Using computerized vision analysis to spot fake art

[nsane.forums]

In the world of art, the authenticity of a work can make millions of dollars worth of difference. Although it's possible to detect recent frauds based on objective measures like chemicals in the paints, it often requires subjective judgements to determine the difference between the work of a true master having an off week and the product of one of that master's students. A paper that will appear in PNAS suggests that a technique borrowed from vision research may help take some of the subjectivity out of this sort of analysis.

The study of artwork through math and statistics is known as "stylometry," and is a relatively recent development—similar methods have been used to analyze literature for much longer. The new paper uses a technique called sparse coding, in which analysts break down works of art into tiny patches and represent them as a series mathematical functions. By comparing the functions produced with authentic artwork to those from possible imitators, they can produce an objective measure of whether the piece in question is real or fake.

View: Original Article