STEVENS’ HANDBOOK OF EXPERIMENTAL PSYCHOLOGYpeople.brandeis.edu/~sekuler/papers/sekulerStevens... · pashler-44104 book January 11, 2002 11:20 122 Motion Perception Figure 4.1 Four

pashler-44104 pash44104˙fm January 11, 2002 14:8

STEVENS’ HANDBOOK OFEXPERIMENTAL PSYCHOLOGYTHIRD EDITION

Volume 1: Sensation and Perception

Editor-in-Chief

HAL PASHLER

Volume Editor

STEVEN YANTIS

John Wiley & Sons, Inc.

iii

pashler-44104 book January 11, 2002 11:20

CHAPTER 4

Motion Perception

ROBERT SEKULER, SCOTT N. J. WATAMANIUK, AND RANDOLPH BLAKE

INTRODUCTION AND OVERVIEW

Gordon Lynn Walls, a comparative anatomist,observed, “If asked what aspect of visionmeans the most to them, a watchmaker mayanswer ‘acuity,’ a night flier ‘sensitivity,’ andan artist ‘color.’ But to the animals which in-vented the vertebrate eye, and hold the patentson most of the features of the human model,the visual registration of movement was of thegreatest importance” (Walls, 1942) p. 342.

The rich and rapidly expanding scientificliterature on visual motion perception sug-gests that Walls was right: To organisms all upand down the phylogenetic scale, visual mo-tion perception is of unmatched importance.Visual motion serves a wide variety of crucialroles: wayfinding (optic flow), perception ofshape from motion, depth segregation, judg-ments of coincidence (time to collision, timeto filling a tea cup), judgments of motion di-rection and speed, and perception of animate,biological activity. Sometimes, the presenceof motion can compensate for deficienciesin other forms of visual information, as Fig-ure 4.1 shows. The three images in the figureare frames from a video showing a person per-forming a common action. Clearly, no singleframe conveys sufficient spatial structure topermit recognition that a person is present, letalone recognition of what the person mightbe doing. However, the complex patterns of

visual motion generated when these framesare displayed as part of a video convey im-mediately that a person is present and thatthe person is in the process of sitting down(Bobick & Davis, 2001).1

Recent decades have produced major ad-vances in understanding of visual motionperception.2 Many such advances have comefrom complementary approaches to analyzingmotion: psychophysical, computational, andneurophysiological. It is now known that thedetection and analysis of motion are achievedby a cascade of neural operations, startingwith the registration of local motion signalswithin restricted regions of the visual field andcontinuing with the integration of those localmotion signals into more global descriptionsof the direction and speed of object motion.Physiological studies of animals—most no-tably cats and monkeys—have revealed someof the neural hardware comprising this hierar-chical processing scheme. Recently, exciting

1To download the video from the Internet, go tohttp://www.cis.ohio-state.edu/∼jwdavis/Archive/blurmotion.mpg orhttp://www.cis.ohio-state.edu/∼jwdavis/Archive/blurmotion.mov.2Previous editions of this handbook paid scant notice tothe topic of our chapter. In the first edition, Graham (1951)spent just six pages on motion perception, emphasizingresearch on apparent motion. In the second edition, morethan three decades later, the coverage was increased byonly ten pages distributed over two chapters (Hochberg,1988; Westheimer, 1988).

121


122 Motion Perception

Figure 4.1 Four still frames cut from a video by Bobick and Davis (2001). Used with permission.NOTE: The video shows a person engaged in a common, everyday activity. The low-pass spatial filteringof video makes it difficult, from any individual frame, to discern the person, let alone what the person isdoing. However, when the video is played at normal rate, the pattern of motions makes both the personand the person’s action immediately apparent.

new techniques including brain imagingand transcranial magnetic stimulation havebeen deployed in concert with psychophysicsto identify neural concomitants of motionperception in the human visual system.

Our goal in this chapter is to highlightsome of these exciting developments. How-ever, limitations on space—together with theexponential growth of the literature on mo-tion perception—forced on us hard choicesabout what to include and what to omit.Thus, this chapter emphasizes motion in thefront-parallel plane, unavoidably deempha-sizing work on motion in depth and “cyclo-pean” motion perception (Patterson, 1999).In addition, the chapter focuses on motionsof objects defined by luminance contrast,with little discussion of important work onthe role of chromatic information in motionprocessing (Dobkins, 2000; Gegenfurtner &Hawken, 1996). The chapter slights numerousinteresting and potentially informative illu-sions of motion (e.g., Hikosaka, Miyauchi, &Shimojo, 1993; Krekelberg & Lappe, 2001;Viviani & Stucchi, 1989). Moreover, our cov-erage primarily focuses on motion perceptionin primates, particularly Homo sapiens. Con-sequently, interesting work on motion per-ception in birds (e.g., Bischof, Reid, Wylie,

& Spetch, 1999; Wylie, Bischof, & Frost,1998), fish (e.g., Albensi & Powell, 1998;Orger, Smear, Anstis, & Baier, 2000) and in-sects (e.g., Dror, O’Carroll, & Laughlin, 2001;Gabbiani, Mo, & Laurent, 2001) has beenleft out. Our chapter does include researchon “atypical observers,” particularly indi-viduals with diminished motion sensitivityconsequent to brain damage.

Stimuli

In introducing the first edition of this hand-book, Stanley Smith Stevens (1951, pp. 31–32) observed that “In a sense there is onlyone problem of psychophysics, the definitionof the stimulus. . . . [T]he complete definitionof the stimulus to a given response involvesthe specification of all the transformationsof the environment, both internal and exter-nal, that leave the response invariant. Thisspecification of the conditions of invariancewould entail, of course, a complete under-standing of the factors that produce and thatalter responses.” We agree. As this chapterunderscores, contemporary research on visualmotion perception has advanced in large mea-sure because researchers are able to gen-erate and deploy suitable stimuli, including


Introduction and Overview 123

innovative computer-generated animations,that simulate complex, real-life events.

Commenting on one aspect of this chal-lenge, Graham (1965, pp. 579–580) cautionedthat “we must take care that parameters are notconfounded, a danger that arises only too read-ily from the fact that velocity itself involvesthe variables of distance and time. In any givenexperiment the variables of time, distance, in-terval between stimuli, and cycle of repetitionof stimuli must be clearly analyzed before wecan be confident that unequivocal conclusionsmay be drawn.”

Researchers have developed many cleverways around the confounds that Stevenswarned against. Consider just two examples.Under normal circumstances, a visual target’smovement always involves a change of thatobject’s shape, position, or both. This con-founding of motion and position change hasmade it difficult to connect psychophysical re-sponses to motion alone. To break the con-found, Nakayama and Tyler (1981) generatedmatrices in which black and white cells al-ternated at random. All cells in a row wereshifted back and forth, left and right; with ap-propriate rates of shift, observers saw oscilla-tory motion. The cells were small (<3 minarc)and spatially quite dense. Moreover, becauseall cells of the same color were indistinguis-hable from one another, the positions of indi-vidual elements could not be tracked. Despitethe absence of position information, observerscould detect the oscillatory motion generatedby shifts of pattern elements.

Consider a second example of a stimulusdesigned to test a hypothesis about motionperception. To explore how the visual sys-tem combines or segregates spatially inter-mingled motions in different directions, Qian,Andersen, and Adelson (1994) created dis-plays whose every local region contained bal-anced, opposite directions of motion. The lo-cally opposed directions tended to cancel oneanother, which caused observers to see no

overall motion. This chapter offers numerousother examples of complex stimuli specifi-cally designed to probe particular aspects ofmotion perception.

Overview of Motion Processing Stages

Where appropriate, this chapter relates psy-chophysical results on motion perception tounderlying neural mechanisms. An interestin establishing such connections drives muchcontemporary research into visual motion. Forthis reason, it will be helpful to provide a broadoverview of the anatomy and physiology ofthose portions of the primate visual system ex-plicitly involved in the analysis of motion in-formation (see Figure 4.2); for a more detailedaccount, see Croner and Albright (1999).

Among the neurons in the visual systemsof primates, cells selectively responsive to thedirection of motion are first encountered inarea V1, the primary visual cortex, which islocated in the occipital lobe. Such neuronsare often described as “tuned” for direction.3

Beginning with the landmark work of Hubeland Wiesel (1968), it has been known thata significant fraction of V1 neurons respondbest when a contour moves through theirreceptive fields in a particular direction;responses are significantly diminished whenmovement is in the opposite direction. Dif-ferent neurons have different preferred direc-tions of motion, with all directions aroundthe clock represented within the ensembleof neurons. This inaugural stage of process-ing comprises a local analysis of motion en-ergy. In this analysis, direction-selective neu-rons act as filters that register the presence ofcomponent features of moving objects within

3As Parker and Newsome put it (1998, p. 229), “A neuronis considered to be ‘tuned’ if the response is strongestto a particular value (or narrow range of values) of thestimulus and declines monotonically as stimulus valuesdepart from this ‘preferred’ value.”



VIP

LIP

7A

PP

MST

PGCortex

TECortex

Dorsal

Ventral

V1 V2

V3V4

MT

TEO

FST

V1VVV

TETETE

PGPGP

Figure 4.2 Diagram illustrating proposed functional and anatomical streams in the primate cerebralcortex.NOTE: Partially separate streams carry information from area V1 either dorsally, toward the inferiorparietal cortex (PG), or ventrally, toward the inferior temporal cortex (TE). Arrows indicate the mainfeedforward projections between areas. Abbreviations used in the diagram: V1, primary or striate cortex;MT, middle temporal area (also known as V5); VIP, ventral intraparietal; LIP, lateral intraparietal; PP,posterior parietal, MST, medial superior temporal; FST, fundus superior temporal; PG, inferior parietalcortex; TE, inferior temporal cortex.SOURCE: After Ungerleider and Haxby (1994).

the local regions of their receptive fields(Emerson, Bergen, & Adelson, 1992).

The outputs of these local filters in areaV1, in turn, activate second-stage analyzersthat integrate motion signals over more ex-tended regions of visual space. This second-stage analysis begins with neurons in the mid-dle temporal visual area, or area MT, as it istypically called. Area MT receives some ofits input directly from area V1 and the restindirectly from area V1 via areas V2 and V3.Nearly all neurons in area MT are selectivefor the direction and speed of stimulus mo-tion, again with the range of preferred direc-tions among neurons spanning 360 degrees.MT neurons have larger receptive fields thando V1 neurons, which means that they canintegrate motion signals over larger regionsof visual space. Moreover, a given MT neu-ron will respond to motion in its preferred

direction regardless of whether those motionsignals are carried by luminance, color, or tex-ture. MT neurons, in other words, exhibit forminvariance (Croner & Albright, 1999), imply-ing that those neurons register motion infor-mation per se. MT neurons, in turn, project tohigher visual areas that encode more complexforms of motion, including expansion and ro-tation (Tanaka & Saito, 1989) and motion-defined boundaries (Van Oostende, Sunaert,Van Hecke, Marchal, & Orban, 1997). Out-puts from area MT also make their way to vi-sual areas in the frontal lobe that are concernedwith the control of eye movements (Bichot,Thompson, Chenchal Rao, & Schall, 2001;Schall, 2000).

A great many studies implicate area MTin the perception of motion. Though neuronsin area MT certainly contribute to the percep-tion of motion, it is clear that this is not the


Introduction and Overview 125

sole site where neurons extract significant mo-tion information. Actually, various aspects ofmotion perception depend on the neural com-putations carried out in different areas of thecortex. Normally, motion perception dependson activity distributed over many areas of thebrain, each extracting somewhat different in-formation from the retinal image. Compli-cating matters, in macaque monkeys, whichhave visual systems that are highly similarto those of Homo sapiens, back-projectionsfrom area MT to area V1 have been demon-strated (Beckers & Homberg, 1992). Initialevidence suggests that in humans this back-projection may be important for consciousawareness of visual motion. To explore thisidea, Pascual-Leone and Walsh (2001) ap-plied brief pulses of magnetic energy4 tospatially restricted regions of the scalps ofhuman observers. This technique is knownas transcranial magnetic stimulation (TMS).When the localized pulses are adjusted induration, frequency, and amplitude and aredelivered to particular regions of the scalp,TMS creates sensations of flashes of light.Called phosphenes, these flashes appear tomove when the pulses are delivered to thescalp overlaying visual area MT, but theyare stationary when TMS is delivered to thescalp that overlays area V1. By applyingseparate TMS pulses asynchronously to areaV1 and area MT, Pascual-Leone and Walshobliterated observers’ conscious experienceof the moving phosphenes that were ordi-narily evoked by MT stimulation. This result

4TMS offers a powerful tool for investigating cognitiveor perceptual neural circuitry (Pascual-Leone, Walsh, &Rothwell, 2000), including circuitry that supports vari-ous aspects of motion perception (e.g., Hotson & Anand,1999; Walsh, Ellison, Battelli, & Cowey, 1998). Whenthe TMS coil is positioned against an observer’s skull,a powerful, focused magnetic field hits and penetratesthe skull. The field penetrates superficial layers of thecerebral cortex, and can temporarily terminate or modifycurrently ongoing neural activity or alter neural activitythat is about to begin.

required the investigators to deliver TMS toarea V1 some tens of milliseconds after areaMT was stimulated. Presumably, the obliter-ation of motion perception is caused by a dis-ruption of a re-entrant: back-projections fromarea MT to area V1. A similar result wasreported by Beckers and Homberg (1992).

The preceding description of the motionpathway was based mainly on physiologi-cal and anatomical studies of nonhuman pri-mates. During the past decade, understand-ing of motion perception’s neuronal substratesin humans has been advanced significantlyby the use of brain imaging techniques, pri-marily functional magnetic resonance imag-ing (fMRI). This growing literature has iden-tified at least a dozen distinct regions in whichneurons respond to visual motion. These re-gions in the human brain stretch from the oc-cipital lobe to the frontal lobe (Culham, He,Dukelow, & Verstraten, 2001; Sunaert, VanHecke, Marchal, & Orban, 1999). Amongthe regions responsive to motion are area V1(which responds to almost any moving pat-tern, as well as to stimulus flicker) and theMT/medial superior temporal (MST) com-plex, located on the brain’s lateral surface nearthe junction of the occipital, parietal, and tem-poral lobes. This region, which we shall referto as MT+, responds weakly to flicker butstrongly to coherent motion, including opticflow patterns (discussed later). Other impor-tant motion areas include area KO (for kineticoccipital), which responds preferentially tomotion-defined boundaries, and area STS (forsuperior temporal sulcus), which is especiallyresponsive to patterns of motion that por-tray biological motion. As appropriate, brainimaging results are introduced throughout thischapter to clarify the neural computations thatmake motion perception possible.

With this overview in place, we can nowexplore several aspects of motion perceptionthat make it so crucially important for guid-ance of people’s everyday activities.



THE LIMITS OF MOTIONPERCEPTION

Motion Detection

Visual motion can be construed as an eventthat unfolds over space and time. Distilled tothe simplest case, motion involves a contin-uous change in the spatial position of a sin-gle object over time; this can be depicted inthe form of a space-time plot in which spa-tial position along one dimension is plotted asthe function of time in Figure 4.3. Intuitively,one might expect that the ease with whichthis kind of simple event can be seen woulddepend on the magnitude of the displace-ment over time and on the rate at which that

space (x)

time

(t)

time

(t)

time

(t)

time

(t)

time

(t)

f

space (x)

a

space (x)

future

past

present

space (x)

c

space (x)

b

d

space (x)

e

Figure 4.3 Illustrative space-time (x-t) diagramsthat are used to represent motion.NOTE: Panel A: Vertical bar moves rightward atconstant speed. Panel B: The space-time repre-sentation of the movement in Panel A. Panel C:Space-time representation for bar moving right-ward at higher speed than in Panel A. Panel D:Space-time representation for bar moving leftwardat same speed as in Panel A. Panel E: Space-timerepresentation for bar that moves rightward, stopssuddenly, and remains stationary. Panel F: Space-time representation for bar that moves rightwardand then abruptly reverses direction.

displacement occurred. There is truth to thisintuition. Consider, for example, the move-ment of a clock’s minute hand. People can-not see the clock hand’s gradual progression,but intuitively we know that it has movedbecause its position has changed over time.Motion perception, however, need not involveany kind of intuitive process; motion is a di-rect experience, uniquely specified by the vi-sual system (Exner, 1888; Nakayama, 1981;Thorson, Lange, & Biederman-Thorson,1969). But how sensitive is the system thatgenerates this experience? What is the lowerlimit for detection of motion? Measurementswith a single moving object show that anobject must traverse at least 1 minarc for asensation of motion to be experienced. (Thisdistance is approximately the edge-to-edgelateral separation between adjacent alphanu-meric characters on this page when viewedat arm’s length.) This value varies, however,with the object’s duration, velocity, and lumi-nance, as well as with the region of the retinastimulated. An up-to-date summary of this lit-erature is given by Tayama (2000).

Very seldom, however, are people calledupon to detect the motion of a single objectappearing in isolation. Instead, most everydaydetection of motion involves the detection ofan object’s (or a group of objects’) motionrelative to another object (or another set ofobjects). And at this challenge—detectingrelative motion—humans excel. We are vastlybetter at detecting relative motion thanwe are at detecting absolute motion (e.g.,Leibowitz, 1955). This is strikingly demon-strated in a study by Lappin, Donnelly, andKojima (2001) in which observers viewed anarray of three horizontally aligned “blobs”(circular, Gaussian luminance patches). In onecondition, all three blobs moved laterally backand forth in unison (in-phase motion), andin another condition the center blob’s direc-tion of motion was opposite to that of theflanking blobs (antiphase, or relative, motion).


The Limits of Motion Perception 127

Motion acuity—the smallest detectable mo-tion displacement—was actually better forthe antiphase condition compared with thein-phase condition. In other words, tiny mo-tion displacements visible in the antiphasecondition were impossible to see when therewas no relative motion. This finding dovetailsnicely with earlier results showing that hu-mans are keenly sensitive to shearing motion(Nakayama, 1981; Snowden, 1992), whichalso entails relative motion instead of over-all rigid displacement. This direction con-trast amplifies differences in motion vectorsin neighboring regions of the visual field. Onemust keep in mind that in the research de-scribed later, motion performance is measuredfor moving targets that appear within somebackground framework.

Intuition suggests that motion might ren-der an object less detectable. (Consider, forexample, the difficulty that a person experi-ences when trying to read a newspaper in amoving subway.) Under most circumstances,however, exactly the opposite is true. Espe-cially for objects with significant energy atlow spatial frequencies, motion can render anotherwise invisible object visually conspicu-ous (Robson, 1966). To experience this foryourself, hold an object between the uniformsurface of a wall and a light source in orderto create a large, very faint shadow on thewall. When the nonmoving dim shadow fadesto invisibility, move the occluding object andnotice how the shadow abruptly springs intoexistence. The familiar textbook example ofseeing the retinal blood vessels in your owneye by jiggling (i.e., moving) a light sourceon your sclera is another instance of motion’sability to reveal what otherwise would havebeen undetectable.

To characterize more precisely the optimalstimulus for the motion system, researchershave exploited the lower envelope principle.Barlow first called attention to the principle40 years ago and most recently framed it this

way: “Sensory thresholds are set by the classof sensory unit that has the lowest thresholdfor the particular stimulus used and are lit-tle influenced by the presence or absence ofresponses in the enormous number of otherneurons that are less sensitive to that stimulus”(Barlow, 1995, p. 418). Discussing various ap-plications of the principle to sensory systems,Parker and Newsome (1998, p. 242) noted that“in its pure form, the lower envelope principlemeans literally that a single neuron governsthe behavioral threshold. The development ofthe lower envelope principle has been verymuch a reaction by neurophysiologists to theformerly prevalent notion that single neuronsare inherently unreliable devices.”

In an attempt to apply the lower enve-lope principle to motion, Watson and Turano(1995) measured the minimum contrast atwhich observers could discriminate directionof movement. Their test stimuli were drawnfrom a family of patterns known as Gaborfunctions. Each Gabor function comprisesa sinusoidal grating that has been multi-plied, point by point, by the values of a two-dimensional Gaussian function. This multi-plication modulates the sinusoid’s contrast,producing a pattern whose contrast falls offsmoothly in all directions from a maximumat the pattern’s center.5 In search of the mosteasily seen moving stimulus, Watson andTurano independently varied the spatiotempo-ral characteristics of both the sinusoidal grat-ing and its modulating Gaussian function. Thestimulus yielding the lowest contrast thresh-old was a sinusoidal grating with spatial fre-quency of 3 cycles/degree and drift rate of5 Hz, with a width and height of 0.44 degrees

5These functions bear the name of Dennis Gabor, aHungarian engineer and applied physicist. Gabor won the1971 Nobel Prize in Physics for his work on wavefrontreconstruction in optical holography. In vision research,Gabor functions are used as stimuli; also, they are goodapproximations to the spatiotemporal receptive fields ofmany visual neurons.



visual angle and a duration of 0.13 s. Applyingthe lower envelope principle, these parametersdescribe the most sensitive of the neural filtersin the direction-extracting system.

Stimuli like those used by Watson andTurano (1995) and others have the advan-tage of limiting stimulus energy to particularbands of spatial and temporal frequencies. Al-though such grating patterns are often used toexplore motion perception, they bring somedisadvantages. For example, they confoundmotion with orientation and can appear tomove only in the directions perpendicular totheir orientation. To get around these limita-tions, researchers devised an entirely differentclass of motion stimuli, arrays of irregularlyspaced moving elements such as blobs or dots.These computer-generated stimuli are com-monly known as random dot cinematograms(RDCs), and there are several different algo-rithms for generating them. Distilled downto their essence, most RDCs consist of“signal” dots that move in a given direc-tion (or within a given range of directions)and are intermingled randomly with “noise”dots that move in random directions.6 Whenthe proportion of signal dots is high, dotsin the RDC appear to move coherently inthe general direction of those signal dots;when signal dots comprise only a small frac-tion of the RDC, the sense of motion co-herence is weak or, in the limit, absent en-tirely. Motion threshold is defined as the min-imum percentage of signal dots necessary fordetection of coherent motion. It should bestressed that the information supporting de-

6The terms signal and noise, commonplace in psy-chophysics, derive from engineering and communicationsciences. There, the task facing a biological or machinedetector is portrayed as the extraction of a message (sig-nal) from a stream of input, some portion of which (noise)is irrelevant or even antithetical to the extraction of themessage. These definitions enable investigators to char-acterize a detector’s sensitivity in terms of the ratio ofsignal to noise that just allows signal extraction.

tection performance in these stochastic stim-uli must be extracted globally: Observers can-not perform well simply by attending to asingle dot or to a restricted region of thedisplay.

