeer that Sotheby's will offer for auction in July, after a 10-year investigation into its legitimacy. The research involved modern techniques like infrared and X-ray imaging, but the ultimate determination, Mr. Rubinstein said, was "based more on greater knowledge than serious technology." An example is the analysis of thread counts in Vermeer's canvases. "That knowledge wasn't available 20 years ago," he said. "It's not complicated technologically, but it's just that nobody had ever done it before."

Nonetheless, he said, "The ultimate decision about whether something is by artist X or not is a subjective one, and has to be based on what you see with your eyes."

Dr. Thomas Wessel, a Cologne, Germany, art historian for AXA Art Insurance, a company that insures many museums and private collections, finds such new technology fascinating from a scholarly perspective, but says that "in terms of authentication, it doesn't change a thing." He points to a more practical reason, however: auction houses would be unlikely to embrace a new technique that "sharpens the consciousness of how many fakes they may have sold," he said.

Indeed, a paradoxical effect of the proliferating technologies is that they can raise more questions than they answer. The art scholar Gary Schwartz, who has written widely on Dutch and Flemish masters, mentions the example of infrared reflectography, a technique used to reveal underdrawing beneath the surface of a painting. "They thought that I.R.R. would provide clinching proof for old-style attributions," he said. "It took them 30 years to realize that it was answering questions they had not yet asked."

The high stakes involved create an atmosphere in which art experts tend to rely on established traditions, rather than rock the boat with contentious new insights. "They put themselves in for an awful lot of autosuggestion by pinning their reputations and staking their professional status on attributions," said Mr. Schwartz. "It becomes hard to see things in another way."

This situation makes some experts less willing to put their necks on the line by making attributions. "It is a growing tendency," says the Swiss art dealer and van Gogh expert Walter Feilchenfeldt. "There also is this American way of suing if you have an opinion that doesn't suit, let's say, the owner of a work of art. At that moment, science stops. It's like Galileo, when he said the world moves around the sun, and he was sentenced for saying so."

But according to Mr. Rubinstein of Sotheby's, this heightened sensitivity is just a reflection of the times. "Like every other aspect of life, things are more litigious now, so we have to be more careful than our predecessors half a century ago were," he said.

The knowledge resulting from new scholarly and technical research calls for finer nuances. "The process of attribution has become much more rigorous over the last generation," he said. "Things that even maybe 30 years ago might have been cataloged for sale, or even in a museum collection, as `by artist X' might now be described as `attributed to,' because we are constantly refining our understanding of what each artist did."

In the world of van Gogh scholarship, one voice rings loudest: the Van Gogh Museum in Amsterdam. "They're definitely considered to be the authoritative voice," Mr. de Pury said.

Which is why the people behind the Authentic project are so keen to collaborate with them. "We really depend on the museum," Mr. Berezhnoy said. "Only in close cooperation with museums can we make something that really works."

That cooperation is now under way, since curators at the Van Gogh Museum recently agreed to provide the researchers with a series of high-quality transparencies that will help with their experiments.

Why the interest? "Common sense," said Dr. Louis van Tilborgh, a curator at the Van Gogh Museum who has been working with the Maastricht team. "We know that the computer can do things that people can't, so the question is can you also use it for research into van Gogh."

He is not looking for results by tomorrow. "We didn't enter into this project in order to immediately have success," he said, "but to see if it can do things for us that we cannot do, or can do them better and faster. So now we're testing the waters, and whether that will lead to anything, we don't know yet."

Mr. Rubinstein, who hasn't seen the Authentic system, wondered how it would account for spontaneous shifts in a painter's style. "An artist sometimes just does something out of the blue in a creative or imaginative way, which maybe extends even to aspects of brush strokes and not just composition," he said, "and this might confuse a purely empirical study of the type that a computer could do."

But Authentic's goal isn't to provide the easy answers most people want. "We do not make binary decisions," said Mr. Berezhnoy, referring to simple yes or no answers about a work's authenticity. "We analyze paintings and give numbers and numerical facts to art experts to make their life easier. We are like forensic experts, but much cheaper because we need only images for our methods."

Dr. Postma realizes this noncommittal approach can be frustrating. "Like with the weather, you want to know is it going to rain or not?" he said. "People will always want to know whether it's genuine or fake."

Even against this backdrop, Dr. Postma believes the computer's march into the museum is inevitable. "I think it will be commonplace a few years from now," he said. "Art experts will rely on computers to support their theories. And it's cheaper, it's faster and it's more reliable."  

Douglas Heingartner is a journalist based in Amsterdam.

Subscribe Today: Home Delivery of The Times from $2.90/wk.