High-Quality Rendering and Visualization for Artistic Manipulations

Computer-generated imagery is crucial in many contexts: visualization of medical data, architectural design, simulations, movies and also games. Creating images on a computer, however, is far more complicated than it might seem at first glance.

The real world exhibits many complicated structures, such as skin and rough surfaces, which possess details so minute that they may often only be discerned in close-up, though they impact the overall appearance of the surface even from far away. Faithful representations of such detail use a vast amount of memory to the extent of easily exceeding the capacity of standard computing equipment. Visualizing this content is thus quite challenging, but remains of high importance in domains such as diagnosis from medical data.

Besides the problem of raw geometric complexity, realistic image synthesis relies on adherence to physical principles.

Accurate simulations quickly become computationally prohibitive in a real-time context, even for high-performance hardware. Our solution is thus tailored to create realistic and detailed imagery at a low cost. To this extent, we exploit sample-based representations captured from the real-world or obtained by transforming a complex virtual scene on the fly.

Such a structure promises many efficiency and cost advantages through techniques such as hierarchical processing, fast mapping and scheduling of multiple processors, computational coherence, and integration of real-world information.

We also often find that we may utilize adaptive methods, which when feasible, adopt simpler calculations with an equivalent outcome. Intelligently sampling only information from the large input datasets that is actually visible and has an influence on the image promises to facilitate a significant leap towards truly efficient large-scale rendering.

 

Project team

Principal Investigator
Prof. Dr. Elmar Eisemann

Researchers
Oliver Klehm
Dr. Matthias Holländer