Humans exhibit remarkable sensitivity tocoherent motion in RDCs. Under optimal con-ditions, observers can detect signal percent-ages as small as 5% (Scase et al., 1996), andthis holds for signal dots portraying trans-lational motion, rotational motion, and ex-pansion and contraction (but see Ahlstrom &Borjesson, 1996; Blake & Aiba, 1998). Ab-solute threshold values vary with display size(Burr, Morrone, & Vaina, 1998), dot density(Watamaniuk, 1993), and exposure duration(Watamaniuk, Sekuler, & Williams, 1989),but not with direction of motion (Gros, Blake,& Hiris, 1998). Visual cues that allow segmen-tation of signal dots from noise dots—such ascolor or contrast polarity—can substantiallyenhance detection of motion (Croner & Al-bright, 1997). Prior exposure, or adaptation,to strongly coherent motion in a given direc-tion temporarily elevates coherence thresh-olds for directions of motion within roughly40 degrees of the adapting direction, withthe elevation in threshold being largest at theadapting direction. Thresholds are also af-fected by higher-order variables such as vi-sual attention to a particular direction of mo-tion (Raymond, 2000; Raymond, O’Donnell,& Tipper, 1998), a point discussed later in thischapter.

In an influential series of experiments,Newsome and colleagues used RDCs to testmotion detection in monkeys. They recordedneural responses from single cells in areasMT and MST of the monkey’s brain whilethe monkey tried to detect motion (see areview of this work by Parker and Newsome,1998). In general, the monkey’s behavioralthreshold for detecting motion was very closeto the neural threshold derived for some in-dividual directionally selective neurons. The



correspondence between neural thresholdsand behavioral thresholds points to a tightlinkage between neural activity in areas MTand MST and the monkey’s performance, with“neural” thresholds corresponding closely tobehavioral thresholds (Celebrini & Newsome,1994; Newsome, Britten, & Movshon, 1989).

This linkage was further strengthened bythe finding that electrical stimulation of neu-rons in area MT (Salzman, Murasugi, Britten,& Newsome, 1992) or in area MST (Celebrini& Newsome, 1994) can bias a monkey’s per-ceptual report of motion direction in RDCs.Thus, direct stimulation of MT neurons tunedto leftward motion increased the probabilitythat the RDC would appear to move in thatdirection. Direct electrical stimulation of par-ticular clusters of MT neurons, then, was per-ceptually equivalent to the effect normallyproduced by an RDC moving in the neu-rons’ preferred direction with a particular co-herence level. Lesion studies lend additionalsupport to the idea that area MT participatesin motion perception. Lesions encompassingarea MT, area MST, or both areas producepermanent impairments in the ability to ex-tract coherent motion from RDCs (Newsome& Pare, 1988; Rudolph & Pasternak, 1999).Comparable deficits in motion sensitivity toRDCs are found in human patients with dam-age to an area of the brain homologous to areaMT in primates (Schenk & Zihl, 1997).

Trajectory Detection

Algorithms that generate RDC stimuli ordi-narily prevent any signal dot from movingin a constant direction throughout the entireanimation sequence. This is done to preventobservers from basing judgments on any sin-gle dot’s trajectory, instead forcing judgmentsto arise from the integration of many motionvectors. In the natural environment, however,there are many instances in which it is impor-tant to detect one object’s movement in the

presence of other, distracting moving objects,and, for that matter, despite temporary occlu-sions in the target object’s path. To take an un-pleasant example, a lion in hunting mode canvisually track the path of one particular zebrain a herd, even when all other members of theherd are moving about in random directions,and even though the target zebra is temporar-ily obscured by vegetation or other opaqueobjects. To learn how vision manages suchfeats, Watamaniuk, McKee, and Grzywacz(1995) measured observers’ abilities to detectthe presence of a single dot moving on a fixedstraight path in a field of noise dots whosedirections changed randomly over time. Thesignal dot was identical to the noise dots inluminance, size, and speed. For a stimulus du-ration of 500 ms, motion was detected 90% ofthe time, even when there were as many as 250noise dots. Under such conditions, the propor-tion of signal dots was minute: only 0.4%.

Watamaniuk and McKee (1995) also foundthat a single moving dot’s trajectory is easilyseen even when that trajectory is interruptedby a series of opaque occluders (Figure 4.4).Detection of the dot’s motion across threepath segments, each separated by an occluder

Figure 4.4 Diagram of display used byWatamaniuk and McKee (1995) to examineeffects of occlusion on trajectory detection.NOTE: See text for further details.



1 degree wide, was essentially as good as de-tection of the same motion over a single unin-terrupted path of equivalent length. Therefore,the motion signal generated by the movingobject was essentially unaffected by tempo-rary interruptions of that signal when the ob-ject was occluded. However, when the dot’strajectory was momentarily interrupted by adifferent kind of occluder, detection of themoving dot fell dramatically. Here, occlud-ing regions were filled with random-directionmotions (noise) similar to the random mo-tions in the display’s other regions. The dotdisappeared when it entered a noisy occlud-ing region and reappeared when it left thatregion. Within each noisy occluder the virtualpath of the signal dot was probably maskedor distorted by similarly directed motion vec-tors in the noise. These noise directions ledthe directional signal astray, reducing the pre-cision of matches from one segment of thedot’s trajectory to the next. (This is an in-stance of the so-called motion correspondenceproblem, which is discussed in a subsequentsection.) Because the reduction in detectabil-ity persisted even when noise-filled occluderslay in a depth plane different from the regionscontaining the trajectory, it seems that trajec-tory detection operates prior to the assignmentof depth to image components.

Grzywacz, Watamaniuk, and McKee(1995) proposed a model that can accountfor many observations on trajectory detec-tion. In their model, local connections amongmotion mechanisms enhance responses thatresult when mechanisms are stimulated in se-quence and roughly in the direction of theirdirectional tuning. These connections are spa-tiotemporal analogues to the spatial asso-ciation fields that have been implicated incontour integration (Field, Hayes, & Hess,1993; Geisler, Perry, Super, & Gallogly, 2001;Sigman, Cecchi, Gilbert, & Magnasco, 2001).From a perceptual perspective, the connec-tions postulated by Grzywacz et al. promote

what Gestalt psychology dubbed “good con-tinuation”: Perception will favor trajectoriesthat are smooth over trajectories that havelarge and abrupt changes in direction. Suchspatiotemporal preferences were demon-strated in Metzger’s (1934) observations withobjects moving on independent but intersect-ing paths.

Because of the spatiotemporally tuned lo-cal connections in Grzywacz et al.’s (1995)model, each successively stimulated motionmechanism will produce a response that islarger than that produced by the previouslystimulated mechanism. Eventually, the re-sponse grows large enough to be accuratelydetected, even though other motion detec-tors are responding to the background mo-tion noise. Because local connections involvemechanisms with a range of similar direc-tional tuning (spanning about ±30 degrees),the model can account also for the detec-tion of curved trajectories as well as straightones (Grzywacz et al., 1995; Verghese,Watamaniuk, McKee, & Grzywacz, 1999).

Grzywacz et al.’s (1995) trajectory networkmodel postulates facilitation of signals in se-quentially stimulated motion mechanisms, aresult observed by Verghese et al. (1999), whocompared the detectability of two kinds oftrajectories. In one, the moving elements fol-lowed a single continuous trajectory of lengthL; in the other, the same elements traced out nseparate trajectories, with a gap between suc-cessive segments. Because each segment wasL/n long, their summed lengths were the sameas the length of the single uninterrupted tra-jectory. Verghese et al. wanted to know howthe motion system would sum the motion sig-nals contained in these segments. For a com-putational benchmark, they drew upon the no-tion of probability summation. In its simplestform, probability summation predicts that de-tection of a signal should be equal to the sumof the square of the number of independentstimulus elements (Graham, 1989; Watson,



1979). In Verghese et al.’s experiments, thenumber of independent stimulus elements isn. This computational rule gave a good ac-count of detection when trajectory segmentswere so short that each segment was likelyto be detected by only a single mechanism.With longer trajectories, however, probabilitysummation failed badly. For example, a single200-ms trajectory was approximately threetimes more detectable than were two spatiallyisolated 100-ms trajectories presented one af-ter another, which kept L constant. Thus, thedetection of an extended trajectory cannotbe explained by activation of a series of inde-pendent motion detectors whose outputs aresummed linearly. Instead, the result pointsto significant spatiotemporally tuned interac-tions among local motion units, of the kindpostulated in Grzywacz et al.’s trajectory net-work model.

Motion Discrimination: Direction,Speed, and Coherence

Having highlighted key determinants ofmotion detection, we turn now to motion dis-crimination. In tasks used to measure discrim-ination, observers must not only detect themotion (i.e., see that motion is present) butalso judge one or more of the motion’s essen-tial characteristics, such as direction, speed, orcoherence.

In experiments on motion perception, si-nusoidal grating stimuli were used to inves-tigate discrimination of direction of motionat or near the contrast detection threshold forthat moving stimulus. In general, the con-trast at which observers could just detectthe presence of a moving stimulus was alsosufficient to identify its direction of motion(Derrington & Henning, 1993; Levinson &Sekuler, 1975; Watson, Thompson, Murphy,& Nachmias, 1980). Such results suggestedthat visual mechanisms that extract motionsignals are labeled for direction. In this

context, a neural mechanism is said to belabeled for some elementary sensation, Φ,if activity in the mechanism is sufficient togenerate the experience of Φ (Watson &Robson, 1981). Although much has beenlearned about motion from experiments usinggrating stimuli, an exclusive reliance on suchstimuli encounters substantial limitations. Ifan observer cannot see the ends of the grat-ing (the usual case in motion experiments),unambiguous motions can be produced onlyin the two directions orthogonal to the grat-ing’s orientation, an ambiguity known as theaperture problem (discussed later). Thus, cre-ating other directions of motion requires achange in the orientation of the grating, thusconfounding direction and orientation. Partlyto avoid this potential confound, Westheimerand Wehrhahn (1994) measured direction dis-crimination for a single moving spot. Theyfound that at high speeds (30 deg/s) directiondiscrimination was equal to that for orienta-tion discrimination of a static line of the samelength as the distance traveled by the movingspot and presented for the same duration. Thissuggests that at high speeds the moving spotcreated an oriented smear (a virtual line) onwhich observers could have based their judg-ments (work by Geisler, 1999, makes a simi-lar point). This potential problem is avoided,however, by using RDCs in which the direc-tions of individual dots change frequently orin which each dot has a limited lifetime.

Williams and Sekuler (1984) created RDCsin which individual elements were assignednew directions of motion in each frame. Whendirections were drawn from a distribution ofdirections spanning 180 degrees or less, ob-servers saw the entire fields of dots move inthe direction of the distribution’s mean, eventhough the random movements of individualdots were also visible. This meant that the mo-tion mechanisms responsible for this globalmotion integrated direction information overmuch of the motion display, if not over the



entire display. Using a RDC in which each dotchose a new direction of motion each framefrom a distribution of directions spanning30 degrees, Ball, Sekuler, and Machamer(1983) determined that the direction tuningof these motion mechanisms is broad: Twodirections of motion had to be separated by120 degrees before they no longer stimulatedthe same mechanism.

With such broadly tuned motion mecha-nisms to draw upon, how well can observersdiscriminate small differences in direction?To answer this question, Watamaniuk et al.(1989) measured direction discrimination forfields of random dots. When all dots movedin the same direction, observers reliably dis-criminated directions that differed by only1 degree. Moreover, this threshold was rel-atively resistant to the inclusion of randommotions that potentially interfere with signalextraction. When similar measures were madewith RDCs whose elements moved in a rangeof directions, once the range of directionsin the RDC exceeded 30 degrees, directiondiscrimination thresholds increased with di-rection range. In addition, direction discrimi-nation improved as exposure duration length-ened, up to at least 300 ms to 400 ms, and assize of the stimulus increased, up to at least a10-degree diameter (Watamaniuk & Sekuler,1992; Watamaniuk et al., 1989). Thus, the mo-tion system is robust in the presence of randommotions, and it can produce precise discrim-inations using mechanisms that are broadlytuned.

RDCs can generate motion percepts simul-taneously on two different spatial scales. Inparticular, the perception of global flow cancoexist with the perception of the small ran-dom motions of individual dots. Watamaniukand McKee (1998) showed that direction in-formation was encoded independently on thetwo spatial scales, global and local. In theirRDCs, a single central dot moved in a con-stant direction while the remaining 100 to

150 dots were assigned a new direction ofmotion in each frame from a distribution span-ning 160 degrees. The direction of global flowand the direction of the constant-direction dotwere always similar, but both varied fromtrial to trial. After a brief presentation of theRDC, a tone told the observer which motion,global or local, was to be judged. Under theseconditions, observers judged either directionjust as well as if they had been told in ad-vance which direction to judge. This suggeststhat visual information on different spatialscales is processed simultaneously with littleinterference.

As indicated earlier, motion detection isisotropic. For direction discrimination, how-ever, performance is anisotropic, varyingstrongly with the test direction. Discrimina-tion thresholds are lowest (i.e., performance isbest) for directions at and near the “cardinal”directions of up, down, left, and right (Gros,Blake, & Hiris, 1998). This oblique effectcould reflect a disproportion in the numberof neurons tuned to cardinal directions; al-ternatively, it could arise from narrower di-rectional tuning for neurons tuned to cardinaldirections. The absence of an oblique effectfor the detection of motion suggests that thesecond interpretation is correct, but the ques-tion remains open.

Direction Change

In the natural world, objects often change di-rection, and responses to such changes canbe quite important. For example, directionchanges by inanimate objects may result froma collision; direction changes by living crea-tures may convey biologically significantinformation, such as the information pro-duced by a series of hand gestures. Direction-tuned neurons in area MT of primate cortexhave been shown to be efficient encoders ofsuch changes (Buracas, Zador, DeWeese, &Albright, 1998). Synchronized changes in



direction of motion among an array of smallobjects promote perceptual grouping of thoseobjects into a coherent shape (Lee & Blake,1999).

Dzhafarov, Sekuler, and Allik (1993) pro-posed a formal computational model for re-sponses to changes in speed and direction ofmotion. The model, which has been extendedby Mateeff, Genova, and Hohnsbein (1999),incorporated a process that normalized allstimulus velocity signals registered prior toany change.7 Velocity normalization is an in-stance of visual adaptation processes that re-duce redundancy in neural responses by mini-mizing the total neural activity elicited by anystimulus input (Barlow, 1990). For Dzhafarovet al.’s implementation, assume that somestimulus has an initial velocity V0, whichchanges abruptly to a different velocity, V1.As a result of normalization, this change fromV0 to V1 is detected as though the changewere the onset of an initial motion with ve-locity |V1 − V0|. In other words, the actualvalue of V0 is irrelevant; all that matters is theabsolute difference between V1 and V0. Thisbasic computation successfully predicted ob-servers’ speeds of response to change with aconsiderable variety of values for V0 and V1.A. B. Sekuler and R. Sekuler (1993) exam-ined this normalization process further. In anattempt to disrupt the extraction of informa-tion about V0, they injected transients such

7Normalization refers to various linear operations thattransform some data set, D, into a new set, D′, while pre-serving particular types of numerical relationships amongthe set’s members. Of two common forms of normal-ization, subtractive normalization of D can be describedas D′ = D − k; divisive normalization is represented byD′ = D/k. Subtractive normalization, which is the formused in Dzhafarov et al.’s model, preserves the relativemagnitudes of members of D; divisive normalization pre-serves the proportional relationships among members ofD. Heeger (1994; Heeger, Simoncelli, & Movshon, 1996)describes another type of normalization, which operatesin the visual cortex to suppress partially the responses ofindividual neurons.

as temporary occlusion or disappearance intothe trajectory prior to the change to V1. Byinjecting a transient at various times duringand after V0, they were able to interrupt orfreeze normalization. Even a relatively briefdisappearance of the moving target reset nor-malization entirely, erasing all the velocity in-formation that had been extracted up to thatpoint.

Speed Discrimination

During their daily routines, humans frequentlymake judgments about the speeds of mov-ing objects. Consider, for example, the sim-ple acts of judging the speeds of nearby carswhen changing lanes on the highway or run-ning across the lawn to intercept a small childwho is crawling toward a street. How do hu-mans judge speed in these kinds of situations?Because objects moving at different speedscover different distances in any given tem-poral interval, observers could use distancetraveled as a cue to speed. Alternatively, thetime needed to travel some criterion distancemight also be used as a cue to speed. So howcan one measure speed discrimination with-out confounding influences of time and dis-tance? The usual experimental approach is torandomize presentation time over a range solarge that duration and therefore distance cuesbecome unreliable (McKee & Watamaniuk,1994).

Speed discrimination thresholds are typ-ically presented as Weber fractions (�V/V)that specify the proportional difference inspeed needed to produce reliable discrimina-tion. The smallest increment in speed that canbe reliably detected (�V) is divided by themean or base speed (V). Most studies havereported Weber fractions in the range 0.04to 0.08 with various types of stimuli, includ-ing moving lines, dot fields, and sinusoidalgratings (Bravo & Watamaniuk, 1995; Brown,1961; De Bruyn & Orban, 1988; McKee, 1981;



Nakayama, 1981; Orban, de Wolf, & Maes,1984; Pasternak, 1987; Turano & Pantle,1989; Watamaniuk & Duchon, 1992). TheWeber fraction’s constancy means that thesmallest detectable increment in speed in-creases with the base or starting speed.

Although the Weber fraction for speed dis-crimination is fairly constant over a variety oftest conditions, perceived speed can be alteredby any number of stimulus parameters. Forexample, Katz, Gizzi, Cohen, & Malach(1990) reported that drifting stimuli that areonly briefly presented appear to move fasterthan do stimuli that are presented for longerdurations. A grating’s drift or movement ratetakes degree/second as its units, where“degree” signifies degrees of visual angle. Itis important to distinguish drift rate from arelated variable, temporal frequency, whichtakes units of Hz or, equivalentally, cycles/s.The relationship between a grating’s drift rateand the temporal frequency produced by thatdrift takes account of the grating’s spatialstructure:

Temporal frequency (Hz)= Drift rate(deg/s)

× Spatial frequency (cycles/deg)

Perceived speed of movement (perceived driftrate) varies with a grating’s spatial frequency,which takes units of cycles/degree. Sinusoidalgratings presented at lower temporal frequen-cies (and hence lower spatial frequencies) ap-pear to be slower than gratings moving atthe same speed but with higher temporal fre-quency (Diener, Wist, Dichgans, & Brandt,1976; McKee, Silverman, & Nakayama,1986; Smith & Edgar, 1991). Turning to othervariables that affect perceived speed, whengratings of different contrast move at thesame physical speed, a lower-contrast grat-ing appears to move more slowly (Stone &Thompson, 1992; Thompson, 1982). Further-more, objects seen in the periphery appear tomove more slowly than do foveal objects of

the same speed (Johnston & Wright, 1986;Tynan & Sekuler, 1982). Finally, the percep-tion of an object’s speed is adversely affectedat low luminance levels, which correspond torod-mediated, scotopic vision. Gegenfurtner,Mayser, and Sharpe (2000) showed that amoving object’s perceived speed is consid-erably slowed in rod-mediated vision, com-pared to its perceived speed in cone-mediatedvision. To understand the likely basis for thisresult, note that differences in constants of rodtime and cone time indicate that rods aver-age information over longer periods than docones. Gegenfurtner et al. speculated that therods’ extended temporal averaging attenuatesmotion signals that would be generated in de-tectors that are tuned to high velocities. Thereduction in such signals causes the reduc-tion in perceived speed under rod-dominatedconditions. Grossman and Blake (1999) foundthat the perception of biological motion andstructure from motion were also impaired un-der scotopic viewing conditions. Such find-ings have clear implications for driving safetyat night on poorly illuminated roads.

IDEAL OBSERVERS ANDMOTION ANALYSIS

Early studies established that the visual sys-tem was an extraordinarily efficient detec-tor of light (Hecht, Shlaer, & Pirenne, 1942;Rose, 1948). To assess the efficiency of visionwhen it performs other tasks, researchers of-ten turn to ideal-observer models. In its mostcommon form, an ideal observer comprises amathematical model of a theoretical observerwho has complete and perfect knowledgeof all relevant stimulus and task statistics;in addition, this theoretical ideal makes sta-tistically optimal decisions when transform-ing sensory information into psychophysi-cal responses. Ideal-observer models affordinteresting benchmarks for the fallibility of


Ideal Observers and Motion Analysis 135

human observers, in comparison with the the-oretical limit represented by an ideal observer.Ideal observer models have been used to com-pare human and ideal performance for taskssuch as detecting changes in spatial patterns(e.g., Barlow, 1978; Barlow, 1980; Barlow& Reeves, 1979; Burgess & Barlow, 1983;Burgess, Wagner, Jennings, & Barlow, 1981;van Meeteren & Barlow, 1981).

Random and unpredictable variability inthe stimulus limits an ideal observer’s perfor-mance because it subverts the observer’s oth-erwise perfect knowledge of the stimulus to bedetected. Such random variability is knownas noise.8 Ideal-observer models try to pre-dict how humans and ideal observers mightperform when each must extract informationfrom a noisy stimulus. Increasing stimulusnoise leads any observer—human as well asideal—to make more errors, such as failuresto detect the stimulus, misclassifications ofa stimulus, or declarations that a stimulus ispresent when it actually is not. How closelya human observer approximates the theoret-ical ideal defines the human observer’s effi-ciency. This statistic is given by the square ofthe ratio of human performance to ideal per-formance, where both performance measuresare expressed as d’ values (Chap. 2, Vol. 4 ofthis series). Thus, if a human observer’s per-formance is ideal, efficiency is 1.0. Detailed

8In virtually any psychophysical experiment, valuable in-formation can be gained from measurements made withstimuli to which various amounts of noise have beenadded (Pelli & Farell, 1999). Noise can assume differ-ent forms, depending on the observer’s task. For exam-ple, when the task involves detection of a static form, thenoise usually comprises independent, random luminancevalues added to each element in the stimulus display(Bennett, Sekuler, & Ozin, 1999; Gold, Bennett, &Sekuler, 1999); when the task involves detection of sym-metry in a pattern, noise may be introduced by randomlyaltering the position of each of the pattern’s elements(Barlow & Reeves, 1979); or, when the task requires iden-tification of global motion direction in an RDC, noise canbe generated by randomizing the directions comprisingthe RDC (Watamaniuk, 1993).

comparisons of human and ideal observerscreate valuable diagnostic opportunities foridentifying and quantifying components thatlimit human performance (Geisler, 1989).

