Behavior Research Metlwds, Instruments, &: Computers

1986, 18 (6), 487-492

SOFTWARE TECHNIQUES

The use of computer graphics animation

in motion perception research
DENNIS R. PROFFITI'
University of Virginia, Charlottesville, Virginia

and

MARY K. KAISER
NASA Ames Research Center, Moffett Field, California

The advantages and limitations of using computer-animated stimuli in studying motion per-
ception are discussed. Most current programs of motion perception research could not be pursued
without the use of computer graphics animation. Computer-generated displays afford latitudes
of freedom and control that are almost impossible to attain through conventional methods. There
are, however, limitations to this presentational medium. At present, computer-generated displays
present simplified approximations of the dynamics in natural events. We know very little about
how the differencesbetween natural events and computer simulations influence perceptual process-
ing. In practice, we tend to assume that the differences are irrelevant to the questions under
study and that findings with computer-generated stimuli will generalize to natural events.

This paper is divided into two parts. In the first part, Much of the work in motionperception has focusedon
we discuss some of the advantages of employing identifying minimal conditions for perceiving variousen-
computer-animated graphics in motion perception vironmental properties. For example, researchers have
research. In the secondpart, we discusssome of the limi- investigated the perception of form (three-dimensional
tations inherent to this presentational medium. Although structure) from motion information. Probably the most
many research programs could not possibly be pursued dramatic demonstrations that motion is a minimalcondi-
today without computer animation, we suggest that ex- tion for perceiving form are the kineticdeptheffect (Wal-
isting graphics displays nevermodel perfectly the environ- lach & O'Connell, 1953/1976) and point-light walkerdis-
mental dynamics that they are intended to simulate. Lit- plays (Johansson, 1973). In the original kinetic depth
tle is known about how these differences may influence effect demonstrations, shadowsofunfarniliar wire forms
perceptual processing. were projected onto a screen. Without motion, these
shadows appeared as two-dimensional configurations of
ADVANTAGES OF lines; however, when the wire forms were rotated, their
COMPUTER-GENERATED STIMULI three-dimensional structures were immediately evident.
Johansson's point-light walkers were made by attaching
Muchof the progress in motion perception research was small lights to the joints of people and then filming the
made possible by the development of computer graphics people as they walked in the dark. As with the forms from
animation technologies. Johansson (1950), in studiesem- the kinetic depth effect study, static frames from these
ploying dynamic point-light stimuli displayed on an os- displays appearedas meaningless two-dimensional arrays
cilloscope, not only influenced the conceptual understand- of dots; however, a very brief viewing of a moving se-
ing of the role that motionplays in organizing the visual quence allowedthe observer to identify the projection as
world, but also established the directions of future a locomoting person.
research methodology. More recently, researchershave increasingly turned to
the use of computer-generated dynamic displays. Braun-
This research was supported by Grant NCAl-8? from NASA, Grant stein (1976) developed a computer-based methodology for
HD-16195 from NICHD, and Grant INF-85-014 from the Virginia creating complexkinetic depth effect displaysconsisting
Center for Innovative Technology. Stephen Ellis, Scott Fisher, David of point-lights. Cutting (1978) created a general program
Gilden, Jeffrey Lande, and Susan Whelan providedvaluable criticisms
on an earlier versionof this paper. Reprint requests shouldbe addressed for generating point-light walkers. Bertenthal, Proffitt,
to Dennis Proffitt, Department of Psychology, Gilmer Hall, Univer- and Keller (1985) wrote a very general animation pro-
sity of Virginia, Charlottesville, VA 22903-2477 gram for the Apple microcomputer, interfaced to a Texas

