NOVA recently broadcast an exciting episode on the mathematical/computer wizardry that researchers are developing to smoke out art forgeries.
Princeton wavelet pioneer Ingrid Daubechies is featured in the segment along with two electrical engineering students — Shannon Hughes and Eugene Brevdo.
NOVA challenged Daubechies’ team and two other teams (one based at Penn State and the other at Maastricht University) to train their algorithmic prowess on two seemingly identical paintings. One painting was a genuine Van Gogh and the other a highly skilled forgery. The teams were not told which was which. Each team used digital image analysis models they had developed to try to determine which was the fake. Then the teams gathered together and, while the cameras were rolling, each team announced which painting it had identified as the forgery. Did the Princeton team correctly identify the fake? You will have to watch the show to find out. J.F. Hannan interviewed the Princeton team for the Times of Trenton.
You can see some wavelet detective work in action in this visualization showing a rotating, three-dimensional arrangement of 73 paintings. The paintings are arranged according to “dissimilarity distances,” which distinguish paintings by Van Gogh from those by other painters. The red dots mark the paintings that are not by Van Gogh. Visit Daubechies’ website for more detailed information and to download the Proceedings of the May 2007 First International Workshop on Image Processing for Artist Identification.
