Graphics Reference
In-Depth Information
students. he results showed a significantly greater success rate with factorial suns as
compared to snowflakes and star glyphs.
Several evaluations of Chernoff faces have been reportedsince their introduction.
One interesting set of experiments was reported in Morris et al. (
),who focused
on studying the effectiveness and preattentiveness of different facial features. Sub-
jects were shown images containing varying numbers of faces, and they were asked
to determine if a face with a designated feature existed. he amount of time required
to complete each task was measured and analyzed. Not surprisingly, the amount of
time needed was proportional to the number of glyphs on the screen. However, the
authors determined that it is not likely that preattentive processing was involved, as
tests done with short duration, even with small numbers of glyphs, yielded poor re-
sults. heir conclusion was that, because glyph analysis with Chernoff faces was be-
ing done sequentially, they are unlikely to provide any advantage over other types of
multivariate glyphs.
Astudythat placedChernofffacesahead ofseveralotherglyphtypeswasreported
byWilkinson(
).Inthisstudy,subjectswereaskedtosortsetsofglyphsfrommost
similar to least similar. he glyphs used were Chernoff faces (Chernoff,
), Blobs
(Andrews,
), castles (Kleiner and Hartigan,
), and stars (Siegel et al.,
).
he results were that faces produced results with the best goodness of fit to the real
distances, followed by stars, castles, and blobs. he author felt that the memorability
of the faces helped users to better perform this type of task.
Summary
7.9
Glyphs are a popular, but insu
ciently studied, class of techniques for the visualiza-
tion of data. In this paper, we've discussed the process and issues of glyph formation
and layout, including the identification of problems of bias due to perceptual limi-
tations and dimension ordering. We also presented techniques for evaluating the ef-
fectiveness of glyphs as a visualization method and some results obtained from eval-
uation.
Many avenues exist for future development and application of glyphs for data and
information visualization. here is a continuing need for glyph designs that mini-
mize bias while maximizing the accuracy of communicating data values. While the
majority of recent designs have been tailored to particular domains and tasks, we be-
lieve there is still room for work on general-purpose glyph designs. Scalability is also
a big issue, as most glyph methods in use are limited either by the number of data
recordsordatadimensionsthatcanbeeasilyaccommodated.Giventhegrowthinthe
sizeanddimensionalityofcommondatasets,novelmechanismsareneededtoenable
userstoexplorelargerandlargeramounts ofdata. Workonaggregation glyphs(Yang
etal.,
)orothermultiresolution strategies maybethekeytotheproblemofscale.
Finally, rigorous evaluation is essential to help identify the strengths and weaknesses
of each proposed glyph in terms of user perception, analysis tasks, and data charac-
teristics. Most efforts at evaluation to date have been ad hoc or very limited in scope.
