1Photorealistic rendering of graphical models contin-
ues to be a topic of considerable research effort in
the computer graphics community. However, photo-
realism is not the only criterion for judging the value
of an image. In visualization, where rendering is the
conduit through which abstract, non-realistic forms
(e.g. isosurfaces or streamtubes) are made perceiva-
ble to users, “photorealism” may be neither the most
useful, nor indeed appropriate, property. This article
describes the authors’ long term research vision in
an area which, in their view, should gain a lot of im-
portance in the future. The direct inspiration for this
line of work came when one of the authors (IH) vis-
ited an exhibition on Japanese prints in Amsterdam
but, indirectly, the visit the authors made some years
ago to the old imperial city of Kyoto had a great in-
fluence, too; the reader will soon understand why.
Consider the paintings in Figure 1 and Figure 2.
The first is a reproduction of a painting of Johannes
Vermeer, one of the outstanding Dutch painters of
the 17th century. His “The Little Street” is a typical
example of Dutch, but also of classical European
painting. The second is a detail taken from a land-
scape painting by the Chinese artist Hsia Kuei, dated
to the 12th century. The artistic techniques used in
the paintings are quite different, as will be discussed
shortly. If we consider these paintings as forms of
communication, then a natural concern is how the
different styles of representation affect that commu-
nication. A full account of these issues lies in the ar-
eas of art theory and visual perception. Here, we
simply use the example as an analogy, to observe
that computer graphics, which today plays a major
role in visualization and visual communication with-
in the information society, has concentrated much on
one particular approach to rendering information, so
called photorealistic rendering.
Vermeer tried to represent reality on the canvas,
with all the intricate effects of lights, of shadows, of
reflections, etc. Such minute details as the texture of
the brick walls or the garment of people are also rep-
resented with great care, although they are hardly
noticeable to the naked eye. This attempt for realism
has been one of the main characteristics of European
painting up to the beginning of the 20th century. This
approach to art is in sharp contrast with the art of
China and Japan. The contrast between the two
paintings is striking. Clearly, Hsia Kuei, as most tra-
ditional Chinese and Japanese painters, did not try to
reproduce nature. The picture conveys an impres-
sion of the landscape; only parts of the contours, of
the main lines of objects (of the hills, the trees, etc.)
are represented. The whole of the picture is remark-
ably void of details. Nevertheless, the “message”,
the “information content” is there, and the undenia-
ble aesthetic beauty of this painting is just as appeal-
ing as the one of Vermeer’s*.
Some European artists, like Dürer or Leonardo
da Vinci, and indeed Vermeer himself, too, conduct-
ed life experiments to understand the propagation of
light, human vision, the nature of shadow, etc. In do-
ing so, they became precursors of an early form of
experimental mathematics; for example, modern
projective geometry, the rules of perspective map-
pings, grew out of these experiments. This tradition
Figure 1. Johannes Vermeer, “The Little Street”
* Note that, in this text, the example of Far Eastern paint-
ing is used as a contrast; one could also have referred to
various schools of modern European art, or to cartoon and
caricature drawings. Beyond issues of personal taste, a
reason why Chinese/Japanese paintings might be an inter-
esting point of departure are the traditions and philosoph-
ical background underlying the art, which may help in
developing new methods.
Minimal Graphics
Ivan Herman
CWI
Amsterdam
David Duke
University of Bath

2of developing models of representation based on
physical reality has continued into the twentieth cen-
tury and became the foundation of traditional ren-
dering in computer graphics, too. However, in
looking at other artistic traditions, we might be en-
couraged to ask deeper questions, about how infor-
mation is represented in an image. Apart from
enriching our understanding of how we select and
represent representations of visualization data, these
questions prompt a re–evaluation of the principles of
rendering with application to graphics in general.
Motivations
Why should artistic traditions be of interest to infor-
mation scientists? Traditional computer graphics, as
it developed in the past 15 to 20 years, may be con-
sidered as a direct continuation of traditional Euro-
pean painting, at least up until the end of the 19th
century: the goal is to reproduce nature through im-
ages generated by computer graphics. The ideal is
“photorealism”, or its generalisation into concepts
of “virtual reality”, “virtual humans”, etc. It is not
the goal of this paper to criticise these lines of re-
search, which are stimulating, exciting, and full of
challenging research problems. However, one
should not forget an essential issue. A significant
goal of computer graphics is to help the human ob-
server to understand information through pictorial
means. In some cases (e.g., a virtual walk–through
of a building) photorealism has a clear role, but par-
ticularly in visualization one should realise that this
is not necessarily the case. The example of Chinese/
Japanese painting shows that information about
one’s environment can also be conveyed without a
strive for photorealism, judiciously choosing instead
a level of graphical information which is enough to
communicate the intended message. In addition, this
can be done without losing the expressiveness and
the aesthetic beauty of the image.
