Simon Breslav from the Environments and Ergonomics team likes to draw and did some work with his colleagues at the University of Toronto on teaching the computer how to draw. The computer can study the line styles on a drawing and apply it to a new 3D model creating a hand drawn look similar to the artists original creation. This can be useful for both replication and restoration of historic works as well as new applications such as animated cartoons.
The group had artists hand shade 3D models of different objects and taught the computer to analyse what the artists had done. From the drawing they pulled out the following information:
- Hatching level: whether a region contains no hatching, single hatching, or cross-hatching.
- Orientation: the stroke direction in image space
- Cross-hatching orientation: the cross-hatch direction when present
- Thickness: the stroke width
- Intensity: how light or dark the stroke is
- Spacing: the distance between parallel strokes
- Length: the length of the stroke
These factors can be visualized:
And then synthesized - both making for interesting drawings themselves:
The results of the learning and application are pretty impressive as you can see below:
This work is the first of its kind in learning the complexities and intricacies in the human artistic process. Future studies may include stroke textures, stroke tapering and randomness in strokes (such as wavy or jittered lines).