876 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984 Opponent-movementmechanisms in human vision C. F. Stromeyer III, R. E. Kronauer, and J. C. Madsen Division of Applied Sciences, Harvard University, Cambridge, Massachusetts 02138 S. A. Klein College of Optometry, University of Houston, Houston, Texas 77004 Received September 13, 1983; accepted April 9, 1984 A vertical grating that sinusoidally reverses contrast can be synthesized from two identical component gratings that move with equal velocities in opposite directions (leftward and rightward). Such a counterphase grating is used as a suprathreshold masking pattern. When the mask is of low spatial frequency and is modulated rapidly, a test pattern consisting of an increment of the rightward component and an equivalent simultaneous decrement of the leftward component is highly detectable compared with simultaneous increments or decrements of both compo- nents. The visibility of the opponent-movement test signal is strongly facilitated by high-contrast masks. This facilitation is accompanied by a high sensitivity for judging the direction of motion of the test. These results show that certain detection mechanisms are highly sensitive to the difference of the rightward and leftward components. However, when the mask is of threshold contrast, the rightward- and leftward-moving test components appear to be detected independently. A high-contrast grating that rapidly moves in one direction strongly masks gratings moving in the same or opposite direction; this shows that moving patterns are not detected by unidirectional mech- anisms when contrast is clearly suprathreshold. The results may be explained by a model with mechanisms that are excited by one direction of motion and inhibited by the opposite direction. 1. INTRODUCTION To determine if humans have unidirectional visual mecha- nisms that respond to motion in one direction but not in the opposite, Levinson and Sekuler 1 measured threshold sum- mation with moving gratings. The visibility of a grating moving in one direction was not affected by a grating of the same spatial and temporal frequency moving in the opposite direction, except through probability summation. A sta- tionary grating flickering in counterphase (sinusoidally re- versing contrast) had approximately twice the contrast threshold of a similar grating that moved in one direction at the same rate as that with which the counterphase grating reversed contrast. The counterphase grating can be decom- posed analytically into two matching gratings that move in opposite directions at the same rate. The threshold mea- surements thus suggest that the visual system similarly de- composes the counterphase grating. The two moving com- ponents of the counterphase grating appear to be detected independently rather than summated to determine threshold. The detection mechanisms are thus selectively sensitive to motion in one of two opposite directions. The independence, however, appears to break down at slow velocities. 2 , 3 For a low spatial frequency of 2 cycles/deg, there is little summation between oppositely moving patterns for a wide range of temporal frequencies (1.5-12.4 Hz). 3 This implies that the two patterns are detected independently. At a higher spatial frequency of 6 or 8 cycles/deg, there also is little summation at fast rates (8-12 Hz) but relatively strong summation at slow rates (1.5-4 Hz). 2 , 3 This strong summa- tion shows that fine slow-movingpatterns are not detected by highly direction-selective mechanisms. Experiments on the identification of the direction of motion lead to similar conclusions. Watson et al. 3 and Mansfield and Nachmias 4 showed that observers could discriminate whether fast-moving patterns of low and high spatial frequency moved leftward or rightward at the threshold for detecting the presence of the patterns. Direction discrimination was much poorer for fine slow-movingpatterns, even with retinal sta- bilization. 4 These studies show that patterns moving at relatively high velocity are detected by mechanisms that are largely sensitive to motion in one direction. At threshold levels these mech- anisms appear to act as undirectional mechanisms that re- spond to motion in one direction but are unaffected by motion in the opposite direction. Thus a threshold grating moving in the opposite direction does not appear either to facilitate or to inhibit the response to the preferred direction of motion. In this study we examine the behavior of these direction-se- lective mechanisms at a suprathreshold response level by employing a masking paradigm. For the first experiments, the mask was a counterphase flickering grating. The test signal was a variation in the relative strength of the rightward- and leftward-moving components of the mask. The purpose of the experiment was to determine whether there are mech- anisms that are highly sensitive to the difference between rightward- and leftward-moving components. High sensi- tivity to such differential motion would indicate clearly that the detection mechanism is not unidirectional (as defined above), since opposite directions of motion affect the re- sponse. In the second series of experiments, the mask was a grating that drifted in one direction. The test signal was a drifting grating that moved in the same direction as the mask or in the opposite direction. If there are unidirectional motion mechanisms that operate at suprathreshold contrast levels, then the test pattern that moves in the opposite direction will be unaffected by the mask. 0740-3232/84/080876-09$02.00 ) 1984 Optical Society of America Stromeyer et al.
Transcript
876 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984
Opponent-movement mechanisms in human vision
C. F. Stromeyer III, R. E. Kronauer, and J. C. Madsen
Division of Applied Sciences, Harvard University, Cambridge, Massachusetts 02138
S. A. Klein
College of Optometry, University of Houston, Houston, Texas 77004
Received September 13, 1983; accepted April 9, 1984
A vertical grating that sinusoidally reverses contrast can be synthesized from two identical component gratings thatmove with equal velocities in opposite directions (leftward and rightward). Such a counterphase grating is usedas a suprathreshold masking pattern. When the mask is of low spatial frequency and is modulated rapidly, a testpattern consisting of an increment of the rightward component and an equivalent simultaneous decrement of theleftward component is highly detectable compared with simultaneous increments or decrements of both compo-nents. The visibility of the opponent-movement test signal is strongly facilitated by high-contrast masks. Thisfacilitation is accompanied by a high sensitivity for judging the direction of motion of the test. These results showthat certain detection mechanisms are highly sensitive to the difference of the rightward and leftward components.However, when the mask is of threshold contrast, the rightward- and leftward-moving test components appear tobe detected independently. A high-contrast grating that rapidly moves in one direction strongly masks gratingsmoving in the same or opposite direction; this shows that moving patterns are not detected by unidirectional mech-anisms when contrast is clearly suprathreshold. The results may be explained by a model with mechanisms thatare excited by one direction of motion and inhibited by the opposite direction.
