light adaptation in motion direction judgments

Takeuchi et al. Vol. 18, No. 4 /April 2001 /J. Opt. Soc. Am. A 755

Light adaptation in motion direction judgments

Tatsuto Takeuchi

Department of Psychology, University of California at Berkeley, Berkeley, California 94720, and NTT CommunicationScience Laboratories, Atsugi-shi, Morinosato-Wakamiya 3-1, Kanagawa 243-0198, Japan

Karen K. De Valois

Department of Psychology and Vision Science Group, University of California at Berkeley, Berkeley, California 94720

Isamu Motoyoshi

NTT Communication Science Laboratories, Atsugi-shi, Morinosato-Wakamiya 3-1, Kanagawa 243-0198, Japan

Received July 10, 2000; revised manuscript received October 23, 2000; accepted October 23, 2000

We examined the time course of light adaptation in the visual motion system. Subjects judged the directionof a two-frame apparent-motion display, with the two frames separated by a 50-ms interstimulus interval ofthe same mean luminance. The phase of the first frame was randomly determined on each trial. The gratingpresented in the second frame was phase shifted either leftward or rightward by p/2 with respect to the gratingin the first frame. At some variable point during the first frame, the mean luminance of the pattern increasedor decreased by 1–3 log units. Mean luminance levels varied from scotopic or low mesopic to photopic levels.We found that the perceived direction of motion depended jointly on the luminance level of the first framegrating and the time at which the shift in average luminance occurs. When the average luminance increasesfrom scotopic or mesopic to photopic levels at least 0.5 s before the offset of the first frame, motion in the 3p/2direction is perceived. When average luminance decreases to low mesopic or scotopic levels, motion in the p/2direction is perceived if the change occurs 1.0 s or more before first frame offset, depending on the size of theluminance step. Thus light adaptation in the visual motion system is essentially complete within 1 s. Thissuggests a rapid change in the shape (biphasic or monophasic) of the temporal impulse response functions thatfeed into a first-order motion mechanism. © 2001 Optical Society of America

OCIS codes: 330.5020, 330.5510, 330.4150, 330.7320.

1. INTRODUCTIONIn the natural environment, the ambient light level maychange by a factor of 108 between day and night.1 Evenunder daylight conditions, the average luminance levelfluctuates between photopic and mesopic levels.2 Rapidlight adaptation, the ability of the visual system to adjustits visual sensitivity ( gain) and spatiotemporal responseproperties (dynamics) to the prevailing illuminationquickly, is essential in such an environment. Many ear-lier studies examining the time course of light adaptationhave measured detection thresholds during or followingchanges in background luminance.1,3–7 They have shownthat the visual system can recalibrate its sensitivity rap-idly to the new light levels.

The visual system contains specialized mechanisms toanalyze the velocity of moving objects,8 but motion per-ception is not invariant with luminance level.9–15 For ex-ample, it has been shown that the perceived direction orvelocity of motion can depend on the state of lightadaptation.10–13 The present study was designed to in-crease our understanding of the time course of light adap-tation in the visual motion system. We ask how soon mo-tion perception becomes stable after the backgroundluminance level changes.

For this purpose, we took advantage of a visual motionillusion known as motion reversal.16–19 When a patternis presented first in one position, then another, an ob-

0740-3232/2001/040755-10$15.00 ©

server may experience a strong sensation of motion.This is traditionally called apparent motion. If the twopresentations are separated by a brief interstimulus in-terval (ISI) of the appropriate duration, however, and ifthe interval is filled with a blank screen equated in space-averaged luminance to the pattern display, then the ap-parent direction of motion may be reversed. We reportedin a previous study that the perception of motion reversaldepends on the average luminance level.10 We showedthat the probability of seeing reversed motion decreasesas adapting level falls, finally disappearing completely atscotopic adaptation levels, at which point veridical motionis perceived. By measuring how quickly the perceived di-rection changes after a sudden change of background lu-minance, we hope to estimate the time course of light ad-aptation and describe the temporal responsecharacteristics (dynamics) of the visual motion system.

2. GENERAL METHODSA. SubjectsFive subjects, KM, IS, EK, HI, and TT (two female andthree male, 18–34 years old), participated in the experi-ment. TT is an author of this report. The other subjectswere paid volunteers and were unaware of the purposeand ongoing results of the experiments. All subjects hadnormal or corrected-to-normal vision.

2001 Optical Society of America

756 J. Opt. Soc. Am. A/Vol. 18, No. 4 /April 2001 Takeuchi et al.

B. ApparatusStimuli were generated by a visual stimulus generatorVSG (Cambridge Research Systems) controlled by aPentium-based computer and were displayed on a 20-in.RGB monitor (SONY Multiscan20se). The refresh rate ofthe monitor was 100 Hz, and the spatial resolution was1024 3 768 pixels. The gray-scale resolution was 15 bitsper gun. The monitor was calibrated with a TOPCONBM-5 colorimeter, and its output was linearized ( gammacorrected) under software control. For all experimentsusing luminance-varying stimuli, the space-averagedchromaticity (CIE 1931) of the display was x 5 0.29, y5 0.30. Subjects observed the display through a 2-mmartificial pupil, with head position maintained by chinand head rests. Viewing distance was 57 cm. The roomwas darkened and light shielded, with no other source ofillumination present. Three average luminance levels,30.0 cd/m2 [2.0 log photopic trolands (log Tp),]0.3 cd/m2 (0.0 log Tp), and 0.03 cd/m2(21.0 log Tp), rang-ing from photopic to low mesopic or scotopic levels,1,20,21

