Can a computer learn to paint with your style?
DeepArt is a free online computer algorithm that claims to transform your photos into painterly images, using your own painting style to guide the computer.
I thought I'd try it out by uploading a photo (left) and a closely related plein-air oil painting (right) to see what the algorithm comes up with.
After waiting about 10 hours, I got an email saying my image was ready:
The result is pretty disappointing, with strange dark slashes in the sky that weren't in either the photo or the painting. The rest of the building looks really crudely painted.
I tried it once more with a photo and a painting that were really different.
The output uses the color scheme of the painting, and grafts paint-like textures to match the relative values. Beyond that it didn't make any choices that I would regard as artistic. Although the result is marginally more interesting than a Photoshop filter, it doesn't look like a painting. The algorithm doesn't do well with faces, which require a particular attention to the eyes and mouth.
Despite the shortcomings of this algorithm, it's easy to imagine the power of future software that not only patches together stylistic fingerprints, but also uses strategies of machine perception and image parsing.
You can also try their Visual Turing Test.