1. INTRODUCTION
To determine if humans have unidirectional visual mecha-nisms that respond to motion in one direction but not in theopposite, Levinson and Sekuler1 measured threshold sum-mation with moving gratings. The visibility of a gratingmoving in one direction was not affected by a grating of thesame spatial and temporal frequency moving in the oppositedirection, except through probability summation. A sta-tionary grating flickering in counterphase (sinusoidally re-versing contrast) had approximately twice the contrastthreshold of a similar grating that moved in one direction atthe same rate as that with which the counterphase gratingreversed contrast. The counterphase grating can be decom-posed analytically into two matching gratings that move inopposite directions at the same rate. The threshold mea-surements thus suggest that the visual system similarly de-composes the counterphase grating. The two moving com-ponents of the counterphase grating appear to be detectedindependently rather than summated to determine threshold.The detection mechanisms are thus selectively sensitive tomotion in one of two opposite directions.
The independence, however, appears to break down at slowvelocities. 2,3 For a low spatial frequency of 2 cycles/deg, thereis little summation between oppositely moving patterns fora wide range of temporal frequencies (1.5-12.4 Hz). 3 Thisimplies that the two patterns are detected independently. Ata higher spatial frequency of 6 or 8 cycles/deg, there also islittle summation at fast rates (8-12 Hz) but relatively strongsummation at slow rates (1.5-4 Hz).2,3 This strong summa-tion shows that fine slow-moving patterns are not detectedby highly direction-selective mechanisms.
Experiments on the identification of the direction of motionlead to similar conclusions. Watson et al. 3 and Mansfield and
Nachmias 4 showed that observers could discriminate whetherfast-moving patterns of low and high spatial frequency movedleftward or rightward at the threshold for detecting thepresence of the patterns. Direction discrimination was muchpoorer for fine slow-moving patterns, even with retinal sta-bilization. 4
These studies show that patterns moving at relatively highvelocity are detected by mechanisms that are largely sensitiveto motion in one direction. At threshold levels these mech-anisms appear to act as undirectional mechanisms that re-spond to motion in one direction but are unaffected by motionin the opposite direction. Thus a threshold grating movingin the opposite direction does not appear either to facilitateor to inhibit the response to the preferred direction of motion.In this study we examine the behavior of these direction-se-lective mechanisms at a suprathreshold response level byemploying a masking paradigm. For the first experiments,the mask was a counterphase flickering grating. The testsignal was a variation in the relative strength of the rightward-and leftward-moving components of the mask. The purposeof the experiment was to determine whether there are mech-anisms that are highly sensitive to the difference betweenrightward- and leftward-moving components. High sensi-tivity to such differential motion would indicate clearly thatthe detection mechanism is not unidirectional (as definedabove), since opposite directions of motion affect the re-sponse.
In the second series of experiments, the mask was a gratingthat drifted in one direction. The test signal was a driftinggrating that moved in the same direction as the mask or in theopposite direction. If there are unidirectional motionmechanisms that operate at suprathreshold contrast levels,then the test pattern that moves in the opposite direction willbe unaffected by the mask.
0740-3232/84/080876-09$02.00 ) 1984 Optical Society of America
Stromeyer et al.
Vol. 1, No. 8/August 1984/J. Opt. Soc. Am. A 877
2. THEORY
The luminance profile of a moving vertical sine-wave gratingis represented by
L(x, t) = Lo[l + m cos(27rfx ± 27rct)],
where Lo is the mean luminance, m is the contrast of thegrating, f is the spatial frequency in cycles per degree, x is thehorizontal position of the grating in degrees, w is the rate ofhorizontal movement in hertz, and t is time. Two gratingsof the same spatial frequency, moving at the same rate inopposite directions, are represented by
L(x, t) = Lo[l + mieft cos(27rfx + 27rcot)
+ mright cos(27rfx - 27rot)].
If the modulation of each grating is equal to m', a counter-phase flickering grating is formed. This grating is a standingwave that does not drift in either direction but simply reversescontrast as a sinusoidal function of time; it can be representedby
L(x, t) = Lo[1 + 2m' cos(27rfx)cos(2lrwt)].
The peak contrast of the counterphase grating is twice thecontrast of the two moving components that compose thecounterphase grating.
In the first experiments, the mask was a vertical counter-phase grating consisting of identical leftward- and right-ward-moving components. The test signals were equal si-multaneous increments of both components (designated+R+L), a simultaneous decrement of both components(-R-L), or an increment of the right component and anequivalent simultaneous decrement of the left component(+R-L). Unidirectional mechanisms sensitive to rightwardmotion only will not respond to variations in the leftward-moving component. Thus, if the detection mechanisms areunidirectional and equally sensitive to increments and dec-rements, all three test signals will be equally detectable.
The test signal was produced by adding a counterphasegrating to the counterphase mask grating with the appropriatespatial and temporal phase. Consider the test signal +R-L.The test plus mask pattern can be represented by
The test signal produces an increment (+A) of the rightcomponent of the mask and an equal decrement (-A) of theleft component. This expression can be rewritten as
L(x, t) = Lo[1 + 2m' cos(27rfx)cos(27rcot)
+ 2A sin(27rfx)sin(27rwt)],
in which m' = mright = mieft. The first part of the expressionrepresents the counterphase mask grating; the second part,the test grating. Thus the test signal +R-L is simply a sec-ond counterphase grating that is identical in spatial andtemporal frequency to the mask but shifted by 7r/2 rad inspace and time. The test signal produces an increment of theright component of the mask and a decrement of the leftcomponent, the contrasts of which are each equivalent to onehalf of the contrast of the test grating. If the test signal isequal in contrast to the mask, the resultant pattern is a simple
rightward-moving grating: the rightward-moving compo-nents of the mask and test patterns summate completely,whereas the leftward-moving components exactly cancel.Thus, whereas a counterphase grating consists of equal left-ward- and rightward-moving components, a moving gratingconsists of two equal counterphase gratings of appropriatespatiotemporal phase.
3. METHODS
A. Apparatus and StimuliVertical sine-wave gratings were presented on a Textronix 602display monitor with white phosphor. The mean luminanceof the field was 19 cd/M 2. The field was 90 in diameter witha dark surround and a small central-fixation point. Thedisplay was viewed from 53 cm with the left eye. The displayframe rate was 208 Hz.
