http://vfacstaff.ltu.edu/lshamir/publications/wm_pollock.pdf
Lior Shamir, a computer scientist at Lawrence Technological University, has developed an algorithm for authenticating Jackson Pollock “drip” paintings by analyzing their fractal dimensions. The computer is correct 93% of the time. Counterfeit Pollocks are difficult for even pundits to distinguish from the genuine.
Pollock’s paintings obey fractal geometry. Moving around a large canvas laid on the ground, the artist let paint fly from all angles, using his whole body. Human motion is known to display fractal properties when people restore their balance, says Taylor, and films of Pollock seem to show him painting in a state of ‘controlled off-balance’. Second, the dripping and pouring itself could be a chaotic process. The machine extracted 4,024 image content descriptors from each Pollock painting and found fractal features and numerical image content descriptors such as Zernike polynomials, Haralick textures, and Chebyshev statistics. The machine compares the images with the Weighted Nearest Neighbor classification such that the Fisher discriminant scores of the content descriptors were used as weights.
Shamir obtained 26 paintings known to be by Pollock and a second set of counterfeits. The pieces were normalized to contain 640,000 pixels and then divided up into 16 equal-sized areas. Twenty works were used to train the software, then the remaining six were used for testing purposes. The analysis was then repeated multiple times to provide a greater statistical power. Given these paintings as input, the computer was accurate over 90 percent of the time when asked to determine whether a painting was real or fake. But previous computer analyses had suggested Pollock’s style evolved over time, so Shamir went back and found 26 paintings that were done in the first half of the 1950s. With these as the training set, the accuracy reached 100 percent. This shows the true power of machine learning. It is amazing how a computer can quantify the details at the pixel by pixel level once a painting has been digitized and “see” details and patterns that we do not consciously detect.
Alex Tsai's AI Blog 