Several researchers have pointed out key requirements for the next generation of visualization systems.
Foley and Ribarsky [14] have suggested a general visualization environment in which all the
modules are in a feedback loop that processes through the user's brain to refine the visualization
iteratively or to focus on certain aspects of the data. Flexibility and integration with data analysis are two
key components in the model of Foley and Ribarsky. Another aspect that has received special attention
refers to the use of effective and efficient graphical representations capable of displaying high dimensional
data. Glyphmaker [33] is a visualization system in the direction suggested by the model of Foley
and Ribarsky.
Lee and Grinstein [24] has pointed out the need of integrating database and visualization systems in
order to provide more flexible and powerful data selection support. In this integration, the visual nature of
the database exploration is emphasized. Users are also part of the feedback loop in
the iterative process of data exploration. Users perceive visual data representations and use them to guide
the exploration process. Exbase [23] is a system for exploring databases developed on the basis
of this integration. It provides support for the visual exploration of databases by integrating visualization
techniques and a real database management system. The search for structure inside the database is
focused on manual mechanisms but it does not exclude support for automatic discovery tools.
Goldstein, Roth and Mattis have proposed a framework for interactive data exploration [16].
There are several key points in that framework. First, they point out that, given the growth of data sets in
both complexity and amount of information, the design of graphic displays is requiring more expertise from
users. Thus, systems for assuming the responsibility of automatically designing graphic displays are a
fundamental requirement for the next generation of visualization systems. Second, visualization systems
must provide explicit support for the data archaeology process. The data archaeology process refers to the
use of users' discovered relationships to guide the data exploration process. This process, iterative and
interactive in nature, is initiated and controlled by people. Third, tools in the visualization system must be
oriented to provide support for the kind of subtasks usually involved exploring large databases, including
explicit support for a dynamic specification of the user's information-seeking goals. Visage [37], a
user interface environment for exploring information, is based on these principles and on a new paradigm
for user interfaces that they have called an information-centric approach.
Robertson and De Ferrari [35] have proposed a reference model for data visualization within
which the scope and limitations of existing tools and systems can be identified. In this model, the user is
again part of the exploration data analysis loop. The end goal is the automatic generation of visual
representations given a description of all the important data characteristics and the user's interpretation
aims. User interpretation aims refers to the specification of characteristics of the data or relations
between data variables the user is interested in analyzing by mean of visual representations.
There are four basic components in the Robertson and De Ferrari model: 1) the data model; 2) visualization
specification; 3) visualization representation; and 4) the matching procedure. The model also specifies
desired requirements for all of these components. The data model should be precise and functionally-rich
with the data structure and the data fields explicitly specified. The visualization specification should
support both user directives (requirements explicitly defined by the user) and user interpretation aims
(requirements implicitly defined by user specified criteria). In reference to the visual representation,
Robertson and De Ferrari suggest that many possible visual representations should be described based on
some criteria. There are currently two approaches for describing visual representations. The bottom-up or
expressiveness/effectiveness approach defined by Mackinlay [26] and extended by other
researchers [38], and the top-down approach defined by Robertson [34] which is
based on scene properties and criteria for matching scene properties to data variables. The matching
procedure refers to mechanisms for encoding and decoding information. Robertson and De Ferrari suggest
that for each visualization technique those mechanisms must be explicitly stated by giving a formal model
which describes the encoding mechanism and explicitly states what decoding capabilities it assumes.
The iconographic technique intends to exploit the texture perception mechanism of the human visual
system. Experiments developed by Julesz [22], demonstrated that differences in textons
(characteristic features of the surface) or their densities can be preattentively detected by the human
visual system. The work developed by Julesz shows that texture is a statistical property of textons. The
perception and discrimination of textures seems to be based on the density (first order statistics) of
textons.
