Where Computer Science Meets Abstract Art


Parker Hayes ’17 (left) and Jeonguk Choi ’18

Two computer science majors are spending the summer learning about two of the twentieth century’s most innovative artists: Wassily Kandinsky and Jackson Pollock. Parker Hayes ’17 and Jeonguk Choi ’18 are working on computer programs that can imitate the work of the two painters. No easy task, they both agreed. Both students are working under the supervision of Assistant Professor of Digital and Computational Studies and Computer Science Mohamed T. Irfan, who specializes, among other things, in the image analysis of art.

Imitating Pollock
“We have a long way to go,”  said Choi, who has spent much of the past few weeks staring intently at the paintings of Jackson Pollock, a major figure in the abstract expressionist movement, known for his unique and distinctive style of drip painting. Choi, who was awarded a Kibbe Science Fellowship, has written a program that generates a passable Pollock to the untrained eye. As he activates the computer program, even splashes of color appear on the plasma screen, as curved lines shoot off in apparently random directions. But Choi knows it’s some way from passing the authentication test. “For example if you study the corners, you see the lines have a very homogenous curvature due to the limit on control points. We tried to make them more varied but it didn’t work yet.

No. 5. Jackson Pollock. 1948

No. 5. Jackson Pollock. 1948

“I’ve been trying to figure out the characterstics of the curves that Pollock paints, the thickness of the brush strokes, and all the mathematical properties of his work.” One aspect of the paintings that amazed Choi, and also amazes art critics, is Pollock’s use of fractals: these are naturally occurring, repetitive, physical patterns found, for example, in cloud formations and tree branches. “There’s a well-known algorithm for detecting fractal structures in an image. Using this algorithm, fractals were found in Pollock’s paintings, which is extraordinary because the concept of fractals wasn’t really defined until the 1960s, after Pollock had died.” Choi said the algorithm he employs for imitating Poillock’s paintings makes sure that the “resulting image preserves fractal properties. The next step would be to extend this automatic process to preserve other known mathematical structures, in addition to fractals, in Pollock’s paintings.”

Before undertaking this fellowship, Choi said he had no particular interest in Pollock. “I just thought it was someone throwing paint on the canvas, anyone can do it.” But over the summer his opinion changed: “The more I study Pollock, the more his paintings mean to me. I would definitely go to a Pollock exhibition, given the chance, which is not something I would have said a year ago.”

The Appeal of Kandinsky
Parker Hayes was also drawn to the idea of integrating art and computer science. “It was fun,” he said, adding that he might even take an art history class next year “because I now appreciate art and the work that goes into it much more, especially Kandinsky, who wrote books about his work specifically describing the patterns that he followed.” Like Pollock, Kandinsky is known as a painter of abstract art, but unlike Pollock, Kandinsky’s work is known for its geometric precision.


Composition viii. Wassily Kandinsky. 1923

“For example,” said Hayes, “if you look at Kandinsky’s use of circles, they are generally on the top portion of his paintings, rising like bubbles, with the bigger circles rising higher.” Colors and shades meanwhile represent different feelings and emotions, said Hayes, and it was all clearly defined by Kandinsky, who was known as an art theorist as well as a painter.

“I’m writing a computer program that can take in images of his paintings,” said Hayes, “and then ultimately recreate Kandinsky-like art, using computational algorithms to map out lines, triangles, squares, circles, where they are on the canvas (and where they are not), what angles are used, and whether or not they’re overlapping.” Hayes, the recipient of a Surdna Foundation fellowship, said this automatic mapping process requires an understanding of Kandinsky’s geometric compositions. “For that, we have written a program that automatically identifies all straight line segments in an image. The program can be applied beyond Kandinsky’s paintings to any geometric compositions with straight lines,” said Hayes.

Hayes and Choi are among the estimated 200 Bowdoin students working on campus over the summer engaged in faculty-mentored research.