Isotropic clustering for hierarchical radiosity - implementation and experiences

Although Hierarchical Radiosity was a big step forward for finite element computations in the context of global illumination, the algorithm can hardly cope with scenes of more than medium complexity. The reason is that Hierarchical Radiosity requires an initial linking step, comparing all pairs of initial objects in the scene. These initial objects are then hierarchically subdivided in order to accurately represent the light transport between them. Isotropic Clustering, as introduced by Sillion, in addition creates a hierarchy above the input objects. Thus, it allows for the interaction of complete clusters of objects and avoids the costly initial linking step. In this paper, we describe our implementation of the isotropic clustering algorithm and discuss some of the problems that we encountered. The complexity of the algorithm is examined and clustering strategies are compared