Underlying Chinese and Japanese art is an aes-
thetic of visual simplicity. Our objective is to find a
new approach to rendering that incorporates this
aesthetic. For a lack of a better name, the term “min-
imal graphics” has been chosen for the following re-
search goal: based on some model of information
(which may be either a traditional geometric model
of a full scene or something different) one should
produce images which strive for a minimum level of
complexity for a task, i.e. which should be as simple
as possible, but which should convey the intended
information to a human observer. Furthermore, (al-
though this is even more difficult to describe in al-
gorithmic terms) the generated images should be
“pleasing” to the human eye.
Although research could be motivated by a sheer
intellectual challenge or aesthetic requirements,
computer graphics has always been driven by prac-
tical needs, too. So is minimal graphics of any prac-
tical interest? Here are three reasons:
1) Schuman et al.[4] have described how archi-
tects, when talking to their clients in the early
phase of design, prefer to use sketches rather
than photorealistic images. Sketches seem to
have an affective quality that encourages inter-
action, as they convey a sense of only partial
commitment to a design. Would minimal
images be better at encouraging dialogue and
interaction than photorealistic alternatives?
2) Application development for mobile computing
via devices such as mobile phones and PDA–s is
constrained by the limited displays of these
devices. If images are to be used to convey
information within applications, the rendering
processes will need to take the display limita-
tions into account. We suggest that the research
goal of minimal graphics is essentially what is
needed for more effective use of these devices.
3) New kinds of input and output device are
becoming available to support human-computer
interaction. Haptic devices are one example of
Figure 2. Hsia Kuei: A Pure and Remote View of Rivers and Mountains.

3this. Although these are still expensive and lim-
ited, they are becoming more widely accessible.
Haptic rendering has significant timing con-
straints; variants of minimal graphics might be
more adaptable to this technology than algo-
rithms derived from photorealistic approaches.
Non-photorealism and Minimal Graphics
Artistic considerations have already influenced the
development of new rendering methods in computer
graphics. The majority of this work has concentrated
on the reproduction of various artistic techniques and
tools, such as pen and ink, pencil, brush, and paint ef-
fects. Non–photorealistic (NPR) techniques have also
been developed that utilise principles of human vi-
sion. However, the emphasis has been on low–level
aspects of cognition, e.g., the use of textons, shading
etc. The aim of minimal graphics is to use models of
human information processing that encompass not
just low–level vision, but also structural recognition,
interpretation, and even affective properties of imag-
es to produce a general purpose foundation for ren-
dering graphical information.
Artistic considerations have already made signifi-
cant contributions to visualization. There is no place
to give a detailed overview here, let us just refer to the
paper of Laidlaw in the March/April issue of IEEE
CG&A[2] as one example of the potential interaction.
Fundamentals of Minimal Graphics
The extraction of a minimalist information from a
model seems to be, at first glance, a geometric task.
One would try to extract and use, for example, geo-
desic or other characteristic curves, use some sort of
silhouette detection algorithm, or special forms of
dithering (see a more complete report on the topics[1]
for further examples). However, these approaches do
not “simplify” the image, and the results lack the
“symbolic”, abstract nature that minimal graphics is
seeking to achieve. Other techniques could comple-
ment these approaches, for example the use of some
sort of smooth (not necessary convex) hull of the 3D
objects, which could be used for the final image.
Wavelet–like encoding could be envisaged; multires-
olution methods in modelling objects might help in
extracting the “sweep” of a curve or a surface.
However, adapting existing techniques from
graphics is not sufficient. When trying to formulate
the issues raised by minimal graphics, one soon real-
ises that “abstract” and “minimal” are not concepts
that cannot be described in purely algorithmic terms.
Rather, minimal means that the image is just rich
enough for the human mind to complete the represen-
tation through cognitive processing. This might seem
as if we were posing additional demands of comple-
tion upon the human, so incurring a usability deficit.
However, perception is not just a matter of comple-
tion. In practice, a great deal of irrelevant informa-
tion, or noise, must be disposed of before completion
of what remains, the signal, can take place. A number
of optical illusions have been described which exem-
plify how human cognition is capable of “complet-
ing” an image.* Figure 3, for example, shows the so–
called Kanisza triangles: the three wedges in the
black circles create an illusionary white triangle. To
take another example, consider the image of the
Duomo of Milan (Figure 4). The façade of the
building has a very complicated edge, consisting of a
complex pattern of stone carving. Nevertheless, the
human mind clearly perceives a triangular façade, by
“smoothing” the edges in the image. Looking at a
cloud in the sky, the contours of a fractal image: these
are all examples of the same effect. Generalising from
these examples, it seems that the human cognitive
process is somehow able to fill in some “emptiness”
(the “triangle” in the middle of the Kanisza figure, the
empty space at the edge of the Duomo). This duality
between “empty” vs. “full” seems to play an essential
role in the way humans perceive their environment.
This is why sketch images can show the Duomo of
Milan as a simple triangle with some additional orna-
ments: minimal graphics should be able to generate
similar sketchy images automatically.
