Electronic Proceedings of the
ACM Workshop on Effective Abstractions in Multimedia
November 4, 1995
San Francisco, California

Since Less is often More:
Methods for Stylistic Abstractions in 3D-Graphics

Antonio Krüger
Stuhlsatzenhausweg 3
66123 Saarbrücken

Thomas Rist
Stuhlsatzenhausweg 3
66123 Saarbrücken

ACM Copyright Notice

Table of Contents

1. Introduction

More than two decades of research in computer graphics led to advanced rendering techniques which, together with enriched object models, enable the production of photorealistic images - even on today's low-cost standard hardware. Of course, in many areas (e.g., entertainment and advertisement) photorealism is the goal to strive for. However, there are as many other occasions where graphics is right in place but photorealism is not since such images provide too much information. A good example where this observation can be made is the field of technical documentations. Authors of device descriptions and instruction manuals preferably rely on other graphical forms than photographs for the presentation of objects and the illustration of their maintenance. Inspite of reconstructing the detailed visual appearence of an object, methods for filtering, generalization, and customization of information are applied in order to generate an effective illustration. For depicting objects this includes, among other things:
While human illustrators intuitively or conciously follow such basic rules of communication, they have to be implemented somehow in order to make automated presentation systems work.

In this paper we present a computational approach for achieving stylistic abstractions in 3D-graphics.

< -- Table of Contents
< -- Introduction

2. Technical vs. Stylistic Abstractions in 3D-Graphics

The generation of graphical object depictions can be understood as mapping a source of information (e.g., a 3D-object model, a propositional object description, or even graphical material) onto graphical presentation. By the term graphical abstraction we refer to cases where some information available in the source is not conveyed in the resulting graphics. Afterwards we distinguish between two classes of graphical abstractions.

Technical abstractions are caused by some technical reasons during the process of image generation from a given object model. Typical technical reasons are the use of simplified object and lightening models to spare with limited computational resources. So for example an object depiction produced with the Gouraud shading method can be considererd as being more abstract than a Phong-shaded picture of the same object under the same viewing and lightening conditions. In the area of computer graphics, the process of image generation is usually described in terms of a rendering pipeline [1], or within a more general reference model such as the ISO-standard [2]. Technical abstractions may be identified systematically at each single step of such a pipeline or at each layer of a reference model - depending on the particular algorithms used to perform the transformations the step or layer stands for. Doing this, however, is beyond the scope of the current paper.

Instead of doing so, we contrast technically caused abstractions to the notion of stylistic abstraction. With this we mean abstractions which are technically avoidable but which are made intentionally in order to achieve a certain presentation goal or to follow a superordinate design directive. Presentation goals where graphical abstraction may come in include the following: ``Draw the viewer's attention to a certain object property'', ``Show an object in its spatial context without de-focussing it'', and ``Show a prototype representer of a set of instances''.

Examples of superordinate design directives are the two advices mentioned in the introductory section: not to overload graphics, and to enable a proper distinction between important and less relevant aspects. Further directives are the wish to keep a presentation as simple as possible, and to increase the degree of uniformity over a series of pictures. Whether or not such directives can be followed usually depends on parameters, e.g., purpose, viewer profile, presentation context etc.

In order to enable a graphics generation component which satisfies presentation goals and follows design directives two things have to be done.

  1. we must provide operationalizations of methods for performing stylistic abstractions, and
  2. we must specify the conditions under which a certain technique should be applied to accomplish a given presentation goal or design directive

< -- Table of Contents
< -- Technical vs Stylystic Abstractions

3. Methods for Stylistic Abstractions

Our intuitive understanding of applying a stylistic abstraction method is, that we should get a picture which is simplified in some aspect. Consequently we first introduce our notion of simplicity.

3.1 Meassuring Simplicity

In our approach simplicity is a partial order relation for graphical object depictions. However, such statements like ``line drawings are more simple than shaded pictures'' are insufficient because they are too general. A better solution is to base the order on a comparison of the amount of information conveyed in different depictions of the same object. In particular, we compare the aspects: object composition, object shape, and material object properties.

Given two depictions G1 and G2 of one and the same object O, we say that G1 is more simple than G2 with respect to O's composition, if G1 shows less parts of O than G2 does. Concerning O's shape we say that G1 is more simple than G2, if the number of lines forming the contour of O's depiction in G1 is lower than in G2. If G1 is a black&white picture and G2 shows O coloured, we say G1 is more simple than G2 with respect to O's material object colour. Of course, there are other plausible criteria on which the comparison of a single aspect could be based. If, for example, O has a convex silhouette in G1 but a concave one in G2 we regard G1 to be more simple than G2 too, with respect to the encoding of O's shape.

Now we can turn to the question of how to produce object depictions which are simplified according to a certain aspect. Since we are concerned with graphics generation from 3D-object models, abstraction methods can manipulate either the 3D-object models, the projection functions or the object depictions, so that we obtain a three layered abstraction process (cf. Fig. 1). Each layer functions as a filter which reduces the information in the final graphics. In the following, we give examples of abstraction methods for all three layers.

