Feature Analysis in Multifields - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 186

Derived Properties such as gradient, local density, vorticity or information theoretic
complexity are computed directly from the input data, then visualized separately.
Distributions compute statistics of one or more properties of the data which are
visualized separately, while Abstract Structures show summary information about
relationships deduced from the data.
All these methods depend on implicit or explicit understanding of features, so that
feature analysis in multifields depends strongly on interactive and human-centred
methods. Moreover, it is characteristic in all methods that the result is confirmed by
visualizing any features detected directly, and usually interactively.
In addition to this spectrum of methods, it is often the case that researchers in the
application domain have an existing test for features which can be exploited directly.
Moreover, visual fusion, interactive methods, derived properties and clustering are
canvassed elsewhere in this volume, and are largely omitted, except where explicit
feature extraction is used.
As a result, it is convenient to discuss multifield feature detection and analysis in
the following categories:
Section 18.2 Scalar Features in Reduced Domains
Section 18.3 Scalar Features in the Range
Section 18.4 Manifold Features
Section 18.5 Overlapping Scalar Features
Section 18.6 Joint Feature Analysis
18.2 Scalar Features in Reduced Domains
A related approach is to choose a feature in one property of the multifield, then
restrict another property to that feature, and analyse its restriction.
Bremer, Weber et al. [ 2 ] apply this to combustion simulations. One property
(temperature) is restricted to define an isosurface. Features are then identified and
tracked over time for a second property (combustion rate) restricted to the underlying
isosurface. Subsequently, Bremer, Weber et al. [ 3 ] analysed the topology of one
property (fuel consumption), but annotated the features so discovered with values
derived from the other properties.
Ropinski et al. [ 34 ] extract aortic arches from mouse PET/CT scans, then use
image-processing techniques to recognize standard features and register the scans.
Once this is done, however, secondary properties such as vessel diameter are mapped
onto the detected geometry for human visualization.
To date, however, there has been relatively little work performed under this head-
ing. Instead, many researchers have concentrated on detecting features in the range
of the multifield, and we consider these approaches next.