Two identical counterphase flickering gratings were pro-duced with phase-locked function generators. The patternswere displayed with a phase difference of 7r/2 rad in space andtime. The mask consisted of either one or both of thesegratings, that is, either a counterphase flickering grating ora drifting grating (the sum of the two counterphase gratings).The same two signals from the function generators also wentto a separate summing amplifier, an attenuator, and a phaseshifter. These latter signals formed the test grating, which,like the mask, was a counterphase flickering grating or adrifting grating. When the mask was a single counterphasegrating, the test was also a counterphase grating. The coun-terphase test was added to the counterphase mask to produceequal increments or decrements of right and left components(+R+L or -R-L) or an increment of the right component anda simultaneous equal decrement of the left component or viceversa (+R-L or -R+L). When the mask was a driftinggrating, the test was also a drifting grating. For one experi-ment, the mask and the test moved with different velocities,and a synchroresolver was used to produce the test.5
B. Procedure and Definition of MaskingFor each run a single mask was continuously presented at aconstant contrast. The observer initially viewed the mask foiseveral minutes. A run of 100 trials was then conducted inwhich a single test signal was presented at four contrast levels(including blanks) with equal probability. The observerpresented the test pattern for 1.0 sec (square envelope) bypressing a button while fixating a point in the center of thefield. He then rated the visibility of the test pattern on awhole-number scale of 1-5 and was immediately informedwhat the contrast level had been. Receiver-operating-char-acteristic (ROC) curves were fitted to the ratings by a maxi-mum-likelihood estimation to determine the detectability d'of the different contrast test patterns.6 7 Each estimate wasbased on several runs. The threshold was arbitrarily definedas the test contrast value that corresponded to d' = 1.0 on aline fit by least squares. This rather low d' value was usedbecause many of the conditions did not include high d'values.
Masking is defined as the ratio of the contrast thresholdwith the mask to that without the mask, TITo. Thus a maskvalue greater than 1.0 indicates that the mask raises the
Stromeyer et al.
878 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984
threshold, whereas a value less than 1.0 indicates that themask lowers the threshold.
The strength of the mask itself is expressed by the ratioMITo, where M is the mask contrast and To is the thresholdcontrast of the unmasked test (for d' = 1.0). The mask con-trast is normalized by the test threshold, as the two patternsare similar.
4. RESULTS
A. Masking by Counterphase Gratings: Evidence forDifferential Motion MechanismsIn the first series of experiments the mask was a counterphasegrating that was continuously presented. The test signal waspresented for 1.0 sec and was (+R+L), (-R-L), or (+R-L).Unidirectional mechanisms that respond to only rightwardmovement, for example, would be sensitive to increments ordecrements of the right component only; variations in thecontrast of the left component would not affect the mecha-nisms. The three test signals would be equally visible if de-tected by unidirectional mechanisms, provided that the in-crements and the decrements are about equally effective.
1. Masking at 0.8 Cycle/Deg and 20 HzMechanisms Sensitive to Differences of Leftward- andRightward-Moving Components. A low-spatial- and high-
2.0
1.0
di2.0
1.0
1.0 10.0
TEST CONTRAST ()Fig. 1. Figures 1-3 show the detectability d' 1 S.E. of various testpatterns as a function of their contrast in the presence of a contin-uously present mask. The mask and test patterns were verticalsine-wave gratings of the same spatial frequency. The test was alwayspresented for .0 sec. For Fig. 1, the mask grating was 0.8 cycle/deg,26% contrast, in 20-Hz counterphase flicker. The test was a similarcounterphase pattern added to the mask in a given spatial and tem-poral phase: test grating with mask at zero contrast (); test addedto mask in phase to produce contrast increment (A) or decrement ()of mask (+R+L or -R-L, respectively); test added to produce in-crement of right component () of mask and equivalent decrementof left component (+R-L) [for observer CFS, the same test was usedwith the mask contrast at 52% (X)]. The high-contrast mask reducesthe visibility of the +R+L and -R-L tests and strongly facilitatesdetection of +R-L.
temporal-frequency (high-velocity) pattern was selected tofavor stimulation of transient motion-selective mecha-nisms.2 '3'8 The counterphase mask and test were 0.8 cycle/degand 20 Hz. The mask contrast was 26%, approximately 10-20times threshold. The circles in Fig. 1 show, for two observers,the detectability d' of the counterphase test signal with nomask present as a function of test contrast. Upward- anddownward-pointing triangles show d' for the same test gratingadded to the mask so as to produce increments (+R+L) anddecrements (-R-L) of the mask, respectively. The maskconsiderably reduces the visibility of these latter test patterns.Squares show d' for the test signal +R-L, a simultaneousincrement of the right mask component and decrement of theleft component. Crosses indicate results for observer CFS forthis same test signal when the mask contrast was increasedto 52%. The test signal produces a change in the left and rightmask components that is one half of the test contrast (seeSection 2).
The high-contrast mask strongly facilitates detection of the+R-L test while reducing the visibility of the +R+L and-R-L tests. The results thus demonstrate a striking phaseselectivity, as the visibility of the same test grating changesradically depending on the phase in which the pattern is addedto the mask.
Certain features of the results of Fig. 1, as well as other re-sults obtained with 0.8-cycle/deg counterphase masks, arelisted in Table 1. For example, consider results for observerJCM (second row from bottom). Figure 1 shows that theunmasked threshold To (corresponding to d' = 1.0) is -1.2%contrast (circles). Thus the strength of the mask, MITo, is22 (third column in Table 1). The fourth column showsmasking (TRL/To) of the differential (D) test signal +R-L.The threshold for this test signal is -0.45% contrast (Fig. 1,squares). Since the threshold is lower than on the blank field,the masking effect is less than 1.0; it is 0.40 (fourth column).The fifth column shows the masking of the test signals +R+Land -R-L that produces the same () increments or decre-ments of both the rightward and leftward components. Thecontrast threshold for these signals is -4.1% (Fig. 1, triangles);the masking effect is 3.5 (fifth column). The sixth columnshows the ratio SID or the ratio of the masking effects in thefifth and fourth columns; in the presence of the mask, thethreshold of the +R+L and -R-L tests are about 9 timeshigher than thtthreshold of the +R-L test. This SD ratiois about 16 for observer CFS. Observers are thus highlysensitive to the difference between the contrast of the right-ward- and leftward-moving components. Unidirectionalmechanisms clearly cannot account for these results.
