Abstract
The Bidirectional Reflectance Distribution Function, or BRDF, is used to model the reflectance properties of opaque materials. BRDFs are used in many fields, including applied physics, remote sensing, computer vision, and computer graphics. In computer graphics, artists specify BRDFs for each object in a virtual environment, and then render images of this environment. Traditionally, artists specify BRDFs through analytical models. However, tuning the parameters of these analytical models is time consuming, and often appropriate analytical BRDF models are not available. Increasingly, artists are turning to measured BRDF data to make this process easier. However, most existing measured BRDF data only has 3 color channels. Very little multispectral BRDF data is available, and the little data that is available has sparse angular sampling.
We present an image based BRDF measurement system capable of measuring multispectral BRDFs with dense angular sampling. We have measured several multispectral BRDFs, and we provide numerous plots, visualizations, and renderings of this data. We also present an analysis of several techniques used for calibrating BRDF measurement systems and algorithms for evaluating a BRDF from unstructured samples. Furthermore, we propose a new algorithm for extracting BRDF measurements from a set of images based on nonlinear optimization. This technique is simpler compared to techniques presented by previous authors.
Additionally, we conducted a large scale statistical analysis of 59 million real world reflectance spectra. Using Principal Component Analysis(PCA), we validate the work of previous researchers and demonstrate that the space of real world spectral reflectances is spanned by a low-dimensional linear subspace. We also analyze this dataset using Non-Negative Matrix Factorization(NMF) and discuss the implications of our results for future multispectral BRDF measurement systems.
