Related Work

  Computer graphics algorithms that imitate non-photographic techniques such as painting or pen-and-ink are referred to as non-photorealistic rendering (NPR). The various NPR methods differ greatly in style and visual appearance, but are all closely related to conventional artistic techniques (e.g., [6,8,10,13,14,16,20,26]). An underlying assumption in NPR is that artistic techniques developed by human artists have intrinsic merit based on the evolutionary nature of art. We follow this assumption in the case of technical illustration.

NPR techniques used in computer graphics vary greatly in their level of abstraction. Those that produce a loss of detail, such as semi-randomized watercolor or pen-and-ink, produce a very high level of abstraction, which would be inappropriate for most technical illustrations. Photorealistic rendering techniques provide little abstraction, so photorealistic images tend to be more confusing than less detailed human-drawn technical illustrations. Technical illustrations occupy the middle ground of abstraction, where the important three-dimensional properties of objects are accented while extraneous detail is diminished or eliminated. Images at any level of abstraction can be aesthetically pleasing, but this is a side-effect rather than a primary goal for technical illustration. A rationale for using abstraction to eliminate detail from an image is that, unlike the case of 3D scene perception, the image viewer is not able to use motion, accommodation, or parallax cues to help deal with visual complexity. Using abstraction to simplify images helps the user overcome the loss of these spatial cues in a 2D image.

In computer graphics, there has been little work related to technical illustration. Saito and Takahashi [19] use a variety of techniques to show geometric properties of objects, but their images do not follow many of the technical illustration conventions. Seligmann and Feiner present a system that automatically generates explanation-based drawings [21]. Their system focuses primarily on what to draw, with secondary attention to visual style. Our work deals primarily with visual style rather than layout issues, and thus there is little overlap with Seligmann and Feiner's system, although the two methods would combine naturally. The work closest to our own was presented by Dooley and Cohen [7] who employ a user-defined hierarchy of components, such as line width, transparency, and line end/boundary conditions to generate an image. Our goal is a simpler and more automatic system, that imitates methods for line and color use found in technical illustrations. Williams also developed similar techniques to those described here for non-technical applications, including some warm-to-cool tones to approximate global illumination, and drawing conventions for specular objects [25].