A Generalised-Mutual-Information-Based Oracle for Hierarchical Radiosity

摘要

One of the main problems in the radiosity method is how to discretise a scene into mesh elements that allow us to accurately represent illumination. In this paper we present a new refinement criterion for hierarchical radiosity based on the continuous and discrete generalised mutual information measures between two patches or elements. These measures, derived from the generalised entropy of Harvda-Charvat-Tsallis, express the information transfer within a scene. The results obtained improve on the ones based on kernel smoothness and Shannon mutual information.