In the first application of ideal-observeranalysis to visual motion perception,Watamaniuk (1993) devised an ideal-observermodel that discriminated the direction ofglobal flow generated in RDCs. In speciallyconstructed RDCs, the directions in whicheach dot moved over successive frames werechosen randomly, with replacement, from aGaussian distribution of directions. The un-usual algorithm for generating the movementsor elements made the stimulus noisy: The al-gorithm introduced a random discrepancy be-tween the actual directional content of thestimulus and the nominal, or average, direc-tional content represented by the sampleddistribution. To vary the magnitude of thisdiscrepancy, in different conditions direc-tions were drawn from one of five Gaussiandistributions with different standard devia-tions. The larger the standard deviation, thegreater the mean absolute discrepancy be-tween actual and nominal-direction informa-tion in the stimulus. This introduced randomsampling noise into the stimuli. Because theideal observer had to rely on its knowledgeonly of the nominal stimulus, introduction ofvariability (noise) into the actual stimulus re-duced the observer’s performance.

Direction discrimination was measured foreach direction distribution for a range of stim-ulus durations, stimulus areas, and spatialdensities of dots. As expected, human per-formance was always poorer than the idealobserver’s performance. Efficiency generallydecreased with increased stimulus area or den-sity and remained constant as duration in-creased from 100 ms to 500 ms. However,the data clearly showed that efficiency in-creased as stimulus noise increased, reach-ing averaged values of 0.35. This suggeststhat the human visual system was influenced



by the random noise less than the ideal ob-server was. Note that high efficiency doesnot mean high level of performance. Becauseefficiency is a ratio of human performanceto ideal performance, high efficiency can beobtained at any level of performance. Infact, Watamaniuk (1993) found highest effi-ciency for direction discrimination when av-erage performance was at a d’ of about 0.75,which translates to a percent correct discrim-ination of about 70%. Finally, Watamaniukidentified several factors that underminehuman performance, including the limitedspatial and temporal summation of humanvision.

Watamaniuk’s (1993) ideal observer wasdesigned to discriminate one motion from an-other, but a comparable ideal observer couldbe designed for another task: to detect motion.Because the visual system exploits a differentsource of neural information both for directiondiscrimination and for motion detection (Hol& Treue, 2001), specifying an ideal-observeranalysis for a new task can be far from a triv-ial matter. Barlow and Triparthy (1997) ap-plied an ideal-observer analysis to the detec-tion of coherent motion embedded in randommotion noise. Human as well as ideal ob-servers received two-alternative forced choicetests. They had to identify which of two in-tervals contained coherent motion rather thancompletely random directional noise. Noisewas introduced into the stimulus by makingthe frame-by-frame positioning of the coher-ently moving dots less precise. As precisiondeclined, efficiency increased, reaching val-ues of approximately 0.3. This result pointsto the relatively coarse spatial resolution ofhuman vision and shows that adding spatialposition noise affects the ideal observer morethan it affects the human. This coarse spatialresolution represents a low-pass spatial filter-ing operation, which, as Barlow and Triparthyspeculated, enhances sensitivity to naturallyoccurring motion.

Extending this analysis of motion detec-tion, Baddeley and Triparthy (1998) examinedsome limitations that might possibly under-mine the performances of human observers.Using a novel statistical procedure that exam-ined the frame-by-frame movements of eachdot, Baddeley and Triparthy determined thatwhereas an ideal observer bases decisions onthe motions of all dots in the display, humanobservers seem to use only a proportion ofthe dots in the display. The same frame-by-frame analysis allowed them to rule out otherpossible limiting factors, including the ideathat human observers differentially weighteddirectional information generated at variouslocations in the visual field.

OPTIC FLOW AND STRUCTUREFROM MOTION

Motion affords potentially powerful infor-mation about the three-dimensional shapesof moving objects as well as about an ob-server’s own movements within the environ-ment populated by those objects. In partic-ular, the movements of objects within anenvironment create spatiotemporal changesin the light distribution on the retina of astationary observer. Likewise, an observer’smovements through a stationary environmentchange his or her own retinal images. Suchspatiotemporal changes, whatever their ori-gin, are termed optic flow, and they consti-tute significant sources of visual information.For example, optic flow provides informationabout the speed, direction, and path of an ob-server’s movements; it can also provide in-formation about the three-dimensional struc-ture of the environment (Koenderink, 1986).In the natural environment, an otherwise cam-ouflaged object—such as an edible insect—stands out conspicuously when it moves rela-tive to its background. Any creature that pos-sesses the neural machinery to extract and reg-


Optic Flow and Structure from Motion 137

ister the presence and shape of this target cansecure a meal courtesy of visual motion.

Historically, research on optic flow hastended to bifurcate into distinct branches. Onebranch has focused on the use of optic flow insteering an observer’s movement and head-ing; the other branch has focused on the useof optic flow in revealing the shape and struc-ture of moving objects. In discussing the usesof optic-flow information, we will respect thisbifurcation, but it is worth remembering thatboth uses of optic flow arise ultimately fromthe same source.

Optic Flow Supports Perceptionof Heading

Conventionally, the patterns of retinal imageflow produced by self-motion are representedby an instantaneous velocity field, as illus-trated for simple translatory movement in thetop panel of Figure 4.5. Each vector signifiesthe velocity (direction and speed) in the reti-nal image of an environmental element. Forthe case illustrated, the observer’s gaze is as-sumed to coincide with the direction in whichthe observer is moving. This creates a radialpattern of optic flow in which the focus ofthe flow corresponds to the observer’s head-ing. Note that the representation contains noinformation about the physical attributes ofthe elements, such as their color, shape, andsize. Instead, they are treated as uniform en-tities, known as tokens. The lower panel ofFigure 4.5 represents the velocity field result-ing from an observer’s circular movement par-allel to the ground plane. This velocity fieldwould be generated, for example, on the retinaof a driver whose automobile made a smoothturn.

Although snapshot representations likethose in Figure 4.5 omit information suchas acceleration-produced changes or tempo-rary occlusions (Warren, Blackwell, Kurtz,Hatsopoulos, & Kalish, 1991), they are still

A

B

Figure 4.5 Optic-flow patterns.NOTE: Panel A: Optic-flow pattern produced whenan observer translates along a straight path; di-agram assumes that observer is directed towarddestination point. Panel B: Optic-flow pattern pro-duced when an observer translates along a curvedpath.SOURCE: From Warren, Blackwell, and Morris(1988). Used with permission.

useful in understanding optic flow’s possiblerole in steering and wayfinding. Koenderink(1986) provided a thorough and accessiblemathematical account of the optic-flow dy-namics that result from various basic typesof observer movement. Transformations gen-erated by such movement, no matter howcomplex, can be decomposed into four ba-sic components: translation, isotropic expan-sion, rigid rotation, and shear. Summing thesebasic components in varying amounts can re-produce the original complex optic-flow field.As a result, visual mechanisms specializedfor extracting these basic components couldgenerate signals that, in aggregate, would rep-resent the complex flow field and, therefore,an observer’s movements. The receptive fieldproperties of neurons in area MST make those



neurons well suited to extracting optic-flowinformation related to an observer’s heading.The participation of such neurons in naviga-tion has been confirmed empirically. Directelectrical stimulation of local clusters of neu-rons in area MST of monkeys altered the mon-keys’ judgments of heading direction (Britten& van Wezel, 1998). With human observersand human brains, functional neuroimagingreveals large areas of MT+ that respondto particular components of optic flow—such as either circular or radial motion—butnot to simple translation motions in a front-parallel plane (Morrone, Tosetti, Montanaro,Fiorentini, Cioni, & Burr, 2000).

Research into optic flow’s role in wayfind-ing has addressed two key issues. First, psy-chophysical experiments had to determinewhether observers could extract headinginformation from displays containing onlyoptic-flow information. Second, knowingthat such extraction was indeed feasible,researchers have tried to identify any condi-tions under which such optic-flow informa-tion is actually used. We consider these twolines of investigation in turn.

Accuracy of Heading Judgments Basedon Optic Flow

For modes of human locomotion such aswalking, running, skiing, and driving, Cutting(1986, pp. 277–279) estimated the accuracywith which people needed to assess their ownheading in order to avoid obstacles. For exam-ple, suppose a person who is 1.75 m tall andweighs 65 kg runs steadily at 5 m/s (approx-imately 11 mph). Assuming that the runnerhas normal reaction time, in order to swerveto avoid an obstacle such as a tree, the runnerwould have to judge his or her own headingto an accuracy of at least 1.6 deg visual an-gle. If the same person walked rather than ran,far less precise heading judgments would stillafford a margin of safety. For downhill skiing,in which velocities may be nearly three times

that of running, the margin of safety narrowsto 0.78 deg visual angle, which heightens ski-ing’s potential risk.

How do such theoretical estimates squarewith what human observers can actually do?Empirical measurements of such judgmentsshowed that with paths generated by puretranslation across a ground plane, human ob-servers achieve the performance level thatCutting stipulated (Warren, Blackwell, &Morris, 1988). Warren et al. created two-dimensional displays in which movements ofdisplay elements simulated an optic-flow fieldproduced by an observer’s movement. Whenviewing random dot displays that representvarious directions of self-movement throughthe environment, observers can judge direc-tion to an accuracy of about 1 deg visual angle(Warren et al., 1988). This level of visual pre-cision is maintained whether the observer’sgaze remains fixed or shifts with eye move-ments of the sort that observers might makewhile walking through a real environment.Furthermore, direction judgments are rela-tively robust in the face of visual noise addedto the otherwise consistent motions of simu-lated flow fields (van den Berg, 1992).

Most experiments on the accuracy of head-ing judgments have used displays that simu-late retinal consequences of simple, straightpaths through the environment. In an impor-tant departure from this approach, Warren,Mestre, Blackwell, and Morris (1991) sim-ulated self-motion on a curved path. Af-ter viewing random dot fields that simulatedsome curved path through the environment,observers were shown a distant target and hadto judge whether the path they had seen wouldhave taken them to the left or to the right of thetarget. Warren et al. measured heading thresh-olds for paths with varying radii of curvatureand densities of dots in the flow field. Pathswith typical curvature supported thresholds of1.5 deg; performance was unchanged by evendramatic reductions in the number of random



elements in the display. This finding carriestheoretical weight. Neural network-based ex-plorations predicted that a pair of elementsseen in two successive views would providevisual information sufficient to support per-ception of an observer’s curved path (Warren,Blackwell et al., 1991). As a public servicefor drivers and downhill skiers, we note thataccuracy of path judgments is severely de-graded at small radii of curvatures (sharperturns).

Under many conditions, then, observerscan judge their own paths quite accuratelybased on optic flow. However, examinationof this ability has been limited to movementover single, simple paths. Does optic flow playa role when people attempt to navigate com-plex, multilegged paths? To answer this ques-tion, Kirschen, Kahana, Sekuler, and Burack(2000) asked people to navigate computer-generated synthetic environments with andwithout salient optic flow. Using the arrowkeys on a computer keyboard to control theirown simulated self-movement, Kirschen’sparticipants made repeated trips over what hadoriginally been a novel path. Trips grew fasteras participants learned the environment’s lay-out. Because these test environments were se-ries of identically textured virtual corridorsand intersections, participants needed to con-struct some mental representation of the en-vironment in order to perform the task. Byvarying the rates at which the display was up-dated, the researchers created optic flow thatwas either smooth or choppy. The choppy con-dition created the impression of a series ofseparate still views of the environment. Theavailability of smooth optic flow promotedfaster learning of complex paths, mainly bypreventing disorientation and backtracking. Ina second experiment, participants navigatedwithin a virtual city-block environment, ex-periencing two different kinds of optic flowas they went. Smooth optic flow enhancedobservers’ ability to navigate accurately to

the remembered position of target objects.Kirschen et al. concluded that when othercues (e.g., distinctive landmarks) are not avail-able, optic flow can be a significant aid tonavigation.

Mere availability of reliable optic-flow in-formation does not guarantee that all ob-servers will be equally able to exploit suchinformation. Here we consider two classes ofobservers for whom the quality of headingjudgments is diminished.

Ball and Sekuler (1986) showed thathealthy older people in their 70s and 80s hadelevated direction discrimination thresholds.Although these measurements involved ran-dom dot motion in a front-parallel plane ratherthan heading-related judgments, the older ob-servers’ elevated thresholds might be a signof some general age-related decline in motionperception, particularly perception of multi-element displays and tasks requiring head-ing judgments. Warren et al. (1988) measuredheading thresholds for two groups of people,young (mean age about 20 years) and old(mean age of late 60s). Observers saw opti-cal velocity fields that would be produced byobserver translation or by observer movementalong a curved path. After each display, whichlasted about 4 s, a vertical line was presented,and observers had to judge whether the head-ing direction that they had seen would havetaken them to the left or right of the line.With straight paths, young observers’ head-ing thresholds averaged 1.1 deg visual angle,whereas older observers’ heading thresholdswere significantly higher, 1.9 deg visual angle.With curved paths, the corresponding thresh-old values averaged 1.4 and 2.9 deg visualangle. After ruling out ocular and other pe-ripheral causes, Warren et al. suggested thatthese substantial age-related declines in head-ing acuity resulted from changes in high-levelvisual processing.

Patients with Alzheimer’s disease, a pro-gressive, degenerative disease of the brain,



often have difficulty finding their way aroundtheir surroundings, even when those surround-ings are familiar. Two groups of researchershave connected this difficulty to subnormalprocessing of information contained in op-tic flow. If optic-flow information actuallyguides navigation, at least under some cir-cumstances, then failure to process that in-formation fully could produce spatial confu-sion and loss of one’s bearings. Rizzo andNawrot (1998) showed that patients with mildto moderate Alzheimer’s disease have partic-ular trouble extracting form or shape informa-tion from stimuli in which the form is definedby movement (shape from motion). More-over, Tetewsky and Duffy (1999) showed thatsome patients with Alzheimer’s disease areimpaired in extracting directional informationfrom optic-flow stimuli. Many of these samepatients showed correspondingly poor perfor-mance on a test of spatial navigation (wayfind-ing) ability.

Is Optic Flow Normally Usedto Guide Locomotion?

Observers’ abilities to exploit the informa-tion in optic flow in order to judge head-ing does not guarantee that such ability isactually used to control locomotion. To seewhether the information was used, Warren,Kay, Zosh, Duchon, & Sahuc (2001) allowedindividuals to walk freely in a very largeroom while wearing a head-mounted displaythat afforded wide-field stereoscopic visionof computer-generated imagery. At the sametime, head position was tracked, and head-position signals were used to update the com-puter imagery at 60 Hz. While viewing severaldifferent sorts of display images, observers at-tempted to walk as quickly as possible to a vis-ible target such as vertical line or a doorway.Under normal viewing conditions outside thelaboratory, walkers could base their locomo-tion on information other than the optic flowgenerated by their own movements. There-

fore, a walker could register the position of aseen target in egocentric coordinates and thenwalk toward that target’s position, trying to re-main centered on the target. Ordinarily, the op-tic flow generated by this egocentering strat-egy would be identical to the flow generated ifan observer used the flow itself to control loco-motion. Warren et al. broke this normal corre-lation by feeding participants optic-flow infor-mation that deviated systematically from whattheir movements alone would have generated.In a control condition, when no optic-flow in-formation was provided, observers’ paths in-dicated that they had walked in the egocen-tric direction of the target. Thus, they homedin on the target by keeping it centered withrespect to their bodies. In other conditions,optic flow was introduced, with the unusualarrangement that the focus of expansion wasoffset from the walking direction (by 10 degvisual angle). As additional optic flow wasintroduced into the displays—by adding tex-ture to floors and ceilings—observers’ navi-gation behavior changed dramatically. Now,instead of walking toward the target as ob-servers had been instructed to, they tended tofollow the optic flow, which was intention-ally misleading. It appears, then, that undernormal conditions the visual system dependson both optic flow and egocentric localization,using the two sources of information in a com-plementary fashion. Warren et al. noted thatwhen flow is reduced (e.g., on a grass lawnor at night), reliance on flow is also reduced,but that in environments that afford consid-erable optic flow (e.g., forested areas), loco-motor behavior is likely to be dominated byflow.

Optic Flow Supports Collision Avoidance

When an observer is on a collision course withan object, the object generates a characteris-tic spatiotemporal expansion on the observer’sretina. This fact holds equally well, of course,



for a moving observer and a stationary object,or vice versa. If the rate of movement is con-stant, the retinal angle subtended by the ob-ject grows nonlinearly in a nearly exponentialfashion.

Animals as dissimilar as fiddler crabs,chicks, monkeys, and human infants all tryto avoid looming patterns created artificiallyon television screens (Schiff, 1965). Thisis true even for newborn infants who havenever before encountered a looming stimulus.Apparently, learning plays little role in thisbehavior.

The rate of retinal image expansion speci-fies the time to collision, that is, the momentat which a moving observer would reach theobject. If at time t an observer D meters awayfrom the object starts moving steadily towardthe object, time to collision is given by traveldistance divided by travel rate, D/R. Recog-nizing that solving for time to collision wouldrequire information about both D and R, Lee(1980) suggested that vision exploited anotherdynamic source of information about time toarrival, which would make it unnecessary toknow either D or R.

Tau, the variable that Lee (1980) intro-duced, is the ratio between the current retinalimage size and the rate of change in that imagesize. If the visual system computed a valueapproximating tau, time to collision wouldbe given by tau’s reciprocal. Note that thiscalculation’s result is independent of objectsize.

The connection between imminent colli-sion and the rate of expansion of retinal imagesize holds for any moving creature that has itseyes open—including birds that fly along andthen dive at great speeds in order to catch a fishin the water below. Among the best studied ofthese diving birds is the gannet, a large web-footed seabird with a sharply pointed beak.Gannets fly along with their wings spreadwide until just before their diving bodieswould hit the water’s surface, at which point

they tuck their wings tight to their sides. Tim-ing is everything. If a gannet performs thismaneuver too late, the impact with the watercould be quite damaging; if a gannet folds itwings prematurely, its body will be buffetedby cross-winds that will alter the point of en-try into the water. Although we have no wingsthat need folding, the human visual systemcarries out similarly complex computations.For example, information about the rate of ex-pansion can be used to control the braking ofan automobile (Lee, 1976), the split-secondchanges in gait needed when running acrossrough terrain (Warren, Young, & Lee, 1986),or the various movements and adjustmentsof the hand that are required for catchinga ball (Savelsbergh, Whiting, & Bootsma,1991). This coupling between optical expan-sion and action is not performed consciously.People succeed in these tasks despite be-ing unaware—or being unable to articulate—what they are doing (Savelsbergh et al., 1991).

Although tau may be helpful in many cir-cumstances in which people must recognizecollision time, it cannot be the only effectivecue to collision. In fact, tau would fail un-der a number of conditions (Tresilian, 1999).For example, if the approach rate is not con-stant, tau evaluated at any single moment failsto give the correct collision time. Gravity-induced acceleration also undermines tau’susefulness as a predictor of when a fallingbody will strike some surface. Additionally,with a very slow approach to an object, the rateof image expansion could become so small asto drop below threshold.

Tresilian (1999) discusses tau’s limita-tions, and catalogs other cues that observerscould and, in fact, do use. Human observersseem to be quite flexible in making useof available information to solve the col-lision problem. Thus, Schrater, Knill, andSimoncelli (2001) showed that observers canestimate expansion rates in the absence ofoptic-flow information by using only gradual



changes in the scale of random texture ele-ments. In a simulated ball-hitting task, Smith,Flach, Dittman, and Stanard (2001) demon-strated that observers optimize performanceby adjusting the relative weights given to cuessuch as the approaching object’s angular sub-tense and rate of expansion.

It is both plausible and theoretically at-tractive to postulate that observers adjustcue weights to match task demands and toreflect the reliability and availability of var-ious cues. This theoretical proposition, how-ever, raises a series of theoretical questionsabout the control architecture that might beused to integrate task-specific processes(Hildreth, personal communication, May2001). For example, do observers always gen-erate the values of all potential informationsources and then use an optimization strategyto select the most reliable? Or do observersactually generate only one or two preselectedweights, based on the characteristics of the sit-uation and task? Such questions must be askedand answered in order to clarify the real valueof postulating observers’s flexibility in choiceof strategy.

Optic Flow Supports Perceptionof Object Structure

Kinetic Shape

In the laboratory, motion is a potent speci-fier of shape. This potency was demonstratedin studies by Regan (1989), who created dis-plays in which alphanumeric characters weredefined by clusters of dots moving in differ-ent directions. To illustrate, imagine a densearray of tiny dots, each of which moves. Nowsuppose that dots in a subset of those dots,which fall within the boundaries of a virtualshape, all move in the same direction, whiledots outside the shape’s boundaries move in adifferent direction. (It is important to realizethat the region of the virtual shape itself is not

necessarily moving; only the dots definingthat area move.) People readily see a figuredefined by those common motion vectors,and they can judge with excellent accuracythe shape of that figure. Called “kinetic formperception,” this ability is conserved in theface of substantial amounts of random motionnoise.

The perception of biological motion pro-vides another compelling example of vision’sability to recover object information from mo-tion. When an animal’s body moves, the bodydeforms; that is, various parts of the bodymove relative to one another. These charac-teristic relative movements are signatures ofnormal biological motion. In fact, when a hu-man moves with stiff or locked body joints,reducing the normal movement of body partsrelative to one another, the body’s movementlooks unnatural, artificial. Because normal bi-ological motion involves deformation of thebody, biological motion is classified as non-rigid motion. Although there are many nonbi-ological sources of nonrigid motion (e.g., themovement of a flag waving in the breeze), per-ception of human body movement has drawnthe most interest and research.

In studies of biological motion, the activ-ity and identity of an animate creature are ex-tracted quickly and compellingly from merelya dozen or so “light points” placed strate-gically on the creature’s body (Johansson,1973). In animation sequences that representthe points’ movements over time, no singlepoint conveys information about the object orevent being depicted. Individual points merelyundergo translational or elliptical motions, orboth. The lights promote perception best whenthey are placed on joints, the parts of thebody whose movements are most diagnos-tic. Perception of a biological organism thatis engaged in an activity requires global in-tegration of motion signals over space andtime. As a result, the perception of such an-imation sequences is literally the creation of



motion information (in the same sense thatperception of an object in a random-pointstereogram is the creation of binocular dis-parity information). Even a brief view of apoint-light display allows an observer to iden-tify the gender of the person in the display(e.g., Kozlowski & Cutting, 1977; Mather &Murdoch, 1994), the activity in which the per-son is engaged (Johansson, 1973), or the per-son’s emotional state (e.g., Brownlow, Dixon,Egbert, & Radcliffe, 1997). Human infants asyoung as 3 months of age can perceive bio-logical motion (e.g., Fox & McDaniel, 1982),as can adult cats (Blake, 1993). Perceptionof biological motion is remarkably robust inthat observers can readily discriminate bio-logical from nonbiological motion even whenthe points’ contrast changes randomly overtime or when the points are defined entirelyby texture and not by luminance (V. Ahlstrom,Blake, & Ahlstrom, 1997).

