Article Dans Une Revue Journal of the Optical Society of America. A Optics, Image Science, and Vision Année : 2024

Data-driven correction for the masking model of Smith

Résumé

To render realistic material appearances, physically based models often rely on the microfacet theory. These models require several parameters that drive the distribution of microfacet orientations, their reflectance, and a geometric attenuation factor. The latter accounts for self-masking and self-shadowing; it must be managed carefully when physical plausibility is required. The masking term proposed by Smith [ IEEE Trans. Antennas Propag. 15 , 668 ( 1967 ) IETPAK 0018-926X 10.1109/TAP.1967.1138991 ] is widely used for its accuracy when employed with theoretical distributions. However, it does not ensure exactness when compared with the masking of measured microsurfaces. We have conducted an in-depth study of the error associated with isotropic roughnesses, based on a ray-casting measurement with mesh-based surfaces. This article proposes a correction function that can be added to the theoretical masking term at a very low computation cost while bringing the masking closer to the ground truth. Our correction term is built from a linear combination of two Johnson SB distributions, parameterized according to statistical features of the microsurface. We show that the resulting masking term always reduces the error when compared to the original Smith term alone. This improvement is illustrated in the whole bidirectional reflectance functions with rendered images.

Fichier principal
Vignette du fichier
output.pdf (6.08 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04868067 , version 1 (12-12-2025)

Licence

Identifiants

Citer

Elsa Tamisier, Mickaël Ribardière, Daniel Meneveaux, Sébastien Horna, Pierre Poulin. Data-driven correction for the masking model of Smith. Journal of the Optical Society of America. A Optics, Image Science, and Vision, 2024, 41 (9), pp.1777. ⟨10.1364/JOSAA.530840⟩. ⟨hal-04868067⟩
270 Consultations
169 Téléchargements

Altmetric

Partager

  • More