Machine Learning and the Old Masters

In his book Secret Knowledge painter David Hockney advanced the idea that the Renaissance masters had worked from 3-D images of their subjects that were created by curved mirrors. Disputing this theory provided David G. Stork his entry into the computer analysis of art.  Now Chief Scientist at Ricoh Innovations, Stork uses a combination of optics and high-powered computing to get inside a painting and create a virtual world from which it was drawn.  Among the more famous subjects on which he has trained his "computer vision" are Vermeer’s "Girl with a Pearl Earring," Jan Van Eyck’s "Portrait of a Cardinal," and Veláquez’s "Las Meninas." 

 

Many of this wide-ranging physicist’s thirty-some patents involve neural network technology, and his publications center on statistical analysis of images, machine learning, and neural network optimization.  His computer analysis of paintings focus–sorry, but there’s no way to skirt this pun–on the sources and behavior of light in the two-dimensional painting surface and use what the computer "learns" to simulate the fully dimensional world from which the painting was made.  Stork posits, for instance, the real-world version of the daylight streaming through the window of the tavern in Caravaggio’s "Calling of Saint Matthew" was actually the glow of a lantern.