Vision Res. Vol. 32, No. 5, pp. 8154321, 1992 Printed in Great Britain. All rights reserved
0042-6989/92$5.00+ 0.00 Copyright 0 1992Pergamon Press plc
Effects of Colour Substitutions Upon Motion Detection in Spatially Random Patterns M. J. MORGAN,*?
R. CLEARY*$
Received 20 March 1991; in revised form 25 October 1991
To investigate the effects of colour upon motion detection, the short-range motion displacement limit (D,,,,,,) was determined using two-frame kinematograms in which the two classes of square comprising the pattern dilfered both in luminance and in colour. In the second motion frame, the squares retained either the same luminance and colour as in the first frame, or they changed their colour while retaining their luminance. The experiment was repeated at three different viewing distances to investigate the effects of element angular size. Two of the four observers had normal trichromatic colour vision; the other two were dichromats (protanopes). For the trichromatic observers, the change of colour between frames made motion displacements harder to detect when the squares were large, but not when they were small. The result accords with an input of colour into motion d&&ion at low but not at high spatial frequencies. For the dichromats, the colour change had little e&t at any of the viewing distances, thus mling out the possibility that the deleterious effects of colour substitution upon motion detection in trichromats was due to chromatic abberation or other artefacts. Motion
Colour
Dichromacy
Spatial frequency
INTRODUCTION There is some evidence that purely chromatic information has at most a minor influence upon the mechanisms responsible for short-range motion detection in random-dot patterns. The initial psychophysical evidence was that motion could not be detected in isoluminant coloured random-dot kinematograms (Ramachandran & Gregory, 1978). Physiological evidence points to the magnocellular pathway as being primarily responsible for motion analysis through its projection to the cortical visual area V5 (see reviews by Livingstone & Hubel, 1988; Zeki, 1980; Zeki & Shipp, 1988). Neurones in the magnocellular pathway are chromatically broad-band, in contrast to the wavelength selective neurones in the parvocellular pathway (Wiesel & Hubel, 1966; Derrington, Krauskopf & Lennie, 1984). Schiller, Logothetis and Charles (1990) have recently shown that selective lesions to the magnocellular pathway disrupt motion analysis but leave thresholds for colour discrimination unaffected, while lesions to the parvocellular pathway have the opposite effect. There are thus some grounds for the conclusion that motion is analysed primarily by neural mechanisms that do not respond to purely chromatic contrast (Livingstone & Hubel, 1988). *Department of Pharmacology, University of Edinburgh Medical School, 1 George Square, Edinburgh, Scotland. tTo whom all correspondence should be addressed. SPresent address: Caspe Research, King Edward’s Hospital Fund, 14 Palace. Court, Bayswater, London W2 4HT, England. “R
32,5--B
Other evidence, however, points to an influence of colour upon motion, albeit a weaker e&e& than luminance (Cavenagh, Boeglin & Favreau, 1985; Derrington & Badcock, 1985; Troscienco & Fahle, 1988; Gorea & Papathomas, 1987). Papathomas, Gorea and Julesz (1991) have recently shown that colour can resolve ambiguities in apparent motion, and that it can even dominate weak luminance cues when colour and luminance are in conflict. However, the stimuli used by Papathomas et al. (1991) consisted of large (0.38 x 0.29 deg) spatially-separated squares, and the possibility cannot be ruled out in these circumstances that colour was affected a long-range, attentional tracking process, rather than a low-level short-range process. Short-range motion (Braddick, 1974, 1980; Anstis, 1980) is the type of motion detection held to be used by the observer to solve the correspondence problem in random-dot kinematograms, where there are no obvious shape cues to guide the correspondence process from frame to frame. In these circumstances there is an upper spatial limit (Q_) to the size of displacement that can be detected. Displacements greater than D,, are seen as incoherent motion over the pattern rather than as a coherent shift of all the elements. In the Papathomas et al. study, it is not clear whether there would have been a D, effect, since data were collected with a single displacement. In the experiment reported here, we investigate the effects of colour in a motion detection task where short-range detection is unambiguously involved. The stimulus was a two-frame random-square kinematogram, in which there was a clear D,, effect with
815
M. J. MORGAN and R. CLEARY
816
luminance-de~ned elements. The patterns were composed of abutting squares differing both in luminance and colour. For example, the first frame of the twoframe sequence could consist of relatively high luminance green squares, and relatively low luminance red squares. In the control condition, the squares in the second frame maintained both their colour and luminance values. In the colour”substitution condition, the colour of the squares was reversed, but they maintained their luminance values (see examples in Fig. 1). To an achromatic system the two conditions should be identical and there should be no effect upon D=,. As one way of controlling for unwanted changes in luminance in the colour-substitution condition, two dichromats were included in the study, as well as normal t~chromats. Dichromats have only one class of cone in the red-green region of the spectrum, and thus the exchange of red and green should be much less visible to them than to trichromats. Any luminance changes, however, would be at least as great for the dichromats. The logic here is the same as in an experiment where colour was used to camouflage texture (Morgan, Mollon & Adam, 1989), and in which colour camouflage was found in trichromats, but not in dichromats. The dichromats are an especially valuable control for unwanted luminance transients caused by chromatic abberration since this pre-reeeptoral effect will be as great for dichromats as for trichromats. METHODS Apparat~
and st~~~~~
Random-square patterns were generated on a Barco (Model 33) RGB colour monitor controlled by a
graphics processor (Ikon Pixel Engine, Digisoive Ltd). The kinematogram consisted of a two-frame motion sequence, in which each frame was presented for 80 msec with an interframe interval of no greater than 18 msec. A single frame was generated by displaying a sub-area of a much larger, precomputed, pattern within a fixed window in the centre of the screen. Motion was produced by scrolling the second frame within this stationary window, the direction of scrolling being either upwards or downwards. Each of the two frames consisted of 16 x 16 contiguous squares, each of which was in turn composed of 4 x 4 identically-~oloured pixels. Each square in the pattern subtended 0.325 deg at the standard viewing distance of 2.28 m. (It should be noted that the squares comprising these kinematograms are larger than those typically employed in dete~inations of &a. The reason was to maximise the effects of the stimulus upon chromatic pathways by moving the stimuli to lower spatial frequencies.) Each square in the first frame was randomly chosen to be either green (50 cd/m’) or red (20 cd/m2). The Michelson contrast of the stimulus was therefore 0.42 with a mean luminance of 35 cd/m2. The corresponding squares in the second frame either had the same luminance and colour (control condition) as in the first frame or were reversed in colour (i.e. green became 20 cd/m*, and red became 50 cd/m’). Both of the possible combinations of colour and luminance were used, resulting in two control and two experimental conditions. The room in which the experiment took place was dimly lit by a desk-top lamp directed at the white ceiling. This background illumination produced a veiling luminance of approx. 3 cd/m2 on the face of the display. Procedure
FIGURE 1. The figure illustrates symbolicatly the stimub us& in the experiments. The white squares denote hi&h Ia ekiacnts and the dark squares denote low luminan~ elements. Ektnents marked “R” are red and those marked “G” are gmen. In the actual display each frame consisted of 16 x 16 elements, Each stimulus c~@?istedof a two-frame motion sequence (first frame, ieft; sezcond frame, right). Top: the stimuli maintain their colour and their hnniuan~ between frames (control condition). Bottom: the stimuli change theirc&our but not their luminance between frames.