were used. The average luminance level was manipu-lated both by software changes and by the use of appro-priate neutral density filters placed just distal to the ar-tificial pupil. As described below, the average luminancechanged during stimulus presentations. Sudden lumi-nance changes were accomplished by changing the con-tent of the lookup table controlling the voltage output tothe display monitor. We restricted the contrast range ofthe stimuli to 123 direction discrimination threshold orless and the range of the change in the background lumi-nance to 3 log units or less. These values are within thelimits imposed by the monitor. In most experiments, weused lower contrasts (63 direction discrimination thresh-old) to reduce the effects of inherent nonlinearities of thevoltage output in the monitor (see Subsection 4.B).

C. Visual StimuliA luminance-varying vertical sine-wave grating subtend-ing 8.0 (H) 3 3.0 (V) deg was presented in the center ofthe display. The spatial frequency of the grating was 0.5c/deg. The edges of the stimulus were tapered by aGaussian function with a space constant of 1.0 deg (anelongated Gabor pattern). The remainder of the screenwas a uniform field of luminance equal to the space-averaged luminance of the grating. The motion sequencecomprised two frames of the grating, either with or with-out a 50-ms ISI. The phase of the first frame was ran-domly determined on each trial. The grating presentedin the second frame was phase shifted either leftward orrightward by p/2 with respect to the grating in the firstframe. When the ISI was presented, a blank screenequated to the grating in space-averaged luminance andchromaticity was inserted between the two gratings’frames. The onset of the first frame and the offset of thesecond frame were ramped by a Gaussian function ( s t5 200 ms); the offset of the first frame and the onset ofthe second frame were step functions. The duration ofthe first frame (excluding the onset ramp) was 2.5 s, andthe duration of the second frame (excluding the offsetramp) was 1.0 s (see Fig. 3 below). Since the average lu-minance was changed during the presentation of the firstframe, and the timing of the luminance change is the ex-

perimental variable in experiment 3, we made the dura-tion of the first frame longer than the duration of the sec-ond frame. A small fixation cross was presented at thecenter of the display for 500 ms before the onset of eachstimulus, and subjects were instructed to fixate the centerof the display. In all experiments, the task of the subjectwas to judge the direction of motion (left or right). Sub-jects were required to respond within 2.5 s after the offsetof the second frame. If the subjects failed to respondwithin this time limit, the trial was repeated. Subjectsreported that the procedure was not difficult, and fewerthan 1% of trials were repeated in each experiment.

In experiment 1, we measured the contrast sensitivityfor direction discrimination in order to equate the effec-tive visibility of the stimulus in terms of multiples ofthreshold under different average luminance conditions.In experiment 2, we measured the perceived direction ofmotion at different background luminance levels. In ex-periment 3, we estimated the time course of light adapta-tion in the motion system by measuring how rapidly per-ceived direction changes after the average luminancelevel changes.

3. EXPERIMENT 1: CONTRASTSENSITIVITY FOR DIRECTIONDISCRIMINATIONA. MethodsSince contrast sensitivity varies with average luminancelevel, it is important to make stimuli equally visible un-der different background luminance conditions. One wayto equate visibility under different conditions is to set theluminance contrast of the various stimuli to equal mul-tiples of the relevant threshold. Thus we began by deter-mining contrast sensitivity for direction discrimination ofthe two-frame sine-wave gratings described above. NoISI was inserted between the first and second frames.

We used a two-alternative, temporal forced-choice pro-cedure. In one of two intervals, the motion was leftward(p/2 shift leftward); in the other interval, it was rightward(p/2 shift rightward). The subject, by pressing one of twomouse buttons, indicated in which interval the leftwardmotion appeared. The two intervals were separated by a1-s blank field of the same space-averaged luminance, andthe onset of each interval was marked by an auditory cue.No feedback was given.

Pattern contrast was varied by using a staircase algo-rithm designed to converge at a 79% correct level.22 Con-trast decreased after three consecutive correct responsesand increased after one wrong response. When the sub-ject reported the p/2 direction, the response was judged tobe correct. The size of the contrast increments or decre-ments decreased as the staircase depth increased, being0.4 log unit in the beginning and falling to a terminalvalue of 0.1 log unit. The threshold for a given staircaserun was computed as the mean of the contrasts of the fi-nal six out of nine turning points. Five staircases wererun for each subject to determine each threshold.

Three staircases in which background luminance levelswere different (2.0, 0.0, or 21.0 log Tp) were interleavedin a single session. Following the subject’s response, thedisplay was filled with a uniform field whose luminance


was equal to the space-averaged luminance of the gratingto be presented in the next trial. The duration of the uni-form field was 3.0 s. We interleaved staircases at differ-ent luminance levels to determine thresholds under con-ditions in which the background luminance varies in timeand subjects do not fully adapt to a single luminancelevel. The contrast threshold measured under these con-ditions would be expected to differ from that measuredwhen the subjects are adapted to a single luminance levelthroughout a session. We will discuss this further below.

