“I’ve always been interested in using computers to analyze art. AI can do a lot to help art historians better understand a particular artist or genre of art. But while current AI models are pretty good at analyzing photos, for example, art—as an expression of human imagination, intelligence, and creativity—is different. Artists have different styles and different ways of expressing themselves. In that realm of abstraction, I don’t think today’s AI is capable of understanding every aspect of a piece of art.

“My group is developing a tool to help art historians better understand the directional organization of brushwork in well-known works of art. We applied computational techniques to digital images of Impressionist paintings. Beginning at sampled points on a particular painting, our method traces streamlines forward and backward along the dominant local directions within the painted surface until that directional continuity breaks down. This gives us a map of the artists’ brushwork directional flow. In Monet’s Haystack series, for example, our analysis shows how the artist directed his brushstrokes to create the haystack shapes and the light that falls on them. 

“The directional map created makes it easier to discern patterns in the brushwork that are difficult to follow with the naked eye, and can help art historians compare hundreds, potentially thousands, of paintings by one artist, or compare the techniques of different artists. We are currently preparing an online demo for anyone studying artworks—and anyone interested in learning how a particular artist worked—to upload a painting and create a streamline map of the piece.”