The perception of biological motion maybe mediated, at least in part, by unique mo-tion mechanisms. First, information specify-ing biological motion can be summed overmuch longer temporal intervals than can infor-mation for simple translational motion (Neri,Morrone, & Burr, 1998). Second, damage tospecific regions of the brain can impair per-ception of biological motion while leaving in-tact other forms of motion perception (Cowey& Vaina, 2000). Conversely, damage to otherregions of the brain impairs perception oftranslational motion but has no influenceon perception of biological motion (Vaina,Lemay, Bienfang, Choi, & Nakayama, 1990).Finally, functional neuroimaging (Grossman,Donnelly, Price, Morgan, Pickens, Neighbor,& Blake, 2000) has identified regions of thebrain, located on the posterior STS, that are ac-tive during the viewing of biological motion,but not during the viewing of the same lo-cal motion vectors scrambled in space. Thesespecialized regions are located anterior andsuperior to area MT+.

Kinetic Depth

In addition to object shape, motion informa-tion can also convey the three-dimensionalspatial structure of surfaces. A single viewof a stationary two-dimensional projection ofa three-dimensional object affords little un-ambiguous depth information. When the ob-ject is set into motion, however, its projec-tion can produce a clear impression of theobject’s depth and spatial structure. Wallachand O’Connell (1953) provided an early re-port of this phenomenon, which they dubbedthe “kinetic depth effect.” Today, the preferredterm for the general class of phenomena is“structure-from-motion,” which encompassesnot only the emergence of depth from mo-tion but also the generation of surfaces andother object-related properties. Specially con-structed motion displays have been crucial tounderstanding the computational and neuralbases of structure from motion. Figure 4.6 il-lustrates the construction of one such stim-ulus display. A computer is programmedto create a flat, two-dimensional projection

Two-dimensional

projection

Transparentcylinder

Figure 4.6 Two-dimensional projection (right)of transparent rotating cylinder (left).NOTE: The dots on the cylinder’s surface create in-termingled left- and right-moving dots in the two-dimensional projection.



of a revolving, speckled, transparent, verti-cal cylinder. Viewing this two-dimensionalprojection, an observer sees intermingledrightward-moving and leftward-moving dots(arising from front and rear surfaces of thecylinder). Additionally, the speeds of individ-ual dots will vary systematically across thedisplay, and the fastest-moving elements willbe projections of dots that lie closest to theleft and right edges of the cylinder’s projec-tion. This two-dimensional projection typi-cally produces a strong percept of structurefrom motion, giving rise to a bistable percept:a cylinder whose direction of rotates alternatesbetween front leftward and front rightward(Andersen & Bradley, 1998).

In humans, the ability to exploit motionas the source of three-dimensional shape in-formation seems to develop very early. Us-ing a habituation technique, Arterberry andYonas (2000) showed that infants as young as8 weeks can perceive three-dimensional ob-ject shapes defined only by optic flow. Thisearly access to structure from motion is fortu-nate because mechanisms that extract static-form information are relatively immature inearly infancy.

Having seen the potential importance ofshape from motion, one can ask how that in-formation is actually extracted from the opticarray. Originally, it was thought that three-dimensional structure in displays such as theone in Figure 4.6 resulted from a series ofmatches, over successive samples, of the spa-tial locations at which elements are located(Ullman, 1984). Such an algorithm, in whichspatial positions were matched, ignored con-tributions from local motions, whose centralrole in structure from motion has been provenconclusively (Andersen & Bradley, 1998).

To explain the neural basis for structurefrom motion, Nawrot and Blake (1991b)devised a neural network incorporating in-teractions between stereopsis and direction-selective neurons (1991a). Because some

neurons in area MT are tuned not only formotion but also for stereo depth, Nawrot andBlake’s model incorporated units tuned tocombinations of disparity (e.g., near vs. far)and of direction of motion (e.g., left vs. right).The display shown in Figure 4.6 would acti-vate two different types of units in a networkof neurons: units sensitive to the conjunctionof near disparity and rightward motion andunits sensitive to far disparity and leftwardmotion. Mutual inhibition among units tunedto a given depth plane put them into an oppo-nent relationship, and activation of one type ofunit (e.g., near-depth leftward) tended to re-duce responses in units signaling the oppositedirection of motion (in this case, near-depthrightward). These within-plane interactionsensured that activity at a given depth planewas associated with only one direction of mo-tion. In addition, units tuned to the same di-rection of motion in near- and far-depth planesalso exerted mutual inhibition on one another.These between-plane interactions promotedsegregation of different directions of motioninto different depth planes. Together, these op-ponent arrangements keep the network fromgenerating contradictory percepts within anylocal area, just as in the physical world anysingle, local region on the cylinder’s surfacecannot simultaneously move leftward andrightward.

Nawrot and Blake (1991b) assigned theirunits nonlinear stimulus-response activationfunctions. This nonlinearity, together with ex-citatory connections among similarly tunedunits, allows structure from motion to buildup over time, which is consistent with humanobservers’ experiences with many such dis-plays. For example, under many conditionsthe full structure of a display such as the cylin-der in Figure 4.6 can take as long as a sec-ond or more to emerge fully. Similar coop-erative interactions among like-tuned neuralunits have been implicated in other aspects ofmotion perception (Nawrot & Sekuler, 1990;


Motion Transparency 145

Williams, Phillips, & Sekuler, 1986). Finally,the model’s combination of noise (moment-to-moment variability in response) and op-ponent organization caused it to reproduceanother notable feature of human perceptionof structure from motion: a perceptual bista-bility produced by ambiguous displays suchas the cylinder. Confirming this general idea,analogous bistability can be seen in the be-havior of MT neurons in the brains of mon-keys who view ambiguous two-dimensionaldisplays such as the projection of the cylinder(Andersen & Bradley, 1998).

Nawrot and Blake’s (1993) model makessome interesting predictions. First, becausekinetic depth and stereoscopic depth are com-puted within the same neural network, themodel predicts that it should be possible tocreate stimulus conditions in which the twoforms of depth—kinetic and stereoscopic—will produce equivalent perceptual experi-ences. That prediction was confirmed in aseries of experiments. In one experiment, ob-servers viewed two successive animations de-picting a rotating sphere. In one display, theimpression of a three-dimensional sphere wascreated solely by kinetic depth; in the otherdisplay, retinal disparity was used to portraythe sphere. For some disparities, these twodisplays were indistinguishable under forced-choice testing. A related experiment revealedthat a weak sense of depth from retinal dis-parity could be reinforced or, alternatively,canceled by a strong kinetic-depth stimulusthat itself contained no explicit disparity in-formation. This finding, too, points to a com-mon neural substrate for kinetic depth and dy-namic stereopsis. A kinetic-depth stimulus’scapacity to bias stereoscopic depth is compa-rable to the effect of direct, localized electri-cal stimulation of direction-selective MT neu-rons. In monkeys, stimulation of area MT hasbeen shown to bias perceptual judgments ofdepth (DeAngelis, Cumming, & Newsome,1998).

MOTION TRANSPARENCY

When an observer views a display in whichtwo or more velocities are spatially intermin-gled, the percept can be one of transparency(two or more separate patterns movingthrough one another) or coherence (patternscohering and moving in a single direction).Gibson, Gibson, Smith, and Flock (1959) re-ported that observers could detect two over-lapping planes in a display in which the planesmoved at different speeds. Subsequently,Andersen (1989) demonstrated that observerscould accurately identify the presence of upto three planes in similar displays with a du-ration of 2 s. When duration is reduced toonly 250 ms, observers can distinguish be-tween a display comprising as many as fivetransparent sheets of dots moving in differ-ent directions from a display of dynamic ran-dom noise (Mulligan, 1992). Furthermore,with translatory movements the process ofsegregating a display into different planes isfast: With just a 60-ms exposure, observerscan correctly identify a two-plane stimulus.However, if more complex motion patternssuch as expansion/contraction and rotationsare superimposed and presented for brief du-rations (85 ms followed by a random noisemask), observers cannot accurately identifythe component motions (De Bruyn & Orban,1993).

In transparent displays, the dots definingeach separate plane usually move in phase,as an intact pattern in a given direction andspeed. However, transparency can also beobserved if individual dots in the displayalternate between two different speeds (Bravo& Watamaniuk, 1995). In these cases, trans-parency is perceived, as evidenced by ob-servers’ abilities to judge accurately thespeeds of individual component motions.

Although two spatially superimposed setsof dots moving in different directions can gen-erate perceptual transparency, the two sets of



motion signals do not operate without mu-tual interactions. Specifically, the perceiveddirections of the two sets of dots appearto be further apart than they actually are,a phenomenon referred to as motion repul-sion (e.g., Hiris & Blake, 1996; Marshak &Sekuler, 1979). If the directions of the super-imposed sets of dots differ by less than ap-proximately 90 deg, the perceived directionof each motion will be pushed away fromthe direction of the other motion. The mag-nitude of this repulsion effect depends onthe speeds of the component motions and onthe density of the moving elements (Dakin& Mareschal, 2000; see also Lindsey, 2001).In addition, Snowden (1990) and Verstraten,Fredericksen, van Wezel, Lankheet, and vande Grind (1996) showed that sensitivity tothe direction of each motion plane decreasesas the speeds of the two motion planes be-come more similar. Thus, although the perceptof transparency shows that vision segregatescomponent motions from one another, whenthe two motion signals are sufficiently simi-lar, each can influence the perception of theother.

The percept of transparency requires a sin-gle local region to contain more than a singledirection of motion. But what is the size ofsuch transparency-producing regions? Qian,Andersen, and Adelson (1994) addressedthis issue by generating direction-balanceddisplays in which dots moving in oppositedirections (left and right) were either paired,so that a dot from each set was in closespatial proximity to the other, or unpaired.The observer’s task was to judge the de-gree of transparency in the display. Whenthe paths of the briefly presented dot pairscrossed or when they were separated ver-tically by 0.2 deg or less, the percept oftransparency was abolished. In both cases inwhich transparency is abolished, the oppo-sitely moving dots were close to one another.A similar lack of transparency is perceived if

two oppositely moving sine-wave gratings aresuperimposed.

Transparency can be restored to dot andgrating displays if the components are suffi-ciently separated in depth, or if componentgratings differ in spatial frequency by abouttwo octaves.9 Curran and Braddick (2000)refined this work, showing that if paired dotsmoved in directions separated by just 60 degto 120 deg, rather than in opposite directions,then the percept was that of coherent globalflow in the direction of the dots’ vector aver-age. Presumably, paired directions differingby only 60 deg to 120 deg does not triggerthe inhibition that would be generated hadthe directions opposed one another, that is,had they differed by 180 degrees. Similarly,Lindsey and Todd (1998) found that motionsignals embedded in random noise were moreeasily detected when the component motionsmoved at right angles to one another, ratherthan in opposite directions. These data areconsistent with the existence of a suppres-sive stage of motion processing in which op-positely tuned motion detectors inhibit eachother locally. When the directions of spa-tially proximate motions are opposite oneanother, directional signals are perfectly bal-anced, and mutual, direction-selective inhibi-tion results in no net perceived motion (Qian,Andersen, & Adelson, 1994). When spatiallyproximate signals differ by just 90 deg, theabsence of directional inhibition enables ob-servers to see the motion easily, as Lindseyand Todd found.

Neurophysiology provides further supportfor this hypothesis. Qian and Andersen (1994)recorded responses from V1 and MT neu-rons during presentation of paired or unpairedmotion stimuli and found that V1 cells

9A difference of two octaves means that the spatial fre-quencies differ by a factor of four. With a difference thislarge, the two gratings will be initially registered by differ-ent spatial frequency tuned visual mechanisms (Graham,1989).


How the Visual System Measures Motion: Three Problems to Solve 147

responded equally well to both types of stim-uli, but that MT cells responded better to stim-uli in which local motions were unpaired.Snowden, Treue, Erickson, and Andersen(1991) reported data that were consistent withthese results. But what about the brains ofhumans? Does the human motion complex(area MT+) behave as its nonhuman pri-mate homologues do? Does area MT+ exhibitmotion opponency? To answer this, Heeger,Boynton, Demb, Seidemann, and Newsome(1999) compared multiunit recordings fromneurons in monkey cortex to the fMRI acti-vation patterns generated in the human brain.A multiunit recording collects responses fromlocal aggregates of neurons, rather than fromindividual neurons in isolation. This approachwas meant to enhance comparability withfMRI measurements, which represent aggre-gated neural activity. Heeger et al. usedequivalent stimuli for both measures of neu-ral activity, and with both species. In humansas well as in monkeys, area V1 showed noevidence of motion opponency: Responses topaired and unpaired stimuli (dot patterns orgratings) were essentially the same. However,area MT in the monkey and its homologuein humans showed considerable motion op-ponency. In particular, the fMRI activation inarea MT+ was far stronger with nonpaireddot stimuli than with dot patterns in which lo-cal, opposite motions were pitted against oneanother. Thus, area MT+ is a site at whichdirection opponency could initiate the assess-ment of transparency.

HOW THE VISUAL SYSTEMMEASURES MOTION: THREEPROBLEMS TO SOLVE

This section provides an overview of threemajor problems that vision must solve in or-der to provide observers with useful accountsof motion in the visual field.

Direction Selectivity/Reichardt Detectors

The first computational account of motionperception arose five decades ago, from thecollaboration of Bernhard Hassenstein, abiologist, and Werner Reichardt, a physicist(Borst, 2000). Their product was a simplemultiplicative correlation detector made upof two oppositely tuned subunits. To under-stand the detector’s operation, imagine thata spotlight moves across the retina, succes-sively stimulating different groups of adjacentphotoreceptors one after another. To simplify,assume that the direction of the moving spotcaused the spot to fall first on photoreceptorA and then, after some delay �t, on photore-ceptor B. As a result, the luminance signalelicited from A precedes the signal generatedfrom B by �t. This delay depends on two vari-ables, the spatial separation between A and B,and the speed with which the spot moves. Nowto the detector’s circuitry: For one of the detec-tor’s subunits the luminance signal generatedin photoreceptor A is multiplied by a delayedluminance signal from a second neighboringphotoreceptor set, B. This basic operation isreplicated in the detector’s other subunit, butin mirror-symmetrical fashion: The two pho-toreceptors are switched, and the delay is nowapplied to the signal from the previously non-delayed photoreceptor. Because of the delays,a spot that reaches first A and then B generatesa response that is larger in the second subunitthan in the first; the same spot traveling at thesame speed but in the opposite direction gen-erates a response that is larger in the first sub-unit than in the second. In other words, the nu-merical difference between the two subunits’responses is directionally selective: Motion inone direction generates a positive difference,whereas motion in the opposite direction gen-erates a negative difference.

The model’s simple circuit guarantees thatmotion sensitivity will reflect a stimulus’stemporal and spatial parameters, which is



certainly true of vision. In its first tests, themodel was applied to insect vision to exploitas a behavioral index of motion perception theoptomotor reflex of the beetle Chlorophanus.The model’s success inspired a good deal ofresearch, including work on higher animals.It also promoted the creation of other modelsthat performed a similar computation usingdifferent circuitry (Sperling & Lu, 1998).

Although the Hassenstein-Reichardt mo-tion circuit has many virtues, it also has oneproperty that could be considered a flaw:It fails to distinguish between two classesof stimuli that are physically quite differentfrom one another. In particular, the circuitwould give equivalent responses to (a) a spotthat moved smoothly with the proper veloc-ity from the receptive fields of one subunit’sreceptors to the receptive fields of the othersubunit’s receptors and (b) a spot that waspresented to one set of receptive fields, thenextinguished, and, after a delay, presented tothe other receptive fields. With proper delaysbetween presentations of the spot, this lat-ter sampled or stroboscopic motion stimuluswould be indistinguishable from its smoothcounterpart. Scaled up to an entire human vi-sual system, this perceptual “error” becomesquite important because it allows the sam-pled images that comprise film and video se-quences to mimic smooth motion. The resultof such a sampling process is known as “ap-parent motion,” a designation meant to con-trast with smooth or “real” motion. The qual-ity of apparent motion (i.e., how smooth themotion appears) varies with a number of pa-rameters, particularly the rate at which thestimulus is sampled in both space and timedomains. As the interval lengthens betweensuccessive frames of display, the samplingrate is said to decrease. Intuitively, as sam-pling rate increases—and successive framescome closer together in time—the appearanceof the sampled stimulus approaches that of asmoothly-moving stimulus.

Watson, Ahumada, and Farrell (1986) de-veloped a simple model that predicts whetherany spatial and temporal sampling rate wouldproduce the appearance of smooth motion.Their model defines a spatiotemporal rangeof each observer’s window of visibility. Theboundaries of this window, a region in jointspatial- and temporal-frequency space, definethe spatial- and temporal-frequency limits ofthe observer’s sensitivity to energy in the stim-ulus. When the stimulus is sampled in time,as for video or film or computer displays, thesampling process generates energy at tempo-ral frequencies in addition to the fundamen-tal frequency. A low sampling rate producesenergy over a range of low temporal frequen-cies; a high sampling rate produces energyover a range of high temporal frequencies.As a result, the higher the sampling rate, themore likely it is that the resulting energy willfall outside the window of visibility, whichrenders them invisible and perceptually in-consequential. Therefore, two stimuli—onesmoothly moving and the other representingsampled motion—will appear identical if theirspectra within the window of visibility areidentical; portions of their spectra that lie out-side the window are irrelevant. Using two dif-ferent strategies for sampling stimuli, Watsonet al. confirmed the essential validity of theirelegantly simple model.

Following Hassenstein and Reichardt,most studies of motion perception have ex-amined responses to drifting modulations ofluminance (or chromatic contrast). Thesestimuli, termed first-order stimuli or Fourierstimuli, would evoke responses in visualmechanisms that are responsive to spatiotem-poral variations in luminance or chromaticcontrast. Such stimuli correspond to a dom-inant species of spatiotemporal modulationthat are encountered every day, but such stim-uli do not exhaust the possibilities. Somestimuli, termed second-order or non-Fourierstimuli, would elude detection by such



mechanisms (Chubb & Sperling, 1988;Pantle, 1973). Nevertheless, perceptually,such stimuli elicit strong motion responses. Itis worth emphasizing that second-order mo-tion is not merely a creation of the laboratory.A well-camouflaged creature moving againsta background with the same texture markingsas the creature’s own will generate second-order motion only. The same is true for thewaving of tree branches when wind blowsthrough a forest, or wheat stalks waving ina wheat field.

Many psychophysical studies comparingfirst- and second-order motion have demon-strated clear distinctions between the two. Thedistinction between the two classes of motionstimuli has gained increasing theoretical sig-nificance from reports that localized lesions ofthe brain can selectively impair either first- orsecond-order motion perception, while spar-ing the nonaffected form (Vaina, Cowey, &Kennedy, 1999; Vaina, Grzywacz, LeMay,Bienfang, & Wolpow, 1998).

The Correspondence Problem

Some of the computational prerequisites ofmotion perception reflect computational ob-stacles to perception more generally. To ap-preciate this point, consider the problem ofmotion correspondence. It has long been rec-ognized that the proximal stimulus underspec-ifies the distal stimulus. In describing thisfact, Helmholtz (1866) noted that at any in-stant the distribution of light on the retina (theproximal stimulus) is consistent with an in-definitely large combination of stimulus ob-jects and patterns of object illumination (dis-tal stimuli). To resolve, or at least reduce, theproximal stimulus’s massive underspecifica-tion of the distal stimulus, the visual systemexploits various sources of supplementary in-formation, including constraints and regular-ities embodied in the physics of the naturalworld.

Although motion perception must over-come the underspecification that is commonto all perception, as a spatiotemporal pro-cess, motion faces some additional, uniquesources of underspecification. An early stageof motion extraction requires correspondencematching. Some local pattern—for example,a luminance perturbation centered on retinalcoordinates x0, y0—is detected at time t0 andmatched at later time, t1, to the pattern lo-cated at new coordinates, x1, y1. As this de-scription implies, motion depends on a matchor correspondence in space and time. AsAttneave (1974) pointed out, when a stim-ulus comprises more than one element thatmoves, identification of correspondences overtime becomes a challenge.

Measurements of element positions in suc-cessive samples insufficiently determine themotion correspondences of the samples.10

Simple demonstrations of this point are shownin Figure 4.7. If there are n elements in eachof two samples, then there are at least n!sets of correspondences consistent with thesamples. (This calculation assumes that oneand only one element in the second sample ismatched to each element in the first sample.)Dawson (1991) argued that to resolve this mo-tion correspondence, the visual system ex-ploits a trio of global spatiotemporal con-straints that mimic the properties of motionin the natural world. (“Global” implies thatthe constraints are applied simultaneously tothe entire field of dots, or to large portionsof the field.) These constraints are known asthe nearest neighbor principle (minimize themean displacements between points matched

10To illustrate the correspondence problem, we have usedexamples in which stimuli are time-sampled at a relativelylow rate. For historical reasons, when such stimuli gen-erate the experience of motion, that motion is often des-ignated as apparent motion. It is important to note thatdespite our limited choice of illustrative examples, thecorrespondence problem confronts any biological motionsystem.



a

b

e

c d

gf

Figure 4.7 Illustration of motion’s correspon-dence problem.NOTE: In each panel, the three black squares rep-resent items presented in the first frame of a cin-ematogram, and the three white squares representitems in the second frame. Panels illustrate vari-ous ways that items from the two frames could bematched.SOURCE: After Dawson (1991).

between successive samples), the smooth-ness principle (because natural surfaces tendto vary smoothly, motion arising from thedisplacement of elements on such surfacesshould be as smooth as possible, minimizing

1 2 3 4

Figure 4.8 Four frames of a quartet display.NOTE: In each frame, two tokens are black, and two are gray; the quartet of tokens rotates rigidly by 45deg between successive frames. In the absence of color-based correspondence matching, the display’smotion would be ambiguous, as likely to move clockwise as counterclockwise. However, color-basedmatching generates consistent motion in a clockwise direction, and the probability of seeing motion inthe direction dictated by color matches increases with the perceptual difference between the two pairs oftokens.

abrupt changes in velocity among neighbor-ing elements), and the element integrity prin-ciple (because surfaces do not tend to pop inand out of existence, elements on such sur-faces should persist over time; i.e., one ele-ment should not split into two, and two ele-ments should not fuse into one).

