Van Gogh Trains Forgery Recognition System

A Research team at the Penn State University has used 101 Vincent van Gogh paintings to teach their computers to recognize art forgeries. The unique style of van Gogh’s work and the large number of paintings he created, made the choice for van Gogh obvious.

As museums continue to digitize their art collections, the artwork becomes available In high-resolution digital images. As a result, more and more forgers are able to copy paintings in every detail. A good reason for James Wang, a researcher at Penn State University, to train a computer to recognize fake paintings.

How does one teach a computer to distinguish a forgery from a genuine painting? Exactly the same way as you would teach a human – by studying a lot of paintings for a long time.

Vincent van Gogh was an obvious choice for the research team, he was a very productive artist and there are several paintings that experts cannot agree if they are genuine.

The research team analyzed 101 high-resolution scans of paintings (made in shades of grey), which were either by van Gogh or in his style. The van Gogh and Kroller-Muller Museums in the Netherlands provided the scans. By dividing each image into sections the professionals study the artist’s distinctive brush strokes based on patterns and geometric contours.

82 of the paintings had been attributed to van Gogh, and 6 others were knowns to be painted by others. Another 13 paintings, which experts cannot agree on, were added to the research population. 23 of the selected paintings were used to ‘train’ the computer in recognizing the way van Gogh painted.

After the training, the computer managed to recognize some paintings. After only two known van Gogh paintings had been used for training, the system did not recognize a fake painting. But after the image analysis had been intensified, the forgery was shown to be different.

Wang describes the results as ‘encouraging, but not perfect’. Further development of these computerized image-processing systems ‘can help us learn about artists’ techniques and styles, says Dr. Wang.

Computers will probably remain one of the tools to detect forgeries. The researchers claim that they don’t want to replace art historians, their methods aren’t sophisticated enough top do so, “Art experts will always have the final say.”

June 8, 2008.