Abstract: We present a new clustering algorithm for global illumination in complex environments. The new algorithm extends previous work on clustering for radiosity to allow for nondiffuse (glossy) reflectors. We represent clusters as points with directional distributions of outgoing and incoming radiance and importance, and we derive an error bound for transfers between these clusters. The algorithm groups input surfaces into a hierarchy of clusters, and then permits clusters to interact only if the error bound is below an acceptable tolerance. We show that the algorithm is asymptotically more efficient than previous clustering algorithms even when restricted to ideally diffuse environments. Finally, we demonstrate the performance of our method on two complex glossy environments.

One-line summary: Extending the clustering methods known from diffuse global illumination ("radiosity") to glossy global illumination.

Published in: ACM Transactions on Graphics (TOG), volume 16, number 1, pages 2-33. ACM, January 1997.

Download paper here: tog97.pdf.

Back to Per's publication page.