B. ResultsThe average direction discrimination threshold contrastsfor five subjects were 1.4% (at 2.0 log Tp), 2.7% (at 0.0 logTp), and 5.0% (at 21.0 log Tp). In the next experiments,the Michelson contrast of the stimulus was set to equalmultiples of the contrast values threshold determined in-dividually for each subject.

4. EXPERIMENT 2: DIRECTIONDISCRIMINATION MEASUREMENTS UNDERDIFFERENT BACKGROUND LUMINANCESA. MethodsTo clarify how the perceived direction of motion with anISI depends on the average luminance, we measured di-rection discrimination under three different backgroundluminance levels (2.0, 0.0, and 21.0 log Tp). Two ISI val-ues, 0 and 50 ms, were selected on the basis of our previ-ous study.10 A blank field of the same space-averaged lu-minance as the grating was presented during the ISI.Under photopic conditions, the p/2 shift direction shouldbe dominant when the ISI is 0 ms, and when the ISI is 50ms the 3p/2 shift direction should be dominant (motionreversal). The method of constant stimuli was used formeasurements of direction discrimination.

The two-frame sine-wave stimulus was displayed 500

ms after an auditory cue. The subject’s task was to indi-cate the perceived direction of motion (leftward or right-ward) by pressing the appropriate button. As in experi-ment 1, the average luminance varied from trial to trial.Successive trials were separated by 3.0 s, during whichthe display was filled with a uniform field equal in lumi-nance to the space-averaged luminance of the grating tobe presented on the next trial. One experimental blockconsisted of 48 trials, including all combinations of two di-rections of motion (left or right), two ISIs (0 and 50 ms),three background luminance levels (2.0, 0.0, and 21.0 logTp), and four contrast levels (three, six, nine, and twelvetimes thresholds). Each subject completed 20 blocks.

B. ResultsFigure 1 shows the perceived direction of two-frame ap-parent motion with and without an ISI, averaged acrossall subjects. Percent response is plotted as a function ofthe multiple of direction discrimination contrast thresh-old measured in experiment 1. Figure 1(A) shows datafrom trials on which the ISI was 0 ms. Subjects per-ceived motion in the p/2 direction on a majority of trials,independent of the Michelson contrast of the stimuli andof the average luminance. Figure 1(B) shows the datawhen the ISI was 50 ms. When the average luminancewas 2.0 log Tp, motion reversal (3p/2 direction) was per-ceived on nearly 80% of trials. When the average lumi-nance was 0.0 log Tp, the subjects perceived the veridicalmotion direction (p/2 direction) on approximately 70% oftrials. When the average luminance was 21.0 log Tp,veridical motion was perceived on 80% to 90% of trials.

These results are consistent with those of our previousstudy,10 in which we showed that the perceived directionof two-frame motion with an ISI depends on the averageluminance, and motion reversal disappears as the aver-age luminance decreases. It should be noted, however,that the subjects in the earlier study were fully adapted

Fig. 1. Averaged responses of five subjects for perceived direction of two-frame apparent motion in experiment 2. (A) Percent responsein the p/2 direction is plotted as a function of the multiple of direction discrimination contrast threshold. A response greater than 50%indicates that the veridical motion direction (p/2 direction) was perceived on the majority of trials; less than 50% means that motionreversal (3p/2 direction) was perceived on the majority of trials. Symbols identify the average luminance level: 2.0 log Tp (solidcircles), 0.0 log Tp (open squares), 21.0 log Tp (solid triangles). No ISI was inserted. Each data point is based on 200 trials. (B) TheISI was 50 ms. Other conditions were as in (A).


to the background luminance before judging the directionof motion. The results of experiment 2 show that theperceived direction is sensitive to the current backgroundluminance and follows the rule found by Takeuchi and DeValois10 even without full light adaptation. Because ofthis sensitivity to the current luminance level, we will beable to take advantage of motion reversal to examine thetime course of light adaptation in the motion system inexperiment 3.

Figure 1 also shows that those tendencies are invariantfor different luminance contrasts. This may reflect thecharacteristic contrast response saturation seen in themotion system at low contrast levels.23,24 Perfect controlof the subjects’ adaptation state is difficult for the experi-ments in our study. Even though we equated for mul-tiples of threshold contrast based on values measured inexperiment 1, there may remain a residual mismatch ineffective luminance contrast in experiments 2 or 3. How-ever, since the effect of contrast on the perceived directionappears to be quite small [Fig. 1(B)], a mismatch in lumi-nance contrast should not significantly affect the results.On the basis of these results, we set the luminance con-trast of stimuli in the next experiment at 63 directiondiscrimination threshold.

Figure 2 replicates a part of Fig. 1(B), plotting the per-ceived direction of motion at 63 direction discriminationthreshold as a function of average luminance. As shownby the vertical arrows in Fig. 2, the perceived direction ofmotion depends on the average luminance. In the next

experiment, we estimate the time course of light adapta-tion by measuring how soon the perceived direction of mo-tion changes after a sudden change of background lumi-nance between the luminance levels represented in Fig. 2.