The luminance of the differently coloured squares was measured using a microscope to focus the image of a small area of the screen upon the active surface of a PIN diode with a filter matched to the CIE luminosity curve. The current output of the PIN was converted to voltage and amplified x 10 by a d.c. amplifier with a 10 Hz low-pass titer. To establish the equivalence between voltage and luminance, the luminance of a homogeneous white screen was measured with an SE1 photometer, and the voltage output from the detector was measured using the same screen. Because of basic deficiencies in the CRT display, we found it necessary to establish these luminance values using exactly the same random patterns as in the experiment, since values changed when either the size or the number of squares changed. We do not claim that the resulting luminance matches were perfect, but reasons will be given later why small differences cannot explain our data. In dichromats it is not appropriate to match &ours by the CIE luminosity curve, since their sensitivity profile is necessarily different from that of the trichromat. For these subjects, therefore, the colours were matched subjectively in brightness. Additional observations were also undertaken in one of the trichromats
817
EFFECTS OF COLOUR ON MOTION DETECTION
(MM) using colours matched subjectively in brightness. The technique for brightness matching was to view a random red/green square pattern and to adjust the luminance of one of the colours until the borders were minimally distinct (Boynton, 1979). In these subjectivematch experiments, the red and green values in the first frame were the same as in the main experiment (i.e. green = 50 cd/m*; red = 20 cd/m*). The values in the second frame were such as to maintain the same (subjectively determined) luminance values but with the colours reversed. Since the protanopes are less sensitive than usual to red, this necessarily meant increasing the intensity of the red relative to the values used for the trichromats. The photometrically-determined luminance contrast in the second frame was thus 0.9 for observer RC and 0.84 for observer CC. The upper threshold for direction discrimination (D,,) was measured using the method of constant stimuli. A series of motion displacements ranging from 1 to 8 pixels was chosen, and on each trial one of these was selected randomly without replacement until there had been 10 trials at each displacement. On each trial
the direction of motion (up or down) was randomly selected and the observer pressed a button to indicate the perceived direction of motion. From the resulting psychometric function, the value of displacement producing 80% correct was found by interpolation, and this was defined as D,. Each condition was repeated 4 times, giving a total of 40 trials at each displacement. The whole experiment was replicated at 3 different viewing distances (2.28, 1.14 and 0.54 m). Subjects
The subjects were two males with normal trichromatic vision (AG, MJM) and two protanopes (RC, CC), who were characterised by their colour matching behaviour. Both dichromats had been subjects in an earlier study of colour camouflage (Morgan et al., 1989) and had shown the behaviour typical of other dichromats. RESULTS
The results of the experiment are shown in Figs 2 and 3. Figure 2 shows how the values of D, are affected by
9
t
RC
- . :: t
5-4--
n
3-2-1 --
1
0, FULL
1/2D
1/4D
FULL
6 1,‘ZD
1/4D
FRAME 1 /FRAME 9--
AG
0-o 0-e A-A
a--
9
A/A a/e A/B
MM
a 7 6
a===--=0
5 k 4
A
\
3 \ 2
FULL
1/2D
* A
1/4D
VIEWING
DISTANCE
FIGURE 2. The figure shows results for four observers in Expt 1. The task was to detect the direction of displacement in random-dot kinematograms composed of red and green elements. The red and green elements also differed in luminance. The colours and luminancts of the two classes of element were either the same in the two frames of the motion sequence (conditions A/A and B/B); or the colours of the elements but not their ltinanccs were exchanged between frames (conditions A/B and B/A). The experiment was conducted at 3 different viewing distances (horizontal axis). The two observers in the top panels (CC and RC) were dichromats, and those in the lower panels (MM, AG) were normal trichromats. Values on the vertical axis are Da, the upper threshold for motion direction detection in units of screen pixels, which subtended a visual angle of 4.9 arc min at the full viewing distance.
