Importance sampling for physically-based hair fiber models

Abstract

We present a new strategy for importance sampling hair reflectance models. To combine hair reflectance models with increasingly popular physically-based rendering algorithms, an efficient sampling scheme is required to select scattered rays that lead to lower variance and noise. Our new strategy, which is tied closely to the derivation of physically-based fiber functions, works well for both smooth and rough fibers based on the Marschner et al. model and also for Lambertian fibers. It should be directly usable with future hair reflectance models that allow for more general cross-sections and more complex surface properties, provided the lobes are derived in a similar, separable fashion. Our strategy includes lobe selection and can efficiently sample complex lobe shapes like the Marschner TRT function. The scheme is easy to implement and requires no precomputation, allowing fully heterogeneous variation of all fiber parameters.