Fig. 2. Averaged responses of five subjects for perceived direc-tion of apparent motion in experiment 2. Data showing percentresponse in the p/2 direction at 63 direction discrimination con-trast threshold were taken from Fig. 1(B). Percent response inthe p/2 direction is plotted as a function of average luminance(log Tp). The ISI was 50 ms. Arrows indicate the differencesbetween responses in the p/2 direction at different average lumi-nances.

Fig. 3. Upper row, average luminance and contrast profiles of the two-frame motion stimulus with an ISI in a single trial, the stimulusthat was used in experiment 3. At some variable point in frame 1, specified by the timing value t, the average luminance (A) decreasedor (B) increased while luminance contrast was maintained at a constant multiple of direction discrimination threshold. The physicalcontrast thus changed when the background luminance was shifted. Lower row, schematic description of a space–time plot of thestimulus. (A) Luminance decrements, (B) luminance increments. For clarity, the scale of the temporal domain does not reflect theactual values used: The durations of the first and second frames were equated, the duration of the ISI was exaggerated, and the tem-poral window in each frame was not represented in the figure.


5. EXPERIMENT 3: ESTIMATION OF THETIME COURSE OF LIGHT ADAPTATIONBY CHANGING THE AVERAGE LUMINANCEA. MethodsFigure 3 shows the average luminance and contrast pro-files of the motion display (upper row) and a schematic de-scription of a space–time plot of the stimulus (lower row)in a single trial. As in the previous experiments, thestimulus consisted of two frames of sinusoidal gratingsseparated by an ISI. The duration of the full-contrastportions of the first and second frames was set to 2.5 sand 1.0 s, respectively. At some variable point duringthe first frame, the average luminance decreased [Fig.3(A)] or increased [Fig. 3(B)] by 1–3 log units; at the sametime, grating contrast either increased or decreased tomaintain 63 the direction discrimination threshold con-trast measured in experiment 1. The timing value t isthe delay between the shift in average luminance and theoffset of the first frame. The average luminance shiftedbetween 2.0 and 21.0 log Tp, between 2.0 and 0.0 log Tp,or between 0.0 and 21.0 log Tp.

The ISI between the first frame and the second framewas either 0 or 50 ms. A sudden shift in average lumi-nance causes masking, in which a transient temporal re-sponse is generated, reducing the visibility of the stimuliand of stimulus motion. This masking effect is not re-lated to light adaptation or motion computation. To es-timate how long the masking continues after the shift inaverage luminance, we examined conditions in which noISI was inserted, while the other conditions were thesame as shown in Fig. 3. When the ISI was 0 ms, thetiming value t varied from 0.02 to 2.4 s in 18 steps. Asdescribed later, we found that if the timing value t wassmaller than 0.1 s, the directional judgment became al-most impossible (see Fig. 4 below). Therefore, when theISI was 50 ms, we varied the timing value t from 0.1 to2.4 s in 14 steps.

In a trial, 500 ms after the auditory cue the two-framesine-wave stimulus was displayed. The subject’s taskwas to indicate the perceived direction of motion (leftwardor rightward) by pressing the appropriate button. Suc-cessive trials were separated by 3.0 s, during which thedisplay was filled with a uniform field whose luminancewas the same as the initial space-averaged luminance ofthe grating to be presented on the next trial.

Three conditions of background luminance shift andtwo ISI durations (0 and 50 ms) were used in separate ex-perimental sessions. When the ISI was zero, each ses-sion comprised 72 trials presented in random order: tworepetitions of all combinations of 18 timing values t andtwo directions (left or right). Each subject completed 30sessions. When the ISI was zero, only the luminanceshift between 2.0 and 21.0 log Tp, the largest luminanceshift, was examined. When the ISI was 50 ms, each ses-sion comprised 56 trials presented in random order: tworepetitions of all combinations of 14 timing values t andtwo directions (left or right). Each subject completed 30sessions. Similar procedures were repeated for each ofthe three conditions of luminance shifts.

B. ResultsFigure 4 shows the percent response as a function of tim-ing value t when no ISI was inserted. The average lumi-

nance was decreased from 2.0 to 21.0 log Tp (solid circles)or increased from 21.0 to 2.0 log Tp (open circles). Theresults show that when t was smaller than 0.1 s, directiondiscrimination performance was poor. The deteriorationof performance can be considered to be a masking effectproduced by changing the average luminance level by3 log units. When t was larger than 0.1 s, motion direc-tion was judged veridically almost 100% of the time.

Our results show that although masking by a suddenluminance shift decreases performance, the effect isprominent only within 0.1 s after the shift in average lu-minance. Poot et al.25 showed similar results regardingmasking, although the task and the background lumi-nance level used were quite different. Based on the re-sults shown in Fig. 4, we varied the timing value t from0.1 to 2.4 s on the trials when the ISI was 50 ms, wherethe data should not be contaminated by the effects ofmasking.