Movement Discrimination. Is this high sensitivity to thedifference of rightward and leftward components accompa-nied by the perception of motion? The counterphase testgrating was set to 1.3% contrast for observer CFS. A d' valueof 0.23 + 0.14 was obtained when the pattern was presentedon a blank field. Thus the pattern by itself was barely visible.Next, the same test pattern was added to the 26% contrastcounterphase mask so as to produce an increment of the rightcomponent and an equivalent decrement of the left compo-nent (+R-L, rightward motion) or vice versa (-R+L, left-ward motion). The observer discriminated perfectly the di-rection of motion for a 100-trial run. When the test signal wasturned off, the mask grating appeared to move in a directionopposite the motion produced by the test signal. (ObserverJCM also saw this effect clearly.) The test signal thus pro-
) I I I I I I I Il_C FS
JCM
I I I I II I I I I I I I I
Stromeyer et al.
Vol. 1, No. 8/August 1984/J. Opt. Soc. Am. A 879
Table 1. Masking of 0.8-Cycle/Deg Counterphase Test Patterns by Counterphase Mask of Same Spatial andTemporal Frequencya
Masking at Different TestsS
Counterphase Mask Strength [Average of DifferentialFrequency Contrast of Mask D (TR+L/To) and Masking
a Observers CFS and JCM (JCM's values in parentheses).
duces a clear sensation of motion and the well-known after-effect of seen movement.
Near-Threshold Masking. To examine if opposite direc-tions of motion are processed independently at lower con-trasts, the same counterphase mask (0.8 cycle/deg and 20 Hz)was reduced from 26 to 5.2% contrast. The mask strength(MTU) was 2.2 and 4.4 for CFS and JCM, respectively. Thecrosses in Fig. 2 show d' for a test signal that was an actualrightward-moving grating added in phase with the right-handcomponent of the counterphase mask; squares show the de-tectability of the previous counterphase signal +R-L. Themask strongly facilitates detection of the latter signal, for themasking effect TR-L/To is considerably less than 1.0; the fa-cilitation factor is 0.14 and 0.25 for observers CFS and JCM,respectively (Table 1). When both test patterns have thesame contrast, the rightward-moving test grating producesa 2-arbitrary-unit increase in the right component of the mask,whereas the +R-L signal produces a 1-unit increase in theright component and a 1-unit decrease in the left component.The two test signals thus produce equal differences in themagnitude of right versus left components. The rightward-moving test signal is only slightly more detectable. The re-sults demonstrate that the visibility of the test patterns islargely governed by the difference between the magnitudesof right and left components. If detection is determined bya single direction of motion, then there should be a twofoldhorizontal separation between the curves in Fig. 2.
The experiment was repeated with the mask reduced to 2%,which was slightly subthreshold for CFS and 70% abovethreshold for JCM (Table 1). Figure 3 shows that the +R-Ltest (squares) must now be set to about twice the contrast ofthe rightward-moving test grating (crosses) to be equallyvisible. When the signals are set in such a 2:1 contrast ratio,they produce equal increments of the right component of themask. The decrement of the left component produced by the+R-L test signal now appears to have little or no effect.
Since the left component of the mask is near threshold, itwould be surprising if a decrement did have a strong effect.These results are similar to the results of Levinson andSekuler, 1 who showed that components moving only in onedirection appear to affect the threshold. However, at contrastlevels only slightly higher, the differences between the left andright components largely determines the threshold (Fig. 2).Although the mask is near threshold, it still strongly facilitatesdetection of the +R-L signal, as shown by the fact that themasking effect TR-L/To is considerably less than 1.0 (Table1). The results suggest that this facilitation is largely due tothe fact that the +R-L test produces an increment in therightward-moving mask component.
2.0 - CFS JCM
d'1.0
0.2 0.4 0.6 0.2 0.4 0.6
TEST CONTRAST (%)Fig. 2. Mask grating at 0.8 cycle/deg in 20-Hz counterphase flickerat 5.2% contrast (2-4 times threshold). Test gratings were incrementof right component of mask (X, +2R) or increment of right componentand equivalent decrement of left component (, +R-L). By defi-nition, when the two test patterns are of equivalent contrast (abscissa)the +2R test signal produces a twofold greater change in the contrastof the rightward-moving components than does the +R-L test. Thetwo test patterns are about equally detectable, indicating that thedifference between the leftward- and rightward-moving componentsdetermines detection.
Stromeyer et al.
880 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984
3.0 CFS JCM
2.0
1.0
I I I,, ,,11 ., , ,,, ,,II
0.2 0.4 0.8 0.2 0.4 0.8
TEST CONTRAST (%)Fig. 3. Similar to Fig. 2 but with the mask contrast reduced to nearthreshold, 2% contrast. The contrast of the +R-L test () must beabout twice the contrast of the +2R test (X), indicating that therightward-moving components largely determined detection.
2. Masking at 0.8 Cycle/Deg and 5 and 10 HzThe first. experiment was repeated with the counterphaseflicker rate reduced to 5 and 10 Hz. The degree of maskingfor the various test patterns is shown in Table 1. The maskagain elevates the threshold of the +R+L and -R-L testsignals. The thresholds for these two test patterns are quitesimilar, and Table 1 thus shows the average masking effect forthe patterns. At 10 Hz, the mask facilitates detection of the+R-L test pattern (TR-L/To < 1.0). At 5 Hz, the +R-Lsignal is only slightly more visible than the +R+L and -R-Lsignals, and the differential masking (SID) is considerablyreduced from that observed at 10 and 20 Hz.
3. Counterphase Masking at Higher Spatial FrequenciesMeasurements similar to those of the first experiment weredone at a relatively high spatial frequency of 6 cycles/deg andat 2.2 cycles/deg, midway between 0.8 and 6 cycles/deg on alogarithmic scale.
