It works like this: At the sight of a suspicious piece of art, the investigator snaps a photo using a cell phone. Software in the phone automatically calls a number that connects to a central computer server. The digital picture is transferred and an image analysis program compares the snapshot with photos in the database.
It takes about 30 to 40 seconds for the software to transfer the image, identify the shape, outline, color and texture of objects in the original image and return a top ten list of closest matches to the detective.
"If the painting is clearly registered as stolen, the investigator has to confiscate it. But the exact procedure depends on the law of the appropriate country where the auction takes place," said Nickolay.
Julian Radcliffe, chairman of the Art Loss Register headquartered in London, has looked into image recognition technology as a way to spot stolen art, but expressed skepticism. "None of the image matching is good enough to repace the art historians we use," he said.
According to Radcliffe, searches for stolen art happen weeks before an auction. Art historians, who typically speak three or four different languages and are well-versed in the jargon of fine art, use details from auction catalogs to search their computer database for matches.
Nickolay and Veenhuis are currently investigating other uses for the image analysis system, such as exposing counterfeits.