100
-100
80 --
,lOO-
80 --
80 -1
60-40 .20--
-201 0
; 2
: 4
: 6
: 8
: IO
; 12
;
14
; 16
%2QT 18 0
i 2
; 4
: 6
I 8
I 10
; 12
; 14
: 16
,204 18 0
: 2
I 4
I 6
: 8
I 10
; 12
I 14
-MM. FULL BO-1
60 --
2
60s-
-207
0
A-A
B-/A
6
8
,
AG FULL
2
4
6
8
80-60--
4
1
80.T
60.-
CC FULL
1
40--
40s20.o--
-209-+TT0 4
FIGURE 3. The figure shows individual psychometric functions for obacrvera ia different conditions of Expt 1. Each panel presentsthedataforasi~eobsarverataa~~di~.Wtthin~plaelthe~tcum,rsfatothe relations between the two frames, using the same symbols and conventions as in Fig. I. The horizontal axis repraentr the size of motion displacement, and the vertical axis shows the % errors in motion direction discrimination. The symbols have the same meaning as in Fig. 1. Each row of figures shows the data from a difkrent observer. The error bars an derived from the standard deviation of the binomial distribution with n = 40 (the number of trials on which the error statistic is based). 818
; 16
819
EFFECTS OF COLOUR ON MOTION DETECTION
viewing distance and by the relationship between the two frames. In a type A frame the green squares are more luminous than the red; in a type B frame the red squares are more luminous. In the A/A and B/B conditions the frames are identical apart from the motion step. In the A/B condition the colour values of the squares but not their luminance values are changed between frames. The B/A condition is the same as A/B except that the order of the two frames is reversed. Results are shown separately for four observers, two of whom (MM and AG) are trichromats, and two of whom (CC and RC) are dichromats. Figure 3 shows the psychometric functions from which the D, values in Fig. 1 are derived. Each panel shows the data from one observer at a particular viewing distance. Within each panel the curves refer to the A/A, B/B, A/B and B/A conditions, using the same symbols as in Fig. 1. Figure 4 shows the results for one observer only (MM) in the “subjective match” condition, where the luminance equivalence between the two colours was established by direct matching rather than photometrically (A/B subj). Also shown in the same figure for comparison are the results for the subjective match (A/B obj) and for the no-swap condition (A/A). The results can be summarised as follows: (i) For the two trichromats, colour-substitution made the motion harder to detect. This is reflected in the lower condition. values for D,, in the colour-substitution Because D,, values run in the opposite direction to most other thresholds, we make the same point in a way less likely to cause confusion: at large displacements, there were more errors in the colour-substitution condition than in the control condition. (ii) The effect of colour substitution depended upon the viewing distance. At the closest viewing distance, motion was generally harder to detect, and this was
10,
9
I
FRAME l/FRAME
MM
8 _s?
7i
‘\\ A
2
O-O
A/A
0-e
A/B
(SUBJ
A--h
A/B
(~BJ)
d.f. Distance Swap Swaprdistance
MM
AG
RC
CC
2118 20.4t 54.6t 14.0t 5.4* l/18 39.87 60.07 0.78 0.67 2/18 13.8t 16.0t 0.5 0.65
*P < 0.05; tP < 0.001.
particularly true in the colour-substitution condition. There was little effect of colour substitution at the greatest viewing distance. (iii) Colour substitution did not affect the performance of the two dichromatic subjects. To test the significance of these effects, a parametric Analysis of Variance was carried out on the data. Only the A/A and A/B conditions were included because the B/B and B/A were not experienced by observer CC, who was available for a limited time. The analysis was possible because there were 4 replications (independent thresholds) under each condition, giving an error term against which main effects and interactions could be tested. To begin with, an independent analysis was carried out on each observer (Table 1). This revealed that in the two trichromats (MM, AG) the main effects of colour-swap and distance, and the interaction between colour-swap and distance, were all highly significant (P < 0.001). In the two dichromats, however, only the effect of distance was significant. This is consistent with the interpretation that colour swap affected trichromats but not dichromats, and that the effects of colour swap for the trichromats was greater at the smallest viewing distance. To check the significance of differences between the trichromats and dichromats pairwise comparisons were carried out (Table 2). In every case, the colour blind variable (tri- vs dichromat) interacted significantly with the colour swap variable (S*C). This is consistent with our interpretation that colour-swap affected the trichromats, but not the dichromats. The three-way interaction (colour blind/swap/distance) was not so clear. In the case of one of the dichromats (RC) the three-way interactions were both significant; but in the case of the other dichromat (CC) both three-way interactions failed to reach conventional levels of significance.
O-------O
DISCUSSION
e
Eflects of viewing distance
\
Values of D,, were reduced as the subject approached the display. However, this is only true if D,, is expressed
\ A\.
01
TABLE 1. F-ratios for individual observers
TABLE 2. F-ratios for pairwise comparisons FULL
VIEWING
1/2D
1/4D
DISTANCE
FIGURE 4. The figure shows the results for one observer only (MM) in the “subjective match” condition, where the luminance equivalence between the two colours was established by direct matching rather than photometrically (A/B subj). Also shown in the same figure for comparison are the results for the objective match (A/B obj) and for a no-swap condition (A/A). The measure of performance is the upper motion displacement threshold (D,).
d.f. l/36 Swap (S) Distance (D) 2136 Colour blind (C) l/36 S*D 2136 D*C 2136 l/36 s*c S*C*D 2/36 *P < 0.01; tP < 0.001.
AG/RC
AG/CC
17.3’ 61.57 189t 6.0* 5.6* 30.1t 7.7;
18.lt 32.6t 1.3 6.1’ 2.9 7.9* 2.2 (0.12)
MM/RC 16.0t 34.3t 1lot 6.6* 0.6 27.17 4.2t
MM/CC 17.6t 19.47 0.2 6.7. 0.5 8.1* 2.78 (0.075)
820
M. J. MORGAN and R. CLEARY
in terms of displacements on the display screen (pixels). If D,,, is converted into visual angle, the opposite result holds: values of D,, increase as the subject gets nearer to the display, at least in the non-colour-swap condition. There are several possible explanations, not necessarily mutually exclusive. First, it is known that D,,, increases as the low spatial frequency content of the stimulus is enhanced by low-pass filtering (Chang & Julesz, 1983; Cleary & Braddick, 1990). As the observer approaches the screen, the spatial frequency content of the display is scaled to lower values. A second possible explanation, following Baker and Braddick (1985) is that the whole display subtends a larger visual angle from the closer viewing distance, with the result that it invades the peripheral visual field, where values of D,, are higher. If the effects of spatial frequency shift and peripheral viewing exactly compensated for the increased visual angle of the motion displacement at closer viewing distance, we would obtain scale-invariance (Koenderink, 1977). If true scale invariance held, thresholds would be constant over viewing distance when expressed in units of displacement on the screen. This is not the case, particularly at the closest