The field of non-photorealistic rendering (NPR), a subfield of computer graphics, develops methods for the automatic production of abstract renderings ranging from pure technical illustrations to artistically looking images that, e.g., mimic oil paintings or watercolor. The methods allow visual abstraction of given geometric data in various ways. In this project we want to transfer abstraction methods from NPR to information visualization and scientific visualization. This can be done for all those visualizations that work with geometric representations (surfaces, bodies, glyphs, volumetric data etc.). Secondly, we want to investigate how to compute, given an input scene and an intended degree of abstraction, a valid visual abstract representation in a particular abstraction style.pMost abstraction methods in their current form do not have a proper parameterization to control their output complexity. For others the relation between parameters and output complexity is highly non-linear. However, many applications in visualization (and also computer graphics) would need such a quantitatively defined degree of abstraction. In real time scenarios only a limited amount of geometric elements can be drawn which results in a certain level of abstraction, for small and mobile devices visualizations have to be shown with not too many details, in other applications models of varying complexity have to be combined and need a similar degree of abstraction.pIn the past, a great number of abstraction styles have been developed, ranging from pen-and-ink to watercolor, from very loose to precise, from pure black and white to colorful. In this project we will select styles that encompass most popular forms of visual abstraction. Measures for abstraction will be developed that deal with number, distribution and form of drawing primitives (dots, lines, curves, objects etc.) and their drawing complexity. Furthermore, different forms of model representations and their abstraction will be used: surface/triangle-oriented models as well as volumetric data. Thus we will be able to support many kinds of visualizations. The following scientific questions will be addressed.