By emphasizing global motion matching,Dawson’s (1991) model assumes that prop-erties such as chromatic or shape similaritiesexert only a minor influence on motion corre-spondence. Although some empirical resultshave been taken as justifying that assumption,we believe that the preponderance of evidenceleaves little doubt that motion correspondenceis powerfully influenced by the similaritybetween stimuli. For example, A. B. Sekulerand Bennett (1996) examined motion cor-respondence generated by stimuli of differ-ing calibrated discriminability. Manipulatingrelative phase—a crucial determinant of formperception—they assessed strength of corre-spondence produced by quartets of stimuli.Each quartet comprised four stimuli that wereevenly spaced on a virtual circle. To empha-size the figural relationships among membersof the quartet, we can designate the four stim-uli as A, B, A, B.

Figure 4.8 illustrates the entire quartet ro-tated rigidly over 45 deg between successivedisplay frames. Consider the four succes-sive frames represented in that figure. If there



were no color-based correspondence match-ing, the motion produced by the display wouldbe ambiguous (i.e., as likely to move clock-wise as counterclockwise). However, color-based matching would generate consistentmotion in a clockwise direction. The elementsin A. B. Sekuler and Bennett’s (1996) quar-tets were compound sinusoidal gratings, eachcomprising a fundamental component and oneharmonic component, at twice the frequencyof the fundamental. A. B. Sekuler and Bennettmanipulated the contrast and relative phase ofthe harmonic component in order to generatepairs of stimuli that varied in discriminability.When such stimuli were placed in quartet con-figuration (as in Figure 4.8), the probability ofseeing motion in the direction dictated by fig-ural matches varied with the perceptual differ-ence between the two pairs of compound grat-ings. Tse, Cavanagh, and Nakayama (1998)described a novel class of displays in whichfigural parsing and matching are requisites forperception of motion. (“Parsing” refers to thesegmentation, spatial isolation, and identifi-cation of separate stimulus elements.)

Before leaving the topic of motion corre-spondence, we should take special note ofbistable motion sequences (i.e., stimuli inwhich the correspondence problem has twoequally likely outcomes). An example of thisunusual situation is illustrated in Figure 4.9,which shows two successive frames of anima-tion sequence in which a pair of discs movesback and forth between two positions. Notethat the upper disc in frame 1 could correspondeither to the left-hand disc or to the right-handdisc in frame 2 (and likewise for the lowerdisc in frame 1). Because the two alterna-tive motion paths are exactly equal in length,the motion system has no basis for decidingwhich pattern of motions is correct: Motioncorrespondence is entirely ambiguous. Whenstimuli like this are viewed for an extendedperiod, people perceive both possible paths(Anstis & Ramachandran, 1987; Kramer &

1 2

Figure 4.9 Two successive frames of a displaythat illustrates feature matching in the solution ofa correspondence problem.NOTE: The upper disc in frame 1 could correspondeither to frame 2’s left-hand disc or to its right-handdisc (and likewise for the lower disc in frame 1).Because the two alternative motion paths are equalin length, the motion system has no basis for de-ciding which pattern of motions is correct: motioncorrespondence is entirely ambiguous. When suchstimuli are viewed for an extended period, bothpossible paths are seen in alternation.

Yantis, 1997; Pantle & Picciano, 1976;Ramachandran & Anstis, 1985), with eachpattern of motion dominating for a few sec-onds and then giving way to the alternativepattern. Such ambiguous motion displays areuseful tools for studying the stimulus factorsthat influence solutions to the correspondenceproblem (Francis & Kim, 1999; Yu, 2000).Multimodal stimuli, such as auditory sig-nals, can influence the perception of bistablemotion displays (Lewkowicz, in press; R.Sekuler, Sekuler, & Lau, 1997; Watanabe &Shimojo, 2001).

The Aperture Problem

As described earlier, local motion is extractedfrom the retinal image by neurons in areaV1. The receptive fields of such neurons canbe construed as apertures, spatially delim-ited windows within which neurons registerthe presence of motion in a given direction.If an extended moving edge or bar is seenwithin such an aperture, then regardless of the



direction in which the edge or bar actuallymoves, the neuron will respond as though theedge or bar were moving perpendicular tothe orientation in the neuron’s receptive field(Bradley, 2001). To take a specific instance,imagine that a smooth, featureless, verticallyoriented bar moves obliquely (up and to theleft) at a constant velocity. Any small recep-tive field positioned along the contour’s lengthsignals only the motion component that is per-pendicular to the contour’s orientation, in thiscase the leftward motion component. Becausethe contour extends beyond the boundariesof the small receptive field, the contour’s up-ward motion component produces no infor-mation that changes with time (Pack & Born,2001).

Because each directionally selective (DS)unit sees only what is happening within itsown receptive field, the resulting signals fromDS units are necessarily ambiguous. This am-biguity, now known as the aperture problem,was pointed out more than 60 years ago byHans Wallach (Wuerger, Shapley, & Rubin,1996). Figure 4.10 illustrates what Wallachhad in mind. The circular area in the figure rep-resents the receptive field of a DS neuron. Itspreferred direction of movement is rightward.As the vertical edge of a large black line movesrightward at an appropriate speed through thereceptive field, the neuron responds strongly(Panel A). However, this is not the only direc-tion of movement that could evoke such a re-sponse. As long as the black line is large com-pared to the aperture (the receptive field), thesame local spatiotemporal event—movementat the same velocity (meaning the same speedand direction)—could be generated by anynumber of other combinations of direction andspeed, some of which are suggested in PanelsB and C. This equivalence means that the neu-ron’s response is inherently ambiguous. Be-cause the neuron’s view of the world is limitedto the confines of its receptive field, the neu-ron responds exactly the same way to each of

timeactual

direction

A

B

C

registereddirection

Figure 4.10 Illustration of the aperture problem:The ambiguity of directional information within areceptive field aperture.NOTE: The circular area represents the recep-tive field of a directionally selective neuron tunedto rightward motion. As the vertical bar movesrightward through the receptive field, the neu-ron signals rightward motion (Panel A). How-ever, movement of the same bar downward andto the right (Panel B), or movement upward to theright (Panel C) also generates rightward movementwithin the receptive field. The equivalence of theseand other directions of actual movement rendersthe neuron’s response ambiguous.SOURCE: Modified from R. Sekuler and Blake(2001).

these different velocities of movement. As aprelude to explaining the visual system’s strat-egy for resolving this ambiguity, consider oneside effect of this ambiguity:

Movement seen within any single recep-tive field could have arisen from a vari-ety of distinctly different visual events. Thisambiguity opens the door to numerous il-lusions of perceived motion. With some ofthese, observers watch as a line or bar movesthrough an aperture whose shape strongly in-fluences the perceived direction of movement.For example, suppose that an L-shaped aper-ture is cut in a piece of paper. Suppose alsothat a long oblique line (at 45 deg) behindthe aperture moves steadily straight down-ward (Figure 4.11A). Initially, an observersees an oblique line that moves downward.Then, once the line reaches the base of the



time

a

b c

Figure 4.11 Wallach’s demonstration of the in-fluence of aperture shape on perceived motion.NOTE: Panel A: Motion of an obliquely orientedline into, through, and then out of an L-shaped aper-ture. Panel B: The line initially appears to movevertically downward; then, when the line enters theaperture’s horizontally oriented portion, its appar-ent direction abruptly changes to horizontal. PanelC: See text for an explanation.SOURCE: Modified from R. Sekuler and Blake(2001).

L-shaped aperture, the movement abruptlychanges (Figure 4.11B). Now the observersees a rightward-moving oblique line.Wallach investigated this and several dozenrelated aperture illusions (Wuerger et al.,1996). To appreciate the basic elements of hisapproach, return to the L-shaped aperture.

One can approximate the L-shaped aper-ture’s upright portion by a set of colinearcircular receptive fields (Figure 4.11C). Anoblique, downward-moving line would tra-verse each of these fields, one after another.However, this pattern of stimulation under-specifies the distal stimulus; in fact, the pat-tern of stimulation could have arisen fromany number of distal stimuli. For example,it could have been produced by several dif-ferent but similarly oriented lines, each ofwhich traverses just one receptive field andthen disappears, just as the next line appearsat the top of the next receptive field and be-gins its own descent. Or, the same patternof stimulation across all the receptive fieldscould have resulted, as it did in Wallach’s

demonstration, from a single line that movedfrom top to bottom, entering and then exit-ing one receptive field after another. Giventhe limited information available to it, howdoes the visual system select the scenariothat most likely was responsible for that in-formation? In the spirit of Gestalt psychol-ogy, Wallach proposed that the single per-ceptual choice made in such multiple-choicesituations tends to be the simplest global mo-tion. In this case, an observer sees a singlesteadily moving line rather than a series ofdifferent lines in succession. An alternativeview might portray the single perceptual out-come as the product of a Bayesian perceptualinference (Knill, Kersten, & Yuille, 1996).Choosing between these alternative accountsrequires additional, carefully designed empir-ical measurements.

Perceived motion, then, is not determinedsolely by local responses to stimulus ve-locity that are generated within separate re-stricted regions of the field. Instead, localmeasurements of velocity are integrated atsome place in the visual system at which thelocal velocity-related signals from area V1 arecollected and combined. Such a combinationof signals is the nervous system’s standardoperating procedure for resolving neural am-biguity in the signals of individual neurons.When the response of one neuron is ambigu-ous, the nervous system can diminish the am-biguity by aggregating outputs from a numberof differently tuned neurons. For visual mo-tion, the nervous system reduces ambiguityabout individual, local, spatiotemporal eventsby channeling outputs from the first stage ofdirection processing to its second stage of neu-ral processing in area MT.

Some neurons in area MT receive inputfrom V1 neurons with different preferred di-rections of motion. As a result, the directionalselectivity of these MT neurons differs quali-tatively from that of their predecessors in areaV1. Furthermore, these differences enhance



the importance of area MT’s contribution tothe perception of motion. To see this, returnto Wallach’s demonstration with two over-laid, moving gratings. When analogous two-component grating displays are imaged withinthe receptive fields of V1 neurons, the neu-rons tend to respond to the separate compo-nents. Some of Wallach’s psychophysical ob-servations foreshadowed these physiologicalresults. He showed observers patterns com-posed of two different line gratings such asthose shown Figures 4.12A and 4.12B. Whenthe two are overlaid, as in Figure 4.12B, theygenerate a series of diamond-shaped struc-tures. When the bars of the two gratings movedownward at the same rate, the display’s ap-pearance fluctuates between (a) a field of dia-monds that moves downward (Figure 4.12C)and (b) two line gratings that move at different

a

c

b

d

Figure 4.12 Superimposition of two diagonal,moving gratings (A and B) produces a diamondpattern whose motion is bistable.NOTE: When either moving grating is presentedalone, it appears to move in a consistent direction,which is indicated by the arrows in A and B. Whenthe gratings are superimposed, they cohere to forma single downward-moving pattern (Panel C), orthey appear to slide over one another, in two dif-ferent superimposed directions.SOURCE: Demonstration devised by Wallach. Dia-gram modified from R. Sekuler and Blake (2001).

velocities, sliding over each other in the pro-cess (Figure 4.12D). As Wallach found, therelative potency of the two alternative percep-tual outcomes varies with a number of factors,including the angle between the two line grat-ings. This bistability is reminiscent of Rubin’svase/face illusion and of various phenomenaof binocular rivalry (see R. Sekuler & Blake,2001).

This perceptual bistability points to some-thing quite important about motion percep-tion. In order to parse the display into twogratings that move in different directions, thenervous system requires mechanisms that sep-arate the different directions within each localregion. However, to produce the alternativeoutcome—the moving diamond—the nervoussystem needs a second mechanism that col-lects dissimilar directional signals from re-gions across the display and binds those sig-nals into coherent perceptual wholes ratherthan independent, isolated elements. This maysound challenging, but it is utterly necessary.In the example shown in Figure 4.12, thesedirections would be obliquely downward tothe left and to the right. When the same dis-plays are presented in the receptive fields ofMT neurons, however, many of those cells re-spond to the motion not of the components,but of the overall moving pattern (Movshon,Adelson, Gizzi, & Newsome, 1986). In thecase illustrated in Figure 4.12, this wouldbe coherent motion directly downward. It isthought, then, that important aspects of mo-tion perception arise from two stages of pro-cessing: one stage in which local motion vec-tors are extracted, and a second stage in whichthose vectors are sorted into object-relatedcombinations. As Braddick put it, “[W]e donot live in a world of independent fragments,but of objects and surfaces that move coher-ently” (1993, p. 263). Vision needs some wayto combine velocity signals that belong to thesame moving object. It also needs to separatevelocity signals that do not belong together,


Motion Aftereffects 155

that is, signals that arise from different objects(Snowden & Verstraten, 1999). Obviously, thecharacteristics of the world mandate preciselythe sort of complex behavior that Wallachdemonstrated.

Several computational strategies could beexploited in order to resolve the ambi-guity represented by the aperture problem.Fennema and Thompson (1979) were the firstto observe that the ambiguity could be over-come by combining two or more ambiguousmeasurements of local motion. Movshon et al.(1986) demonstrated that some neurons inarea MT seem to perform precisely such a dis-ambiguating computation, a computation thatproduces what is known as an intersection ofconstraints (Adelson & Movshon, 1982).

Pack and Born (2001) demonstrated a dy-namic form of disambiguation in macaqueMT neurons. As expected from the influenceof the aperture problem, MT neurons ini-tially responded primarily to the componentof motion perpendicular to the orientation ofa moving contour. Then, over approximately60 ms, the neurons’ responses shifted steadily.At the end of this dynamic process, neurons’responses began to signal the actual directionof the stimulus, no matter what the stimu-lus’s orientation. This shift in neural activ-ity had a striking correlate in the monkeys’oculomotor behavior, in smooth pursuit eyemovements. The initial velocity of pursuit eyemovements deviates in a direction perpendic-ular to local contour orientation, suggestingthat the earliest neural responses influencethe oculomotor behavior. These results sug-gest that the primate visual system derives aninitial estimate of motion direction by inte-grating ambiguous and unambiguous local-motion signals over a fairly large spatial re-gion and then refines this estimate over time.Lorenceau, Shiffrar, Wells, and Castet (1993)demonstrated an analogous result with humanobservers: The perceived direction of a fieldof moving bars is initially perpendicular to the

bars’ orientation but shifts slowly toward theactual direction of movement.

Geisler (1999) called attention to yet an-other avenue to the disambiguation of mo-tion signals. With a stimulus that moves rel-atively rapidly, the visual system’s temporalintegration produces an oriented virtual con-tour, which Geisler labeled a visual streak. (Itis important to appreciate the virtual charac-ter of such streaks; they should not be con-fused with the actual streaks, or trails, some-times generated as a wake when a bright targetmoves rapidly across the display of a cathoderay tube.) The orientation selectivity of neu-rons in area V1 means that motion streaks,which accurately reflect the direction of con-tour motion, could generate reliable orienta-tion signals that can help disambiguate di-rection of motion (see also Jancke, 2000).Geisler buttressed his ingenious hypothesiswith measurements of motion thresholds withsimultaneous orientation masking and afterorientation adaptation.

MOTION AFTEREFFECTS

When one stares at motion in one direction forsome period of time and then gazes at a sta-tionary scene, the static scene appears to movein the direction opposite to the previouslyviewed motion. Although prolonged exposureto motion has other consequences—such asdirection-selective threshold changes and al-terations in perceived velocity—illusory post-exposure motion defines what is known asthe motion aftereffect (MAE).11 Reports ofthis visual phenomenon can be traced backto Aristotle (about 330 B.C.) and TitusLucretius Carus in about 56 B.C. (Wade &

11Even very brief exposures to motion can generatesubstantial consequences, such as twofold, directionallyselective decreases in sensitivity or clear alterations inperceived direction (Pantle, Gallogly, & Piehler, 2000).



Verstraten, 1998). The best-known early de-scription of the phenomenon was given byAddams (1834), whose observation of theeffect at Scotland’s Falls of Foyer gave thephenomenon its common name, the waterfallillusion. Wade and Verstraten (1998) providedan excellent historical treatment of MAE,which is a widely used tool for studying mo-tion processing more generally. In fact, theexplosive growth in the literature on MAE,as documented by Mather, Verstraten, andAnstis (1998, p. viii) parallels the growth inpublished works on many aspects of visualmotion.

The conditions under which MAE occurswere extensively studied by Wohlgemuth(1911). The strength of MAE can be assessedby any of a number of measures, includingapparent speed, apparent direction, and du-ration (Pantle, 1998). Using MAE durationas an index of strength, Wolgemuth foundthat maintaining fixation during the adapta-tion period increased the effect and that adapt-ing one eye would produce an aftereffect inthe other, an example of interocular transfer.This surprising finding was initially reportedby Dvorak (1870) and has been replicatedrecently by others (e.g., Raymond, 1993;Steiner, Blake, & Rose, 1994; Symons,Pearson, & Timney, 1996; Wade, Swanston,& De Weert, 1993). The interocular effects areusually only half as strong as the monoculareffects.

Many motion studies generated adaptationwith drifting sinusoidal grating, whose ef-fect was measured with a stationary test grat-ing. Not unexpectedly, the MAE was tunedto spatial frequency: The strongest effect oc-curred when adapting both the grating andthe stationary test grating had the same spa-tial frequency (Bex, Verstraten, & Mareschal,1996; Cameron, Baker, & Boulton, 1992;Thompson, 1998). The effect also showedtemporal frequency tuning; the strongest ef-fect was reported at a temporal frequency of

5 Hz, regardless of spatial frequency (Pantle,1974).

Although the aftereffect is evoked usu-ally with a stationary test stimulus, more re-cent experiments have used dynamic RDCsas test stimuli (e.g., Blake & Hiris, 1993;Ledgeway, 1994). Using such test stimuli, af-tereffect strength can be measured by hav-ing a proportion of dots in the RDC move inthe adapted direction to null the illusory mo-tion. Moreover, adapting to non-Fourier mo-tion does not produce an aftereffect with statictests (McCarthy, 1993; Nishida, Ashida, &Sato, 1994; von Grunau, 1986), but an afteref-fect is observed when the test stimulus is dy-namic, such as a pattern of flickering elements(Ledgeway, 1994; McCarthy, 1993; Nishida& Sato, 1995). Some investigators feel thatdynamic, rather than static, test patterns pro-vide the most accurate reflection of the prop-erties of motion mechanisms (Hiris & Blake,1992).

The MAE ordinarily lasts for a relativelybrief time following adaptation, typically de-caying after about 1 s to 15 s. Although thepassage of time strongly influences the de-cay of the aftereffect, other factors also playa role. In fact, certain conditions of test view-ing can freeze the aftereffect’s normal decay,causing the aftereffect to be stored (Spigel,1960, 1962a, 1962b, 1964). Immediately af-ter adaptation, if one closes one’s eyes for aperiod that exceeds the MAE’s expected du-ration, then the MAE will still be seen whenthe eyes are reopened. In other words, closingthe eyes slows the MAE’s normal decay. Thisphenomenon is often described as an exampleof aftereffect storage. Such storage has someinteresting characteristics that have changedresearchers’ views of the MAE. For exam-ple, suppose that after adaptation the MAE istested with a dynamic field (randomly movingdots). If the MAE is allowed to run its coursewith this dynamic test field and if a static testfield is then presented, a MAE will be seen on


Motion Aftereffects 157

that field. Most surprising, the duration of theMAE on the static field will be little affectedby the intervening dynamic MAE. Reversingthe order of the two test fields (i.e., the staticMAE followed by the dynamic MAE), theduration of the dynamic MAE is affected bythe intervening static MAE (Verstraten et al.,1996). This relative independence of the staticand dynamic MAEs, and other divergencesbetween these two types of test fields, has en-couraged the idea that adaptation can occurin at least two distinct regions of the brain.Specifically, Nishida et al. (1994) suggestedthat the static MAE is caused by adaptationin the primary motion detectors, whereas thedynamic MAE reflects adaptation in a highercortical area, such as area MT.

Early modern accounts of the MAE as-signed a central role to the fatiguing of cellsduring adaptation. These accounts exploitedtwo key ideas: first, that neurons in the mo-tion system fire spontaneously in the absenceof motion, and second, that perceived direc-tion of motion reflects the relative firing ratesof neurons tuned to opposite directions. Sup-pose that the spontaneous activity of fatiguedcells is significantly reduced for a brief timeafter adaptation. Therefore, the presentationof a stationary stimulus generates an imbal-ance in the spontaneous firing of the motioncells, and the nonadapted cells have a propor-tionately higher rate. According to this hy-pothesis, the brain takes this imbalance assignaling the presence of actual motion in a di-rection opposite to the earlier adapting direc-tion (Barlow & Hill, 1963; R. Sekuler & Ganz,1963; Sutherland, 1961). Such illusory mo-tion is, of course, what people perceive. Thismodel was labeled the ratio model because itheld that perceived direction was controlledby the ratio of responses among motion cellsthat are tuned to different directions of motion.A somewhat modified version of this theory,dubbed the distribution shift model (Mather,1980), recognized that adaptation would

affect a range of directionally tuned mecha-nisms, not just the one mechanism that wasmost sensitive to the adapting direction. Theresult would be a more widespread change inpostadaptation spontaneous activity.

By incorporating various forms of inhibi-tion, recent theories of the mechanisms of theMAE depart from earlier emphases on fatigue.This new emphasis reflects in part a recogni-tion of the central role that inhibition playsin the cortex and, therefore, in visual com-putations. For example, Barlow (1990) pro-posed that the MAE results from a buildupof mutual inhibition among populations ofmotion-sensitive neurons during adaptation.12

This mutual inhibition remains for some timeafter the adapting stimulus is removed. As aresult, the pattern of responses among the mo-tion cells is modified such that a neutral (sta-tionary) stimulus is perceived as moving inthe opposite direction. Because this inhibitorybuildup could occur at any site where motion-sensitive cells are found, adaptation can oc-cur at one or more sites in the motion system.Nishida and Sato’s (1995) MAE results, men-tioned earlier, are consistent with this idea.