Table 2 shows masking effects at 2.2 cycles/deg and 10 Hzfor observer CFS and 20 Hz for CFS and JCM. All masks arewell above threshold. The mask slightly increases the visi-bility of the +R-L counterphase test signal while slightlyreducing the visibility of the +R+L and -R-L signals. Thedifferential masking effect, SID, at 20 Hz is 2-3-fold (Table
Table 3. Masking of 6-Cycle/Deg Counterphase TestPatterns by Counterphase Mask of Same Spatial and
Temporal Frequencya
Counterphase Mask Strength DifferentialFrequency Contrast of Mask Masking (S/D)
(Hz) (M, %) (M/To) [(TR+L + T-R-L)/ 2 )]/(TR-L)
(10) (21) (-5) (1.30)
15 25 -4 1.03(15) (17) (-4) (1.45)
a Observers CFS and JCM (JCM's values in parentheses).
2) and is substantially reduced from the 9-16-fold effect ob-served at 0.8 cycle/deg and 20 Hz (Table 1).
The results in Table 3 show that the differential maskingeffect for +R-L versus +R+L and -R-L test signals is fur-ther reduced at 6 cycles/deg. The former signal is only slightlymore visible than the latter, and thus the SID ratio is onlyslightly greater than 1.0. The masks were 4-5 times thresh-old, as determined by the method of adjustment, yet theyappeared to have rather high contrast.
In summary, the results with counterphase patterns showthat observers are highly sensitive to the differences betweenthe contrast of suprathreshold left- and right-moving com-ponents when the components are of low spatial frequency andmoderately high temporal frequency. At a relatively highspatial frequency of 6 cycles/deg the enhanced sensitivity tothe difference between the left- and right-moving componentsessentially disappears.
B. Masking by Moving Gratings: Evidence againstUnidirectional Motion Mechanisms at High ContrastThe previous results show that, for low spatial and hightemporal frequencies, there are mechanisms that are highlysensitive to the difference between the contrast of rightward-and leftward-moving components. Unidirectional mecha-nisms sensitive to only leftward or rightward movement wouldbe strongly stimulated by the counterphase mask, whichconsists of both rightward- and leftward-moving components.These mechanisms might be strongly desensitized by the maskpattern. Thus, in the first experiments, unidirectionalmechanisms may have been desensitized so that only differ-ential motion mechanisms were used to detect the +R-L testsignal.
The next experiments examine whether sensitive unidi-rectional mechanisms also exist. If there are such mecha-
Table 2. Masking of 2.2-Cycle/Deg Counterphase Test Patterns by Counterphase Mask of Same Spatial andTemporal Frequencya
Masking at Different TestsS
Counterphase Mask Strength [Average of DifferentialFrequency Contrast of Mask D (TR+L/To and Masking
a Observers CFS and JCM (JCM's values in parentheses).
Stromeyer et al.
Vol. 1, No. 8/August 1984/J. Opt. Soc. Am. A 881
nisms, a test grating moving in one direction may be little af-fected by a mask grating moving in the opposite direction.
1. High-Contrast MaskingThe mask was a continuously presented grating of 0.8 cycle/deg and 26% contrast that moved rightward at 20 Hz. Themask was well above threshold; the mask strength (MITo) was29 and 49 for observer CFS and JCM, respectively.
The test pattern was also 0.8 cycle/deg and moved leftwardor rightward at 10 or 20 Hz. The masked thresholds (T/TO)are shown in Fig. 4. The horizontal axis specifies the testvelocity, with plus and minus values representing rightwardand leftward motion, respectively. The moving test patternswere presented for 1.0 sec with a square temporal envelope.The lines are fitted through the 0- and 10-Hz data. The mostimportant result is that tests that move leftward, opposite themask, are strongly masked. The strong masking for both testdirections presumably involves motion-selective mechanisms,for the same mask had little effect on the visibility of a sta-tionary test pattern that linearly turned on and off with agradual 2-sec triangular envelope, shown at 0 Hz. This pat-tern was used to stimulate sustained mechanisms selectively.The masked threshold is only slightly greater than 1.0.
The dashed line in Fig. 4 shows that masking is partiallyvelocity selective for the test patterns that move oppositethe mask direction; masking increases with the test velocityand is maximal when the test moves at 20 Hz, as the maskdoes. (However, for observer CFS, the masking is not sig-nificantly different for the 10- and 20-Hz leftward-moving testpatterns.) A comparison of the dashed and solid lines showsthat masking is direction selective for the 10-Hz test patterns.The solid line also emphasizes that the masking of the 20-Hzrightward-moving test pattern seems to be too low for a ve-locity-selective mechanism. This is not surprising since, whenthe test and the mask move in the same direction at the same
10
(9z A-6(/4
I h/'K I
v -20 -10 0 10 20 HzLeftward ' Rightward
TEST PATTERNFig. 4. The mask was a continuously presented grating of 26% con-trast and 0.8 cycle/deg that moved rightward at 20 Hz. The bars showthe strength of masking (T/TO) for a 0.8-cycle/deg drifting test gratingthat was presented for 1.0 sec, moving rightward at two rates, +10 and+20 Hz, or leftward at -10 Hz and -20 Hz. When the 20-Hz testmoved rightward with the mask, the test and mask were summed inphase. The pattern at 0 Hz was a 0.8-cycle/deg stationary grating thatlinearly turned on and off with a 2-sec triangular envelope.
0I-
Z 1.0
s4
C,)
0~~~~~~~~~1 2 3 4 5 6
MASK STRENGTH (M/To)Fig. 5. Effects of masking with low-contrast masks of 0.8 cycle/deg,moving rightward at 20 Hz. The mask was continuously presented.The test was a 1.0-sec-duration increment of the mask contrast. Thevertical axis shows the contrast of the test pattern normalized to itsthreshold (To for a d' value of 1.0), and the horizontal axis representsthe mask contrast, also normalized by To. Masks of even 5-6 timesthreshold strongly facilitate detection of the test pattern.
rate, the test might be detected by the change in contrast, forthe gratings are summed spatially and temporally in phase.The threshold was found to be about 0.10 of the mask contrast,in agreement with the expected 0.10-contrast-incrementWeber fraction. The same pattern of results was obtainedwith observer CFS using a 0.8-cycle/deg 26%-contrast maskmoving at 5 Hz with a 5-Hz test. When the test grating wasadded 90 deg out of phase to the mask (to minimize the con-trast change), the masking increased more than fivefold.
