Machine Vision System Learns to Recognize Artists’ Styles

Have you ever gotten back from an estate sale across town, only to start wondering whether the oil painting you spent $10,000 on is really van Gogh’s Still Life: Vase with Five Flowers?

Wonder no more!

Computer science professors at the University of Haifa in Israel have devised a mathematical program that enables machine vision equipment to judge whether a piece of art is a van Gogh masterpiece or a knockoff from a lesser light.

The software permits artificial systems to learn an artist’s creative style, not merely how a particular image is supposed to appear. After samples of an artist’s works are presented before the system’s camera, the program breaks down the recorded images as a series of mathematical symbols.

Developed by professor Daniel Keren, the software can then recognize further pieces of art as being by the same artist, even if it has not “seen” them previously. For example, once the system has seen a few of van Gogh’s sunflower still lifes, it begins to identify the painter’s characteristic swirls and thereafter will recognize that the same artist created The Starry Night or Wheat Field with Crows once it views those paintings.

According to Keren, however, this type of computer-aided vision has more catching up to do with human recognition skills. For example, computers still have difficulty identifying how many human faces are in a painted or photographed scene, or even when an image is of a face.