Bandwidth selection and reconstruction quality in point-based surfaces

Hao Wang, Carlos E. Scheidegger, Cludio T. Silva

Research output: Contribution to journalArticle

14 Scopus citations

Abstract

We investigate the influence of bandwidth selection in the reconstruction quality of point-based surfaces. While the problem has received relatively little attention in the literature, we show that appropriate selection plays a significant role in the quality of reconstructed surfaces. We show how to compute optimal bandwidths for one class of moving least-squares surfaces by formulating the polynomial fitting step as a kernel regression problem for both noiseless and noisy data. In the context of Levin's projection, we also discuss the implications of the two-step projection for bandwidth selection. We show experimental comparisons of our method, which outperforms heuristically chosen functions and weights previously proposed. We also show the influence of bandwidth on the reconstruction quality of different formulations of point-based surfaces. We provide, to the best of our knowledge, the first quantitative comparisons between different MLS surface formulations and their optimal bandwidths. Using these experiments, we investigate the choice of effective bandwidths for these alternative formulations. We conclude with a discussion of how to effectively compare the different MLS formulations in the literature.

Original languageEnglish (US)
Article number4745637
Pages (from-to)572-582
Number of pages11
JournalIEEE Transactions on Visualization and Computer Graphics
Volume15
Issue number4
DOIs
StatePublished - Jul 1 2009
Externally publishedYes

Keywords

  • Bandwidth selection
  • MLS surfaces
  • Surface reconstruction

ASJC Scopus subject areas

  • Software
  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Computer Graphics and Computer-Aided Design

Fingerprint Dive into the research topics of 'Bandwidth selection and reconstruction quality in point-based surfaces'. Together they form a unique fingerprint.

  • Cite this