487 Copyright 1986 Psychonomic Society, Inc.

488 PROFFITT AND KAISER

Instruments TMS 9918A video display processor; the pro- to the perception of real objects and events. One of the
gram allows a user to create point-light projections (limit important environmental properties that is absent in pic-
is 32 points) of rigid or jointed objects capable of all six tures is, of course, motion; however, the substitution of
degrees of freedom movement. (All of the demonstrations motion for pictorial cues does not necessarily make dy-
in Johansson's, 1950, book can be easily created with this namic computer displays more ecologically valid.
program.) Whenever people look at a computer-animated display,
There is now a long list of research topics in which dy- they are presented with conflicting information about
namic computer-generated displays are used. This list in- depth relationships. All of the primary depth cues specify
cludes studies on perceiving: ego motion, texture segre- that the transforming projections are two-dimensional.
gation, form, form change, object displacement, and Moreover, unless the display's motions are yoked to the
dynamics (the recovery of mass and force information observer's head movements (Rogers & Graham, 1979),
from kinematics). It has also been found that infants as the absence of motion parallax will further define two-
young as 3 months of age will attend with interest to dimensional aspects of the display. At odds with this in-
computer-generated point-light displays and can extract formation are the display's dynamics specifying three-
some structure from them (Bertenthal, Proffitt, & Cut- dimensional structures.
ting, 1984). A recent issue of Perception (Ramachandran, What is the influence of this dual awareness on the per-
1985), devoted to motion perception, contained a wide ceptual processing of computer-animated displays? Our
variety of research reports employing computer anima- research suggests that a person's ability to extract dynamic
tion. This issue even provided an Apple disk that allowed information from motion displays is related to the degree
the reader to view many of the dynamic stimuli that were of naturalness found in the simulation (Kaiser, Proffitt,
discussed. & Anderson, 1985). We suspect that too great a reliance
Computer animation has numerous advantages over on dynamic computer displays may result in an under-
other techniques for creating minimal motion displays. estimation of people's sensitivities to motion-specified in-
Programmed displays are far more flexible and easier to formation.
create than are physical mechanisms constructed to
produce desired motions. Computer displays can be eas- Scaling Perspective, Depth, and Size
ily programmed to display violations in natural dynamics A perceived object in a computer-animated display does
(events in which the laws of physics are violated). This not appear to be located on a monitor's screen; it appears
capability proves to be very useful in assessing a person's to be somewhere behind the screen. This indeterminacy
visual sensitivities to natural dynamics and is, of course, of absolute depth (observer-relative depth) creates a set
extremely difficult to achieve using real objects. Finally, of difficult problems in programming a natural-appearing
computer-generated stimuli provide the researcher with simulation.
exact knowledge of the display's parameters. An exam- Suppose that the researcher wants to generate a per-
ple of the importance of such tractability can be seen in spective projection of a rigid object undergoing rotation.
research on perceiving point-light walkers. Proffitt, Ber- To compute the perspective transformation, the researcher
tenthal, and Roberts (1984) created a computer display must specify, as one of the parameters, the distance be-
that contained all of the information previously thought tween the geometric eyepoint and the simulated object.
to be effective in the perception of Johansson's (1973) (Hagen, 1980, is an excellent source of articles on the
naturally produced point-light walker. After 1V2 min of geometry of perspective projection.) Should the viewer
viewing, only about a third of their subjects recognized assume that this distance is equal to the distance between
that this display could be seen as a projection of a person the eye and the screen, the viewer will be in for quite a
walking. We still do not know all of the parameters of surprise, since the simulated object will appear to deform
information that people use when extracting the human drastically as it rotates.
form from Johansson's original, naturally produced point- A rigid object will appear to deform as it rotates unless
light walker displays. Computer simulations can, in cases either the perceived viewing distance is specified ac-
such as this, serve as empirical tests of processing models. curately or the perspective transformation is not salient.
Determining appropriate viewing distances, thus, becomes
LIMITATIONS OF an empirical, rather than a purely geometrical problem.
COMPUTER-GENERATED STIMULI Moreover, perceived viewing distance may vary across
individuals and may not remain stable over time within
Dual Awareness an individual. With regard to the salience of perspective
As with viewing static pictures, viewing computer- transformations, many researchers present displays in
animated displays gives the observer a dual awareness: parallel projection and obtain few reports of nonrigid
(1) an awareness of a transforming two-dimensional pat- motions.
tern appearing on the terminal screen, and (2) an aware- If perspective information is not given, or if it is not
ness of the three-dimensional event that is being simu- sufficiently salient, then depth-order ambiguities arise un-
lated. Gibson (1979) argued that this dual awareness was less the animation program is capable of hidden surface
one of the aspects of picture perception that made it removal. (Depth order refers to whether elements are in
difficult to generalize from research employing pictures front of or behind other elements in the display.) Depth-
 GRAPlDCS ANIMATION 489

