Non-Photorealistic Rendering and the Science of Art
Aaron Hertzmann
Abstract
I argue that Non-Photorealistic Rendering (NPR) research will play a key role in the scientific understanding of visual art and illustration. NPR can contribute to scientific understanding of two kinds of problems: how do artists create imagery, and how do observers respond to artistic imagery? I sketch out some of the open problems, how NPR can help, and what some possible theories might look like. Additionally, I discuss the thorny problem of how to evaluate NPR research and theories.
Paper
A. Hertzmann. Non-Photorealistic Rendering and the Science of Art. Proc. NPAR 2010, Meta papers.
[PDF]
Slides
PDF
Figures
Some simple but effective ideas in NPR,
for cartoon illustration, painting, and line drawing.
The algorithms involved are based on simple mathematical or algorithmic ideas, and, given a set of high-level parameters, run completely automatically.
While there is much to criticize in these results, I believe it is nonetheless amazing and intriguing that they can be achieved by simple and automatic algorithms.
A photograph processed by the algorithm of
Winnemöller et al. [2006], and the variant of
Haeberli's method [1990] from
Hertzmann [1998]. Photo by Holger Winnemöller. (Click on the image for full resolution).
Suggestive contour line drawings
[DeCarlo 2003] of a 3D model, stylized
[Goodwin et al. 2007].
Left: Illustration from ``Bone'' [Smith 1998] (BONE® is Copyright © 2010 Jeff Smith).
Middle: A 3D model of the possum we created, and a rendering of the possum with our algorithm described in [Goodwin et al. 2007].
Right:
The line thicknesses in the drawing exhibit a number of properties that can be derived from the Isophote Distance formula.
These are rules that can be checked without knowing the 3D geometry.
(top) Because the wrist has higher curvature than the upper arm, it has thinner strokes. (middle) Occluded strokes are not tapered. (right) Distant objects have thinner strokes than nearby objects of the same type.