viewing distance, where scale invariance breaks down, It would therefore appear that when stimuli are magnified at the eye, there comes a point when further magnification makes motion detection more difficult. This may be because D,, values do not increase linearly with eccentricity, as proposed by Baker and Braddick (1985). Efects of colour substitution The prediction that colour substitution shouId have no effect upon motion detection was confirmed for the greater viewing distances. However, colour substitution did have a deleterious effect at the shortest viewing distance in trichromats. The idea that the motion system is relatively insensitive to chromatic information must therefore be modified. A possible explanation is that there may be a direct input from the parvoceliular pathway into motion-detecting mechanisms. When there is no colour substitution, signals from the parvo- and magno-pathways are in accord, but when colour substitution occurs, noise is added to the parvo-signal. This does not str~ghtfo~ardiy explain the inte~ction with viewing distance. However, it would be reasonable to suppose that purely chromatic information has an input to the motion system only at low spatial frequencies, consistent with the low-pass spatial frequency characteristics of purely chromatic vision (Mullen, 1985; Morgan & Aiba, 1985; Krauskopf & Farell, 1991). Decreasing the viewing distance enhances the low spatial frequency content of the stimuli at the eye, and thus increases the contribution of purely chromatic mechanisms. An alternative explanation we have considered is that the motion signal arises only in neurones of the magnocellular pathway. Although these are chromatically broad-band, they nevertheless receive their input from cones, and have a centre-surround organisation. It is unlikely that the two classes of cones have exactly equal inputs to the centre and surround in all neurones:
presumably there is a random (binomial) scatter in frequencies over neurones, and this may account for the observed variation in chromatic opponency in ma~~ellular neurones in LGN (Derrington et al. 1984). Therefore, colour substitution will be detected by neurones in the magnocellular pathway, and the random changes in effective contrast of the stimulus will make motion harder to detect. The problem with this explanation, however, is that it does not obviously explain the effects of element angular size. The results of the experiment support the view that motion in random-dot kinematograms is primarily detected by achromatic mechanisms, most likely those in the magnocellular system. Colour, however, does influence motion detection at large element sizes, and thus presumably at low spatial frequencies. Our results are consistent with the possibility that magno- and parvo-systems are pooled prior to motion detection, with colour making a substantial contribution to the parvo-signal only at low spatial frequencies (Ingling & Martinez, 1983; Morgan & Aiba, 1985). The view that motion detection depends preferentially on an achromatic system may therefore reflect the spatial frequency content of stimuli used in the experiments, rather than a fundamental property of the visual pathway. If signals from different pathways are pooled prior to the decision process in motion detection, there may be little profit in debating whether there are one or many “motion mechanisms”. The answer to this question depends on what is meant by “a motion mechanism” (Cavenagh & Mather, 1990). The term could refer either to specific mechanisms for extracting a dir~tionaliyselective signal from a particular stimulus dimension, such as luminance, colour or texture; or to the final common pathway where decisions about motion are made.
REFERENCES An&s, S. M. (1980). The perception of apparent movement. Phitosophicat Transactions of the Royal Society of London B, 290, 153-168. Baker, C. L. Br Braddick, 0.1. (1985). ~nt~~tyd~d~t scaling of the limits for short-range apparent motion perception. Vfsiun Research, 25, 803-8 12. Boynton, R. M. (1979). Human color vision. New York: Holt (Reinhardt & Wilson). Braddick, 0. 3. (1974). A short range process in apparent motion. Vision Research, 14, 519-527. Braddick, 0. J. (19gO}.Low-Ievel and high-level processes in apparent motion. P~it~~~~~~t Transections of the Rayat Society af London 3, m, 137-151. Cavern&, P. & Mether, G. (1990). Motion: The long and the short of it. Spa&t Vision, 4, 103-129. Cavenagh, P., BoegIin, J. & Favreau, 0. E (1985). Perception of motion in equih,tmnrous kinematograms. Pffcffptition,14, 151-162. t &nits for q&a1 Cftaag, I. J. 6r Juiesz, 3. ($983). M frequency #temd random-dot cinematograms in apparent motion. Vision Research, 23, 1379-1385. Cleary, R. & Braddick, 0. J. (1990). Directiondiscrimination for band-pass filtered random dot kinematograms. Vision Research, 30, 303-316.
EFFECTS OF COLOUR ON MOTION DETECTION Derrington, A. M. & Badcock, D. R. (1985). The low-level motion system has both chromatic and achromatic inputs. Vision Research, 25, 18791884. Derrington, A. M., Krauskopf, J. & Lennie, P. (1984). Chromatic mechanisms in lateral geniculate nucleus of macaque. Journal of Physiology, 357, 241-265.
Gorea, A. & Papathomas, T. V. (1989). Motion processing by chromatic and achromatic pathways. Journal of the Optical Society of America, Ad, 59C602.
Ingling, C. R. & Martinez, E. (1983) The spatiochromatic signal of the r-g channel. In Mollon, J. D. & Sharpe, L. T. (Eds), Colour vision: Physiology and psychophysics. London: Academic Press. Koenderink, J. J. (1977). Current models of contrast processing. In Spekreijse, H. & Tweel, L. H. (Eds) Spatial conrrast, report of a workshop. Amsterdam: North Holland. Krauskopf, J. & Farell, B. (1991). Vernier acuity: Effects of chromatic content, blur and contrast. Vision Research, 31, 735-749. Livingstone, M. & Hubel, D. (1988). Segregation of form, colour, movements and depth: Anatomy, physiology and perception. Science, 240, 74&749. Morgan, M. J. & Aiba, T. S. (1985). Positional acuity with chromatic stimuli. Vision Research, 25, 689695. Morgan, M. J., Mollon, J. D. & Adam, A. (1989). Dichromats break colour-camouflage of textural boundaries. Investigative Ophthalmology and Visual Science, 30 (Suppl.), 220.
821
Mullen, K. (1985). The contrast sensitivity of human colour vision to red-green and blue-yellow chromatic gratings. Journal of Physiology, 3.59, 38149.
Papathomas, T. V., Gorea, A. & Julesz, B. (1991). Two carriers for motion perception: Color and luminance. Vision Research, 31, 1883-1891.
Ramachandran, V. S. & Gregory, R. L. (1978). Does colour provide an input to human motion perception? Nature, 275, 55-56. Schiller, P. H., Logothetis, N. K. & Charles, E. R. (1990). Function of colour-opponent and broad-band channels of the visual system. Nature, 343, 68-69.
Troscienco, T. & Fahle, M. (1988). Why do isoluminant stimuli appear slower? Journal of the Optical Society of America, AS, 871380. Wiesel, T. & Hubel, D. (1966). Spatial and chromatic interactions in the lateral geniculate nucleus of the rhesus monkey. Journal of Neurophysioiogy, 29, 1115-l 116. Zeki, S. (1980). The representation of colours in the visual cortex. Nature, 284, 412418.
Zeki, S. & Shipp, S. (1988). The functional connections. Nature, 335, 31 l-3 17.