Although there is no simple, complete model that
accounts for all aspects of visual illusion in terms of
cognitive processes, there are theories that explain
significant aspects of the problem at particular levels
of operational detail, for example from neurological
* The book of Ninio[3] is one of the best collections of
such illusions we know about. Although in French, it is
quite enjoyable for non French readers, too
Figure 3. The Kanisza triangle
Figure 4. The Duomo of Milan

4properties of the pre–cognitive phase, through to
cognitive effects grounded in the interplay between
top-down and bottom-up processing. The image of
the cathedral may be reduced to a triangular outline
by frequency filtering, and similar bottom–up proc-
esses may complete the occluding triangle in the
Kanisza figure, but these cannot be the only process-
es that are in operation.
As said before, perception is not just a matter of
completion. Minimal graphics could aid the user by
minimising irrelevant information. The computa-
tional target is to find algorithms for identifying
‘noise’ within graphical representations. This can
only be done by identifying the task relevant at-
tributes of the original data, coupled with an under-
standing of the ways in which cognitive processes
map images to percepts. In there lies, in our view,
the greatest challenge in minimal graphics.
Visual perception and psychophysics are some
of the original parts of experimental psychology,
and there are many different theoretical approaches
to specific problems such as depth perception, mo-
tion perception, binocular fusion, object recognition
and so forth. It is of course not the goal of minimal
graphics to produce new cognitive theories, but rath-
er to draw on the existing knowledge of how such
processes contribute to our understanding and inter-
pretation of images; Strothotte contains interesting
examples of doing this [5].
We believe that there is an interesting analogy
between the foundations of photorealistic graphics
and those of minimal graphics. The principles of
photorealistic graphics rest on an approximation of
physical reality that is elegantly captured by Kaji-
ya’s well known rendering equation. In contrast,
minimal graphics does not (necessarily) seek to re-
produce physical aspects of reality, and requires in-
stead a model of cognitive information processing.
Kajiya’s equation is not in itself an algorithm for
rendering images, but rather provides the theoretical
foundation for families of approaches (ray tracing,
radiosity) that implement aspects of photorealism.
Similarly, we do not expect or require that the cog-
nitive theories underpinning minimal graphics will
provide an explicit approach to rendering. Rather,
we believe that such a theories will provide the basis
for defining a family of new rendering techniques
that achieve a minimal approach.
Conclusions
In this article we have argued that the conventional
view of rendering, is but one part of a much broader
enterprise of graphics–based communication in
which there is a need to consider fundamentally dif-
ferent approaches to the rendering problem. To
make progress on this enterprise, it is necessary to
understand perceptual and cognitive theories that
explain how graphical information is processed and
understood by humans. What is not simple to state,
of course, is how this new view of the rendering
problem can be addressed. Our comparisons have
focused on non–photorealistic rendering. There are,
of course, other lines of work which deal with the is-
sues of generating effective presentations, these in-
clude work on visual communication and, from an
AI perspective, on presentation planning. The con-
tribution of these areas to minimal graphics is still to
be explored. We are well aware that this paper only
scratched the surface of a significant research effort.
Let us close this note with a traditional Taoist
poem which nicely symbolises some of the ideas
that influenced us:
Thirty spokes share the wheel’s hub
It is the centre hole that makes it useful.
Shape clay into a vessel;
It is the space within that makes it useful.
Cut doors and windows for a room;
It is the holes that make it useful.
Therefore profit comes from that which is there;
Usefulness from what is not there.
— Lao Tse: Tao Te Ching, No. 11
Acknowledgements
This article summarises a number of ideas first ex-
plored in the report[1]. The ideas described here
have been discussed with a number of different col-
leagues and friends, and they have all, in some ways,
contributed to their development. We should men-
tion the names of Phil Barnard (MRC–CBU, Cam-
bridge, UK), Jon May (University of Sheffield, UK),
David Duce (RAL, UK), Patrick Olivier (University
of York, UK), and Pere Brunet (UPC, Spain). This
work was partially supported by the “TACIT” net-
work funded by the European Commission TMR
Programme.
References
[1] I. Herman and D.J. Duke: Minimal Graphics.
Reports of the Centre for Mathematics and
Computer Sciences, Report No. INS–R9903, ISSN
1386–3681, 1999. Also available on–line: http://
www.cwi.nl/~ivan/AboutMe/Publications/INS-
R9903.pdf
[2] D.H. Laidlaw: “Loose, Artistic ‘Textures’ for
Visualization”, In: IEEE Computer Graphics &
Applications, vol. 21, no. 2, March/April 2001, pp.
6-9.
[3] J. Ninio: La Science des Illusions. Odile Jacob,
Paris, 1998.
[4] J. Schumann, T. Strothotte, A. Raab, and S. Laser:
“Assessing the effect of nonphotorealistic rendered
images in CAD”, In: Proceedings of the CHI'96
Conference, ACM Press, 1996.
[5] C. Strothotte and T. Strothotte: Seeing between
Pixels. Springer–Verlag, Berlin, Heidelberg, 1997.