Reproduction and application of stylistic means in non-photorealistic computer graphics

Non-photorealistic rendering (NPR) has been receiving an increasing attention within the last 15 years. In contrast to photorealistic computer graphics, which is bound by a physically correct simulation of light transport and material properties, NPR encompass a rich tool set enabling expressive and personal interpretation by individual rendering styles, and is thereby able to communicate information in a focused and efficient way. Therefore, visual perception, aesthetic requirements, or traditional artwork can be the driving force in non-photorealistic rendering. This dissertation thesis explores different aspects of a NPR pipeline designed for computer generated watercolor illustrations. The digital reproduction of such a complex painting style requires a holistic consideration of the painting process including the emulation of the painting medium as well as creative aspects such as extracting essential features, placing accents, or exploiting color contrast. We see our contribution in the improved visual aesthetic and believability of computer generated watercolor renderings won by combining universal principles derived from traditional art work with effective techniques for emulating the tools and the medium. We contribute to the following topics: First, we formulate two approaches for the reproduction of the visual attributes of watercolor. In contrast to physicallybased approaches usually applied in canvas simulations, we concentrate on empirical approaches that allow to process 2d and 3d input data efficiently. Second, we reproduce and apply stylistic means within the painting process to transfer universal principles from traditional art work to computer generated art work in order to improve the visual aesthetic and believability of the watercolor renderings. Finally, we present collaborations that applied the findings of this thesis in film and graphics industry. A first project aims at the integration of a watercolor style into 3d animation pipelines. A second project was initiated with the goal for a simple, one-click solution producing convincing watercolor illustrations from complex 3d cad data.