Acknowledgements-This
logic of cortical
work was supported by a grant from the Medical Research Council.
0042-6989/92$5.00+ 0.00 Copyright 0 1992Pergamon Press plc
Effects of Colour Substitutions Upon Motion Detection in Spatially Random Patterns M. J. MORGAN,*?
R. CLEARY*$
Received 20 March 1991; in revised form 25 October 1991
To investigate the effects of colour upon motion detection, the short-range motion displacement limit (D,,,,,,) was determined using two-frame kinematograms in which the two classes of square comprising the pattern dilfered both in luminance and in colour. In the second motion frame, the squares retained either the same luminance and colour as in the first frame, or they changed their colour while retaining their luminance. The experiment was repeated at three different viewing distances to investigate the effects of element angular size. Two of the four observers had normal trichromatic colour vision; the other two were dichromats (protanopes). For the trichromatic observers, the change of colour between frames made motion displacements harder to detect when the squares were large, but not when they were small. The result accords with an input of colour into motion d&&ion at low but not at high spatial frequencies. For the dichromats, the colour change had little e&t at any of the viewing distances, thus mling out the possibility that the deleterious effects of colour substitution upon motion detection in trichromats was due to chromatic abberation or other artefacts. Motion
Colour
Dichromacy
Spatial frequency
INTRODUCTION There is some evidence that purely chromatic information has at most a minor influence upon the mechanisms responsible for short-range motion detection in random-dot patterns. The initial psychophysical evidence was that motion could not be detected in isoluminant coloured random-dot kinematograms (Ramachandran & Gregory, 1978). Physiological evidence points to the magnocellular pathway as being primarily responsible for motion analysis through its projection to the cortical visual area V5 (see reviews by Livingstone & Hubel, 1988; Zeki, 1980; Zeki & Shipp, 1988). Neurones in the magnocellular pathway are chromatically broad-band, in contrast to the wavelength selective neurones in the parvocellular pathway (Wiesel & Hubel, 1966; Derrington, Krauskopf & Lennie, 1984). Schiller, Logothetis and Charles (1990) have recently shown that selective lesions to the magnocellular pathway disrupt motion analysis but leave thresholds for colour discrimination unaffected, while lesions to the parvocellular pathway have the opposite effect. There are thus some grounds for the conclusion that motion is analysed primarily by neural mechanisms that do not respond to purely chromatic contrast (Livingstone & Hubel, 1988). *Department of Pharmacology, University of Edinburgh Medical School, 1 George Square, Edinburgh, Scotland. tTo whom all correspondence should be addressed. SPresent address: Caspe Research, King Edward’s Hospital Fund, 14 Palace. Court, Bayswater, London W2 4HT, England. “R
32,5--B
Other evidence, however, points to an influence of colour upon motion, albeit a weaker e&e& than luminance (Cavenagh, Boeglin & Favreau, 1985; Derrington & Badcock, 1985; Troscienco & Fahle, 1988; Gorea & Papathomas, 1987). Papathomas, Gorea and Julesz (1991) have recently shown that colour can resolve ambiguities in apparent motion, and that it can even dominate weak luminance cues when colour and luminance are in conflict. However, the stimuli used by Papathomas et al. (1991) consisted of large (0.38 x 0.29 deg) spatially-separated squares, and the possibility cannot be ruled out in these circumstances that colour was affected a long-range, attentional tracking process, rather than a low-level short-range process. Short-range motion (Braddick, 1974, 1980; Anstis, 1980) is the type of motion detection held to be used by the observer to solve the correspondence problem in random-dot kinematograms, where there are no obvious shape cues to guide the correspondence process from frame to frame. In these circumstances there is an upper spatial limit (Q_) to the size of displacement that can be detected. Displacements greater than D,, are seen as incoherent motion over the pattern rather than as a coherent shift of all the elements. In the Papathomas et al. study, it is not clear whether there would have been a D, effect, since data were collected with a single displacement. In the experiment reported here, we investigate the effects of colour in a motion detection task where short-range detection is unambiguously involved. The stimulus was a two-frame random-square kinematogram, in which there was a clear D,, effect with
815
M. J. MORGAN and R. CLEARY
816
luminance-de~ned elements. The patterns were composed of abutting squares differing both in luminance and colour. For example, the first frame of the twoframe sequence could consist of relatively high luminance green squares, and relatively low luminance red squares. In the control condition, the squares in the second frame maintained both their colour and luminance values. In the colour”substitution condition, the colour of the squares was reversed, but they maintained their luminance values (see examples in Fig. 1). To an achromatic system the two conditions should be identical and there should be no effect upon D=,. As one way of controlling for unwanted changes in luminance in the colour-substitution condition, two dichromats were included in the study, as well as normal t~chromats. Dichromats have only one class of cone in the red-green region of the spectrum, and thus the exchange of red and green should be much less visible to them than to trichromats. Any luminance changes, however, would be at least as great for the dichromats. The logic here is the same as in an experiment where colour was used to camouflage texture (Morgan, Mollon & Adam, 1989), and in which colour camouflage was found in trichromats, but not in dichromats. The dichromats are an especially valuable control for unwanted luminance transients caused by chromatic abberration since this pre-reeeptoral effect will be as great for dichromats as for trichromats. METHODS Apparat~
and st~~~~~
Random-square patterns were generated on a Barco (Model 33) RGB colour monitor controlled by a
graphics processor (Ikon Pixel Engine, Digisoive Ltd). The kinematogram consisted of a two-frame motion sequence, in which each frame was presented for 80 msec with an interframe interval of no greater than 18 msec. A single frame was generated by displaying a sub-area of a much larger, precomputed, pattern within a fixed window in the centre of the screen. Motion was produced by scrolling the second frame within this stationary window, the direction of scrolling being either upwards or downwards. Each of the two frames consisted of 16 x 16 contiguous squares, each of which was in turn composed of 4 x 4 identically-~oloured pixels. Each square in the pattern subtended 0.325 deg at the standard viewing distance of 2.28 m. (It should be noted that the squares comprising these kinematograms are larger than those typically employed in dete~inations of &a. The reason was to maximise the effects of the stimulus upon chromatic pathways by moving the stimuli to lower spatial frequencies.) Each square in the first frame was randomly chosen to be either green (50 cd/m’) or red (20 cd/m2). The Michelson contrast of the stimulus was therefore 0.42 with a mean luminance of 35 cd/m2. The corresponding squares in the second frame either had the same luminance and colour (control condition) as in the first frame or were reversed in colour (i.e. green became 20 cd/m*, and red became 50 cd/m’). Both of the possible combinations of colour and luminance were used, resulting in two control and two experimental conditions. The room in which the experiment took place was dimly lit by a desk-top lamp directed at the white ceiling. This background illumination produced a veiling luminance of approx. 3 cd/m2 on the face of the display. Procedure
FIGURE 1. The figure illustrates symbolicatly the stimub us& in the experiments. The white squares denote hi&h Ia ekiacnts and the dark squares denote low luminan~ elements. Ektnents marked “R” are red and those marked “G” are gmen. In the actual display each frame consisted of 16 x 16 elements, Each stimulus c~@?istedof a two-frame motion sequence (first frame, ieft; sezcond frame, right). Top: the stimuli maintain their colour and their hnniuan~ between frames (control condition). Bottom: the stimuli change theirc&our but not their luminance between frames.