Motion aftereffects can also be producedby stimuli comprising two superimposed di-rections of motion. Here, the resulting af-tereffect is opposite the vector sum of thetwo adapting directions (Riggs & Day, 1980;Verstraten, Fredericksen, & van de Grind,1994). Thus, for example, simultaneous adap-tation to leftward motion and to upward mo-tion subsequently causes a stationary test fig-ure to appear to drift down and to the right(the vector sum of the downward componentand the rightward component). When the twoadapting components are unequal in strength

12For Barlow (1990), the MAE exemplified a generalprinciple of neural operation. In his view, the mutual inhi-bition, to which he ascribes MAE, reduces redundancy inthe firing of different sets of neurons, thereby minimizingthe total activity at any site in the visual system.



(e.g., one component is of higher contrast thanis the other), the aftereffect direction shifts to-ward the stronger component, and the size ofthe shift tracks the inequality between stimu-lus components.

MEMORY AND LEARNINGFOR MOTION

Vision’s essential purpose is to guide behav-ior, and that purpose is served best when vi-sion’s current products can be compared tostored representations of vision’ earlier prod-ucts. Such comparisons, or recognitions, en-able animals to prepare situation-appropriatebehaviors before they are necessary. Althoughmuch of the research on human memory hasfocused on memory for symbolic or verbalmaterial, long before words came on the evo-lutionary stage, animals had a crucial need toremember and recognize scenes, objects, andevents that they had encountered earlier. Al-though motion is certainly among the visualattributes that are worth remembering, rela-tively little research has been done on memoryfor motion.

Magnussen and Greenlee (1992) exam-ined observers’ short-term memory for stim-ulus velocity. Using drifting luminance grat-ing, they measured the difference thresholdfor velocity, �V. They explored the retentionof velocity information in memory by varyingthe interstimulus intervals (ISI) separating thefirst (reference) stimulus and the second (test)stimulus. Weber fractions (�V/V) proved tobe independent of ISIs ranging from 1 s to30 s. This invariance showed that memoryfor velocity is quite robust over delays of upto 30 s.

Blake, Cepeda, and Hiris (1997) exploredmemory for direction of movement by us-ing RDCs with 100% coherence. After a 1-spresentation of an RDC, an observer used acomputer mouse to indicate the direction in

which motion had been seen. The mean abso-lute value of observers’ errors was about 5 deg(Blake, Cepeda, & Hiris, 1997). This level ofperformance is remarkably robust when pre-response delays force the perceived directionto be stored and then retrieved from mem-ory. For example, performance is unchangedwhen the subject’s response is delayed by 8 s.The memory on which responses were basedis unlikely to be of an iconic nature; per-formance was unimpaired by random visualnoise interposed between the RDC and thereport of the remembered direction. Nor didmemory depend on stored information fromeye movement, such as observers might makewhile viewing the RDC: Performance wasunaffected when observers executed randomtracking-eye movements before making theirjudgments. Although memory for RDC direc-tion was preserved over short intervals, suchmemory was dramatically undermined whenBlake et al.’s observers saw not just one direc-tion but a succession of different directions oneach trial. Performance fell off substantiallyas the number of seen directions grew. Forexample, average error climbed from 5 degwith just one presentation to 25 deg to 30 degwhen observers saw and tried to rememberseven different directions of motion.

Pasternak and colleagues have studied therole of area MT in remembering the directionof visual motion (Bisley & Pasternak, 2000;Bisley et al., 2001). They trained monkeys tocompare the directions of motion portrayedin two successively presented animation se-quences, the “sample” and the “test.” Task dif-ficulty was manipulated by varying the coher-ence of these random-dot motion displays andthe delay interval between sample and test.During the delay period—while the monkeywas trying to remember the sample directionthat it had seen—small, brief electrical pulseswere delivered directly to clusters of neuronsin area MT. This electrical stimulation, whichartificially activated the recipient neurons,


Knowledge, Attention, and Motion Perception 159

influenced the monkey’s subsequent judg-ment about the test direction (Bisley, Zaksas,& Pasternak, 2001). This result supports thenotion that area MT is involved in the short-term retention and retrieval of informationabout the direction of visual motion. This con-clusion receives further support from the sameresearchers’ studies of the effects of unilaterallesions to motion centers of the monkey cortex(Bisley & Pasternak, 2000).

As mentioned in the introduction to thissection, visual memory is an important guidefor behavior. Memory of what people haveseen allows us to prepare situation-appropriatebehaviors and to execute them in a timely fash-ion. For visual movement, preparation of ap-propriate behavior requires recognition thatsome motion that is being experienced at agiven time has in fact been seen before. Chunand Jiang (1999) showed that with repeatedexposure to particular complex movementsequences, subsequent recognition of thosesequences remains at the implicit level. Intheir experiment, observers searched for a sin-gle rotated-T target embedded in a field ofL-shaped distractors. An animation sequencemade all the search items—target as wellas distractors—move randomly over the dis-play screen; each item’s random trajectorywas independent of the trajectory of the otheritems. Because all the search items started thesequence as crosses, and only slowly mor-phed into their final form—a rotated-T orL distractors—observers had to monitor theentire set of randomly moving items. Unbe-knownst to observers, some of the randomsequences were repeated during the 72 ex-perimental trials. Search time improved forall sequences, repeated as well as nonre-peated, but the improvement was dramaticallystronger with repeated sequences. Becauseexplicit recognition accuracy was no betterthan chance, Chun and Jiang applied the label“implicit” to the perceptual learning that theyobserved. This result supports the notions that

subjects can pick up and encode dynamic reg-ularities or invariances and that they can do sowithout explicitly recognizing those repeatingfeatures.

KNOWLEDGE, ATTENTION, ANDMOTION PERCEPTION

Attention takes many forms, all of whichpromote preferential processing of stimulior stimulus attributes that are relevant toa particular task and inhibit processing oftask-irrelevant stimuli or stimulus attributes(Chap. 6, this volume; Raymond, 2000). Theselectivity that is represented by attentionmodulates the behavioral impact of any mov-ing stimulus—up or down—in accord with anobserver’s task and goals. Performance on anytask, whether in the laboratory or as part of ev-eryday activity, implicitly reflects this modu-lation, which ordinarily operates unnoticed inthe background. In this section, we consider asample of experiments designed especially tohighlight selectivity.

Tracking Multiple Objects

William James distinguished among forms ofattention, noting that whereas people can at-tend to sensory stimuli (e.g., particular lo-cations or objects), they can attend also to“ideal or represented objects” (1890, p. 393).In other words, James recognized that peoplecould attend either to a thing that was physi-cally present or to an object that existed onlyin the mind, which today might be called avirtual object. Attention’s ability to influencevirtual moving objects is demonstrated by re-sults from a procedure devised by Pylyshynand Storm (1988). The procedure measuresthe ability to keep track, over time, of multi-ple, spatially dispersed, independently mov-ing targets. The task can be likened to watch-ing some crows feeding in a field, and then,



Phase of Trial

1. Target Designation(potential targetdiscs blink)

2. Movement(discs moverandomly, 5 s)

3. Probe(the target disc brightens)

Figure 4.13 Typical trial structure for multiple-target tracking experiment.NOTE: During trial’s initial phase (target designation) several discs are singled out by a momentarybrightening. During the second phase (movement) all discs undergo a series of random movements. Inphase 3 (probe) a single disc is brightened, and the observer must judge whether that disc was among theones that had been identified as potential targets during the trials’s initial phase.SOURCE: Modified from R. Sekuler and Blake (2001).

when the crows fly off in different directions,keeping track of each and every crow.

In a typical experiment on multiple-objecttracking, a trial consists of three phases, calledtarget designation, movement, and probe. Fig-ure 4.13 illustrates each of the three phases.Target designation begins with the presenta-tion of 10 targets, such as discs, scatteredabout on a computer display. Then three, four,or five of the 10 discs, chosen randomly, blinkseveral times. This informs the observer ofthe discs that will be targets for that trial. Inthe movement phase, all 10 discs move aboutsmoothly on the screen in various directions,each changing course unpredictably. Afterseveral seconds of movement, the trial’s probephase begins. One of the 10 discs is high-lighted, and the observer must report whetherthat one probe item had or had not been one ofthe targets designated during the designationphase, several seconds before. Performance,summarized as the proportion of correct re-sponses, is most accurate when people haveto keep track of fewer targets (e.g., 3 insteadof 4 or 5). This is consistent with earlier find-

ings that attention loses potency for any oneitem when additional items must also be at-tended to.

Yantis (1992) examined the influence ofperceptual grouping on multiple-object track-ing. In one experiment, for example, targetseither were chosen at random or were chosento lie at the vertices of virtual simple geomet-ric figures, such as diamonds. At the outsetof the experiment, performance was much su-perior when the target to be recognized hadstarted out as part of a nice geometric percep-tual group (during the target designation phaseof a trial). However observers quickly learnedto impose virtual groupings on elements thathad not been part of ready-made, regular geo-metrical groupings. This erased the early ad-vantage found with ready-made groupings.

Unless grouping is maintained during themovement phase, grouping at the start of atrial is no help at trial’s end (the probe phase).By varying the targets’ movements duringthe movement phase of a trial, Yantis con-trolled the likelihood that any existing group-ing would be maintained. In one condition,



targets moved about randomly, which allowedone or more targets occasionally to cross overan opposite edge of the virtual polygon. Thiscrisscrossing destroyed the original grouping,undermining the coherence of the virtual poly-gon and causing elements to lose identity asdesignated targets in that virtual polygon. Inanother condition, movements of targets wereconstrained, ensuring that none ever crossedover an opposite edge of the virtual polygon.Here, movements of individual targets pro-duced moment-to-moment fluctuations in theshape of a virtual figure that would be cre-ated by connecting those targets. However, notone of these fluctuations was drastic enoughto destroy the convexity of the virtual poly-gon. Performance was distinctly better whenthe virtual polygon was preserved. This sug-gests that observers’ attention creates (in thetarget designation phase) and maintains (in themovement phase) an updatable virtual objectthat is used (in the probe phase) to determinewhether the probed target was in the object.

To track multiple moving objects, the brainexploits neural circuits that are ordinarily ded-icated to a different purpose, namely, shift-ing attention from one location in space toanother. Culham et al. (1998) used fMRI toidentify the brain circuits that participated inmultiple-object tracking. Their results pro-vide a clear picture of how the brain managesthis difficult task. Attentive tracking of mul-tiple, independently moving objects is me-diated by a network of areas that includesparietal and frontal regions—known to be re-sponsible for shifts of attention between loca-tions and for eye movements—as well as areaMT and related regions, which, as noted ear-lier, are central regions for processing motioninformation.

Uncertainty and Motion Detection

When it comes to detecting weak motionsignals that are embedded in noise, it helps

greatly to know in advance in which direc-tion the signal dots are moving: uncertaintyabout direction of motion impairs detectionperformance. In one study, Ball and Sekuler(1981) determined the ease with which peopledetected very dim dots moving across a com-puter display. From trial to trial, the directionof the dots’ motion changed unpredictably. Inaddition, during half the trials, no dots at allwere presented; the viewer saw only a blankscreen. The dots were made dim enough thata viewer had great difficulty telling whetheror not any dots were present. Ball and Sekulermeasured the intensity threshold for detectingthe dots under various conditions. Thresholdswere initially determined simply by random-izing from trial to trial the direction in whichthe dots moved. Thresholds were then mea-sured with an explicit cue that reduced theviewer’s uncertainty about direction of mo-tion. This directional cue was a short lineflashed very briefly at different times relativeto the presentation of the dots. The orientationof the line indicated the direction in which thedots, if present at all, might move (recall thatno dots were presented on half of the trials).

Ball and Sekuler (1981) made several note-worthy discoveries. First, when the cue speci-fied the direction of the dots’ motion precisely,the dots were easier to see; that is, the inten-sity threshold was low. Second, the cue wasnot helpful unless it preceded the dots by about500 ms, indicating that selective attention re-quired some time to operate. Third, if the cue’sorientation did not match the dots’ directionprecisely, but only approximated it, the cuecould still lower the detection threshold, butnot as much as when it was precisely accu-rate. Generally, the greater the discrepancybetween the cue’s orientation and the direc-tion of the dots’ motion, the more difficult itwas to see the moving dots. In the extreme,a cue that misdirected the observer’s expecta-tion by 180 deg was worse than no cue at all:Detection fell below a no-cue baseline.



How do directional cues or certainty aboutdirection exert the effect that they do? Cuesor prior knowledge are not part of the stimu-lus, but they certainly do affect the responseevoked by the stimulus. Obviously, some in-formation extracted from the cue must berecoded into a format capable of influenc-ing subsequent processing of the test stimu-lus. After this recoding process, the nonmov-ing cue then seems able to boost selectivelyresponses in particular sets of directionallyselective neurons.

Shulman et al. (1999) extended Ball andSekuler’s (1981) study by measuring brainactivity while people performed the cued-motion detection task. As before, the station-ary cue, presented prior to the moving target,specified the direction of motion that peoplewould see. As revealed by fMRI signals, thenonmoving cue activated brain areas that in-cluded area MT, as well as adjacent regionsthat normally respond to motion. Also, someareas of parietal lobe that are not normally re-sponsive to motion were activated. Together,these motion-sensitive and motion-insensitiveareas constitute a neural circuit that encodesand maintains the cue during the interval be-tween the cue and the onset of motion.

Presumably, prior information about thedirection of motion temporarily boosts the sig-nals of MT neurons that are particularly re-sponsive to that direction of motion (Treue &Maunsell, 1999; Treue & Trujillo, 1999). Thisinternally generated boost response is equiv-alent to what happens when the responses ofparticular directionally selective neurons arestrengthened, either by the presentation of astrong visual stimulus or by direct electricalstimulation, as in the study by Salzman et al.(1992).

A. B. Sekuler, Sekuler, and Sekuler (1990)used direction uncertainty to explore detec-tion of changes in direction of movement.Direction uncertainty refers to an observer’sprior knowledge of stimulus direction; as de-

scribed earlier, uncertainty diminishes the de-tectability of motion, as indexed by elevatedthresholds or lengthened reaction times. Inthis study, observers had to respond as quicklyas possible to a constant relative change instimulus direction: 30 deg clockwise. Theprechange direction either was fixed withina block of trials (certainty condition) or wascompletely random (maximum uncertainty).Generally, responses to change in the cer-tainty condition were considerably faster thanin conditions of uncertainty. However, if theprechange motion lasted 500 ms or longer,observers’ reaction times to change were nolonger affected by uncertainty about initial di-rection. However, for shorter initial durations,reaction time increased with increased uncer-tainty (i.e., increased in the range of possibleinitial directions). A. B. Sekuler et al. pro-posed that the visual system requires approx-imately 500 ms to normalize the initial direc-tion of motion in order to be able to detect thedirection change by essentially converting thenominal task to one of detecting motion onset.

Alais and Blake (1999) used the MAE toprobe the influence of attention on motionperception. As mentioned earlier, when theadapting stimulus comprises two directionalcomponents, the direction of the MAE is usu-ally the vector sum of the two components; butwhen the components are unequal in strength,the resulting MAE tracks that inequality. Byvarying the relative contrasts of the two com-ponent adapting motions, Alais and Blakewere able to manipulate how much attentionan observer had to pay to one of the compo-nents. In their experiments, observers vieweda computer display consisting of two super-imposed fields of moving dots. In one group,all dots moved coherently in a single direc-tion, shown as “upward” in Figure 4.14. Dotsin the other group moved in random direc-tions most of the time, producing no net di-rectional drift. Occasionally, a subset of therandom dots joined forces to move in the same



adaptation brief "rightward"probe

adaptation

time

Figure 4.14 Schematic of stimulus conditions used by Alais and Blake (1995) to test attention’s influ-ence on motion adaptation.NOTE: During most of the adaptation period, a subset of adaptation dots moved coherently upward (opencircles in left panel), and the remaining noise dots moved randomly in all directions (gray circles in leftpanel). Unpredictably, a subset of the noise dots moved briefly rightward (black dots in center panel)and then quickly resumed their normal motions (right panel). Observers had to detect this brief motion,which required considerable attention. Note that throughout the experiment all dots were actually thesame color; differences of shading are used here for illustrative purposes only.SOURCE: Modified from R. Sekuler and Blake (2001).

direction, shown as “rightward” in Fig-ure 4.14. The proportion of dots moving right-ward was only about 25%, making it neces-sary for observers to look carefully to detecttheir presence. On some trials observers wererequired to indicate when this weak coherentmotion was present; on the remaining trialsobservers simply saw the same stimuli but didnot have to perform the detection task and,therefore, did not have to attend selectivelyto the rightward motion. Alais and Blake rea-soned that the first of these conditions woulddemand more attention than would the sec-ond, passive viewing condition. In the passivecontrol condition, the brief insertions of weakrightward motion had little effect on MAE’sdirection; however, when observers had to at-tend to it, the inserted motion dramatically al-tered the aftereffects’ direction, shifting it byabout 20 deg. This same shift, which was me-diated solely by observers’ attention to a weaksignal, was equivalent to the shift that wouldhave been produced by a very powerful signal(motion with dots 70% correlated). Therefore,

attention to motion in one direction boostedthe response to that motion by almost threetimes, rendering a 25% correlated stimulus aseffective as one that was 70% correlated. Ex-trapolating from neural responses within areaMT to changes in degree of stimulus correla-tion, Alais and Blake deduced that the atten-tional effects seen in human observers wereequivalent, on average, to what would be ex-pected from doubling the stimulus correlationin a nonattended stimulus. There is no doubtthat attention can exert a very powerful influ-ence on perceived motion.

Parsing Everyday Activities

Although considerable progress has beenmade toward an understanding of motion per-ception, a number of important questions re-main unresolved. Among them is the questionof how human observers recognize and cate-gorize everyday actions. To clarify the prob-lem, Bobick (1997) has distinguished betweenwhat he calls movements, such as the lifting of



arm, and what he calls actions, which includeinteractions with the environment and otheractors, as well as inferred causal relationshipsamong image components. To take just a fewexamples, actions include a soccer player’sheading the ball, a cook’s making a cheesesandwich, and your making a bed or ironinga shirt. In all these cases, the actor (the soccerplayer, the cook, and you) generates a more-or-less continuous stream of movements. Anobserver’s understanding of the visual infor-mation could begin with a parsing of the actioninto distinct perceptual events (Tong, 2001).One perceptual component event could bedistinguished from a succeeding componentby changes in velocity or in the movementof one body part relative to another. There-fore, motion perception is essential for ourability to parse complex actions. Combiningbehavioral observations and a novel fMRIparadigm, Zacks et al. (2001) examined mo-tion’s possible contribution to action parsing.Participants first watched movies of every-day activities. The movies, 2 min to 5 minlong, showed someone making a bed, fertiliz-ing a houseplant, ironing a shirt, or washingdishes. Movies were viewed three times each,in random order. During the first viewingof any movie, observers merely watched theaction passively. During this viewing, fMRImeasurements were taken. During subsequentviewings of a movie, participants used a but-ton to signal when they thought that one nat-ural and meaningful unit of action had endedand another had begun. During the original,passive viewing of everyday activities, fMRIsignals reflected transient changes occurringin several related regions of the brain, includ-ing area MT+, which participates in motionperception. The onset of transient changes inneural activity did not occur randomly duringthe action but were in temporal register withmoments that observers deemed to be bound-aries between components of the overall ac-tion. Thus, it may be that motion information

plays a key role in the segmentation and un-derstanding of everyday actions.

It is worth noting when these transientchanges in MT+ activation occurred. On aver-age, they began a few seconds before the per-ceived boundary between action components.As a result, it could be that these anticipatorytransient changes in brain activation signifytop-down influences, that is, influences of ob-servers’ familiarity with the actions and there-fore observers’ expectancies about upcomingchanges in motion. Support for this hypothesiscomes from fMRI research with motion thatis imagined but not actually seen (Grossman& Blake, 2001; Kourtzi & Kanwisher, 2000).

SUMMING UP ANDLOOKING FORWARD

Without question, our understanding of mo-tion perception has been pushed to a levelscarcely imaginable just two decades ago.New psychophysical, physiological, and com-putational research tools have made possi-ble huge strides toward unraveling the mys-teries of the visual registration of motion,which Walls (1942) and we consider to bemost important among the talents that com-prise vision. The application of functionalneuroimaging has begun to identify roughbut intriguing relationships between particu-lar sites in the visual brain and performanceof particular tasks. Obviously, this develop-ment has only just started, and many difficultchallenges lie just ahead.

Our present knowledge of links betweenbrain sites and aspects of visual motion per-ception throws only the dimmest of lightsonto the complex neural transformationsand computations that support performanceon various tasks. Motion perception emergesfrom a shifting partnership between exoge-nous influences, represented by stimulusattributes, and endogenous influences,


References 165

including expectation, attention, memory, andlearning. We have relatively little understand-ing of the parameters and limiting conditionsof this partnership. Also, we have no under-standing whatever of the control architecturethat sets and adjusts the relative weights forthe two partners, exogenous and endogenousinfluences. Furthermore, it is not clear to uswhether traditional behavioral paradigms willcontribute much to the development of suchunderstanding.

Vernon Mountcastle (quoted in Shadlen &Newsome, 1996) sketched out an ambitiousand broad agenda not only for the study ofvisual motion but also for the entire field ofsensory science. He urged researchers to studythe complete chain of events that “lead fromthe initial central representation of sensorystimuli, through the many sequential and par-allel transformations of those neural images,to the detection and discrimination processesthemselves, and to the formation of generalcommands for behavioral responses and de-tailed instructions for their motor execution(p. 628).” This last part of Mountcastle’sunfulfilled agenda would connect decisionsabout sensory signals to the preparation andexecution of motor acts that are appropriateto those decisions, an exciting area in whichwork has only just begun (see, e.g., Gold &Shadlen, 2001).

REFERENCES

Addams, R. (1834). An account of a peculiar op-tical phenomenon seen after having looked ata moving body. London and Edinburgh Philo-sophical Magazine and Journal of Science, 5,373–374.

Adelson, E. H., & Movshon, J. A. (1982). Phe-nomenal coherence of moving visual patterns.Nature, 300, 523–525.

Ahlstrom, U., & Borjesson, E. (1996). Segregationof motion structure from random noise. Percep-tion, 25, 279–291.

Ahlstrom, V., Blake, R., & Ahlstrom, U. (1997).Perception of biological motion. Perception, 26,1539–1548.

Alais, D., & Blake, R. (1999). Neural strength ofvisual attention gauged by motion adaptation.Nature Neuroscience, 2, 1015–1018.

Albensi, B. C., & Powell, J. H. (1998). The differ-ential optomotor response of the four-eyed fishAnableps anableps. Perception, 27, 1475–1483.