The Exvis (Exploratory Visualization) visualization system developed at the University of Massachusetts
at Lowell [17] uses icons as basic primitives. These icons have some of the elements that were
identified by Julesz as parameters associated with the characterization of textons. The visual attributes of
icons are driven by the data. With sufficient density, icons form a surface texture display in which
structures in the data are revealed as streaks gradients, or islands of contrasting textures
[12]. Exvis offers several icons to users: the family of "stick figure icons"
[28], the "velcro icon" [29, 12], and the "color icon"
[25, 11]. Each of these icons has several visual attributes. By representing each record
or row of a multidimensional database as an icon whose features (e.g. color, geometry, position) are under
the control of the various fields of the record (dimensions), and by placing those icons densely enough in
the display, a textural representation of the database is obtained. Exvis is a static visualization system in
which users' interactions are basically limited to examining the actual values associated with the icons (in
a one by one fashion). Users can also change the association between fields in the record and icon visual
attributes, but these changes are implemented in a batch fashion.
Most of the work developed with the iconographic technique has been based on the stick figure icon. The
stick figure icon consists of five connected segments, called "limbs". One of the segments, called the
"body", serves as a reference for the various geometric transformations that an icon can undergo. Each
limb has three parameters that can be bound to data attributes: the angle, intensity, and length. The manner
in which limbs are attached to the body or to each other defines the family of stick figure icons. There are
12 possible stick figure icons in Exvis. The assumption is that each of these members has the potential of
resonating to a different type of data structure [28]. Figure 1, shows one member of the
family of stick figure icons. The figure shows the icon in its base configuration (theoretical icon, no data
mapped to limb's parameters) and also a sample icon (data mapped to limbs' parameters). In this
particular icon all the four limbs are attached directly to the body.
Figure 1: One member of the stick-figure icon family
The L-systems-based visualization prototype that Pinkney has developed [30] is a descendant of
the Exvis system. It uses the stick figure and color icons, but these icons are formally represented using
Open L-systems grammar notation [27]. The grammar consists of an axiom, the initial string, and a
variable number of productions (rewriting rules). Each icon is modeled by setting an initial string and a set
of rules defining the way in which the initial string will evolve along the grammar derivation process.
Pinkney uses the bi-directional query mechanism provided by Open L-systems in order to substitute formal
parameters in the grammar productions with actual parameters occurring in words in the derived string.
The L-systems grammar representation is also used to represent interaction techniques. One of the
advantages of open L-systems is that grammars are able to invoke external modules. Pinkney uses this
facility to alter either the derived string or the course of derivation itself in order to implement different
behaviors of icons. Figure 2, taken from [30], shows an overview of the L-systems-based
visualization system developed by Pinkney. As seen in this figure, interactions are implemented in one or
two places. At the level of the grammar, by including productions defining the desired behavior; or at the
level of the string, by changing values of the derived string.
Figure 2: An overview of the L-systems-based iconographic visualization system [30]
The current implementation of the system provides several interactions: 1) the "magnet" interaction; 2)
the "vibrating" interaction; 3) the "comb" interaction; 4) the "zoom" interaction; and 5) the "pinwheel"
interaction. In all of these interactions the mouse is used for specifying a region of interest. Interaction
acts over all icons within the selected region. Users can select either a constant value or a variable (data
attribute) to drive the interaction. Icons within the selected region are transformed in some way (depending
on which interaction) that is proportional to the value driving the interaction. For example in the "magnet"
interaction the mouse acts as a magnet repelling or attracting icons. Users can select if the magnet will
effect the angle of the icons, the position of the icons, or both. The value associated with the interaction is
used to determine the angular and positional response of the icon to the magnet, in the case of magnet
interaction, lower values imply greater effect.
Glyphmaker is an exploratory tool developed at The Georgia Institute of Technology. [33, 14]. It provides a general approach to data visualization and analysis based on the use of glyphs as
graphical elements for representing data. Glyphmaker uses glyphs in order to exploit the ability of the
human eye-brain system to discern finely resolved spatial relationships and differences in shape. The idea
is to bind visual attributes of the glyphs, such as position, size, shape, color, orientation, etc. to data
attributes in order to get a glyph-based representation of the data.
Glyphmaker intends to provide a flexible environment in which even non-expert users can create
customized visualizations using their own graphical representations (glyphs). Three main components
provide the functionality required for such purposes: the read module, the glyph editor module, and the
glyph binder module.
The read module is able to read a variety of data input formats and convert them to the format required by
other modules. Glyphmaker uses a "self-describing" data structure in order to support different input
formats. The "self-describing" structure has two components: a header, which describes the data and its
format, and the data itself. This simple structure allows users to specify their own format in a flexible
manner. For example, headers and data can be placed anywhere in the file. More than one header and its
corresponding data can be placed in the same file with each header followed by the data it describes.