The lower than expected masking for the 20-Hz right-ward-moving test pattern in Fig. 4 is thus presumably due tothe fact that the test was added in phase with the mask. Ifthe test was added in a different phase or in random spatialphase with the mask, the masking would presumably becomegreater and rise toward the solid line in Fig. 4. This wouldcause the masking to become more direction and velocity se-lective. Despite the limitations of phase coherence, the resultsdemonstrate strong masking for test patterns moving in thedirection opposite the mask direction. There are no sensitiveunidirectional mechanisms for detecting these patterns sincethe rightward-moving mask has such a strong effect of ele-vating the threshold.
2. Low-Contrast MaskingSimilar measurements were made using low-contrast maskgratings of 0.8 cycle/deg that moved rightward at 20 Hz. Themask contrast (<4%) ranged from subthreshold to clearlysuprathreshold. The mask was continuously presented. Thetest pattern, also 0.8 cycle/deg and moving rightward at 20 Hz,was added in phase with the mask for 1.0 second. For eachmask contrast, the test threshold contrast was determined ford' = 1.0. Figure 5 shows the test threshold as a function of themask strength with each axis normalized by the test threshold(To). For zero mask, the normalized test threshold (T/TO)is 1.0 by definition. The test grating becomes more visible asthe mask contrast is increased; the heavy diagonal line indi-cates perfect summation, that is, the sum of the mask and testcontrast determines threshold. The masked threshold isconsiderably below the unmasked threshold, even when the
MASK: 20Hz, rightward
C CFS
ED JCM
I / i 2
i-/l l /I M
Stromeyer et al.
r
882 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984
mask is about 6 times threshold. This latter facilitation effectprovides clear evidence that motion-selective mechanisms donot saturate at 5-6 times threshold, in contrast to the con-clusions of several studies. 9 -10
5. DISCUSSION
A. Motion Mechanisms Sensitive to Opposite DirectionsThe results with rapidly flickering counterphase masks of lowspatial frequency show that observers are highly sensitive totest patterns that produce a difference between rightward-and leftward-moving components. Observers are much lesssensitive to simultaneous equivalent increments or decre-ments of both moving components of the counterphase masks.The high sensitivity to differential motion is reduced whenthe modulation rate of the low-spatial-frequency patterns (0.8cycle/deg) is reduced from 20 to 5 Hz or when the spatial fre-quency is raised from 0.8 to 6 cycles/deg. At 6 cycles/deg, thedifferential motion sensitivity is virtually absent.
The studies reviewed in Section 1 show that direction-se-lective mechanisms detect fine patterns (6 cycles/deg andhigher) that move at sufficient velocity,2 3 and the directionof these patterns can be identified at threshold.3 4 We ob-served little differential motion sensitivity at 6 cycles/deg.We hypothesize that the mechanisms tuned to these higherspatial frequencies may be relatively insensitive to the dif-ference between rightward- and leftward-moving patterns,and thus there may be a basic difference in the nature of themotion-selective mechanisms tuned to low versus high spatialfrequencies. Three objections, however, can be raised to thishypothesis. First, a sustained non-motion-selective mech-anism might detect the fine test patterns on the assumptionthat the flickering counterphase mask reduces the sensitivityof the motion mechanisms below the sensitivity of the sus-tained mechanisms. However, the masks were only 4-5 timesabove threshold. Another possibility is that small eyemovements comparable with the spatial period of the finepatterns may reduce the ability to detect +R-L test signals.This could be tested with image stabilization. A third pos-sibility is that our experiments with 6-cycle/deg patterns didnot extend to high enough velocities; however, there was littleevidence for differential masking at 15 Hz. Use of signifi-cantly higher velocities would require rather high mask con-trasts, for at 15 Hz the masks are as high as 25% contrast and,at this level, the masks are only about 4 times threshold.
Our second set of experiments shows that a high-contrastdrifting grating of 0.8 cycle/deg strongly masks test patternsdrifting in either direction, although the masking is partiallydirection selective. Little masking occurred when the testpattern was stationary. Movement-selective mechanismsthus appear to detect the moving test patterns, and thesemechanisms are strongly affected by both directions of mo-tion. The mechanisms are not unidirectional in that motionin opposite directions strongly affects the mechanisms. Theresults thus show that sensitive unidirectional mechanismsfor detecting rapid motion in the direction oppposite the maskdo not exist.
Adaptation studies also support the conclusion that motionanalyzers are not unidirectional. A moving adapting gratingmay selectively elevate the threshold of a subsequently viewedtest grating that moves in the same direction. However, the
threshold for a pattern moving in the opposite direction is alsotypically elevated, although to a lesser degree.10-'2 Thethreshold elevation for both test directions indicates that themechanisms are partially responsive to both directions ofmotion.
Thus, at suprathreshold levels, motion mechanisms sensi-tive to low spatial and high temporal frequencies are stronglyaffected by opposite directions of motion. However, thethreshold measurements of Levinson and Sekulerl and thepresent results obtained with near-threshold backgrounds(Fig. 3) show that, at very low contrast, information aboutrightward and leftward motion may be processed quite in-dependently.