Andersen, G. J. (1989). Perception of three-dimensional structure from optic flow withoutlocally smooth velocity. Journal of Experimen-tal Psychology: Human Perception & Perfor-mance, 15, 363–371.

Andersen, R. A., & Bradley, D. C. (1998). Percep-tion of three-dimensional structure from motion.Trends in Cognitive Science, 2, 222–228.

Anstis, S., & Ramachandran, V. S. (1987). Visualinertia in apparent motion. Vision Research, 27,755–764.

Arterberry, M. E., & Yonas, A. (2000). Perceptionof three-dimensional shape specified by opticflow by 8-week-old infants. Perception & Psy-chophysics, 62, 550–556.

Attneave, F. (1974). Apparent movement and thewhat-where connection. Psychologia, 17, 108–120.

Baddeley, R., & Tripathy, S. P. (1998). Insightsinto motion perception by observer modeling.Journal of the Optical Society of America A, 15,289–296.

Ball, K., & Sekuler, R. (1981). Cues reduce direc-tion uncertainty and enhance motion detection.Perception & Psychophysics, 30, 119–128.

Ball, K., & Sekuler, R. (1986). Improving visualperception in older observers. Journal of Geron-tology, 41, 176–182.

Ball, K., Sekuler, R., & Machamer, J. (1983).Detection and identification of moving targets.Vision Research, 23, 229–238.

Barlow, H. B. (1978). The efficiency of detectingchanges of density in random dot patterns. VisionResearch, 18, 637–650.

Barlow, H. B. (1980). The absolute efficiencyof perceptual decisions. Philosophical Transac-tions of the Royal Society of London (Series B),290, pp. 71–82.



Barlow, H. B. (1990). A theory about the functionalrole and synaptic mechanism of visual after ef-fects. In C. Blakemore (Ed.), Coding and Ef-ficiency (pp. 363–375). Cambridge: CambridgeUniversity Press.

Barlow, H. B. (1995). The neuron doctrine in per-ception. In M. S. Gazzaniga (Ed.), The cognitiveneurosciences (pp. 415–435). Cambridge: MITPress.

Barlow, H. B., & Hill, R. M. (1963). Evidence for aphysiological explanation for the waterfall phe-nomenon and figural aftereffects. Nature, 200,1345–1347.

Barlow, H. B., & Reeves, B. C. (1979). The versa-tility and absolute efficiency of detecting mirrorsymmetry in random dot displays. Vision Re-search, 19, 783–793.

Barlow, H. B., & Tripathy, S. P. (1997). Correspon-dence noise and signal pooling in the detectionof coherent visual motion. Journal of Neuro-science, 17, 7954–7966.

Beckers, G., & Homberg, V. (1992). Cerebral vi-sual motion blindness: Transitory akinetopsia in-duced by transcranial magnetic stimulation ofhuman area V5. Proceedings of the Royal Soci-ety of London, Series B, 249, 173–178.

Bennett, P. J., Sekuler, A. B., & Ozin, L. (1999).Effects of aging on calculation efficiency andequivalent noise. Journal of the Optical Societyof America A, 16, 654–668.

Bex, P. J., Verstraten, F. A. J., & Mareschal, I.(1996). Temporal and spatial frequency tuning ofthe flicker motion aftereffect. Vision Research,36, 2721–2727.

Bichot, N. P., Thompson, K. G., Chenchal Rao, S.,& Schall, J. D. (2001). Reliability of macaquefrontal eye field neurons signaling saccade tar-gets during visual search. Journal of Neuro-science, 21, 713–725.

Bischof, W. F., Reid, S. L., Wylie, D. R., & Spetch,M. L. (1999). Perception of coherent motion inrandom dot displays by pigeons and humans.Perception & Psychophysics, 61, 1089–1101.

Bisley, J. W., & Pasternak, T. (2000). The mutipleroles of visual cortical areas MT/MST in remem-bering the direction of visual motion. CerebralCortex, 10, 1053–1065.

Bisley, J. W., Zaksas, D., & Pasternak, T. (2001).Microstimulation of cortical area MT affectsperformance on a visual working memory task.Journal of Neurophysiology, 85, 187–196.

Blake, R. (1993). Cats perceive biological motion.Psychological Science, 4, 54–57.

Blake, R., & Aiba, T. S. (1998). Detection and dis-crimination of optic flow components. JapanesePsychological Research, 40, 19–30.

Blake, R., Cepeda, N. J., & Hiris, E. (1997). Mem-ory for visual motion. Journal of ExperimentalPsychology: Human Perception & Performance,23, 353–369.

Blake, R., & Hiris, E. (1993). Another meansfor measuring the motion aftereffect. VisionResearch, 33, 1589–1592.

Bobick, A. F. (1997). Movement, activity andaction: The role of knowledge in the percep-tion of motion. Philosophical Transactions ofthe Royal Society of London (Series B), 352,1257–1265.

Bobick, A. F., & Davis, J. (2001). The recognitionof human movement using temporal templates.IEEE Transactions on Pattern Analysis andMachine Intelligence, 23, 257–268.

Borst, A. (2000). Models of motion detection.Nature Neuroscience, 3(Suppl.), 1168.

Braddick, O. (1993). Segmentation versus integra-tion in visual motion processing. Trends in Neu-roscience, 16, 263–268.

Bradley, D. (2001). MT signals: Better with time.Nature Neuroscience, 4, 346–348.

Bravo, M. J., & Watamaniuk, S. N. J. (1995). Evi-dence for two speed signals: A coarse local sig-nal for segregation and a precise global signalfor discrimination. Vision Research, 35, 1691–1697.

Britten, K. H., & van Wezel, R. J. (1998). Electri-cal microstimulation of cortical area MST biasesheading perception in monkeys. Nature Neuro-science, 1, 59–63.

Brown, R. H. (1961). Visual sensitivity to differ-ences in velocity. Psychological Bulletin, 58,89–103.

Brownlow, S., Dixon, A. R., Egbert, C. A., &Radcliffe, R. D. (1997). Perception of movement


References 167

and dancer characteristics from point-light dis-plays of dance. Psychological Record, 47, 411–421.

Buracas, G. T., Zador, A. M., DeWeese, M. R., &Albright, T. D. (1998). Efficient discriminationof temporal patterns by motion-sensitive neu-rons in primate visual cortex. Neuron, 20, 959–969.

Burgess, A. E., & Barlow, H. B. (1983). The effi-ciency of numerosity discrimination in randomdot images. Vision Research, 23, 811–829.

Burgess, A. E., Wagner, R. F., Jennings, R. J., &Barlow, H. B. (1981). Efficiency of human visualsignal discrimination. Science, 214, 93–94.

Burr, D. C., Morrone, M. C., & Vaina, L. M. (1998).Large receptive fields for optic flow detection inhumans. Vision Research, 38, 1731–1743.

Cameron, E. L., Baker, C. L., & Boulton, J. C.(1992). Spatial frequency selective mecha-nisms underlying the motion aftereffect. VisionResearch, 32, 561–568.

Celebrini, S., & Newsome, W. T. (1994). Neuronaland psychophysical sensitivity to motion signalsin extrastriate area MST of the macaque monkey.Journal of Neuroscience, 14.

Chubb, C., & Sperling, G. (1988). Drift-balancedrandom stimuli: A general basis for studyingnon-Fourier motion perception. Journal of theOptical Society of America A, 5, 1986–2006.

Chun, M. M., & Jiang, Y. (1999). Top-down at-tentional guidance based on implicit learning ofvisual covariation. Psychological Science, 10,360–365.

Cowey, A., & Vaina, L. M. (2000). Blindness toform from motion despite intact static form per-ception and motion detection. Neuropsycholo-gia, 38, 566–578.

Croner, L. J., & Albright, T. D. (1997). Image seg-mentation enhances discrimination of motion invisual noise. Vision Research, 37, 1415–1427.

Croner, L. J., & Albright, T. D. (1999). Seeing thebig picture: Integration of image cues in the pri-mate visual system. Neuron, 24, 777–789.

Culham, J. C., Brandt, S. A., Cavanagh, P.,Kanwisher, N. G., Dale, A. M., & Tootell, R. B.(1998). Cortical fMRI activation produced by

attentive tracking of moving targets. Journal ofNeurophysiology, 80, 2657–2670.

Culham, J. C., He, S., Dukelow, S., & Verstraten,F. A. J. (2001). Visual motion and the humanbrain: What has neuroimaging told us? ActaPsychologica, 107, 69–94.

Curran, W., & Braddick, O. J. (2000). Speed anddirection of locally-paired dot patterns. VisionResearch, 40, 2115–2124.

Cutting, J. E. (1986). Perception with an eye formotion. Cambridge: MIT Press.

Dakin, S. C., & Mareschal, I. (2000). The role ofrelative motion computation in “direction repul-sion.” Vision Research, 40, 833–841.

Dawson, M. R. W. (1991). The how and why ofwhat went where in apparent motion: Modelingsolutions to the motion correspondence problem.Psychological Review, 98, 569–603.

DeAngelis, G., Cumming, B. G., & Newsome,W. T. (1998). Cortical area MT and the percep-tion of stereoscopic depth. Nature, 394, 677–680.

De Bruyn, B., & Orban, G. A. (1988). Humanvelocity and direction discrimination measuredwith random dot patterns. Vision Research, 28,1323–1335.

De Bruyn, B., & Orban, G. A. (1993). Segrega-tion of spatially superimposed optic flow com-ponents. Journal of Experimental Psychology:Human Perception and Performance, 19, 1014–1027.

Derrington, A. M., & Henning, G. B. (1993). De-tecting and discriminating the direction of mo-tion of luminance and colour gratings. VisionResearch, 33, 799–811.

Diener, H. C., Wist, E. R., Dichgans, J., & Brandt,T. H. (1976). The spatial frequency effect on per-ceived velocity. Vision Research, 16, 169–176.

Dobkins, K. R. (2000). Moving colors in the limelight. Neuron, 25, 15–18.

Dror, R. O., O’Carroll, D. C., & Laughlin, S. B.(2001). Accuracy of velocity estimation byReichardt correlators. Journal of the Optical So-ciety of America A, 182, 241–252.

Dvorak, V. (1870). Versuche uber Nachbildervon Reizveranderungen. Sitzungsberichte der



Weiner Akademie der Wissenschaften, 61, 257–262.

Dzhafarov, E. N., Sekuler, R., & Allik, J. (1993).Detection of changes in speed and direction ofmotion: Reaction time analysis. Perception &Psychophysics, 54, 733–750.

Emerson, R. C., Bergen, J. R., & Adelson, E. H.(1992). Directionally selective complex cellsand the computation of motion energy in cat vi-sual cortex. Vision Research, 32, 203–218.

Exner, S. (1888). Uber optische Bewegung-sempfindungen. Biologisches Centralblatt, 8,437–448.

Fennema, C. I., & Thompson, W. B. (1979). Veloc-ity determination via scenes containing severalmoving images. Computer Graphics and ImageProcessing, 9, 301–315.

Field, D. J., Hayes, A., & Hess, R. F. (1993). Con-tour integration by the human visual system:Evidence for a local “association field.” VisionResearch, 33, 173–193.

Fox, R., & McDaniel, C. (1982). The perceptionof biological motion by human infants. Science,218, 486–487.

Francis, G., & Kim, H. (1999). Motion parallel toline orientation: Disambiguation of motion per-cepts. Perception, 28, 1243–1255.

Gabbiani, F., Mo, C., & Laurent, G. (2001). In-variance of angular threshold computation in awide-field looming-sensitive neuron. Journal ofNeuroscience, 21, 314–329.

Gegenfurtner, K. R., & Hawken, M. J. (1996). In-teraction of motion and color in the visual path-ways. Trends in Neuroscience, 19, 394–401.

Gegenfurtner, K. R., Mayser, H. M., & Sharpe,L. T. (2000). Motion perception at scotopic lightlevels. Journal of the Optical Society of AmericaA, 17, 1505–1515.

Geisler, W. S. (1989). Sequential ideal-observeranalysis of visual discriminations. Psychologi-cal Review, 96, 267–314.

Geisler, W. S. (1999). Motion streaks provide aspatial code for motion direction. Nature, 400,65–69.

Geisler, W. S., Perry, J. S., Super, B. J., &Gallogly, D. P. (2001). Edge co-occurrence in

natural images predicts contour grouping per-formance. Vision Research, 41, 711–724.

Gibson, E. J., Gibson, J. J., Smith, O. W., & Flock,H. (1959). Motion parallax as a determinant ofperceived depth. Journal of Experimental Psy-chology, 58, 40–51.

Gold, J., Bennett, P. J., & Sekuler, A. B. (1999).Identification of band-pass filtered letters andfaces by human and ideal observers. VisionResearch, 39, 3537–3560.

Gold, J. I., & Shadlen, M. N. (2001). Neuralcomputations that underlie decisions about sen-sory stimuli. Trends in Cognitive Science, 5,10–16.

Graham, C. H. (1951). Visual perception. In S. S.Stevens (Ed.), The handbook of experimentalpsychology (pp. 868–920). New York: Wiley.

Graham, C. H. (1965). Perception of motion. InC. H. Graham (Ed.), Vision and visual perception(pp. 575–588). New York: Wiley.

Graham, N. V. S. G. (1989). Visual pattern analyz-ers. New York: Oxford University Press.

Gros, B. L., Blake, R., & Hiris, E. (1998).Anisotropies in visual motion perception: Afresh look. Journal of the Optical Society ofAmerica A, 15, 2003–2011.

Grossman, E. D., & Blake, R. (1999). Perceptionof coherent motion, biological motion and form-from-motion under dim-light conditions. VisionResearch, 39, 3721–3727.

Grossman, E. D., & Blake, R. (2001). Brainactivity evoked by inverted and imagined bi-ological motion. Vision Research, 41, 1475–1482.

Grossman, E. D., Donnelly, M., Price, P.,Morgan, V., Pickens, D., Neighbor, G., & Blake,R. (2000). Brain areas involved in perception ofbiological motion. Journal of Cognitive Neuro-science, 12, 711–720.

Grzywacz, N. M., Watamaniuk, S. N. J., & McKee,S. P. (1995). Temporal coherence theory for thedetection and measurement of visual motion.Vision Research, 35, 3183–3203.

Hecht, S., Shlaer, S., & Pirenne, M. H. (1942).Energy, quanta, and vision. Journal of GeneralPhysiology, 25, 819–840.


References 169

Heeger, D. J. (1994). The representation of visualstimuli in primary visual cortex. Current Direc-tions in Psychological Science, 3, 159–163.

Heeger, D. J., Boynton, G. M., Demb, J. B.,Seidemann, E., & Newsome, W. T. (1999).Motion opponency in visual cortex. Journal ofNeuroscience, 19, 7162–7174.

Heeger, D. J., Simoncelli, E. P., & Movshon, J. A.(1996). Computational models of cortical visualprocessing. Proceeding of the National Academyof Science USA, 93, 623–627.

Heidenreich, S. M., & Zimmerman, G. L. (1995).Evidence that luminant and equiluminantmotion signals are integrated by direction-ally selective mechanisms. Perception, 24,879–890.

Hikosaka, O., Miyauchi, S., & Shimojo, S. (1993).Visual attention revealed by an illusion of mo-tion. Neuroscience Research, 18, 11–18.

Hiris, E., & Blake, R. (1992). Another perspec-tive on the visual motion aftereffect. Proceedingof the National Academy of Science USA, 89,9025–9028.

Hiris, E., & Blake, R. (1996). Direction repulsionin motion transparency. Visual Neuroscience, 13,187–197.

Hochberg, J. (1988). Visual perception. In R. C.Atkinson, R. J. Herrnstein, G. Lindzey, R. D.Luce (Eds.), Stevens’ handbook of experimentalpsychology: Vol. 1, Perception and Motivation(2nd ed., pp. 195–276). New York: Wiley.

Hol, K., & Treue, S. (2001). Different populationsof neurons contribute to the detection and dis-crimination of visual motion. Vision Research,41, 685–689.

Hotson, J. R., & Anand, S. (1999). The selectiv-ity and timing of motion processing in humantemporo-parieto-occipital and occipital cortex:A transcranial magnetic stimulation study. Neu-ropsychologia, 37, 169–179.

Hubel, D. H., & Wiesel, T. N. (1968). Recep-tive fields and functional architecture of mon-key striate cortex. Journal of Physiology, 195,215–243.

James, W. (1890). Principles of psychology. NewYork: Holt.

Jancke, D. (2000). Orientation formed by a spot’strajectory: A two-dimensional population ap-proach in primary visual cortex. Journal ofNeuroscience, 20, RC86(1–6).

Johansson, G. (1973). Visual perception of biolog-ical motion and a model for its analysis. Percep-tion & Psychophysics, 14, 201–211.

Johnston, A., & Wright, M. J. (1986). Matchingvelocity in central and peripheral vision. VisionResearch, 26, 1099–1109.

Katz, E., Gizzi, M. S., Cohen, B., & Malach, R.(1990). The perceived speed of an object isaffected by the distance it travels. Perception,19, 387.

Kirschen, M. P., Kahana, M. J., Sekuler, R., &Burack, B. (2000). Optic flow helps humanslearn to navigate through synthetic environ-ments. Perception, 29, 801–818.

Knill, D. C., Kersten, D., & Yuille, A. (1996). ABayesian formulation of visual perception. InD. C. Knill & W. Richards (Eds.), Perceptionas Bayesian inference. Cambridge: CambridgeUniversity Press, pp. 1–21.

Koenderink, J. J. (1986). Optic flow. VisionResearch, 26, 161–180.

Kourtzi, Z., & Kanwisher, N. G. (2000). Activa-tion in human MT/MST by static images withimplied motion. Journal of Cognitive Neuro-science, 12, 48–55.

Kozlowski, L. T., & Cutting, J. E. (1977). Rec-ognizing the gender of walkers from dynamicpoint-light displays. Perception & Psycho-physics, 21, 575–580.

Kramer, P., & Yantis, S. (1997). Perceptual group-ing in space and time: Evidence from the Ternusdisplay. Perception & Psychophysics, 59, 87–99.

Krekelberg, B., & Lappe, M. (2001). Neuronallatencies and the position of moving objects.Trends in Neuroscience, 24, 335–339.

Lappin, J. S., Donnelly, M. P., & Kojima, H. (2001).Coherence of early motion signals. VisionResearch, 41, 1631–1644.

Ledgeway, T. (1994). Adaptation to second-order motion results in a motion aftereffectfor directionally-ambiguous test stimuli. VisionResearch, 34, 2879–2889.



Lee, D. N. (1976). A theory of visual control ofbraking based on information about time-to-collision. Perception, 5, 437–459.

Lee, D. N. (1980). The optic flow field: The foun-dation of vision. Philosophical Transactions ofthe Royal Society of London (Series B), 290,169–179.

Lee, S. H., & Blake, R. (1999). Detection of tempo-ral structure depends on spatial structure. VisionResearch, 39, 3033–3048.

Leibowitz, H. (1955). Effect of reference lines onthe discrimination of movement. Journal of theOptical Society of America, 45, 829–830.

Levinson, E., & Sekuler, R. (1975). The indepen-dence of channels in human vision selective fordirection of movement. Journal of Physiology,250, 347–366.

Lewkowicz, D. J. (in press). Heterogeneity andheterochrony in the development of intersensoryperception. Cognitive Brain Research.

Lindsey, D., & Todd, J. T. (1998). Opponent mo-tion interactions in the perception of transpar-ent motion. Perception & Psychophysics, 60,558–574.

Lindsey, D. T. (2001). Direction repulsion in un-filtered and ring-filtered Julesz textures. Percep-tion & Psychophysics, 63, 226–240.

Lorenceau, J., Shiffrar, M., Wells, N., & Castet,E. (1993). Different motion sensitive units areinvolved in recovering the direction of movinglines. Vision Research, 33, 1207–1217.

Magnussen, S., & Greenlee, M. W. (1992). Reten-tion and disruption of motion information in vi-sual short-term memory. Journal of Experimen-tal Psychology: Learning, Memory & Cognition,18, 151–156.

Marshak, W., & Sekuler, R. (1979). Mutual re-pulsion between moving visual targets. Science,205, 1399–1401.

Mateeff, S., Genova, B., & Hohnsbein, J. (1999).The simple reaction time to changes in directionof visual motion. Experimental Brain Research,124, 391–394.

Mather, G. (1980). The movement aftereffect and adistribution-shift model for coding direction ofvisual movement. Perception, 9, 379–392.

Mather, G., & Murdoch, L. (1994). Genderdiscrimination in biological motion displaysbased on dynamic cues. Proceedings of theRoyal Society of London, Series B, 258,273–279.

Mather, G., Verstraten, F. A. J., & Anstis, S. (1998).Preface. In G. Mather, F. Verstraten, & S. Anstis(Eds.), The motion aftereffect: A modern per-spective (pp. vii–xii). Cambridge: MIT Press.

McCarthy, J. E. (1993). Directional adaptationeffect with contrast modulated stimuli. VisionResearch, 33, 2653–2662.

McKee, S. P. (1981). A local mechanism for dif-ferential velocity detection. Vision Research, 21,491–500.

McKee, S. P., Silverman, G. H., & Nakayama, K.(1986). Precise velocity discrimination despiterandom variations in temporal frequency andcontrast. Vision Research, 26, 609–619.

McKee, S. P., & Watamaniuk, S. N. J. (1994).The psychophysics of motion perception. InA. T. Smith, & R. J. Snowden, Eds., Visual de-tection of motion (pp. 85–114). New York City:Academic Press.

Metzger, W. (1934). Beobachtungen uberphanomenale Identitat [Observations on phe-nomenal identity]. Psychologische Forschung,19, 1–60.

Morrone, M. C., Tosetti, M., Montanaro, D.,Fiorentini, A., Cioni, G., & Burr, D. C. (2000).A cortical area that responds specifically to opticflow, revealed by fMRI. Nature Neuroscience, 3,1322–1328.

Movshon, J. A., Adelson, E. H., Gizzi, M. S., &Newsome, W. T. (1986). The analysis of movingvisual patterns. In C. Chagas, R. Gattas, & C. G.Gross (Eds.), Pattern recognition mechanisms(pp. 177–151). New York: Springer.

Mulligan, J. B. (1992). Motion transparency is re-stricted to two planes. Investigative Ophthalmol-ogy and Visual Science, 33, 1049.

Nakayama, K. (1981). Differential motion hyper-acuity under conditions of common image mo-tion. Vision Research, 21, 1475–1482.