Headers can also be placed together at the top of the file with all the data following the last header. Both
headers and data can be written in plain ASCII, but they can also be specified in a straight binary format.
The advantage of the approach taken in glyphmaker to deal with the problem of specifying an input data
format is that it is easy both to create and understand. Its drawback is that only simple data formats can
be specified.
Users of glyphmaker can use a variety of graphic elements to represent data. Users can select graphic
elements from a predefined set of graphic elements or they can develop their own graphic elements by
using the glyph editor module. The glyph editor module provides a library of basic glyphs ( points, lines
spheres, cuboids, cylinders, cones, and arrows) that can be used in combination to develop yet another
variations. The glyph editor provides a 3D graphical environment in which users not only can develop new
glyphs but also see how they will look in the finished visualization. Tools to manipulate these glyphs in
three dimensional space (selecting different views, including perspective, top, side, front, and all four
views simultaneously), and commands for grouping, ungrouping , saving, and loading glyphs are also
available.
Visual attributes of the glyph are associated with data attributes. The binder module is an interface that
allows dynamic and interactive mapping between graphic element attributes and data attributes. The binder
module is more that a simple interface to allow dynamic binding. It also provides the mechanism for
restricting the amount of data to be visualized. Each attribute has a minimum and maximum value that is
initially set to the values specified in the header, which basically means selecting the whole set of records.
Users can change these values (minimum and maximum) for a particular attribute in order to select a
subset of records according to that attribute. Several attributes can be used to define the subset of values
to be visualized. The subset is determined by the conjunction of the ranges of all the active attributes.
Glyphmaker also includes what they call a "conditional box". Using the conditional box, users are able to
focus on specific regions of the data being visualized. The current implementation of the conditional box is
limited to provide an additional way to reclassify the data according as to whether or not it is inside or
outside the conditional box. The extension of this concept to other relations is one of the things that has
been considered for future versions of glyphmaker. For example, the classification could be based on any
property associated with a variable such as range of temperature, energy, force components, etc. More
complex conditions based on algebraic expressions relating two or more variables could also be used.
IVEE (Information Visualization and Exploration Environment) [5], is a system for automatic
creation of dynamic query applications [4]. IVEE intends to demonstrate how a visualization
system can be designed to automate the task of creating both visualizations and manipulation
objects(sliders), allowing users to concentrate directly on the exploration task. IVEE requires that users
specify relations (e.g. job opportunities, home prices, crime index, etc.) and their named attributes (e.g.
year, city, state, population, etc.) in a straightforward text format. Relations might originate from a
database system or spreadsheet program. IVEE uses this specification to classify the data into "data
types" and "size". IVEE uses a very simple characterization in which only two data types are considered:
integers and strings. Size has two components: the first one refers to the total number of values for the
attribute (sample size) and the second one refers to the number of distinct values for the attribute. IVEE
uses this classification and a couple of simple rules to decide which query device will be assigned to
control which attribute. Users can modify an automatically created query device either by changing the
device itself or by changing the layout.
The user interface of IVEE has two main components, a query area holding a number of query devices, and
a visualization area holding a number of visualizations. An important feature of IVEE is that it can hold
several instances of visualizations, databases (relation specifications), and query areas.
The approach followed in IVEE for visualizations is flexible in regard to both the visualization technique
and the graphic elements used to represent objects in a database relation. It provides a rich collection of
visualizations and graphic elements that users can choose from. Defaults are provided for both
visualization and graphic elements in order to make the system easy to get started and so users are not
forced to specify anything about the visualization or graphic elements to get the system running. A unique
graphical element can be specified to represent each data object in a relation. Users can also specify a
graphical element for each object in the database. It is also possible to specify a small number of objects
for representing data objects in the database. In this last case, objects and their corresponding mapping to
database elements must be specified. By default, a square is used to represent each database object, but
users can also specify a single point, a glyph, or a user-defined polygon in two or three dimensions. The
use of glyphs is limited to those from a predefined library that IVEE provides. Graphical objects can be
coded by color, shape, size, brightness, etc.
