Efficient Realistic Image Synthesis Using Extended Hierarchies

Problem Definition

The rendering of three-dimensional models is moving toward greater realism. In applications such as architecture, training simulators, and movie-making this seems obvious. However, the reason low accuracy images of real scenes are used at all is because of the trade-off between the quality of the image and the time it takes to render. That is, realistic images are considered too expensive; but if they could be produced more efficiently, they would be used in the place of less realistic ones.

It is important now to find ways to bring this efficiency to highly realistic rendering. Two obstacles to rendering more realistic images are the need for computing correct interreflections and the need for geometrically complex models. In short, the continuum of reflection effects from sharp mirrors to color bleeding is only included in its entirety in the class of algorithms which account for interreflections between all scene primitives. These algorithms are inherently O(N squared) where N is the number of primitives, and this cost is what makes these algorithms so inefficient. Add to the already high cost the need for models which will have scene primitives in the millions, billions, or more.

My research will not focus on any aid in designing huge scene databases but it does anticipate the inevitability of these environments. I will justify the use of calculating complete interreflections and show why many currently used algorithms are insufficient while a few are the correct ones for true photo-realism (Monte Carlo Path Tracing, Radiosity, and hybrids of the two). My proposal is to explore methods of hierarchically organizing and rendering primitives to reduce the cost of interreflection preserving algorithms.


A completed and published paper that is part of this thesis was co-authored by myself - "Geometric Simpliciation for Inidirect Illumination Calcuations."

For more information, please contact Charlie Patterson.