The luminance of the differently coloured squares was measured using a microscope to focus the image of a small area of the screen upon the active surface of a PIN diode with a filter matched to the CIE luminosity curve. The current output of the PIN was converted to voltage and amplified x 10 by a d.c. amplifier with a 10 Hz low-pass titer. To establish the equivalence between voltage and luminance, the luminance of a homogeneous white screen was measured with an SE1 photometer, and the voltage output from the detector was measured using the same screen. Because of basic deficiencies in the CRT display, we found it necessary to establish these luminance values using exactly the same random patterns as in the experiment, since values changed when either the size or the number of squares changed. We do not claim that the resulting luminance matches were perfect, but reasons will be given later why small differences cannot explain our data. In dichromats it is not appropriate to match &ours by the CIE luminosity curve, since their sensitivity profile is necessarily different from that of the trichromat. For these subjects, therefore, the colours were matched subjectively in brightness. Additional observations were also undertaken in one of the trichromats
817
EFFECTS OF COLOUR ON MOTION DETECTION
(MM) using colours matched subjectively in brightness. The technique for brightness matching was to view a random red/green square pattern and to adjust the luminance of one of the colours until the borders were minimally distinct (Boynton, 1979). In these subjectivematch experiments, the red and green values in the first frame were the same as in the main experiment (i.e. green = 50 cd/m*; red = 20 cd/m*). The values in the second frame were such as to maintain the same (subjectively determined) luminance values but with the colours reversed. Since the protanopes are less sensitive than usual to red, this necessarily meant increasing the intensity of the red relative to the values used for the trichromats. The photometrically-determined luminance contrast in the second frame was thus 0.9 for observer RC and 0.84 for observer CC. The upper threshold for direction discrimination (D,,) was measured using the method of constant stimuli. A series of motion displacements ranging from 1 to 8 pixels was chosen, and on each trial one of these was selected randomly without replacement until there had been 10 trials at each displacement. On each trial
the direction of motion (up or down) was randomly selected and the observer pressed a button to indicate the perceived direction of motion. From the resulting psychometric function, the value of displacement producing 80% correct was found by interpolation, and this was defined as D,. Each condition was repeated 4 times, giving a total of 40 trials at each displacement. The whole experiment was replicated at 3 different viewing distances (2.28, 1.14 and 0.54 m). Subjects
The subjects were two males with normal trichromatic vision (AG, MJM) and two protanopes (RC, CC), who were characterised by their colour matching behaviour. Both dichromats had been subjects in an earlier study of colour camouflage (Morgan et al., 1989) and had shown the behaviour typical of other dichromats. RESULTS
The results of the experiment are shown in Figs 2 and 3. Figure 2 shows how the values of D, are affected by
9
t
RC
- . :: t
5-4--
n
3-2-1 --
1
0, FULL
1/2D
1/4D
FULL
6 1,‘ZD
1/4D
FRAME 1 /FRAME 9--
AG
0-o 0-e A-A
a--
9
A/A a/e A/B
MM
a 7 6
a===--=0
5 k 4
A
\
3 \ 2
FULL
1/2D
* A
1/4D
VIEWING
DISTANCE
FIGURE 2. The figure shows results for four observers in Expt 1. The task was to detect the direction of displacement in random-dot kinematograms composed of red and green elements. The red and green elements also differed in luminance. The colours and luminancts of the two classes of element were either the same in the two frames of the motion sequence (conditions A/A and B/B); or the colours of the elements but not their ltinanccs were exchanged between frames (conditions A/B and B/A). The experiment was conducted at 3 different viewing distances (horizontal axis). The two observers in the top panels (CC and RC) were dichromats, and those in the lower panels (MM, AG) were normal trichromats. Values on the vertical axis are Da, the upper threshold for motion direction detection in units of screen pixels, which subtended a visual angle of 4.9 arc min at the full viewing distance.
100
-100
80 --
,lOO-
80 --
80 -1
60-40 .20--
-201 0
; 2
: 4
: 6
: 8
: IO
; 12
;
14
; 16
%2QT 18 0
i 2
; 4
: 6
I 8
I 10
; 12
; 14
: 16
,204 18 0
: 2
I 4
I 6
: 8
I 10
; 12
I 14
-MM. FULL BO-1
60 --
2
60s-
-207
0
A-A
B-/A
6
8
,
AG FULL
2
4
6
8
80-60--
4
1
80.T
60.-
CC FULL
1
40--
40s20.o--
-209-+TT0 4
FIGURE 3. The figure shows individual psychometric functions for obacrvera ia different conditions of Expt 1. Each panel presentsthedataforasi~eobsarverataa~~di~.Wtthin~plaelthe~tcum,rsfatothe relations between the two frames, using the same symbols and conventions as in Fig. I. The horizontal axis repraentr the size of motion displacement, and the vertical axis shows the % errors in motion direction discrimination. The symbols have the same meaning as in Fig. 1. Each row of figures shows the data from a difkrent observer. The error bars an derived from the standard deviation of the binomial distribution with n = 40 (the number of trials on which the error statistic is based). 818
; 16
819
EFFECTS OF COLOUR ON MOTION DETECTION
viewing distance and by the relationship between the two frames. In a type A frame the green squares are more luminous than the red; in a type B frame the red squares are more luminous. In the A/A and B/B conditions the frames are identical apart from the motion step. In the A/B condition the colour values of the squares but not their luminance values are changed between frames. The B/A condition is the same as A/B except that the order of the two frames is reversed. Results are shown separately for four observers, two of whom (MM and AG) are trichromats, and two of whom (CC and RC) are dichromats. Figure 3 shows the psychometric functions from which the D, values in Fig. 1 are derived. Each panel shows the data from one observer at a particular viewing distance. Within each panel the curves refer to the A/A, B/B, A/B and B/A conditions, using the same symbols as in Fig. 1. Figure 4 shows the results for one observer only (MM) in the “subjective match” condition, where the luminance equivalence between the two colours was established by direct matching rather than photometrically (A/B subj). Also shown in the same figure for comparison are the results for the subjective match (A/B obj) and for the no-swap condition (A/A). The results can be summarised as follows: (i) For the two trichromats, colour-substitution made the motion harder to detect. This is reflected in the lower condition. values for D,, in the colour-substitution Because D,, values run in the opposite direction to most other thresholds, we make the same point in a way less likely to cause confusion: at large displacements, there were more errors in the colour-substitution condition than in the control condition. (ii) The effect of colour substitution depended upon the viewing distance. At the closest viewing distance, motion was generally harder to detect, and this was
10,
9
I
FRAME l/FRAME
MM
8 _s?
