Proceedings: GI 2017

Revectorization-Based Accurate Soft Shadow using Adaptive Area Light Source Sampling

Márcio C. F. Macedo (Federal University of Bahia, Brazil), Antônio L. Apolinário Jr. (Federal University of Bahia, Brazil)

Proceedings of Graphics Interface 2017: Edmonton, Alberta, 16-19 May 2017, 181 - 189

DOI 10.20380/GI2017.23

  • Bibtex

    author = {Macedo, M{\'a}rcio and Apolin{\'a}rio, Ant{\^o}nio},
    title = {Revectorization-Based Accurate Soft Shadow using Adaptive Area Light Source Sampling},
    booktitle = {Proceedings of Graphics Interface 2017},
    series = {GI 2017},
    year = {2017},
    issn = {0713-5424},
    isbn = {978-0-9947868-2-1},
    location = {Edmonton, Alberta},
    pages = {181 -- 189},
    numpages = {9},
    doi = {10.20380/GI2017.23},
    publisher = {Canadian Human-Computer Communications Society / Soci{\'e}t{\'e} canadienne du dialogue humain-machine},


Physically-based accurate soft shadows are typically computed by the evaluation of a visibility function over several point light sources which approximate an area light source. This visibility evaluation is computationally expensive for hundreds of light source samples, providing performance far from real-time. One solution to reduce the computational cost of the visibility evaluation is to adaptively reduce the number of samples required to generate accurate soft shadows. Unfortunately, adaptive area light source sampling is prone to temporal incoherence, generation of banding artifacts and is slower than uniform sampling in some scene configurations. In this paper, we aim to solve these problems by the proposition of a revectorization-based accurate soft shadow algorithm. We take advantage of the improved accuracy obtained with the shadow revectorization to generate accurate soft shadows from a few light source samples, while producing temporally coherent soft shadows at interactive frame rates. Also, we propose an algorithm which restricts the costly accurate soft shadow evaluation for penumbra fragments only. The results obtained show that our approach is, in general, faster than the uniform sampling approach and is more accurate than the real-time soft shadow algorithms.