In Figs. 5 and 6, percent response is plotted as a func-tion of t, which is the delay between the shift in the aver-age luminance and the offset of the first frame. Figures5(A), 5(B), and 5(C) show the results when the average lu-minance decreased from 2.0 to 21.0 log Tp, from 2.0 to0.0 log Tp, and from 0.0 to 21.0 log Tp, respectively. Fig-ure 6 shows the results when the average luminance in-creased from 21.0 to 2.0 log Tp [Fig. 6(A)], from 0.0 to2.0 log Tp [Fig. 6(B)], and from 21.0 to 0.0 log Tp [Fig.6(C)].

The straight dashed lines in each figure represent thepercent responses measured at the background lumi-nance levels [see Fig. 1(B) and Fig. 2] specified in the fig-ure insets. For example, in Fig. 5(A), when the averageluminance level was 2.0 log Tp, 22% of the responses werein the p/2 direction, where motion reversal was dominant(see Fig. 2). When the average luminance was 21.0 logTp, the veridical motion direction was perceived on 83% oftrials. These values are the minimum and the maximumexpected values when the average luminance varies be-tween 2.0 and 21.0 log Tp.

Fig. 4. Averaged data from five subjects for experiment 3 whenthe ISI was zero. Percent response in the p/2 direction is plot-ted as a function of the timing value t. Symbols identify theshift in average luminance: from 2.0 to 21.0 log Tp (solidcircles), from 21.0 to 2.0 log Tp (open circles). Each data pointis based on 600 trials (120 trials per subject).


Fig. 5. Averaged data from five subjects for experiment 3 when the ISI was 50 ms. (A) Percent response in the p/2 direction is plottedas a function the timing value t. The average luminance decreased from 2.0 to 21.0 log Tp. Dashed horizontal lines, percent responsein the p/2 direction at average luminance levels of 2.0 or 21.0 log Tp, taken from Fig. 2; dashed curve, logistic function fitted by aleast-squares method. Each data point is based on 600 trials (120 trials per subject). (B) Results for the condition in which the averageluminance decreased from 2.0 to 0.0 log Tp. Other conditions were as in (A). (C) Results for the condition in which the average lumi-nance decreased from 0.0 to 21.0 log Tp. Other conditions were as in (A).

In Fig. 5(A), as the timing value t increased from 0.1 to2.4 s, we found that the response in the p/2 directiongradually increased from 20% to 80%. Therefore the per-ceived direction changed from the 3p/2 direction (motionreversal) to the p/2 direction as t increased. The smoothcurve on the data is a logistic function fitted by a least-squares method. When the timing value t was small(less than 0.4 s), the response in the p/2 direction was;20%, which is comparable to the percentage obtainedwhen the average luminance level was constant at 2.0 logTp. This suggests that the perceived direction was deter-

mined primarily by the average luminance (2.0 log Tp)during most of the first frame. When t was greater than1 s, however, the response in the p/2 direction was ap-proximately 80%, which suggests that the perceived di-rection of motion was determined largely by the lumi-nance level following the luminance shift to 21.0 log Tp.We assume that the change in perceived direction of mo-tion reflects the course of light adaptation in the visualmotion system. The function in Fig. 5(A), then, wouldrepresent the time course of light adaptation in this sys-tem.


Similar tendencies are shown in Figs. 5(B) and 5(C),where the average luminance decreased by 2 and 1 logunits, respectively. In Fig. 5(B), when the timing value twas small, responses reflected performance at an averageluminance of 2.0 log Tp. When t was larger than 0.8 s,however, the response reflected performance at an aver-age luminance of 0.0 log Tp. Again, the perceived direc-tion gradually changed from motion reversal (3p/2 direc-tion) to the veridical direction (p/2 direction) as tincreased. In Fig. 5(C), although the percent responsewas always larger than 50%, the response in the p/2 di-rection increased from 60% to 80% as the timing value t

increased. In summary, the perception of the veridicalmotion direction (p/2 direction) was dominant if thechange in average luminance occurred approximately 1 sbefore the offset of the first frame. This suggests thatlight adaptation in the visual motion system is largelycomplete within 1 s when the background luminance de-creases over the luminance range that we have used here.

Figure 6 show the data when the average luminance in-creased by 3 log units [Fig. 6(A)], 2 log units [Fig. 6(B)],and 1 log unit [Fig. 6(C)]. In all three luminance-shiftconditions, the percent response in the p/2 direction de-creased as the timing value t increased. Particularly

Fig. 6. Results of experiment 3, in which a 50-ms ISI was inserted. (A) The average luminance increased from 21.0 to 2.0 log Tp.Other conditions were as in Fig. 5(A). (B) The average luminance increased from 0.0 to 2.0 log Tp. Other conditions were as in Fig.5(B). (C) The average luminance increased from 21.0 to 0.0 log Tp. Other conditions were as in Fig. 5(C).


Fig. 7. (A) Averaged estimated latency of light adaptation for five subjects for three luminance decrements, from 2.0 to 0.0 log Tp, from2.0 to 21.0 log Tp, and from 0.0 to 21.0 log Tp. Error bars show the standard deviation. (B) The estimated latency of light adaptationin case of luminance increment. Other conditions were as in (A).

when the size of the luminance increment reached 2 or3 log units [Figs. 6(A) and 6(B)], the perceived directionchanged from the p/2 direction (veridical direction) to the3p/2 direction (motion reversal) as t increased. In allthree cases the change in perceived direction was accom-plished within ;0.5 s before the offset of the first frame.Thus light adaptation appears to be faster in the case ofluminance increment (Fig. 6) than in the case of lumi-nance decrement (Fig. 5).