7i
‘\\ A
2
O-O
A/A
0-e
A/B
(SUBJ
A--h
A/B
(~BJ)
d.f. Distance Swap Swaprdistance
MM
AG
RC
CC
2118 20.4t 54.6t 14.0t 5.4* l/18 39.87 60.07 0.78 0.67 2/18 13.8t 16.0t 0.5 0.65
*P < 0.05; tP < 0.001.
particularly true in the colour-substitution condition. There was little effect of colour substitution at the greatest viewing distance. (iii) Colour substitution did not affect the performance of the two dichromatic subjects. To test the significance of these effects, a parametric Analysis of Variance was carried out on the data. Only the A/A and A/B conditions were included because the B/B and B/A were not experienced by observer CC, who was available for a limited time. The analysis was possible because there were 4 replications (independent thresholds) under each condition, giving an error term against which main effects and interactions could be tested. To begin with, an independent analysis was carried out on each observer (Table 1). This revealed that in the two trichromats (MM, AG) the main effects of colour-swap and distance, and the interaction between colour-swap and distance, were all highly significant (P < 0.001). In the two dichromats, however, only the effect of distance was significant. This is consistent with the interpretation that colour swap affected trichromats but not dichromats, and that the effects of colour swap for the trichromats was greater at the smallest viewing distance. To check the significance of differences between the trichromats and dichromats pairwise comparisons were carried out (Table 2). In every case, the colour blind variable (tri- vs dichromat) interacted significantly with the colour swap variable (S*C). This is consistent with our interpretation that colour-swap affected the trichromats, but not the dichromats. The three-way interaction (colour blind/swap/distance) was not so clear. In the case of one of the dichromats (RC) the three-way interactions were both significant; but in the case of the other dichromat (CC) both three-way interactions failed to reach conventional levels of significance.
O-------O
DISCUSSION
e
Eflects of viewing distance
\
Values of D,, were reduced as the subject approached the display. However, this is only true if D,, is expressed
\ A\.
01
TABLE 1. F-ratios for individual observers
TABLE 2. F-ratios for pairwise comparisons FULL
VIEWING
1/2D
1/4D
DISTANCE
FIGURE 4. The figure shows the results for one observer only (MM) in the “subjective match” condition, where the luminance equivalence between the two colours was established by direct matching rather than photometrically (A/B subj). Also shown in the same figure for comparison are the results for the objective match (A/B obj) and for a no-swap condition (A/A). The measure of performance is the upper motion displacement threshold (D,).
d.f. l/36 Swap (S) Distance (D) 2136 Colour blind (C) l/36 S*D 2136 D*C 2136 l/36 s*c S*C*D 2/36 *P < 0.01; tP < 0.001.
AG/RC
AG/CC
17.3’ 61.57 189t 6.0* 5.6* 30.1t 7.7;
18.lt 32.6t 1.3 6.1’ 2.9 7.9* 2.2 (0.12)
MM/RC 16.0t 34.3t 1lot 6.6* 0.6 27.17 4.2t
MM/CC 17.6t 19.47 0.2 6.7. 0.5 8.1* 2.78 (0.075)
820
M. J. MORGAN and R. CLEARY
in terms of displacements on the display screen (pixels). If D,,, is converted into visual angle, the opposite result holds: values of D,, increase as the subject gets nearer to the display, at least in the non-colour-swap condition. There are several possible explanations, not necessarily mutually exclusive. First, it is known that D,,, increases as the low spatial frequency content of the stimulus is enhanced by low-pass filtering (Chang & Julesz, 1983; Cleary & Braddick, 1990). As the observer approaches the screen, the spatial frequency content of the display is scaled to lower values. A second possible explanation, following Baker and Braddick (1985) is that the whole display subtends a larger visual angle from the closer viewing distance, with the result that it invades the peripheral visual field, where values of D,, are higher. If the effects of spatial frequency shift and peripheral viewing exactly compensated for the increased visual angle of the motion displacement at closer viewing distance, we would obtain scale-invariance (Koenderink, 1977). If true scale invariance held, thresholds would be constant over viewing distance when expressed in units of displacement on the screen. This is not the case, particularly at the closest viewing distance, where scale invariance breaks down, It would therefore appear that when stimuli are magnified at the eye, there comes a point when further magnification makes motion detection more difficult. This may be because D,, values do not increase linearly with eccentricity, as proposed by Baker and Braddick (1985). Efects of colour substitution The prediction that colour substitution shouId have no effect upon motion detection was confirmed for the greater viewing distances. However, colour substitution did have a deleterious effect at the shortest viewing distance in trichromats. The idea that the motion system is relatively insensitive to chromatic information must therefore be modified. A possible explanation is that there may be a direct input from the parvoceliular pathway into motion-detecting mechanisms. When there is no colour substitution, signals from the parvo- and magno-pathways are in accord, but when colour substitution occurs, noise is added to the parvo-signal. This does not str~ghtfo~ardiy explain the inte~ction with viewing distance. However, it would be reasonable to suppose that purely chromatic information has an input to the motion system only at low spatial frequencies, consistent with the low-pass spatial frequency characteristics of purely chromatic vision (Mullen, 1985; Morgan & Aiba, 1985; Krauskopf & Farell, 1991). Decreasing the viewing distance enhances the low spatial frequency content of the stimuli at the eye, and thus increases the contribution of purely chromatic mechanisms. An alternative explanation we have considered is that the motion signal arises only in neurones of the magnocellular pathway. Although these are chromatically broad-band, they nevertheless receive their input from cones, and have a centre-surround organisation. It is unlikely that the two classes of cones have exactly equal inputs to the centre and surround in all neurones:
