This course of research will address the integration of the iconographic technique with other visualization
or analysis tools. This integration might be done using a dynamic approach, based on the interaction model
Pinkney has proposed, or might be done using another approach such as movable regions [39],
the drag and drop scheme used in Visage [37] or any other scheme that could emerge from this
research. Users might use the iconographic display as a starting point in the process of data exploration,
given its capacity of creating integrated visual representations of high dimensional data sets. On the other
hand, the iconographic technique could take advantage of other visualization, analysis, or data mining tools to
determine relationships between a particular pattern in the iconographic display and its interpretation in the
context of the data domain.
The iconographic technique has been used basically with imagery data (coherent data sets)
[28, 29, 25, 12]. Imagery data has a clearly defined spatial
order and all the attributes are numeric and continuous. The iconographic technique has been also applied to
visualize the results of a differential equation [18] and to visualize the FBI homicide database
[11], however those applications have shown that in dealing with non-coherent databases
several special problems must be addressed.
First, in most of the cases there is no continuous variable to be used for positioning icons in the display in
such a way that textural representations become visible. Second, many fields in those databases are
categorical data (nominal or ordinal) with a relatively small number of discrete values, which complicates
the creation of a textural representation. Third, contrary to the imagery data, in which the image provides
clues to guide the mapping process, in non-coherent data sets there is no visual clue that can be taken as
a reference to determine the appropriateness of some particular mapping, so the mapping process is more
difficult. Fourth, very often there are no definable spatial relationships among the different attributes, so
patterns visible in the iconographic representation do not necessarily relationships in the underlying data and
could be, for example, artifacts created by the particular choice for positioning icons on the display.
Conversely, in some cases a particular relation could be hidden by the particular attributes selected for
positioning icons on the display. This research intends to address issues related to the application of the
iconographic technique to non-coherent databases.
This research will develop the concept of "icon operations". Icons can be considered a generalization of
pixels (a symbol in a m 
n box; a pixel corresponds to the particular case in which m and n are equal to
one). In that sense Digital image processing operations could be extended by considering icons instead of
pixels, creating the so-called icon operations. This course of research will explore the use of these icon
operations to make the presence of visual patterns in the iconographic representation more apparent. For
example, regions in which one or more attributes are changing rapidly along one or both attributes used for
positioning the icons on the display could appear more evident if a "icon high pass operation" were used.
The icon high pass filter operation would be equivalent to a spatial high pass filter operation [6] in
which, pixels were substituted by icons, and the transformation was applied over each of the dimensions in
the underlying data. A properly selected group of operations might help in the process of visually exploring
databases using the iconographic technique.