order ambiguities, in tum, can affect the motions that are visual environment (Fisher, 1985). The disadvantages of
seen. In kinetic depth effect displays, for example, the such a system include the cost and awkwardness of high-
perceived direction of motion will spontaneously reverse resolution color displays and the relatively high hardware
unless perspective information or occlusion is provided and software costs. In addition, head-mounted displays
(Braunstein, Andersen, & Riefer, 1982). For point-light may prove inappropriate for some subject populations
displays, depth order can be specified by causing some (e.g., infants) or experimental tasks.
points to disappear and other points to remain visible At some point, the size of the visual field becomes con-
whenever they pass through a particular location. In such strained not only by display technology, but also by limi-
cases, the points that disappear are seen as being behind tations in the computational hardware; the system simply
those that remain visible, and an invisible intermediate cannot compute values for all of the pixels in the scene,
occluding surface is perceptually specified (Proffitt et al., given the required update rate. When this happens, several
1984). remedies are possible. First, the system can convert from
Since the perceived distance from observer to simulated real-time displays to storage of display sequences on some
object cannot be determined geometrically, neither can random-access storage device (e.g., a laser disc). Alter-
the object's absolute size. If apparent distance and size natively, one can create variable resolution displays that
are different from the values intended by the program- compute high-resolution displays only for the area sur-
mer, then the display's dynamics may also appear in- rounding the observer's current fixation point. The latter
appropriate or out of scale. Consider, for example, a simu- solution is fairly elaborate, but is being pursued in con-
lation of a falling object. If the object's perceived size texts requiring high-resolution real-time graphic-animation
and distance are different from the assumed parameters displays (e.g., flight simulators). A third solution that
of the program, then the object will appear to fall either reduces the computational demand is to reduce the up-
too fast or too slow. We have successfully specified size date rate slightly, yet retain the real-time nature of the
and distance in such a display by placing a simulated display. This solution creates a new set of problems that
familiar object, such as a person, in the same depth plane are the next topic of consideration: The quality of mo-
as the falling object. tion in computer-generated displays.

Size of Visual Field Motion Quality

Unlike natural scenes, computer displays subtend a The quality of simulated motions depends primarily
limited area of an observer's field of view. The better upon two things: the particular motion algorithms em-
computer graphics systems in today's market employ ployed and whether or not the system possesses sufficient
1024 x 1024 pixel RGB monitors. If an observer views computational power to execute the algorithms in time for
such a monitor from a distance of about 80 em, the dis- the next screen update. We discuss the issue of motion
play has fairly good resolution (approximately 40 pixels algorithm adequacy below and again under the topic of
per degree), but subtends less than 26° of visual angle. Simulation Dynamics. First, we consider the computa-
The observer has the clear sensation of viewing a win- tional power issue.
dow on a scene rather than the scene itself. As mentioned above, update rate can be increased if
Attempts to enlarge the visual display entail significant scene resolution (or complexity) is reduced. Thus, most
compromises in terms of cost, resolution, color capabil- researchers are left with a direct trade-off between up-
ity, or update rate. Given present technology, it is ex- date rate and computational complexity. If there is no need
tremely costly to employ larger CRT displays; it is un- for real-time animation, the researcher gains a huge ad-
likely that CRT displays will ever become cost-effective vantage. Complex events can be generated one frame at
for wide field-of-view displays. We are currently using a time, with the ensemble of frames later shown at the
two 45-in. rear-projection screens in studies of peripheral desired update rate. Braunstein (1976) used this method
motion information processing; however, all currently to generate many of his depth-from-motion stimuli. Static
available rear- or front-projection systems have several images from a motion sequence were created on a com-
drawbacks, notably lack of resolution, contrast, and puter terminal and recorded frame-by-frame on 16-mm
brightness. In addition, projection systems tend to be cum- movie film. Many expensive systems on the market to-
bersome and difficult to adapt to all experimental situ- day store the image on videotape or laser disc (the latter
ations. having the advantage of rapid random-access capabilities).
Recently, effort has been directed toward the develop- However, all of these techniques require that the
ment of head-mounted display systems. There are several researcher have a well-defined, limited number of se-
advantages of such systems: they are capable of present- quences to be computed. Furthermore, all but the laser
ing binocular displays; a wide variety of apparent display disc make it extremely difficult to alter the order of
sizes can be produced on a single, small screen (although sequence presentation, as one might in any response-
fairly sophisticated optics are needed to produce appropri- dependent experimental design (e.g., staircase method-
ate geometries); and their displays can be yoked to the ologies).
observer's head motions or other monitored activity. In If one wishes to generate stimuli in real time, the trade-
fact, by employing head-tracking technologies, such a off between complexity and update rate remains. The third
head-mounted display can create a 360° stereoscopic factor in this trade-off is cost: the more expensive com-
490 PROFFITI AND KAISER

