Image Processing Reference
In-Depth Information
1. From the domain to the range perspective, we select a subset of data items in a
physical view and examine the selection in range views. This type of analysis
serves to
localize the investigation
to a region of interest such as an inlet or outlet.
2. From the range to the domain perspective, we select a subset of data items in a
range view, such as in a scatter plot, and examine the result in a physical view.
This type of analysis enables
localizing features
. In this case brushing defines a
feature, usually as a set of thresholds, and highlighting in a spatiotemporal view
shows whether the selection corresponds to a localized feature.
3. Within the range perspective, we select subset of data items in a range view and
observe the selection in another range view. This type of analysis provides a means
of performing an interactive
multivariate analysis
, e.g., by brushing in one scatter
plot and examining the selection in another scatter plot of different variables. This
pattern was originally introduced in the field of information visualization [
1
,
31
].
Using one or more of these patterns is the simplest form of IVA, more recently
referred to as “Show & Brush.” It utilizes multiple views, usually at least one range
view for visually correlating multiple fields and one domain view to show properties
in a physical domain context. Though being the simplest form of IVA, this method
already covers a large percentage of use cases in multi-field analysis and serves as
powerful basis for more advanced types of exploration and analysis. This type of IVA
has proven valuable in many application areas, including aeronautical design [
12
],
climate research [
15
,
20
], biomedical visualization [
8
,
25
], the analysis of gene
expression data [
32
], the analysis of combustion engines [
7
,
22
], and the analysis of
simulations of particle accelerators [
27
].
15.2 Additional Concepts
Based on the simple “Show & Brush” paradigm, a few extensions can greatly enhance
the expressiveness of IVA. First, in many cases it is useful to define brushes not as
binary classifiers into two categories “of interest” and “not of interest” but as a means
to map each data item to a
degree of interest
[
6
]. It is possible to define this degree
by specifying two selections (e.g., regions in a scatter plot). All items inside an
inner
range
have a degree of interest of 100 % (i.e., are definitely of interest), and all items
outside an
outer range
have a degree of interest of 0 % (i.e., not of interest). Between
those regions, a transfer function maps the distance of a sample from inner and outer
range to a degree of interest between 100 and 0 %. A linear ramp is a common choice
for this transfer function. More generally, we can utilize fuzzy logic operators to
combine multiple smooth brushes.
This smooth drop-off of a degree of interest makes it possible to transition
seamlessly between data items of interest and those of not interest and use gen-
eralized focus plus context (F+C) methods [
5
,
9
,
26
] to reduce cluttering in resulting
visualizations and draw a user's attention to the most important details. Traditionally,
focus and context methods use space distortion such as a fish-eye lens to assign more
Search WWH ::
Custom Search