presumably there is a random (binomial) scatter in frequencies over neurones, and this may account for the observed variation in chromatic opponency in ma~~ellular neurones in LGN (Derrington et al. 1984). Therefore, colour substitution will be detected by neurones in the magnocellular pathway, and the random changes in effective contrast of the stimulus will make motion harder to detect. The problem with this explanation, however, is that it does not obviously explain the effects of element angular size. The results of the experiment support the view that motion in random-dot kinematograms is primarily detected by achromatic mechanisms, most likely those in the magnocellular system. Colour, however, does influence motion detection at large element sizes, and thus presumably at low spatial frequencies. Our results are consistent with the possibility that magno- and parvo-systems are pooled prior to motion detection, with colour making a substantial contribution to the parvo-signal only at low spatial frequencies (Ingling & Martinez, 1983; Morgan & Aiba, 1985). The view that motion detection depends preferentially on an achromatic system may therefore reflect the spatial frequency content of stimuli used in the experiments, rather than a fundamental property of the visual pathway. If signals from different pathways are pooled prior to the decision process in motion detection, there may be little profit in debating whether there are one or many “motion mechanisms”. The answer to this question depends on what is meant by “a motion mechanism” (Cavenagh & Mather, 1990). The term could refer either to specific mechanisms for extracting a dir~tionaliyselective signal from a particular stimulus dimension, such as luminance, colour or texture; or to the final common pathway where decisions about motion are made.
REFERENCES An&s, S. M. (1980). The perception of apparent movement. Phitosophicat Transactions of the Royal Society of London B, 290, 153-168. Baker, C. L. Br Braddick, 0.1. (1985). ~nt~~tyd~d~t scaling of the limits for short-range apparent motion perception. Vfsiun Research, 25, 803-8 12. Boynton, R. M. (1979). Human color vision. New York: Holt (Reinhardt & Wilson). Braddick, 0. 3. (1974). A short range process in apparent motion. Vision Research, 14, 519-527. Braddick, 0. J. (19gO}.Low-Ievel and high-level processes in apparent motion. P~it~~~~~~t Transections of the Rayat Society af London 3, m, 137-151. Cavern&, P. & Mether, G. (1990). Motion: The long and the short of it. Spa&t Vision, 4, 103-129. Cavenagh, P., BoegIin, J. & Favreau, 0. E (1985). Perception of motion in equih,tmnrous kinematograms. Pffcffptition,14, 151-162. t &nits for q&a1 Cftaag, I. J. 6r Juiesz, 3. ($983). M frequency #temd random-dot cinematograms in apparent motion. Vision Research, 23, 1379-1385. Cleary, R. & Braddick, 0. J. (1990). Directiondiscrimination for band-pass filtered random dot kinematograms. Vision Research, 30, 303-316.
EFFECTS OF COLOUR ON MOTION DETECTION Derrington, A. M. & Badcock, D. R. (1985). The low-level motion system has both chromatic and achromatic inputs. Vision Research, 25, 18791884. Derrington, A. M., Krauskopf, J. & Lennie, P. (1984). Chromatic mechanisms in lateral geniculate nucleus of macaque. Journal of Physiology, 357, 241-265.
Gorea, A. & Papathomas, T. V. (1989). Motion processing by chromatic and achromatic pathways. Journal of the Optical Society of America, Ad, 59C602.
Ingling, C. R. & Martinez, E. (1983) The spatiochromatic signal of the r-g channel. In Mollon, J. D. & Sharpe, L. T. (Eds), Colour vision: Physiology and psychophysics. London: Academic Press. Koenderink, J. J. (1977). Current models of contrast processing. In Spekreijse, H. & Tweel, L. H. (Eds) Spatial conrrast, report of a workshop. Amsterdam: North Holland. Krauskopf, J. & Farell, B. (1991). Vernier acuity: Effects of chromatic content, blur and contrast. Vision Research, 31, 735-749. Livingstone, M. & Hubel, D. (1988). Segregation of form, colour, movements and depth: Anatomy, physiology and perception. Science, 240, 74&749. Morgan, M. J. & Aiba, T. S. (1985). Positional acuity with chromatic stimuli. Vision Research, 25, 689695. Morgan, M. J., Mollon, J. D. & Adam, A. (1989). Dichromats break colour-camouflage of textural boundaries. Investigative Ophthalmology and Visual Science, 30 (Suppl.), 220.
821
Mullen, K. (1985). The contrast sensitivity of human colour vision to red-green and blue-yellow chromatic gratings. Journal of Physiology, 3.59, 38149.
Papathomas, T. V., Gorea, A. & Julesz, B. (1991). Two carriers for motion perception: Color and luminance. Vision Research, 31, 1883-1891.
Ramachandran, V. S. & Gregory, R. L. (1978). Does colour provide an input to human motion perception? Nature, 275, 55-56. Schiller, P. H., Logothetis, N. K. & Charles, E. R. (1990). Function of colour-opponent and broad-band channels of the visual system. Nature, 343, 68-69.
Troscienco, T. & Fahle, M. (1988). Why do isoluminant stimuli appear slower? Journal of the Optical Society of America, AS, 871380. Wiesel, T. & Hubel, D. (1966). Spatial and chromatic interactions in the lateral geniculate nucleus of the rhesus monkey. Journal of Neurophysioiogy, 29, 1115-l 116. Zeki, S. (1980). The representation of colours in the visual cortex. Nature, 284, 412418.
Zeki, S. & Shipp, S. (1988). The functional connections. Nature, 335, 31 l-3 17.
Acknowledgements-This
logic of cortical
work was supported by a grant from the Medical Research Council.