For purposes of comparison between the luminancedecrement (Fig. 5) and the luminance increment (Fig. 6)responses, we estimated the latency of light adaptation byfitting a logistic function and determined the timing valuet that should produce the middle value between the twoasymptotic values on the Y axis. This is in effect a la-tency measure. Figure 7(A) shows that estimated valueof t in the case of luminance decrements. The three barsrepresent the different amounts of luminance decrement.Figure 7(B) shows the comparable values for the case ofluminance increments. The data show the asymmetry oflight adaptation in the motion system, since the latency(value t) was shorter when the average luminance in-creased than when it decreased. Furthermore, the la-tency of light adaptation was more variable for luminancedecrements than for luminance increments. The latencywas longest when the luminance decrement was 3 logunits. In the case of luminance increments, the esti-mated latency was invariant for different magnitudes ofluminance shift.

6. DISCUSSIONWe examined light adaptation of the visual motion sys-tem by using two-frame apparent-motion stimuli. Sub-jects judged the direction of motion after the average lu-minance had decreased or increased at some variablepoint during the first frame. We found that light adap-tation occurred quickly for this task, being essentiallycomplete within 1 s when the shift in average luminancewas 3 log units. In the case of luminance decrements,veridical motion was perceived if the change occurred 1 sbefore the offset of the first frame. In case of luminance

increments, motion reversal was perceived if the changeoccurred 0.5 s before the offset of the first frame. Lightadaptation in the motion system was found to be asym-metric, with a faster response when the average lumi-nance increased.

One frequently used experimental procedure for exam-ining the time course of light adaptation is a probe–flashparadigm.1,3 A test probe stimulus is displayed on aflashed background, and the effect of temporal delays be-tween the onset of the test probe and the flashed back-ground are explored. The subject’s task is to detect thetest probe. With this method, it is possible to measurehow rapidly the subjects adapt to the background lightlevel. Such studies have suggested that light adaptationis a complex process, in which the response of a faster ad-aptation mechanism is almost (but not completely) fin-ished within several hundred milliseconds, while a sloweradaptation response continues for several seconds.6,26,27

The critical differences between those earlier studiesand ours are that we used a stimulus whose luminancecontrast is clearly suprathreshold (6 3 direction dis-crimination threshold contrast), and we asked the sub-jects to report the perceived direction of motion, not to de-tect the presence of a test stimulus. Light adaptation isbelieved to involve mechanisms that adjust sensitivity( gain) and spatiotemporal response properties contingenton the average luminance level.28 Because of the natureof our task, the light adaptation measured in our studymay be considered primarily to reflect the adjustment ofthe temporal response properties of the mechanisms un-derlying extraction of directional information. The stan-dard probe–flash paradigm, in contrast, measures the ad-justment of gain (sensitivity) of the temporal impulseresponse. Thus the probe–flash paradigm and our mo-tion paradigm examine different aspects of the light ad-aptation process. Our results suggest that the temporalproperties of the visual motion system are modifiedwithin 1 s if the luminance shift is within 3 log units (atleast within the luminance range we studied).

Since we used judgments of the perceived direction oftwo-frame apparent motion to estimate the time course oflight adaptation, our results should reflect light adapta-


tion in a mechanism underlying this specific task. Mo-tion reversal has been explained by assuming a first-ordermotion mechanism29 in which a biphasic temporal im-pulse response function feeds into the responsible motiondetector.17,19 Takeuchi and De Valois10 showed that thedisappearance of motion reversal under scotopic adapta-tion can be predicted by the output of a first-order motionmodel if the underlying temporal impulse response func-tion varies from biphasic to monophasic as the average lu-minance level decreases. It has been shown that theshape of the temporal impulse response function becomesmore monophasic (and sluggish) as the average lumi-nance level decreases.30–32 If the shape of the temporalimpulse response function is the decisive factor in deter-mining the perceived direction of motion, then the presentresults can be understood if based on the following as-sumptions. First, the shape of the temporal impulse re-sponse functions (biphasic or monophasic) that feed into afirst-order motion mechanism changes when the averageluminance changes [see Fig. 1(A) of Takeuchi and DeValois10]. Second, this change of temporal impulse re-sponse occurs within 1 s in the case of a luminance dec-rement and within 0.5 s in the case of a luminance incre-ment if the shift in background luminance is within 3 logunits.