Figure 1: A three layered abstraction process.

< -- Methods for Stylistic Abstractions

3.2 Manipulation of 3D-Model

Instead of projecting the most detailed 3D-object model (in our case wireframes) first a geometrically simplified model can be generated which then will serve as the image source. Inspired by Feiners' work on geometric approximations for domain objects [3], we have defined a set of operations aiming at the reduction of geometrical detail through the merging of object-models. Fig. 2 gives an impression of applying merging operators to the standard modelling primitives cube and cylinder. Note that this merging operation does not change object locations; i.e, the piece of space covered by an object involved before the merging will also be covered by the resulting object. The simplification can be seen in the fact that in all three cases a two-parted object constellation has been reduced to a single constituent.

Figure 2: Simplification of graphics by merging 3D-model

< -- Methods for Stylistic Abstractions

3.3 Parametrization of the Projection

Many 3D-graphics systems support different kinds of projection and allow for modification of projection parameters. These features can be exploited for the definition of methods for stylistic abstractions. For example, simplifying an object depiction by abstracting it's colour, a projection can be used that paints object faces with a single colour. Similar methods are using a uniform line-thickness, and dash-lines. A further method suitable for simplifying line drawings is known as hard-edging. Advanced rendering mechanism take this into account and generate non-photorealistic graphics. An example of such a system is the Sketch-Renderer shown in [4].
< -- Methods for Stylistic Abstractions

3.4 Manipulation of Object Depiction

Most of the effects which are produced by the methods before, can be achieved - at least approximately - by methods that operate directly on object depictions. E.g., instead of merging object models a merging operator for depictions can be defined and the result can be superimposed on the original picture. This is useful if the information source is of a 2D nature, for like map data.

However, the reason why we often prefer the stylistic methods is, that we would have more implementational effort if we rely on picture manipulations only. For example, it is easier to map 3D-edges onto 2D-dashed lines instead of redrawing full lines in a projection. Other methods such as using an eraser to reduce the picture contents, usually can't be applied to single picture elements without producing unwanted side effects. So, up to now, our only method belonging to this group is a scaling operation. Although scaling doesn't mean abstraction itself, it may destroy size proportions, e.g. when a series of pictures has to be sized uniformily.

< -- Table of Contents
< -- Methods for Stylistic Abstractions

4. Applying Stylistic Abstractions

Abstraction methods as presented above have been implemented in a system called PROXIMA [5]. Below, we briefly describe three scenarios in which PROXIMA is used for the generation of illustrations.

4.1 Multimedia Information Presentation

Our first application of PROXIMA has emerged from the WIP research project on the automated generation of multimedia presentations. There, we have built a prototype presentation system (also called WIP cf. [6]) that generates illustrated descriptions and instructions for the assembly and maintenance of technical devices (lawn mower, espressomachine, modem etc.). Thus WIP often has to communicate information about physical objects graphically including the previously mentioned presentation tasks.

In the WIP system, PROXIMA appears as a subcomponent of a knowledge-based graphics generator [7]. This generator receives its input from a superordinate presentation planner[8] which is responsible for content selection, the overall structuring of the presentation, and the coordinated distribution of presentation tasks to the media-specific generators (currently a text and a graphics generator).

After receiving a certain presentation task from the planner, the graphics generator applies so-called design strategies in order to map these tasks to a set of constraints on the graphics to be generated. Some of these constraints have to be refined by further design strategies. Finally, all elementary constraints are associated with operators of the underlying graphics realization modules such as PROXIMA. To give an impression of how this process works, have a look at a presentation task, for example:
Depict object X as a background object for Y
One of our design strategies maps the task onto a set of constraints. Besides visibility constraints on the objects Y and X this set includes the particular constraint:
Y should be focussed.
A solution to satisfy this constraint is to de-emphasize the background object X by stylistic abstraction. Fig. 3 illustrates the effect of applying that strategy when depicting a selector switch of a modem together with the circuit board as a background object; (a) without applying abstraction methods; (b) result after applying model simplification and dashing to all objects except the switch and the second LED (which is to indicate the setting of the switch).

