Notes on the Ward BRDF

abstract
the anisotropic BRDF introduced in [ward 1992] has become widely used in computer graphics, but some important implementation details are less widely known. we discuss how to efficiently evaluate the Ward BRDF. then we derive the probability density function for its associated Monte Carlo sampling shceme and the correct weights to use with the generated samples. finally for the isotropic version, we describe how to bound the maximum possible BRDF value over a region of (direction) space.

1 introduction

the ward BRDF (Bidirectional Reflectance Distribution Function) was introduced in [Ward 1992] as an empirical model to fit measured BRDF (i.e. surface reflectance) data. it has several advantages over the prior BRDF modles and has become widely used in the computer graphics community. it uses only a few simple parameters making it easy to model, can be sampled efficiently for Monte Carlo, can model anisotropic surfaces, and as shown to fit reasonably well to measured BRDF data.

the purpose of this paper is calirfy and correct some important implementation details of the Ward BRDF. we dicuss how to efficiently evaluate the BRDF in section 2. monte carlo BRDF sampling is required for many rendering algorithms, and hence, ward provided an efficient sampling with his BRDF. however he did not provide the associated probability density function, which for mathematical accuracy, is needed to correctly weight the generated samples. we both disucss how to derive such probability density functions in general in section 3, and present the specific results for the Ward BRDF in equation 10.

another powerful, though less widely used, BRDF operation is the ability to bound its maximum vlaue over a range of directions. in section 4, we discuss how to cheaply and tightly bound the isotropic ward BRDF over a set of directions defined by a spatial bounding volume.