Improving the Hierarachical Stochastic Radiosity Algorithm

Abstract

Hierarchical subdivision techniques remove the need for a-priori meshing of surfaces when approximating global illumination. In addition they allow progressive refinement which generates better solutions quickly. However, when subdivision is based upon Monte Carlo methods, due to the stochastic nature of such techniques, conclusions cannot be drawn unless a sufficiently large number of samples have been considered. Shadow boundaries are one of the main features such subdivision algorithms are designed to detect, but mesh elements that are in shadow receive less light, and hence are slower to subdivide. In this paper we consider and evaluate methods for modifying the Monte Carlo hierarchical subdivision algorithm to improve the detection of shadow boundaries and caustics.