Radiosity: Textures, Quadratures, and Big Models
Department of Computer Science
Princeton University
rsg@uni.stanford.edu
Radiosity methods compute the global illumination of environments that consist of perfectly diffuse, Lambertian, reflecting and emitting surfaces. The realism of these environments is enhanced by the inclusion of lighting effects such as soft shadows and color bleeding.
Previous radiosity systems limit each surface to constant reflection and constant emission value. First, a framework that removes this constraint, thus increasing the functionality of radiosity systems will be described. The details and performance of a system that allows these functions to be arbitrarily defined through texture maps will then be discussed.
The computation of energy transfer between objects is the most expensive operation in radiosity systems. This is made tractable by projecting the outgoing and incoming energy distributions of the surfaces (called radiosity and irradiance, respectively) onto a finite-dimensional basis. The transfer of energy between these basis functions is defined as a set of integrals that must be, in general, solved numerically. A study of numerical integration techniques that focuses on solving these integrals efficiently and accurately is presented. The results of this study are used to develop an adaptive algorithm that can significantly improve speed without degrading precision.
Finally, future directions for computing illumination in environments with numerous objects will be discussed. Issues such as clustering, interactive walkthroughs, Monte Carlo methods, and more general reflection functions will be touched upon.
Texture work jointly done with Peter Schröder and Pat Hanrahan.