puter systemshave greater computational capabilities and example, the threads in a weaveof cloth will scatter light
software enhancements. Fortunately, the power-cost more narrowly in the directionof the thread than perpen-
trade-off continuesto become more favorableto the con- dicularly. Although some anisotropic models are being
sumer. Today's microcomputers are more powerful developed (e.g., Kajiya, 1985), such surfaces are still
graphicssystems than are minicomputers of a decadeago. quite difficult to simulate. Thus, we find a plethora of
In particular, processors with the Motorola 68000 chip smooth, regular objectsin computergraphicsdemonstra-
(and their 68010 and 68020 successors) provide impres- tions. Finally, texture presents a challenge to efficient
sive performance. In addition, some microcomputers modeling. One wantsto retainthe stochastic nature of the
(e.g., the Commodore Amiga) have dedicated graphics- texture while utilizing a consistent, efficient algorithm.
processing chips. Unfortunately, performance always Fractal geometry has been employed to this end (Man-
seems to lag expectations, and researchers are likely to delbrot, 1983)and has proved effective for a wide class
lag state of the art performance due to economic and of natural objects (e.g., mountains, trees, clouds).
procurement constraints. However, fractal models are not appropriate for all ob-
Researchersdisagree on what update rate is acceptable ject classes, and most of the current fractal algorithms
for dynamic stimuli, and, of course, the rate depends on are computationally expensive.
characteristics of the event. The update rate at whichtime- Realistic shadingis difficultto achievefor similar rea-
sampled motion becomes indistinguishable from smooth sons. In fact, the two issues are related: in order to specify
motiondepends both on the velocity and spatial frequency ambientlight conditions for shading, reflectance proper-
content of the image and on observer factors (Watson, ties must be known (Nishita & Nakamae, 1985). Con-
Ahumada, & Farrell, 1983). Ultimately, updaterates will sider, for example, the ray-tracing method of scene gener-
be limited by display hardware constraints. Researchers ation. In this method, a number of rays originate from
using raster display systems are limitedby the refreshrate eachpixeland are allowed to propagate through the scene,
of the monitor. This rate is generally 60 Hz for systems bouncing from surfaceto surfacein accordancewitheach
in North America and 50 Hz for systemsmarketed over- object's reflectivity (Cook, Porter, & Carpenter, 1984).
seas (although somemanufacturers use nonstandard rates; Thus, ray-tracing algorithms must adequately model in-
e.g., Sun Microsystems has 66-Hz monitors). terreflectionas well as primary lighting sources in order
When depicted objects move at high velocities, to achieve realistic-looking continuous tone represen-
computer-generated displays possess a notable artifact, tations.
most noticeable when comparing a computer-generated A final complexity is introducedwhen dynamicevents
"frame" with a frame from a movie film of the event. rather than staticscenes are generated. Sincerealistic sur-
In the film frame, quickly moving objects will be blurred. face and shadowing algorithms are computationally com-
In the computer-generated frame, however, all objects are plex, it becomes extremely expensive to generate 20 to
clearly defined, which leads to a strobing (or temporal 60 frames for each secondof the event. Ray-tracing tech-
aliasing) that appears quite unnatural. Effective motion- niques, for example, are beyondthe capabilities of all but
blur algorithms havebeen developed to solvethis problem the most complexcomputer graphics systems, and a sin-
(e.g., Max & Lerner, 1985). In brief, these algorithms gle ray-traced frame can require hours of CPU time to
sampleobjectlocation duringeach frame interval and dis- generate. Nonetheless, thesecomplex algorithms are often
tribute object density accordingly. Of course, the im- employed, because adequate interpolation algorithms have
plementation of such algorithms increases the computa- not yet been developed.
tional complexity of the sequence, and, thus far, cannot
be performed in real time. The Adequacy of Simulated Dynamics
The final issue we consider is the adequacy of simu-
Object Realism lated dynamics in computer-generated animation.
The problems of object realism are similar to those of Presently, onlythe mostsimpledynamic events(e.g., col-
motion quality: there are both computational power con- liding balls, rotating objects) are generated from mathe-
straints and adequacy of algorithm limitations. The two matical motion models. Even these models often make
aspects of object realism that we discuss are surface simplifying assumptions, such as the absence of friction
properties and shading. We also limit our discussion to or the useof particle, as opposed to solidbody, mechanics.
computer-generated objects, excluding those impressive Consider, for example, biomechanical motions, such
computer displaysthat are simplydigitizedphotographs. as those presented in point-light walker displays. Fully
Real objectsin the environment possesscomplexvisual adequate mathematical models of biomechanical motion
surface qualities. Texture and reflectance properties are haveyetto be developed, although progressis beingmade
difficult to model realistically for several reasons. First, (Girard & Maciejewski, 1985). At present, the most im-
reflectance properties havebeen studied for onlya limited pressive examples of computer-animated biomechanical
class of materials, with adequate mathematical descrip- forms were created by techniques borrowed from the
tion developed for fewer still. Second, few natural ob- traditional animation arts. For example, rotoscoping, an
jects have smooth surfaces withconstant reflective proper- animation technique developed at the Walt Disney Stu-
ties; most surfaces in the environment are anisotropic, dio, has been employed to capture the dynamics of hu-
meaning that reflection is a function of orientation. For man motion. In rotoscoping, one first films a person per-
 GRAPHICS ANIMATION 491