The basic visualization (default) in IVEE is the starfield [2]. A starfield is similar to a scatterplot
but it has additional features for panning, zooming, details on demand, rotations, etc.. IVEE has extended
the starfield visualization by letting users specify background objects that add context to the starfield
visualization. For example geographic visualizations can be created by specifying a map as a background.
In addition to maps, other more structured objects can be specified to define an appropriate background
(e.g. the structural drawing of a house). Other visualizations are also available: tree-maps, cone-trees,
and visualizations using complex three dimensional objects.
The query device area can hold three different query devices: 1) rangesliders, used for selecting range
criteria for integer attributes or other attributes with an order relation; 2) Alphasliders [1], used
for selecting items (usually non-numerical or categorical attributes) from a long list of strings; and 3)
toggles, used when only a few alternatives exist for an attribute. Query components only support a
conjunctive combination . The combined effect of all the query devices (conjunction) affects the whole
visualization. An interesting approach used for implementing the query device mechanism in IVEE is that
query devices are tightly coupled [3], which means that the manipulation of a query device not
only affects the whole visualization but also the other query devices. The tightly coupled mechanism
forces all the query devices to include in their respective ranges only criteria existing within the remaining
elements of the database.
Visage [37] is a system for exploring information whose user interface is based on a new
paradigm that authors have called an "information-centric paradigm". Information objects, represented as
first class interface objects, can be manipulated using a common set of basic operations universally
applicable across the environment. Operations such as drill-down and roll-up, drag-and-drop moves,
copy, and dynamic scaling can be applied to data objects whether those objects appear in a hierarchical
table, a map, a slide show, a query, another application user interface, or even the desktop environment.
Visage is built on top of SAGE [36] which is a knowledge-based system for automating the design
of visualizations. Visage renders graphic designs created by SAGE, and each element in these designs is
subject to the direct manipulation operations mentioned before. Visage also includes several exploration
operations that are related to querying. For example, it is possible to aggregate database objects into a new
object which will have properties derived from its elements. These operations are specified procedurally,
using a sequence of direct manipulation operations.
Recently Visage has been extended with a visual query language, VQE [9] which allows users to
express more complicated queries than those possible with the procedural approach. The extension of
Visage with VQE allows user to create visualizations from queries and queries from visualizations. The
integration of VQE into Visage enable the representation of queries in an explicit form. Users can
construct complex queries and reuse them later, a capability not possible with the procedural approach.
Queries can also express relationships among object sets, support navigation among objects, and denote
aggregation. A dynamic query interface is used for range selection, and all of these components are fully
integrated into Visage and, hence, can be subject to all of the basic operations available in Visage. For
example, query results can be dragged to any visualization or query device. Visualizations and query
devices in Visage are linked, so changes in any of these components are immediately reflected in each
other.
Exbase [23] is an integrated database and visualization system that enables the visual exploration
of databases. Exbase uses a data model based on the view concept [24]. Database and visualization
can both be queried. Querying a database refers to the selection of a particular set of data objects from the
database, which constitutes the so-called local database view. Querying a visualization refers to making a
subset of the local database view. This constitutes the so-called visualization view. Exbase offers data
selection capabilities from both a textual-query interface, based on SQL, and through graphical tools, at
the visualization level.
Exbase integrates a real database management system ( the current version is based on an extended
version of Orion) with several visualization techniques. Exbase uses an object oriented approach to
provide the support required to integrate other visualization techniques into it. Each visualization technique
is modeled as a class in an abstraction hierarchy. That hierarchy includes a Visualization abstract base
class, which contains a graphical user interface code common to all concrete subclasses. The current
version of Exbase also includes an abstract base class for visualizations based on the Cartesian
coordinate system. This class ( CartesianVis class ) provides support for spatial data selection at the
level of visualization. In the particular case of techniques using a Cartesian coordinate system, concrete
subclasses for each visualization technique need only specify their display controls, mapping functions,
and rendering functions. Other techniques, such as parallel coordinates, which do not use Cartesian
coordinate axes must provide their own spatial selection functions.
One interesting aspect in Exbase is that it can monitor all user interactions with the database and
visualization. By explicitly representing user interactions with databases and visualizations, Exbase is
able to maintain a history of such interactions in order to reuse them immediately or to recall them in future
analysis.