Using Dimensionality Reduction to Create New Materials from Tabular BRDFs

Representing reality in computer graphics requires simulating the appearance of real-world materials. Bidirectional Reflectance Distribution Functions (BRDFs) are commonly used to perform this task, as they represent the appearance of a material through the quotient between the reflected radiance and the incoming irradiance at this point. Acquiring the BRDF of a real material is an expensive task due to its high dimensionality, so strategies to take advantage of already existing measurements are a promising solution. In this paper we present an approach to create BRDFs for new materials from a basis of tabular representations of BRDFs. We apply dimensionality reduction in this basis and then we perform a triangulation in the resulting reduced space. Thus, any position in this reduced space can be used to create a new material: using the triangulation, we estimate coefficients that can be used to find the BRDF of this new material by interpolating materials in the original space. In addition, we present a technique to navigate in this reduced space that enables the creation of several different materials from two materials in the basis. This approach provides a wide range of new materials based on real measurements without the need to acquire new BRDFs. We show that our navigation is coherent and that it yields a smooth transition between materials.