Efficient Hierarchical FE-Techniques for Global Illumination

Philipp Slusallek

Stanford University


One of the core elements of computer graphics is the desire to create convincing photo-realistic images of virtual worlds. Accounting for the subtle but essential lighting effects is a key factor in this process. The proper understanding of the physical process of light transport and its efficient simulation have been the focus of research in this area for the past years.

Mathematically global illumination requires the solution of a Fredholm equation of the second kind for which FE methods have played a primary role. Important issues here are the sheer size of the problem, often already in the order of several hundred thousand elements but increasing steadily, the non-local nature of light transport leading to sparse but highly unstructured matrices of interaction coefficients, and the high cost of computing the coefficients, which requires to compute global visibility.

Hierarchical techniques allow us to explore the unknown structure of a problem in a top-down fashion and to adapt the solution to this structure. However, top-down hierarchical techniques require efficient but reliable predictors for steering the refinement and for computing a solution with a minimum amount of work for a given accuracy.

The talk presents hierarchical techniques for global illumination in diffuse (2D, radiosity) and non-diffuse (4D, radiance) domains, discusses robustness and the efficient computation of bounds on interaction coefficients for refinement, and introduces special iterative solution algorithms that minimize storage requirements when dealing with complex scenes.

Philipp Slusallek is currently a visiting assistant professor at the Computer Graphics Lab of Stanford University. He received a Diploma in physics from the University of Tuebingen, Germany and a PhD in computer science from the University of Erlangen, Germany in 1995. He was responsible for the design of a commercial CAD package and has been leading the Vision project. Vision is a object-oriented, physically-based rendering system that provides an integrated architecture for state-of-art rendering and global illumination algorithms. His research interests include physically-based and realistic image synthesis, image-based rendering, and object-oriented software design.

He can be contacted at slusallek@graphics.stanford.edu.

Edited by Leonidas Guibas