Bilateral blue noise sampling

Jiating Chen (1) , Xiaoyin Ge (2) , Li-Yi Wei (3) , Bin Wang (4) , Yusu Wang (2) , Huamin Wang (2) , Yun Fei (4) , Kang-Lai Qian (4) , Jun-Hai Yong (1) , Wenping Wang (5)

1 THSS - School of Software
2 OSU - Ohio State University [Columbus]
3 Microsoft Research [Redmond]
4 cgcad, Thss
5 Department of Computer Science [Hong Kong]

Blue noise sampling is an important component in many graphicsapplications, but existing techniques consider mainly the spatialpositions of samples, making them less effective when handlingproblems with non-spatial features. Examples include biologicaldistribution in which plant spacing is influenced by non-positionalfactors such as tree type and size, photon mapping in which photonflux and direction are not a direct function of the attached surface,and point cloud sampling in which the underlying surface is unknowna priori. These scenarios can benefit from blue noise sampledistributions, but cannot be adequately handled by prior art. Inspired by bilateral filtering, we propose a bilateral blue noise samplingstrategy. Our key idea is a general formulation to modulatethe traditional sample distance measures, which are determined bysample position in spatial domain, with a similarity measure thatconsiders arbitrary per sample attributes. This modulation leads tothe notion of bilateral blue noise whose properties are influencedby not only the uniformity of the sample positions but also the similarityof the sample attributes. We describe how to incorporateour modulation into various sample analysis and synthesis methods,and demonstrate applications in object distribution, photon densityestimation, and point cloud sub-sampling.

Domaines: Synthèse d'image et réalité virtuelle [cs.GR]

Jiating Chen, Xiaoyin Ge, Li-Yi Wei, Bin Wang, Yusu Wang, et al.. Bilateral blue noise sampling. SIGGRAPH Asia 2013, Nov 2013, Hong Kong, China. ⟨hal-00920673⟩