SmoothIsoPoints: Making PDE-based Surface Extraction
from Point-based Volume Data Fast
Paul Rosenthal
1
, Vladimir Molchanov
2
and Lars Linsen
2,3
1
University of Rostock, Rostock, Germany
2
Jacobs University, Bremen, Germany
3
Westf¨ alische Wilhelms-Universit¨ at M¨ unster, M¨ unster, Germany
Keywords: Surface Extraction, Isosurfaces, Level Sets, Unstructured Point-based Volume Data.
Abstract: PDE-based methods like level-set methods are a valuable and well-established approach in visualization to
extract surfaces from volume data. We propose a novel method for the efficient computation of a signed-
distance function to a surface in point-cloud representation and embed this method into a framework for PDE-
based surface extraction from point-based volume data. This enables us to develop a fast level-set approach
for extracting smooth isosurfaces from data with highly varying point density. The level-set method operates
just locally in a narrow band around the zero-level set. It relies on the explicit representation of the zero-level
set and the fast generation of a signed-distance function to it. A level-set step is executed in the narrow band
utilizing the properties and derivatives of the signed-distance function. The zero-level set is extracted after
each level-set step using direct isosurface extraction from point-based volume data. In contrast to existing
methods for unstructured data which operate on implicit representations, our approach can use any starting
surface for the level-set approach. Since for most applications a rough estimate of the desired surface can
be obtained quickly, the overall level-set process can be shortened significantly. Additionally, we avoid the
computational overhead and numerical difficulties of PDE-based reinitialization. Still, our approach achieves
equivalent quality, flexibility, and robustness as existing methods for point-based volume data.
1 INTRODUCTION
Many modern simulation methods generate unstruc-
tured point-based volume data, i. e., scalar fields
where the data points may have an arbitrary distribu-
tion in a 3D space and do not exhibit any connectivity.
A major group of such data stems from Lagrangian
numerical simulations of natural phenomena such as
fluid dynamics. They allow for the reproduction of
complex natural phenomena by not only simulating
the evolution of data at the sample points but also sim-
ulating the flow of the sample points under respective
forces. Hence, data points move over time, change
their neighborhoods, and are distributed with a highly
varying density. Such simulations are typically car-
ried out with millions of particles.
Level-set methods have a large variety of applica-
tions. In scientific visualization, they are used to ex-
tract boundary surfaces of features from volume data.
Typically, the algorithms operate on rectilinear cells
and a given initial level-set function is modified to ex-
plicitly or implicitly minimize a given energy func-
tional.
An approach generalizing the basic idea of level
sets to work directly on unstructured point-based vol-
ume data was presented in (Rosenthal and Linsen,
2008b). They do not resample the data over a struc-
tured grid using scattered data interpolation, which
inevitably introduces resampling errors, which can
grow enormously for data sets with highly varying
point density, unless the sampling points are chosen
very densely which lets the data size explode. un-
structured points avoids this source of error.
The level-set approach models the evolution of a
surface by applying forces in normal direction. To
allow for easy change of the topology of the sur-
face, it is implicitly represented as the zero-level set
of a level-set function. The evolution is carried out
by transforming the level-set function using an itera-
tive numerical integration scheme, which implicitly
transforms the surface. The approach typically in-
duces complex calculations and the evaluation of par-
tial differential equations at each sample location and
each iteration step. When dealing with large data
sets and small time-integration steps, this can lead to
enormous computation times until convergence of the
level-set process. The calculations for unstructured
point-based data are even more complex than those in
the gridded case which was the major drawback of the
method by Rosenthal and Linsen. Later, the approach
was modified to utilize a narrow-band technique and
significantly speed up the computations (Rosenthal
Rosenthal, P., Molchanov, V. and Linsen, L.
SmoothIsoPoints: Making PDE-based Surface Extraction from Point-based Volume Data Fast.
DOI: 10.5220/0006537200170028
In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 3: IVAPP, pages 17-28
ISBN: 978-989-758-289-9
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
17