Nakayama, K., & Tyler, C. W. (1981). Psycho-physical isolation of movement sensitivity by


References 171

removal of familiar position cues. VisionResearch, 21, 427–433.

Nawrot, M., & Blake, R. (1991a). The interplaybetween stereopsis and structure from motion.Perception & Psychophysics, 49, 230–244.

Nawrot, M., & Blake, R. (1991b). A neural-network model of kinetic depth. Visual Neuro-science, 6, 219–227.

Nawrot, M., & Blake, R. (1993). On the perceptualidentity of dynamic stereopsis and kinetic depth.Vision Research, 33, 1561–1571.

Nawrot, M., & Sekuler, R. (1990). Assimilationand contrast in motion perception: Explorationsin cooperativity. Vision Research, 30, 1439–1451.

Neri, P., Morrone, M. C., & Burr, D. C.(1998). Seeing biological motion. Nature, 359,894–896.

Newsome, W. T., Britten, K. H., & Movshon, J. A.(1989). Neuronal correlates of a perceptual de-cision. Nature, 341, 52–54.

Newsome, W. T., & Pare, E. B. (1988). A selectiveimpairment of motion perception following le-sions of the middle temporal visual area (MT).Journal of Neuroscience, 8, 2201–2211.

Nishida, S., Ashida, H., & Sato, T. (1994). Com-plete interocular transfer of motion aftereffectwith flickering test. Vision Research, 34, 2707–2716.

Nishida, S., & Sato, T. (1995). Motion aftereffectwith flickering test patterns reveals higher stagesof motion processing. Vision Research, 35, 477–490.

Orban, G. A., de Wolf, J., & Maes, H. (1984). Fac-tors influencing velocity coding in the humanvisual system. Vision Research, 24, 33–39.

Orger, M. B., Smear, M. C., Anstis, S. M., &Baier, H. (2000). Perception of Fourier andnon-Fourier motion by larval zebrafish. NatureNeuroscience, 3, 128–1133.

Pack, C. C., & Born, R. T. (2001). Temporal dy-namics of a neural solution to the aperture prob-lem in visual area MT of macaque brain. Nature,409, 1040–1042.

Pantle, A. (1973). Stroboscopic movement basedupon global information in successively pre-

sented visual patterns. Journal of the OpticalSociety of America, 63, 1280.

Pantle, A. (1974). Motion aftereffect magnitudeas a measure of spatio-temporal response prop-erties of direction-selective analyzers. VisionResearch, 14, 1229–1236.

Pantle, A. (1998). How do measures of the mo-tion aftereffect measure up? In G. Mather,F. Verstraten, & S. Anstis (Eds.), The motionaftereffect: A modern perspective (pp. 25–39).Cambridge: MIT Press.

Pantle, A., Gallogly, D. P., & Piehler, O. C. (2000).Direction biasing by brief apparent motionstimuli. Vision Research, 40, 1979–1991.

Pantle, A., & Picciano, L. (1976). A multistablemovement display: Evidence for two separatemotion systems in human vision. Science, 193,500–502.

Parker, A. J., & Newsome, W. T. (1998). Senseand the single neuron: Probing the physiologyof perception. Annual Review of Neuroscience,21, 227–277.

Pascual-Leone, A., & Walsh, V. (2001). Fastbackprojections from the motion to the pri-mary visual area necessary for visual awareness.Science, 292, 510–512.

Pascual-Leone, A., Walsh, V., & Rothwell, J.(2000). Transcranial magnetic stimulation incognitive neuroscience: Virtual lesion, chrono-metry, and functional connectivity. CurrentOpinion in Neurobiology, 10, 232–237.

Pasternak, T. (1987). Discrimination of dif-ferences in speed and flicker rate dependson directionally-selective mechanisms. VisionResearch, 27, 1881–1890.

Patterson, R. (1999). Stereoscopic (cyclopean) mo-tion sensing. Vision Research, 39, 3329–3345.

Pelli, D., & Farell, B. J. (1999). Why use noise?Journal of the Optical Society of America A,16, 647–653.

Pylyshyn, Z. W., & Storm, R. W. (1988). Track-ing multiple independent targets: Evidence for aparallel tracking mechanism. Spatial Vision, 3,179–197.

Qian, N., & Andersen, R. A. (1994). Transparentmotion perception as detection of unbalanced



motion signals: II. Physiology. The Journal ofNeuroscience, 14, 7367–7380.

Qian, N., Andersen, R. A., & Adelson, E. H. (1994).Transparent motion perception as detection ofunbalanced motion signals: I. Psychophysics.The Journal of Neuroscience, 14, 7357–7366.

Ramachandran, V. S., & Anstis, S. M. (1985). Per-ceptual organization in multistable apparent mo-tion. Perception, 14, 135–143.

Raymond, J. E. (1993). Complete interocular trans-fer of motion adaptation effects on motion co-herence thresholds. Vision Research, 33, 1865–1870.

Raymond, J. E. (2000). Attentional modulation ofvisual motion perception. Trends in CognitiveScience, 4, 42–50.

Raymond, J. E., O’Donnell, H. L., & Tipper, S. P.(1998). Priming reveals attentional modulationof human motion sensitivity. Vision Research,38, 2863–2867.

Regan, D. (1989). Orientation discrimination forobjects defined by relative motion and objectsdefined by luminance contrast. Vision Research,29, 1389–1400.

Riggs, L. A., & Day, R. H. (1980). Visual after-effects derived from inspection of orthogonallymoving patterns. Science, 208, 416–418.

Rizzo, M., & Nawrot, M. (1998). Perception ofmovement and shape in Alzheimer’s disease.Brain, 121, 2259–2270.

Robson, J. G. (1966). Spatial and temporalcontrast-sensitivity functions of the visual sys-tem. Journal of the Optical Society of America,56, 1141–1142.

Rose, A. (1948). The sensitivity performance of thehuman eye on an absolute scale. Journal of theOptical Society of America, 38, 196–208.

Rudolph, K., & Pasternak, T. (1999). Transient andpermanent deficits in motion perception afterlesions of cortical areas MT and MST in themacaque monkey. Cerebral Cortex, 9, 90–100.

Salzman, C. D., Murasugi, C. M., Britten, K. H.,& Newsome, W. T. (1992). Microstimulation invisual area MT: Effects on direction discrimina-tion performance. Journal of Neuroscience, 12,2331–2335.

Savelsbergh, G. J. P., Whiting, H. T. A., & Bootsma,R. J. (1991). Grasping tau. Journal of Experi-mental Psychology: Human Perception and Per-formance, 17, 315–322.

Scase, M. O., Braddick, O. J., & Raymond, J. E.(1996). What is noise for the motion system?Vision Research, 36, 2579–2586.

Schall, J. D. (2000). From sensory evidence to amotor command. Current Biology, 10, R404–R406.

Schenk, T., & Zihl, J. (1997). Visual motionperception after brain damage: I. Deficits inglobal motion perception. Neuropsychologia,35, 1289–1297.

Schiff, W. (1965). Perception of impending colli-sion. Psychological Monographs, 79, 1–26.

Schrater, P. R., Knill, D. C., & Simoncelli, E. P.(2001). Perceiving visual expansion withoutoptic flow. Nature, 410, 816–819.

Sekuler, A. B., & Bennett, P. J. (1996). Spatialphase differences can drive apparent motion.Perception & Psychophysics, 58, 174–190.

Sekuler, A. B., & Sekuler, R. (1993). Repre-sentational development of direction in motionperception: A fragile process. Perception, 22,899–915.

Sekuler, A. B., Sekuler, E. B., & Sekuler, R.(1990). How the visual system detects changesin the direction of moving targets. Perception,19, 181–196.

Sekuler, R., & Blake, R. (2001). Perception. NewYork: McGraw-Hill.

Sekuler, R., & Ganz, L. (1963). A new aftereffect ofseen movement with a stabilized retinal image.Science, 139, 419–420.

Sekuler, R., Sekuler, A. B., & Lau, R. (1997).Sound changes perception of visual motion.Nature, 384, 308–309.

Shadlen, M. N., & Newsome, W. T. (1996). Mo-tion perception: Seeing and deciding. Proceed-ing of the National Academy of Science USA,93, 628–633.

Shulman, G. L., Ollinger, J. M., Akbudak, E.,Conturo, T. E., Snyder, A. Z., Petersen, S. E.,& Corbetta, M. (1999). Areas involved in en-coding and applying directional expectations to


References 173

moving objects. Journal of Neuroscience, 19,9480–9496.

Sigman, M., Cecchi, G. A., Gilbert, C. D., &Magnasco, M. O. (2001). On a common cir-cle: Natural scenes and Gestalt rules. Proceed-ing of the National Academy of Science USA, 98,1935–1940.

Smith, A. T., & Edgar, G. K. (1991). The separa-bility of temporal frequency and velocity. VisionResearch, 31, 321–326.

Smith, M. R. H., Flach, J. M., Dittman, S. M., &Stanard, T. (2001). Monocular optical con-straints on collision control. Journal of Ex-perimental Psychology: Human Perception &Performance, 27, 395–410.

Snowden, R. J. (1990). Suppressive interactionsbetween moving patterns: Role of velocity. Per-ception & Psychophysics, 47, 74–78.

Snowden, R. J. (1992). Sensitivity to relative andabsolute motion. Perception, 21, 563–568.

Snowden, R. J., Treue, S., Erickson, R. G., &Andersen, R. A. (1991). The response of areaMT and V1 neurons to transparent motion. TheJournal of Neuroscience, 11, 2768–2785.

Snowden, R. J., & Verstraten, F. A. J. (1999). Mo-tion transparency: Making models of motionperception transparent. Trends in Cognitive Sci-ences, 3, 369–377.

Sperling, G., & Lu, Z.-L. (1998). A systems analy-sis of visual motion perception. In T. Watanabe(Ed.), High-level motion processing: Computa-tional, neurobiological and psychophysical per-spectives (pp. 153–186). Cambridge: MIT Press.

Spigel, I. M. (1960). The effects of differentialpost-exposure illumination on the decay of themovement after-effect. Journal of Psychology,50, 209–210.

Spigel, I. M. (1962a). Contour absence as a crit-ical factor in the inhibition of the decay of amovement aftereffect. Journal of Psychology,54, 221–228.

Spigel, I. M. (1962b). Relation of MAE durationto interpolated darkness intervals. Life Sciences,1, 239–242.

Spigel, I. M. (1964). The use of decay inhibitionin an examination of central mediation in move-

ment aftereffects. Journal of General Psychol-ogy, 70, 241–247.

Steiner, V., Blake, R., & Rose, D. (1994). Interoc-ular transfer of expansion, rotation, and transla-tion motion aftereffects. Perception, 23, 1197–1202.

Stevens, S. S. (1951). Mathematics, measurementand psychophysics. In S. S. Stevens (Ed.), Hand-book of experimental psychology (pp. 1–49).New York: Wiley.

Stone, L. S., & Thompson, P. (1992). Humanspeed perception is contrast dependent. VisionResearch, 32, 1535–1549.

Sunaert, S., Van Hecke, P., Marchal, G., &Orban, G. A. (1999). Motion-responsive re-gions of the human brain. Experimental BrainResearch, 127, 355–370.

Sutherland, N. S. (1961). Figural aftereffects andapparent size. Quarterly Journal of Experimen-tal Psychology, 13, 222–228.

Symons, L. A., Pearson, P. M., & Timney, B.(1996). The aftereffect to relative motion doesnot show interocular transfer. Perception, 25,651–660.

Tanaka, K., & Saito, H. (1989). Analysis of mo-tion of the visual field by direction, expan-sion/contraction, and rotation cells clustered inthe dorsal part of the medial superior temporalarea of the macaque monkey. Journal of Neuro-physiology, 62, 626–641.

Tayama, T. (2000). The minimum temporal thresh-olds for motion detection of grating patterns.Perception, 29, 761–769.

Tetewsky, S. J., & Duffy, C. J. (1999). Visual lossand getting lost in Alzheimer’s disease. Neurol-ogy, 52, 958–965.

Thompson, P. (1982). Perceived rate of move-ment depends on contrast. Vision Research, 22,377–380.

Thompson, P. (1998). Tuning of the motionaftereffect. In G. Mather, F. Verstraten, &S. Anstis (Eds.), The motion aftereffect: Amodern perspective (pp. 41–55). Cambridge:MIT Press.

Thorson, J., Lange, G. D., & Biederman-Thorson,M. (1969). Objective measure of the dynamics



of a visual movement illusion. Science, 164,1087–1088.

Tong, F. (2001). Brain at work: Play by play. NatureNeuroscience, 4, 560–561.

Tresilian, J. R. (1999). Visually timed action: Time-out for “tau”? Trends in Cognitive Science, 3,301–310.

Treue, S., & Maunsell, J. H. R. (1999). Effects ofattention on the processing of motion in macaquemiddle temporal and medial superior temporalvisual cortical areas. Journal of Neuroscience,19, 7591–7602.

Treue, S., & Trujillo, J. C. (1999). Feature-basedattention influences motion processing gain inmacaque visual cortex. Nature, 399, 575–579.

Tse, P., Cavanagh, P., & Nakayama, K. (1998). Therole of parsing in high-level motion processing.In T. Watanabe (Ed.), High-level motion pro-cessing: Computational, neurobiological andpsychophysical perspectives (pp. 249–266).Cambridge: MIT Press.

Turano, K., & Pantle, A. (1989). On the mechanismthat encodes the movement of contrast varia-tions: Velocity discrimination. Vision Research,29, 207–221.

Tynan, P. D., & Sekuler, R. (1982). Motion pro-cessing in peripheral vision: Reaction timeand perceived velocity. Vision Research, 22,61–68.

Ullman, S. (1984). Maximizing rigidity: The in-cremental recovery of a 3-D structure from rigidand non-rigid motion. Perception, 13, 255–274.

Ungerleider, L. G., & Haxby, J. V. (1994). “What”and “where” in the human brain. Current Opin-ion in Neurobiology, 4, 157–165.

Vaina, L. M., Cowey, A., & Kennedy, D. (1999).Perception of first- and second-order motion:Separable neurological mechanisms? HumanBrain Mapping, 7, 67–77.

Vaina, L. M., Grzywacz, N. M., LeMay, M.,Bienfang, D., & Wolpow, E. (1998). Perceptionof motion discontinuities in patients with selec-tive motion deficits. In T. Watanabe (Ed.), High-level motion processing: Computational, neu-robiological and psychophysical perspectives(pp. 213–247). Cambridge: MIT Press.

Vaina, L. M., Lemay, M., Bienfang, D. C., Choi,A. Y., & Nakayama, K. (1990). Intact “biologicalmotion” and “structure from motion” perceptionin a patient with impaired motion mechanisms:A case study. Visual Neuroscience, 5, 353–369.

van den Berg, A. V. (1992). Robustness of percep-tion of heading from optic flow. Vision Research,32, 1285–1296.

van Meeteren, A., & Barlow, H. B. (1981). The sta-tistical efficiency for detecting sinusoidal mod-ulation of average dot density in random figures.Vision Research, 21, 765–777.

Van Oostende, S., Sunaert, S., Van Hecke, P.,Marchal, G., & Orban, G. (1997). The kineticoccipital (KO) region in man: An fMRI study.Cerebral Cortex, 7, 690–701.

Verghese, P., Watamaniuk, S. N. J., McKee, S. P.,& Grzywacz, N. M. (1999). Local motion de-tectors cannot account for the detectability of anextended trajectory in noise. Vision Research,39, 19–30.

Verstraten, F., Fredericksen, R. E., & van de Grind,W. A. (1994). The movement aftereffect of bi-vectorial transparent motion. Vision Research,34, 349–358.

Verstraten, F. A., Fredericksen, R. E., van Wezel,R. J. A., Lankheet, M. J. M., & van de Grind,W. A. (1996). Recovery from adaptation for dy-namic and static motion aftereffects: Evidencefor two mechanisms. Vision Research, 36, 421–424.

Verstraten, F. A. J., Fredericksen, R. E., van Wezel,R. J. A., Boulton, J. C., & van de Grind,W. A. (1996). Directional motion sensitivityunder transparent motion conditions. VisionResearch, 36, 2333–2336.

Viviani, P., & Stucchi, N. (1989). The effect ofmovement velocity on form perception: Geo-metric illusions in dynamic displays. Perception& Psychophysics, 46, 266–274.

von Grunau, M. W. (1986). A motion aftereffect forlong-range stroboscopic apparent motion. Per-ception and Psychophysics, 40, 31–38.

Wade, N. J., Swanston, M. T., & De Weert, C. M. M.(1993). On interocular transfer of motion after-effects. Perception, 22, 1365–1380.


References 175

Wade, N. J., & Verstraten, F. A. J. (1998). Intro-duction and historical overview. In G. Mather, F.Verstraten, & S. Anstis (Eds.), The motion after-effect: A modern perspective (pp. 1–23). Cam-bridge: MIT Press.

Wallach, H., & O’Connell, D. N. (1953). Thekinetic depth effect. Journal of experimentalPsychology, 45, 205–217.

Walls, G. (1942). The vertebrate retina and itsadaptive radiation. Bloomfield Hills, MI: Cran-brook Press.

Walsh, V., Ellison, A., Battelli, L., & Cowey, A.(1998). Task-specific impairments and enhance-ments induced by magnetic stimulation of hu-man visual area V5. Proceeding of the RoyalSociety of London, Series B, 265, 537–543.

Warren, W. H., Jr., Blackwell, A. W., Kurtz, K. J.,Hatsopoulos, N. G., & Kalish, M. L. (1991).On the sufficiency of the velocity field for per-ception of heading. Biological Cybernetics, 65,311–320.

Warren, W. H., Jr., Blackwell, A. W., & Morris,M. W. (1988). Age differences in perceiving thedirection of self-motion from optical flow. Jour-nal of Gerontology, 44, P147–P153.

Warren, W. H., Jr., Kay, B. A., Zosh, W. D.,Duchon, A. P., & Sahuc, S. (2001). Optic flowis used to control human walking. Nature Neu-roscience, 4, 213–216.

Warren, W. H., Jr., Mestre, D. R., Blackwell, A. W.,& Morris, M. W. (1991). Perception of circularheading from optical flow. Journal of Experi-mental Psychology: Human Perception & Per-formance, 17, 28–43.

Warren, W. H., Jr., Young, D. S., & Lee, D. N.(1986). Visual control of step length during run-ning over irregular terrain. Journal of Exper-imental Psychology: Human Performance andPerception, 12, 259–266.

Watamaniuk, S. N. J. (1993). An ideal observerfor discrimination of the global direction of dy-namic random dot stimuli. Journal of the OpticalSociety of America A, 10, 16–28.

Watamaniuk, S. N. J., & Duchon, A. (1992). Thehuman visual system averages speed informa-tion. Vision Research, 32, 931–941.

Watamaniuk, S. N. J., & McKee, S. P. (1995).“Seeing” motion behind occluders. Nature, 377,729–730.

Watamaniuk, S. N. J., & McKee, S. P. (1998). Si-multaneous encoding of direction at a local andglobal scale. Perception & Psychophysics, 60,191–200.

Watamaniuk, S. N. J., McKee, S. P., & Grzywacz,N. M. (1995). Detecting a trajectory embed-ded in random-direction motion noise. VisionResearch, 35, 65–77.

Watamaniuk, S. N. J., & Sekuler, R. (1992). Tem-poral and spatial integration in dynamic randomdot stimuli. Vision Research, 32, 2341–2347.

Watamaniuk, S. N. J., Sekuler, R., & Williams,D. W. (1989). Direction discrimination in com-plex dynamic displays: The integration of direc-tion information. Vision Research, 29, 47–59.

Watanabe, K., & Shimojo, S. (2001). When soundaffects vision: Effects of auditory grouping onvisual motion perception. Psychological Sci-ence, 12, 109–116.

Watson, A. B. (1979). Probability summation overtime. Vision Research, 19, 515–522.

Watson, A. B., Ahumada, A. J., Jr., & Farrell, J. E.(1986). Window of visibility: A psychophysi-cal theory of fidelity in time-sampled visual mo-tion displays. Journal of the Optical Society ofAmerica A, 3, 300–307.

Watson, A. B., & Robson, J. G. (1981). Discrimi-nation at threshold: Labeled detectors in humanvision. Vision Research, 21, 1115–1122.

Watson, A. B., Thompson, P. G., Murphy, B. J., &Nachmias, J. (1980). Summation and discrimi-nation of gratings moving in opposite directions.Vision Research, 20, 341–347.

Watson, A. B., & Turano, K. (1995). The optimalmotion stimulus. Vision Research, 35, 325–336.

Westheimer, G. (1988). Vision: Space and move-ment. In R. C. Atkinson, R. J. Herrnstein, G.Lindzey, & R. D. Luce (Eds.), Stevens’ hand-book of experimental psychology: Vol. 1. Per-ception and Motivation (2nd ed., pp. 165–193).New York: Wiley.

Westheimer, G., & Wehrhahn, C. (1994). Dis-crimination of direction of motion in human



vision. Journal of Neurophysiology, 71, 33–37.

Williams, D., Phillips, G., & Sekuler, R. (1986).Hysteresis in the perception of motion direc-tion as evidence for neural cooperativity. Nature,324, 20–26.

Williams, D. W., & Sekuler, R. (1984). Coher-ent global motion percepts from stochastic localmotions. Vision Research, 24, 55–62.

Wohlgemuth, A. (1911). On the after-effect ofseen movement. British Journal of Psychology,(Suppl. 1), 1–117.

Wuerger, S., Shapley, R., & Rubin, N. (1996). “Onthe visually perceived direction of motion” byHans Wallach: 60 years later. Perception, 25,1317–1367.

Wylie, D. R., Bischof, W. F., & Frost, B. J. (1998).Common reference frame for neural coding oftranslational and rotational optic flow. Nature,392, 278–282.

Yantis, S. (1992). Multielement visual tracking: At-tention and perceptual organization. CognitivePsychology, 24, 295–340.

Yu, K. (2000). Can semantic knowledge influencemotion correspondence? Perception, 29, 693–707.

Zacks, J. M., Braver, T. S., Sheridan, M. A.,Donaldson, D. I., Snyder, A. Z., Ollinger, J. M.,Buckner, R. L., & Raichle, M. E. (2001). Hu-man brain activity time-locked to perceptualevent boundaries. Nature Neuroscience, 4, 651–655.

Documents

STEVENS’ HANDBOOK OF EXPERIMENTAL PSYCHOLOGYpeople.brandeis.edu/~sekuler/papers/sekulerStevens... · pashler-44104 book January 11, 2002 11:20 122 Motion Perception Figure 4.1 Four