Improving the Selection of Bases of BRDFs for Appearance Preservation

An important step in the appearance preservation of real materials is the analysis of how they interact with light. Since this phenomena happens at a microscopic level, heuristics with different complexity have been developed to capture and reproduce it. In order to minimize sampling efforts, one of these approaches consists in representing the reflectance of a material as a linear combination of a basis of known reflectance functions. To accomplish realistic and efficient representations, this basis must be expressive and contain a reduced number of elements. This work presents three approaches to select such basis. The first one performs an empirical leave-one-out optimization procedure. The other two are based on classical and evolutionary clustering algorithms. To improve clustering results, a new BRDF-oriented fitness function is designed. These approaches are evaluated using NNLS algorithm to estimate sampled materials and a comparison based on numerical precision is performed.