formingthe desired actions, then each film frame is used natural events are perceptually equivalent. We shudder
as a template to specify body and limb coordinate loca- at the thought of perceptual psychologists' attempting to
tions for eachanimation frame. Although sucha technique agreeon criteriaconcerning when suchperceptual equiva-
produces impressive results for the cartoonist's charac- lence is achieved. Ideally, we should "flight test"
ter (e.g., Snow White) or for the computer animator's computer-generated stimuli to determine whether theirdy-
specialeffects (e.g., Abel Graphic's metallic woman), it namics are discriminable from those seen in natural
affords few advantages to the perceptualpsychologist in- events. In practice, we tend to assume that the differences
terested in the specification of and the observers' sensi- are irrelevant to the questions under study and that find-
tivity to biomechanical kinematics. ings with computer-generated stimuli will generalize to
An approach midway between rotoscope techniques and natural events.
true mathematical motion models is the keyframe tech-
nique. With this technique, criticalpointsof the event se- CONCLUSION
quence are sampled, and intermediate coordinate positions
are calculated based on assumed motion properties and Most current programs of motion perception research
constraints(Steketee & Badler, 1985). As yet, keyframe couldnot be pursued without the use of computer graphics
techniques cannot provide motion parameters that are animation. Computer-generated displaysafford latitudes
sufficiently precise to be used in perceptual research. of freedom and control that are almost impossible to at-
As indicatedabove, the algorithms used by perceptual tain throughconventional methods. We think it is impor-
psychologists in theirdynamic simulations havebeenvery tant, however, to be ever mindful that computer simula-
reductionistic, even for relatively simplephysicalevents tions rarely, if ever, achieve a level of verisimilitude
(e.g., two objectscolliding). There are good reasons for capable of causing an observer to confuse the simulation
employing extreme simplifying assumptions: Precisemo- with reality. All of the limitations discussedabove place
tion modeling of complex physical systems is a hugecom- constraints on the apparentrealismof computer-animated
putational problem. The difficulty of developing adequate displays.
modelsof such systems may be better understood by ex- Whenviewing a computer-generated display, theviewer
amining the problems confronted by other disciplines, always experiences a dual awareness: an awareness of
such as computational fluid dynamics, which have at- both a two-dimensional contrivedpatternand a projected
tempted similar modeling. three-dimensional event. Sensitivity to the dynamics
Computational fluid dynamics (CFD) attempts to nu- manifest in the latter is almost surely influenced by the
mericallymodel fluid and gas flow. This work has been awareness of the former. We recommend cautionin mak-
moststrongly pursued in studies of aerodynamics, but also ing unqualified generalizations about human sensitivities
has applications in streamlining, weather prediction, wake to natural events from studies on perceiving computer-
dynamics, and wind loading studies (Kutler, 1983). At animated displays. Convergentinvestigations employing
institutions suchas the NASA Ames Research Center, the natural objects are always desirable, although, in prac-
ultimate goal of CFD is to allow aerospace engineers to tice, such studies are oftenextremelydifficult to conduct.
optimize designs solely through computational models.
At present, CFD modeling is complemented with wind REFERENCES
tunnel and flight test evaluations. Convergence of CFD
BERTENTHAL, B. I., I'ROFFITI, D. R., a: CUlTING, J. E. (1984). Infant
models with these laboratory and field results provides sensitivity to figural coherence in biomecbanicalmotions. Journalof
a basis for evaluating modeladequacy. The advancement Experimental Child Psychology, 37, 213-230.
of CFD toward adequate models is seen as being paced BERTENTHAL, B. I., I'ROFFITI, D. R., a: KELLER, S. E. (1985). 3-D
by several constraints, notably the development of ap- graphics animation program for the Apple n. Behavior Research
propriate turbulence models, the ability to model dynamics Methods. Instruments, & Computers, 17, 195-202.
BRAUNSTEIN, M. L. (1976). Depth perception through motion. New
in three dimensions instead of reducing the problem to York: Academic Press.
a two-dimensional representation, and the development BRAUNSTEIN, M. L., ANDERSEN, G. L., a: RmFl!R, D. M. (1982). The
of more powerful computer architectures. use of occlusionto resolve ambiguityin parallel projections. Percep-
Despite the computational complexity confronting CFD tion & Psychophysics, 31, 261-267.
COOK, R. L., PORTER, T., a: CARPENTER, L. (1984). Distributed ray
modeling (almost all three-dimensional models exceed the tracing. Computer Graphics, 18, 137-145.
computational power of the Cray X-MP computer sys- CUTTING, J. E. (1978). A program to generate synthetic walkers as dy-
tem), there is the advantage that the criteria for model namic point-lightdisplays. Behavior Research Methods & Instrumen-
adequacy are well defined: a CFD model's performance tation, 10, 91-94.
FISHER, S. S. (1985, December). Virtual interfaceenvironment. Paper
should be functionally equivalent to that found for (1) a
presented at theSpace Station Human Factors ResearchReview, NASA
physical aircraft in the wind tunnel and in the field, and Ames Research Center, Moffett Field, CA.
(2) explicitanalytic solutions where they exist. As might GIBSON, J. J. (1979). The ecological approach to visual perception.
be expected, however, thereis a notable lackof consensus Boston: Houghton Mifflin.
among experts as to when equivalence is reached. GIRARD, M., A MARCII!JEWSKI, A. A. (1985). Computational model-
ing for the computeranimationof leggedfigures. Computer Graphics,
The transportation of the CFD criteria for simulation 19, 263-270.
adequacy to perceptual research would require us to HAGEN, M. A. (Ed.) (1980). The perceptionofpictures (Vol. 1). New
demonstrate that our simulations and the corresponding York: Academic Press.
492 PROFFITT AND KAISER

