A General BRDF Representation Based on Tensor Decomposition

Generating photo-realistic images through Monte Carlo rendering requires efficient representation of lightsurface interaction and techniques for importance sampling. Various models with good representation abilities have been developed but only a few of them have their importance sampling procedure. In this paper, we propose a method which provides a good bidirectional reflectance distribution function (BRDF) representation and efficient importance sampling procedure. Our method is based on representing BRDF as a function of tensor products. Four-dimensional measured BRDF tensor data are factorized using Tucker decomposition. A large data set is used for comparing the proposed BRDF model with a number of well-known BRDF models. It is shown that the underlying model provides good approximation to BRDFs.