B. Opponent-Motion MechanismsIn this section we describe how the main results for hightemporal and low spatial frequency may be explained by anopponent mechanism that is excited by motion in the pre-ferred direction and inhibited by motion in the opposite di-rection. The main results are summarized in Fig. 6. Figure6(a) shows masked thresholds for simply detecting a changein contrast of counterphase (+R+L and -R-L) and movingmasks (Fig. 5) of various strengths. Figure 6(b) shows maskedthresholds for the +R-L test pattern as a function of thestrength of the counterphase masks. In both parts there isa similar dip at low contrast, where the mask produces facili-tation. In Fig. 6(b), this facilitation extends over the full
10 I-
1-°
('5U)
A:
I 10 20 30 40 50 60
11 ~ ~ ~ I ~ ~ I I I I10 20 30 40 50 60
MASK STRENGTH (M/To)Fig. 6. Summary of masking for low spatial frequencies. (a) Thetest patterns simply produced a change in the contrast of a counter-phase or drifting mask. Test patterns: 0, increment in contrast oflow-contrast rightward-moving gratings of 0.8 cycle/deg and 20 Hz(from Fig. 5); A, , increment of high-contrast rightward-movinggrating of 0.8 cycle/deg and 20 Hz (from Fig. 4); X, 0, increments anddecrements of counterphase masks of 0.8 cycle/deg and 5, 10, and 20Hz and 2.2 cycle/deg and 10 and 20 Hz (from Tables 1 and 2). (b) Thetest pattern (+R-L) produced an increment in the rightward-movingcomponent of the counterphase mask and simultaneous equivalentdecrements in the leftward component. The masks were counter-phase gratings of 0.8 cycle/deg and 10 and 20 Hz (from Table 1 andFigs. 2 and 3) and 2.2 cycle/deg and 10 and 20 Hz (from Table 2).
I I I i II I
x A CFS
0 CM
x
(a)
I I I I I I
Stromeyer et al.
0
Vol. 1, No. 8/August 1984/J. Opt. Soc. Am. A 883
masking range. The basic difference is in the slope of therising portion of the curves. The straight line in Fig. 6(a)represents a Weber fraction of 0.14, which agrees withLegge's13 measurements of contrast discrimination of su-prathreshold stationary gratings. The slope of the line in Fig.6(b) is 7 times lower.
The initial dip may be explained by a detection mechanismwith a positively accelerated transducer function. 4"l Ascontrast is gradually increased, fixed contrast incrementsproduce proportionally greater output. When contrast israised above this accelerated region, the transducer functionbends in the opposite direction and becomes compressive.14"15
Thus weak masks will cause the mechanism to operate alongthe steep, sensitive part of the function, and facilitation willresult, whereas strong masks will cause a compressive non-linearity, and the threshold will rise.
The results suggest that, at low contrast along the initial dip,leftward and rightward motion are detected independently.In Fig. 6, the results are quite similar in this initial region.The results are quite different at higher contrasts along therising part of the curves.
Opponent-motion mechanisms that are inhibited by motionin the nonpreferred direction may explain these higher-con-trast results. Three aspects of the results will be considered.First, there is strong masking for a rightward-moving testadded to a similar high-contrast mask. The strong-right-ward-moving mask may push the response up on the less-sensitive compressive portion of the transducer function.Second, adding an equivalent leftward-moving mask to therightward-moving mask changes the mask to a counterphasepattern. This added leftward mask may produce inhibition,thus lowering the response to a more sensitive region of thetransducer function. This would make the mechanism highlysensitive to the +R-L test signal, as in Fig. 6(b). Inhibitionmay also explain a third aspect of the results: the low sensi-tivity to a simple contrast change of the counterphase pattern.The test pattern is relatively ineffective, as it contains bothrightward and leftward motion and thus produces responsesthat partially cancel because of inhibition.
Analogs of these inhibitory effects are found in color vision.Detection of a colored test pattern can be facilitated by addinga widely different colored adapting field to the original field.16The field addend presumably produces an inhibitory responsein an opponent color mechanism, thus placing the mechanismon a more sensitive part of its transducer function. 7 Simi-larly, adding a leftward-moving mask to a rightward maskproduces inhibition and facilitates detection of test movement.In addition to these cancellation effects involving the adaptingfields or masks, there are cancellation effects involving the testpatterns. Superimposing red and green test flashes may makethe combined flash less detectable than either flash by itself.18Similarly, adding a leftward-moving test to a rightward test(on a counterphase background) may produce a large decreasein sensitivity. As in color vision, opponent mechanisms mayserve to magnify sensitivity to small stimulus differences.
The concept of an opponent-motion mechanism has beenproposed by Nakayama and his colleagues1920 to explaincertain features, such as motion hyperacuity and detectionof shear in optical flow. The existence of opponent-motionmechanisms is further supported by Levinson and Sekuler, 2 1who observed that the direction-selective adaptation producedby a rightward-moving pattern (1.75 cycles/deg and 7.9 Hz)
was reduced by simultaneous presentation of a leftward-moving pattern. They hypothesized that motion in thenonpreferred direction produces inhibition. Electrophysio-logical studies of motion detectors in the retina of the pigeon,22
the rabbit,23 and the ground squirrel24 25 demonstrate that thereceptive-field center is excited by motion in the preferreddirection and is inhibited by motion in the opposite direction.Two spots moving in these opposite directions may produceresponse cancellation.2 2 24 Direction-selective simple andcomplex cells in the cat's striate cortex frequently manifestfacilitation for motion in the preferred direction2 6 27 and in-hibition for motion in the opposite direction.2 6 28 29
Reichardt's 30 well-known neural model for motion detectionin insects (e.g., Chlorophanus) is an opponent model since itresponds to the difference between rightward and leftwardmotion. The first stage computes an autocorrelation function(for rigid scenes) with a time delay. The output of this firststage yields a direction-selective response with the same signfor a light or dark bar moving in the same direction. Thesecond stage is a linear operation that takes the differencebetween rightward and leftward first-stage responses. VanSanten and Sperling3l noted that the model does not explainthe results of Levinson and Sekuler' on the detectability ofmoving versus counterphase gratings, for the model yields zeroresponse to counterphase gratings. However, many variantsof Reichardt's model are possible. Instead of the nonlinearautocorrelation, it is possible to have a linear-velocity-tunedmechanism followed by a rectification (so that the responseto light and dark bars is the same). Instead of a feedfor-ward-difference operation, it is possible to have a feedback-difference operation. Any model, however, must be modifiedto take into account the findings that, at low mask contrasts,the system is sensitive to unidirectional velocities rather thanto differential velocities. This modification can be accom-plished by having a threshold element suppress an oppositevelocity of low contrast. A second constraint that our dataplaces on any model is the requirement that the model becompatible with Weber's law, as shown in Figs. 5 and 6.