JOHANSSON, G. (1950). Configuration in event perception. Uppsa1a, of three-dimensional objects: Taking accountof shadowsand inter-
Sweden: Almqvist & Wiksell. reflection. Computer Graphics, 19, 23-30.
JOHANSSON, G. (1973). Visual perception of biological motion PROFFITT, D. R., BERTENTHAL, B. I., &: ROBERTS, R. J., JR. (1984).
·and a model for its analysis. Perception & Psychophysics, 14, Therole of occlusionin reducingmultistability in movingpoint-light
201-211. displays. Perception & Psychophysics, 36, 315-323.
KAISER, M. K., PROFFlTI, D. R., &:ANDERSON, K. (1985). Judgments RAMACHANDRAN, V. S. (Ed.). (1985). Humanmotionperception [Spe-
of natural and anomalous trajectoriesin the presenceand absenceof cial Issue]. Perception, 14(2).
motion.JournalofExperimental Psychology: Learning,Memory, & ROGERS, B., &: GRAHAM, M. (1979). Motion parallax as an indepen-
Cognition, 11, 795-803. dent cue for depth perception. Perception, 8, 125-134.
KAmA, J. T. (1985). Anisotropic reflection models. Computer Graphics, STEKETEE, S. N., &: BADLBR, N. I. (1985). Parametrickeyframeinter-
19, 15-24. polation incorporating kineticadjustment and phrasing control. Com-
KUTLBR, P. (1983). A perspectiveof theoretical and applied compu- puter Graphics, 19, 255-262.
tational fluid dynamics (AlAA Paper No. 83-0(37). Reno, NY: WALLACH, H., &: O'CONNELL, D. N. (1953). The kineticdeptheffect.
AlAA. JournalofExperimental Psychology, 45, 205-217. (Alsoin H. Wal-
MANDELBROT, B. (1983). The fractal geometry of nature. San Fran- lach, (1976). On perception. New York: Quadrangle.)
cisco: W. H. Freeman. WATSON, A. B., AHUMADA, A., &: FARRELL, J. E. (1983). The win-
MAX, N. L., &: LERNER, D. M. (1985). A two-and-a-half-D motion- dow ofvisibility: A psychophysical theoryoffidelity in time-sampled
blur algorithm. ComputerGraphics, 19, 85-94. visualmotion displays (NASATechnicalPaper No. 2211). Moffett
NISHITA, T., &: NAKAMA!!, E. (1985). Continuous tone representation Field, CA: NASA.