The origin of the temporal impulse response functionfeeding into the first-order motion system is not well un-derstood. De Valois and colleagues33,34 recently analyzedthe spatiotemporal receptive fields of directionally selec-tive neurons in the primary visual cortex of macaquemonkey and showed that there appear to be inputs fromboth magnocellular (Mc) and parvocellular (Pc) pathways.Purpura et al.35 earlier reported that both Mc and Pc gan-glion cells of macaque monkey become more sluggish andmonophasic as the background light level decreases. Ifthis is also the case for the human visual system, then theshape of the temporal impulse response function thatfeeds into first-order motion mechanisms may be definedat the level of the retina or the lateral geniculate nucleus.Although the temporal impulse response functions associ-ated with the presumed Mc and Pc inputs to directionallyselective V1 cells appear to be determinedprecortically,33,34 directionally selective neurons them-selves are not found before V1 in macaque monkey.36

Therefore the adaptation responses we have character-ized may also include processes that occur at this level oreven later.

Our results differ from those of previous studies inwhich retinal processes are assumed to be primarily re-sponsible for light adaptation,26,27 although we note thata probe–flash paradigm is quite different from our proce-dure. First, there appears to be a very fast component inlight adaptation that is complete within ;50 ms.6 Sinceour subjects were unable to judge motion direction whenthe shift in background luminance occurred 100 ms orless before the end of the first frame (Fig. 4), we could notevaluate light adaptation within this time range. How-ever, our results show that the perceived direction of mo-tion does not change for several hundred milliseconds fol-lowing the luminance shift, especially in the case of aluminance decrement (Fig. 5). We thus conclude that therapid light adaptation that occurs within 100 ms ( pre-

sumably at a local receptor level) is not reflected in thejudgment of motion direction in our task.

Second, Poot et al.25 have suggested that the timecourse of adaptation is faster following luminance decre-ments than following luminance increments, which is theopposite of our results (see also Yeh et al.37). However,other psychophysical and neurophysiological studies haveshown that light adaptation is faster following luminanceincrements.6,38 This type of asymmetry has been ex-plained by assuming a divisive feedback gain-control pro-cess in which the control signal is low-pass filtered.27,38

The output signal from the feedback loop becomes brieferand faster when the input signal (background luminance)is increased than when it is decreased. Furthermore,Snippe et al.27 have shown that the apparent rapidity ofadaptation to the luminance decrement shown in Pootet al.25 can be explained by assuming a slow process andnonlinearity in the gain-control process, in addition to thedivisive feedback gain-control process noted above. Al-though we should be careful about comparing the resultsof previous studies that are based on the probe–flashparadigm with those from our study that are based onmotion direction judgments, the results suggest thatmechanisms similar to a divisive feedback gain-controlprocess are also functioning in light adaptation in the mo-tion system.

ACKNOWLEDGMENTSPortions of this study were reported at the 2000 annualmeeting of the Association for Research in Vision andOphthalmology. We would like to thank Shin’ya Nishidafor helpful suggestions on the experiments and Yoh’ichiTohkura for his continuing support. This work was sup-ported by NTT and by grant EY00014 from the NationalEye Institute.

REFERENCES1. D. C. Hood and M. A. Finkelstein, ‘‘Visual sensitivity,’’ in

Handbook of Perception and Human Performance, K. Boff,L. Kaufman, and J. Thomas, eds. (Wiley, New York, 1986),Vol. 1, Chap. 5, pp. 1–66.

2. J. H. van Hateren, ‘‘Processing of natural time series of in-tensities by the visual system of the blowfly,’’ Vision Res.37, 3407–3416 (1997).

3. B. H. Crawford, ‘‘Visual adaptation in relation to brief con-ditioning stimuli,’’ Proc. R. Soc. London 134, 283–302(1947).

4. W. S. Geisler, ‘‘Adaptation, afterimages and cone satura-tion,’’ Vision Res. 18, 279–289 (1978).

5. E. H. Adelson, ‘‘Saturation and adaptation in the rod sys-tem,’’ Vision Res. 22, 1299–1312 (1982).

6. M. M. Hayhoe, N. I. Benimoff, and D. C. Hood, ‘‘The time-course of multiplicative and subtractive adaptation pro-cess,’’ Vision Res. 27, 1981–1987 (1987).

7. T. E. von Wiegand, D. C. Hood, and N. Graham, ‘‘Testing acomputational model of light-adaptation dynamics,’’ VisionRes. 35, 3037–3051 (1995).

8. K. Nakayama, ‘‘Biological image motion processing: a re-view,’’ Vision Res. 25, 625–660 (1985).

9. M. Dawson and V. Di Lollo, ‘‘Effects of adapting luminanceand stimulus contrast on the temporal and spatial limits ofshort-range motion,’’ Vision Res. 30, 415–429 (1990).

10. T. Takeuchi and K. K. De Valois, ‘‘Motion-reversal reveals


two motion mechanisms functioning in scotopic vision,’’ Vi-sion Res. 37, 745–755 (1997).

11. T. Takeuchi and K. K. De Valois, ‘‘Velocity discriminationin scotopic vision,’’ Vision Res. 40, 2011–2024 (2000).

12. D. Turner, K. K. De Valois, and T. Takeuchi, ‘‘Speed per-ception under scotopic conditions,’’ Invest. Ophthalmol.Visual Sci. 38, (Suppl.), S378 (1997).

13. K. R. Gegenfurtner, H. Mayser, and L. T. Sharpe, ‘‘Seeingmovement in the dark,’’ Nature (London) 398, 475–476(1999).

14. E. D. Grossman and R. Blake, ‘‘Perception of coherent mo-tion, biological motion and form-from-motion under dim-light conditions,’’ Vision Res. 39, 3721–3727 (1999).

