Low-Discrepancy Blue Noise Sampling

ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2016)

Abdalla G. M. Ahmed1             Hélène Perrier2             David Coeurjolly2            Victor Ostromoukhov2           

 Jianwei Guo3            Dongming Yan3             Hui Huang4,5             Oliver Deussen1,5

1University of Konstanz         2Université de Lyon, CNRS/LIRIS          3Institute of Automation            4Shenzhen University               5SIAT

Starting from a template low-discrepancy (LD) point set (a), we use a segmented table of permutations to rearrange the LD set to match a reference set with the desired target spectrum (b). The permutations are localized and carefully constructed in such a way that they have minimal impact on the discrepancy of the underlying template set. The resulting set (c) inherits the spectral profile of the target set, while still retaining the discrepancy profile of the template set (d).

Abstract
We present a novel technique that produces two-dimensional low-discrepancy (LD) blue noise point sets for sampling. Using one-dimensional binary van der Corput sequences, we construct two-dimensional LD point sets, and rearrange them to match a target spectral profile without loosing their low discrepancy. We store the rearrangement information in a compact lookup table that can be used to produce arbitrarily large point sets. To the best of our knowledge, our construction is the first one that combines blue-noise and low-discrepancy properties at the same time. We evaluate our technique and compare it to the state-of-the-art sampling approaches.

Reference
[To reference our ALGORITHM, API, CODE or DATA in any publication, please include the bibtex below and a link to this webpage.]

Interactive Demo


You can draw samples right from this page, scaled to unit area. Drag to move around, use the mouse wheel or pinch to zoom. What You See Is What You Get!

2757 points
You need to upgrade your browser to view this.
Show Strata

Results
This html bundle contains all the data that was collected while performing tests and comparisons between the LDBN sampler and various other samplers from the state of the art. You can click on any sampler to see all its associated data, discrepancy results, integration variance results, point distribution, behaviour relatively to aliasing (zoneplate tests), and spectral behaviour.

Video

 

Acknowledgment 
We thank the anonymous reviewers for their detailed feedback to improve the paper. Thanks to Jean-Yves Franceschi and Jonathan Dupuy for reviewing an earlier version of the paper. This project was supported in part by Deutsche Forschungsgemeinschaft Grant (DE-620/22-1), French ANR Excellence Chair (ANR-10-CEXC-002-01) and CoMeDiC (ANR-15-CE40-0006), 973 Program (2015CB352501), National Foreign 1000 Plan (WQ201344000169), National Natural Science Foundation of China (61372168, 61620106003, 61331018), GD Leading Talents Plan (00201509), GD Science and Technology Program (2014B050502009, 2014TX01X033, 2015A030312015, 2016A050503036), and SZ Innovation Program (JCYJ20151015151249564).

Downloads

Source Code
ZIP (37 kB)

LDBN
PDF(12.2M)

Copyright © 2016-2018 Visual Computing Research Center