Computer Doesn't Know What It Likes But Knows Art

Teaching computers to recognize the works of specific artists is another step toward reliable image recognition. Cynthia Graber reports.

If you’ve spent time in an art museum, you probably learned to appreciate the styles of different artists. You can walk into a room and say, oh, that’s a Matisse.  You may recognize the colors, the brush strokes, the subject choice—even if the artist incorporated other styles into the painting.

Researchers at the University of Haifa in Israel wanted to teach a computer to do the same thing. The scientists developed a system by which the computer breaks down the image and analyzes various aspects of it. People, flowers and other elements get represented with complex mathematics.

After repeatedly analyzing the brush strokes and content of, say, Salvador Dalí paintings, the computer learned his particular style well enough to look at a new image and determine whether Dalí himself painted it. This worked even when the art contained a variety of influences. The program won’t replace professional expert judgment any time soon. But researchers say it’s another step toward teaching computers how to see the world around them.

—Cynthia Graber