15. W. A. van de Grind, J. J. Koenderink, and A. J. van Doorn,‘‘Motion detection from photopic to low scotopic luminancelevels,’’ Vision Res. 40, 187–199 (2000).

16. O. J. Braddick, ‘‘Low-level and high-level processes in ap-parent motion,’’ Philos. Trans. R. Soc. London Ser. B 290,137–151 (1980).

17. S. Shioiri and P. Cavanagh, ‘‘ISI produces reverse apparentmotion,’’ Vision Res. 30, 757–768 (1990).

18. A. Pantle and K. Turano, ‘‘Visual resolution of motion am-biguity with periodic luminance- and contrast-domainstimuli,’’ Vision Res. 32, 2093–2106 (1992).

19. J. J. Strout, A. Pantle, and S. L. Mills, ‘‘An energy model ofinterframe interval effects in single-step apparent motion,’’Vision Res. 34, 3223–3240 (1994).

20. B. Stabell and U. Stabell, ‘‘Absolute spectral sensitivity atdifferent eccentricities,’’ J. Opt. Soc. Am. 71, 836–840(1981).

21. W. L. Makous, ‘‘Fourier models and the loci of adaptation,’’J. Opt. Soc. Am. A 14, 2323–2345 (1997).

22. H. Levitt, ‘‘Transformed up–down methods in psychoacous-tics,’’ J. Acoust. Soc. Am. 49, 467–477 (1971).

23. M. J. Keck, T. D. Palella, and A. Pantle, ‘‘Motion after-effectas a function of the contrast of sinusoidal gratings,’’ VisionRes. 16, 187–191 (1976).

24. K. Nakayama and G. H. Silverman, ‘‘Detection and dis-crimination of sinusoidal grating displacements,’’ J. Opt.Soc. Am. A 2, 267–274 (1985).

25. L. Poot, H. P. Snippe, and J. H. van Hateren, ‘‘Dynamics ofadaptation at high luminance: adaptation is faster afterluminance decrements than after luminance increments,’’J. Opt. Soc. Am. A 14, 2499–2508 (1997).

26. H. R. Wilson, ‘‘A neural model of foveal light adaptationand afterimage formation,’’ Visual Neurosci. 14, 403–423(1997).

27. H. P. Snippe, L. Poot, and J. H. van Hateren, ‘‘A temporalmodel for early vision that explains detection thresholds forlight pulses on flickering backgrounds,’’ Visual Neurosci.17, 449–462 (2000).

28. R. Shapley and C. Enroth-Cugell, ‘‘Visual adaptation andretinal gain controls,’’ in Progress in Retinal Research, N.Osborne and G. Chader, eds. (Pergamon, London, 1984), pp.263–346.

29. E. H. Adelson and J. Bergen, ‘‘Spatiotemporal energy mod-els for the perception of motion,’’ J. Opt. Soc. Am. A 2, 284–299 (1985).

30. D. H. Kelly, ‘‘Theory of flicker and transient responses. I.Uniform fields,’’ J. Opt. Soc. Am. 61, 537–546 (1971).

31. W. H. Swanson, T. Ueno, V. C. Smith, and J. Pokorny,‘‘Temporal modulation sensitivity and pulse-detectionthresholds for chromatic and luminance perturbations,’’ J.Opt. Soc. Am. A 4, 1992–2005 (1987).

32. R. J. Snowden, R. F. Hess, and S. J. Waugh, ‘‘The process-ing of temporal modulation at different levels of retinal il-luminance,’’ Vision Res. 35, 775–789 (1995).

33. R. L. De Valois and N. P. Cottaris, ‘‘Inputs to directionallyselective simple cells in macaque striate cortex,’’ Proc. Natl.Acad. Sci. U.S.A. 95, 14488–14493 (1998).

34. R. L. De Valois, N. P. Cottaris, L. E. Mahon, S. D. Elfar, andJ. A. Wilson, ‘‘What are the inputs to directionally-selectivecells in Macaque striate cortex?’’ Invest. Ophthalmol. Vi-sual Sci. 41, (Suppl.), S333 (2000).

35. K. Purpura, D. Tranchina, E. Kaplan, and R. M. Shapley,‘‘Light adaptation in the primate retina: analysis ofchanges in gain and dynamics of monkey retinal ganglioncells,’’ Visual Neurosci. 4, 75–93 (1990).

36. D. Hubel and T. N. Wiesel, ‘‘Receptive fields and functionalarchitecture of monkey striate cortex,’’ J. Physiol. (London)195, 215–243 (1968).

37. T. Yeh, B. B. Lee, and J. Kremers, ‘‘The time course of ad-aptation in macaque retinal ganglion cells,’’ Vision Res. 36,913–931 (1996).

38. M. J. M. Lankheet, R. J. A. van Wezel, J. H. H. Prickaerts,and W. A. van de Grind, ‘‘The dynamics of light adaptationin cat horizontal cell responses,’’ Vision Res. 33, 1153–1171(1993).

light adaptation in motion direction judgments

Documents