ACKNOWLEDGMENT
This research is supported by U.S. Air Force contract no.F49620-81-K-0016.
REFERENCES
1. E. Levinson and R. Sekuler, "The independence of channels inhuman vision selective for direction of motion," J. Physiol.(London) 250, 347-366 (1975).
2. C. F. Stromeyer III, J. C. Madsen, S. Klein, and Y. Y. Zeevi,"Movement-selective mechanisms in human vision sensitive tohigh spatial frequencies," J. Opt. Soc. Am. 68, 1002-1005(1978).
3. A. B. Watson, P. G. Thompson, B. J. Murphy, and J. Nachmias,"Summation and discrimination of gratings moving in oppositedirections," Vision Res. 20, 341-347 (1980).
4. R. J. W. Mansfield and J. Nachmias, "Perceived direction ofmotion under retinal image stabilization," Vision Res. 21,1423-1425 (1981).
5. C. Enroth-Cugell and J. G. Robson, "The contrast sensitivity ofretinal ganglion cells of the cat," J. Physiol. (London) 187,517-552(1966).
6. C. F. Stromeyer III, S. Klein, and C. E. Sternheim, "Is spatialadaptation caused by prolonged inhibition?" Vision Res. 17,603-606 (1977).
Stromeyer et al.
884 J. Opt. Soc. Am. A/Vol. 1, No. 8/August 1984
7. S. Klein and C. F. Stromeyer III, "On inhibition between spatialfrequency channels: adaptation to complex gratings," VisionRes. 20, 459-466 (1980).
8. J. J. Kulikowski and D. J. Tolhurst, "Psychophysical evidencefor sustained and transient detectors in human vision," J. Physiol.(London) 232, 149-162 (1973).
9. R. Sekuler, A. Pantle, and E. Levinson, "Physiological basis ofmotion perception," in Handbook of Sensory Physiology, Vol.VIII, Perception, R. Held, H. W. Leibowitz, and H.-L. Teuber,eds. (Springer-Verlag, Berlin, 1978), pp. 67-96.
10. A. Pantle, S. Lehmkuhle, and M. Caudill, "On the capacity ofdirectionally selective mechanisms to encode different dimensionsof moving stimuli," Perception 7, 261-267 (1978).
11. A. Pantle and R. Sekuler, "Contrast response of human visualmechanisms sensitive to orientation and direction of motion,"Vision Res. 9, 397-406 (1969).
12. D. J. Tolhurst, "Separate channels for the analysis of the shapeand the movement of a moving visual stimulus," J. Physiol.(London) 231, 385-402 (1973).
13. G. E. Legge, "A power law for contrast discrimination," VisionRes. 21, 457-467 (1981).
14. C. F. Stromeyer III and S. Klein, "Spatial frequency channels inhuman vision as asymmetric (edge) mechanisms," Vision Res.14, 1409-1420 (1974).
15. G. E. Legge and J. M. Foley, "Contrast masking in human vision,"J. Opt. Soc. Am. 70, 1458-1471 (1980).
16. J. D. Mollon and P. G. Polden, "Further anomalies of the bluemechanism," Invest. Ophthalmol. Vis. Sci. Suppl. 16, 140(1977).
17. E. N. Pugh, Jr. and J. D. Mollon, "A theory of the ir, and 7r3 colormechanisms of Stiles," Vision Res. 19, 293-312 (1979).
18. R. M. Boynton, M. Ikeda, and W. S. Stiles, "Interactions amongchromatic mechanisms as inferred from positive and negativeincrement thresholds," Vision Res. 4, 87-117 (1964).
19. K. Nakayama and J. M. Loomis, "Optical velocity patterns, ve-locity-sensitive neurons, and space perception: a hypothesis,"Perception 3, 63-80 (1974).
20. K. Nakayama and C. W. Tyler, "Psychophysical isolation ofmovement sensitivity by removal of familiar position cues," Vi-sion Res. 21, 427-433 (1981).
21. E. Levinson and R. Sekuler, "Inhibition and disinhibition of di-rection-specific mechanisms in human vision," Nature 254,692-694 (1975).
22. H. R. Maturana and S. Frenck, "Directional movement andhorizontal edge detectors in the pigeon retina," Science 142,977-979 (1963).
23. H. B. Barlow and W. R. Levick, "The mechanism of directionallyselective units in rabbit's retina. II: Directionally selective units,"J. Physiol. (London) 178, 477-504 (1965).
24. C. R. Michael, "Receptive fields of directionally selective unitsin the optic nerve of the ground squirrel," Science 152, 1092-1095(1966).
25. C. R. Michael, "Receptive fields of single optic nerve fibers in amammal with an all cone retina," J. Neurophysiol. 31, 257-267(1968).
26. R. C. Emerson and G. L. Gerstein, "Simple striate neurons in thecat. II. Mechanisms underlying directional asymmetry anddirectional selectivity," J. Neurophysiol. 40, 136-155 (1977).
27. J. A. Movshon, I. D. Thompson, and D. J. Tolhurst, "Receptivefield organization of complex cells in the cat's striate cortex," J.Physiol. (London) 283, 79-99 (1978).
28. A. W. Goodwin, G. H. Henry, and P. 0. Bishop, "Direction se-lectivity of simple striate cells: properties and mechanism," J.Neurophysiol. 38, 1500-1523 (1975).
29. A. W. Goodwin and G. H. Henry, "Direction selectivity of complexcells in a comparison with simple cells," J. Neurophysiol. 38,1524-1540 (1975).
30. W. Reichardt, "Autocorrelation, a principle for the evaluationof sensory information by the central nervous system," in SensoryCommunication, W. A. Rosenblith, ed. (MIT Press, Cambridge,Mass., 1961), pp. 303-317.
31. J. P. H. van Santen and G. Sperling, "Temporal covariance modelof human motion perception," J. Opt. Soc. Am. A 1, 451-473(1984).
