Computer model learns to copy artists' drawing styles
Last year, we heard about an industrial robot that was able to create pencil sketches of human subjects. In that case, it utilized algorithms that identified the boundaries between areas of high and low contrast on each subject’s face. Now, however, scientists at Disney Research, Pittsburgh, have taken things a bit further. They’ve created a computer tool that not only produces digital sketches, but it also copies the style of specific human artists as it does so.
The researchers started with seven artists, each of which used a stylus and tablet to create digital sketches based on photos of 24 male and female subjects. The artists did four sketches of each photo – they were allowed 270 seconds to create one, then 90, 30 and finally 15 seconds. Needless to say, the sketches became increasingly abstract as the time limits were shortened.
As they drew, the artists’ individual strokes were recorded. Factors such as length, curvature, stylus pressure, stylus speed, and directionality were taken into account. The strokes were then divided into two groups – shading strokes and contour strokes. Contour strokes were further sub-divided into complex and simple strokes. The computer noted the order in which the strokes were drawn, too – different artists use different facial features as their starting, middle and end points.
Additionally, the computer model noted each artist’s distinctive tendencies to deviate from the geometry of the faces in the photos. Whether an intentional aspect of their style or simply a mistake, some of the artists repeatedly drew subjects’ eyes closer together than they were in the photos, or gave them squarer jawlines.
Drawing on this database, the computer was subsequently able to create its own sketches of the photographed faces, in the style of each of the seven artists. While the computer and the artists’ sketches were by no means identical, they certainly did look like they could have been drawn by the same “person.” Examples can be seen in the video below.
It is now hoped that the technology could be extended to other types of drawing beyond portraits, perhaps ultimately finding use in fields such as artificial drawing, or animation. It could also be a useful tool for showing artists how they could improve, by providing them with a third-person view of their technique.
Source: Disney Research