BRDF models for accurate and efficient rendering of glossy surfaces

Abstract

This article presents two new parametric models of the Bidirectional Reflectance Distribution Function (BRDF), one inspired by the Rayleigh-Rice theory for light scattering from optically smooth surfaces, and one inspired by micro-facet theory. The models represent scattering from a wide range of glossy surface types with high accuracy. In particular, they enable representation of types of surface scattering which previous parametric models have had trouble modeling accurately. In a study of the scattering behavior of measured reflectance data, we investigate what key properties are needed for a model to accurately represent scattering from glossy surfaces. We investigate different parametrizations and how well they match the behavior of measured BRDFs. We also examine the scattering curves which are represented in parametric models by different distribution functions. Based on the insights gained from the study, the new models are designed to provide accurate fittings to the measured data. Importance sampling schemes are developed for the new models, enabling direct use in existing production pipelines. In the resulting renderings we show that the visual quality achieved by the models matches that of the measured data.

Subjective Evaluation of High-Fidelity Virtual Environments for Driving Simulations

“…This device captured a discrete approximation of the BRDF for a particular material. In order to use this measured BRDF data in the rendering system, the BRDF was fit to an analytical representation by following the procedure by Löw et al [24]. The measured data was fit to a normalised Phong model for the Ray Casting (Section IV-A1) and Ray Tracing (Section IV-A2) methods, and was fit to a Microfacet model for use in the Global Illumination methods (Section IV-A3).…”

BRDF Representation and Acquisition

“…Kurt et al [KSKK10] The multiple specular lobe model can represent layered or mixed materials. Bagher et al [BSH12] Provides accurate fitting for materials in the MERL database Löw et al [LKYU12] Guarantees accurate fitting to measured data for glossy surfaces; descibes 2 models based on the ABC distribution DATA-DRIVEN Matusik et al [MPBM03a] Provides realistic appearance and meaningful parameterisation Matusik et al [MPBM03b] Reduces number of samples to acquire and represent BRDF ISOTROPIC Romeiro et al [RVZ08] Bivariate representation, allows to capture off-specular and retro-reflections. [Arv95] or with the simpler option of a power series in (n · h) with a suitable sequence of exponents [Lew94].…”

“…Low et al [LKYU12] proposed two isotropic models for glossy surfaces, based either on the Rayleigh-Rice light scattering theory (smooth surface BRDF) or on the microfacet theory (microfacet BRDF). Both models make use of a modified version of the ABC model [CTL90,CT91], which was originally formulated to fit the Power Spectral Density of some measured smooth surfaces.…”

An abridged goniometer for material appearance measurements

“…[11][12][13] To select the BRDF model best suited for the set of 36 materials studied in this paper the Ward, 14 Ashikhmin-Shirley, 15 and Smooth Surface BRDF 12 models were evaluated, and the cube root metric was used during the optimization process, as compressive metrics were found to obtain a better perceptual approximation than regular RMS metrics. 12,13 A Lambertian lobe was used to represent the diffuse reflectance (p d in Equation 4) and it was set to the measurements taken at a 45:0 geometry, being commonly used in colorimetry to remove specular reflections. The non-linear optimization of the diffuse reflectance was also tested, leading to an overestimation of the diffuse reflectance for the high gloss materials used, as more weight was given to the specular lobe, and a more accurate approximation of the diffuse reflectance for low gloss materials.…”

“…The Ward and Ashikhmin-Shirley BRDF models have an underlying gaussian distribution, which do not approximate the shape of the measured data well, specially the tail of specular lobes. 12 If the specular lobe is approximated with a gaussian function, the near-diffuse region is constrained by the gaussian distribution to be narrower than the measured data. An example of this can be seen in the top left plot of in Figure 6, where the measured data is approximated with the Ward BRDF model.…”