Image Processing Reference
In-Depth Information
Chapter 16
Visual Exploration of Multivariate
Volume Data Based on Clustering
Lars Linsen
Abstract The attribute space of a multivariate volume data set is hard to handle
interactively in the context of volume visualization when more than three attributes
are involved. Automatic or semi-automatic approaches such as involving clustering
help to reduce the complexity of the problem. Clustering methods segment the
attribute space, and the segmentation can be exploited for visual exploration of the
volume data. We discuss user-guided and automatic clustering approaches of the
multi-dimensional attribute space and visual representations of the results. Coor-
dinated views of object-space volume visualization with attribute-space clustering
results can be applied for interactive visual exploration of the multivariate volume
data and even for interactive modification of the clustering results. Respective meth-
ods are presented and discussed and future directions are outlined.
16.1 Introduction
Volume visualizations rely on some segmentation of the given volumetric domain.
Typical examples are the choice of an isovalue for isosurface extraction from a scalar
field or the application of a one-dimensional transfer function to the range of a scalar
field for direct volume rendering. Such a segmentation of the volumetric domain is
implicitly given by segmenting the range of the scalar field. Some early approaches
extended this idea to the segmentation of 2D or even 3D spaces formed by the range
and some derived properties such as magnitude of first- and second-order derivatives.
The segmentation in these spaces are performed interactively by providing respective
interactionmethods andwidgets. This is possible as long as the interaction takes place
in 2D or, with some limitations, in 3D visual spaces. These visual spaces are often
obtained by histogram computations and their visualization by using color coding.
Interaction mechanisms, then, allow the user to select regions of interest and assign
parameters for the volume visualization methods to them. For example, the idea of
using multi-dimensional transfer functions to direct volume rendering goes back to
( B )
Jacobs University, Bremen, Germany
e-mail: l.linsen@jacobs-university.de
L. Linsen
 
Search WWH ::




Custom Search