Figure 3: Depiction of a modems' circuit board; detailed and stylistically abstracted

The example also shows that some of the knowledge about when to use stylistic abstractions is directely coded in the design strategies of WIP's graphics designer. This approach is suitable for capturing particular design cases. A more general mechanism to decide automatically which stylistic abstractions should be applied to which objects, is a build-in facility of PROXIMA and is based on default-rules. For example, if PROXIMA is requested to simplify a certain object, it relies on the rule:

IF X is to be simplified in the picture P
THEN unless specified otherwise apply the same simplification methods to all other objects Z in P which belong to the same superior object as X

< -- Applying Stylistic Abstractions

4.2 Semi-Automated Illustration Design

Our second application is a workbench for semi-automated illustration design called AWI [9] . The motivation for such a workbench is based on the fact that the results of fully automated generators are often not satisfying, but that in many cases only minor changes are necessary to make them perfect. Using AWI, illustration design and realization is a collaboration between the user and the system. Concerning the generation of stylistic abstractions AWI allows the user to intervene during an incremental abstraction process in order to modify the parameters of the abstraction methods, to indicate objects which should also be simplified, or the other way round, to indicate objects that should be excluded from further simplifications. Fig. 4 presents three snapshoots taken during a session with AWI, (a) user selects a LED to exclude it from further simplifications (the LED appears in reverse-video); (b) result after AWI has applied several abstraction steps, and (c) since the user considered AWI`s result as oversimplified, she requests AWI to undo the last merging of the LED row.

Figure 4: Interactive generation of stylistic abstractions

< -- Applying Stylistic Abstractions

4.3 Hypergraphics

In our recent extensions of the WIP system we deal with the design of interactive multimedia presentations. To actively involve the addressee in a presentation, she will be encouraged by the system to explore visual material by herself [10]. Our notion of stylistic abstraction enables the realization of an interesting new interactive exploration technique. In analogy to a hypertext presentation, an object can be depicted first at a high level of abstraction. If the addressee is interested in more details about a certain part of it, she just clicks on the graphical presentation and receives a refined depiction. Technically speaking, we first generate a simplified object depiction using PROXIMA (cf. Fig 5a). Then during the exploration (cf. Fig 5b), PROXIMA is requested to undo applied abstraction operators on selected parts (cf. Fig 5c).

Figure 5: Refinement of a hypergraphics on demand

< -- Table of Contents
< -- Applying Stylistic Abstractions

5. Conclusions

Overall the information reduction in 3D-graphics using abstraction techniques as presented above has several advantages. First it can be used to tailor the degree of detail of a graphics to the particular needs of individual viewers. Second, the techniques can be used to save computational efforts, which is especially important in realtime application design (cf. [11]). Such an application is for example Virtual Reality. There, complex 3D-models have to be simplified in order to obtain satisfying throughput. Our 3D merging operators can be used to produce geometric object descriptions of different degrees of detail, coded for example in VRML (cf. [12]).

Another application where graphical abstraction are helpfull is the domain of cartography. A central problem in map designing is to choose an appropriate degree of detail. For example, it is reasonable that the degree of detail of a town map for pedestrians is higher than it is for car drivers (cf. Fig. 6A 6B). A travel information system could generate dynamically such customized maps using 2D merging operators similar to our 3D-model operators. This research line is currently followed in the context of a collaboration project with Siemens AG (for more details see MOFA-Homepage).

Figure 6: (A) City map for pedestrians, all public paths shown. (B) Less detailed Map for car drivers, level of detail has been decreased by merging adjacent buildings.

< -- Table of Contents
< -- Conclusions

5. References

[1] J. Foley, A. van Dam, S. Feiner, J. Hughes: Computer Graphics: Principles and Practice, 2nd Edition, Addison Wesley, 1990.
[2] ISO/IEC IS 11072, Information processing systems - Computer Graphics - Computer Graphics Reference Model, 1992.
[3] T. Strothotte, B. Preim, A. Raab, J. Schumann, D. R. Forsey: How to Render Frames and Influence People. Department of Simulation and Graphics, Otto-von-Guericke University of Magdeburg, Germany, 1995.
[4] S. Feiner: APEX: An experiment in the automated creation of pictorial explanations. IEEE Computer Graphics and Applications 5, 1985.
[5] A. Krüger: PROXIMA: Ein System zur Generierung graphischer Abstraktionen, Document D-09-95, German Research Center for AI, Saarbrücken, 1995.
[6] W. Wahlster, E. André, W. Finkler, H.-J. Profitlich, T. Rist: Plan-Based Integration of Natural Language and Graphics Generation. AI Journal 63, 1993.
[7] T. Rist, E. André: From Presentation Tasks to Pictures: Towards a Computational Approach to Graphics Design. In: Proc. of ECAI, 1992.
[8] E. André, T. Rist: The Design of Illustrated Documents as a Planning Task. In: M.T. Maybury: Intelligent Multimedia, AAAI Press, 1993.
[9]. T. Rist, T. Krüger, G. Schneider, D. Zimmermann: AWI - A Workbench for Semi-Automated Illustration Design. In: Proc. of the Workshop: Advanced Visual Interfaces (AVI '94), 1994.
[10]. E. André, T. Rist: Multimedia Presentations: The Support of Passive and Active Viewing. In: Working Notes of AAAI Spring Symposium on Intelligent Multi-Media Multi-Modal Systems, 1994.
[11] J. K. Strosnider and C. J. Paul: A Structured View of Real-Time Problem Solving, AI-Magazine, Summer 1994.
[12] VRML Specifications
< -- Table of Contents