30 J. Opt. Soc. Am. A/Vol. 9, No. 1/January 1992 J. Skon and L.

Arend

Color changes in red and white projections at

low luminance

Joy Skon
3 Chauncy Street, Cambridge,Massachusetts 02138

Lawrence Arend
Eye Research Institute, Harvard Medical School, 20 Staniford Street, Boston, Massachusetts 02114

Received August 21, 1990; revised manuscript received August 13, 1991; accepted August 13, 1991
In previous studies photographic projections containing only mixtures of red and white light produced a wide
variety of hues. In the present experiment similar patterns were produced on a color monitor. Squares with
four combinations of red and white light were studied at two mean luminances. Subjects reported that a chro-
maticity that appeared pink at our higher luminance appeared yellow at low luminance.

INTRODUCTION matching remembered Munsell standard colors. In the

present experiment we utilized a color judgment method
This paper reports explorations of perception of red and in which the subjects made quantitative estimates of hue
white patterns. Early researchers' discovered that ele- and saturation.
mentary color photography was possible with only two The main objective of the research presented herein
primary colors. Land demonstrated in 19592that a sur- was to quantify colors perceived in simple red-and-white
prising proportion of the hues of the spectrum can be pro- images of squares presented at two luminance levels, with
duced by the combination of lights that alone appear red the lower intensity case in the mesopic range. The main
and white. Land made red and green color separation conclusion of Pearson et al. was corroborated by our re-
records of multicolored natural scenes in the form of black sults. At reduced luminance, yellow and violet colors, as
and white positive photographic slides and projected them well as the expected red and cyan, were perceived in red-
in registration through red and neutral-density (ND) and-white patterns of squares.
filters, respectively. By this method much of the gamut
of hues in the original scene was reconstructed. Land
built a large collection of black and white color separation METHODS
slide pairs of a variety of subjects by this red-and-green Patterns
decomposition. We simulated Land's original photographic method of pro-
He also composed slide pairs of images consisting of ar- ducing red-and-white images. We used a Tektronix
rays of squares, with the two photographic densities in 690SR color monitor, controlled by a VAX computer and
each square chosen randomly from a gray scale. Red- Adage image processor. Spectra of the phosphors of the
and-white pictures projected from these records did not monitor are shown in Fig. 1. The red phosphor alone and
possess the range of hues attainable from natural scenes. the D65 alignment white of the monitor were used to ap-
However, in a prior study the first author' found a subtle proximate' the red and white primaries in Land's projec-
increase in the range of hues when the intensity with tions. Our red primary had lower purity and shorter
which these red-and-white patterns were projected was dominant wavelength than Land's, and his white was yel-
reduced to a low level by the interposition of a 3.0-log-unit lower than ours.
ND filter. In the present study we extended that result Qualitatively, we sought to produce four discrete mix-
using a video monitor to simulate the projected red-and- tures of the red and white primaries. We chose four red/
white patterns. white ratios and three luminances to produce our basic
Pearson et al.4 had described this effect of mean lumi- patterns, shown schematically in Fig. 2. The upper des-
nance, and Wheeler had done extensive research,5 6 on per- ignations represent relative luminances of the red and
ception in red and white in the domain of both simple and white primaries, and the lower numbers are the total lu-
complex images of squares. Pearson et al. reported a minances (no ND filter).
much higher incidence of yellow and purple-blue Munsell We used three spatial arrangements of the four colors
hues in subjects' color matches to cells in 10 x 10 ma- (Fig. 2). In the contiguous condition the four 1.4° squares
trices of red and white random squares when the patterns (la, 2a, 3a, 4a) were presented in a cluster on a dark back-
were projected at mesopic levels. ground (solid lines). To eliminate contrast interactions
The study of Pearson et al. was criticized by McCann among the patches, we also used a single patch condition.
and Auerbach7 on the grounds that considerable uncer- Each of the four squares was presented on the dark sur-
tainty was introduced into the results by their method of round by itself. In our third, separated condition (dotted
0740-3232/92/010030-04$05.00 X 1992 Optical Society of America
J. Skon and L. Arend Vol. 9, No. 1/January 1992/J. Opt. Soc. Am. A 31

Procedures
The subjects judged each of the four patches in each of the
three spatial configurations at each of the two luminance
levels. The low-luminance condition began with 10 min of
dark adaptation, and both levels were preceded by 3 min
of adaptation to the geometric mean luminance. The
LIJ ..
0::
order of presentation of the 12 stimuli was determined
randomly. Each patch was judged once in each of two ses-
sions. A cursor signaled the square to be judged.
TI In each trial the observers described the appearance of
LIJI- 0.2- the designated patch with a color estimation technique
similar to that of Boynton and Gordon.9 Observers first
reported the relative proportions of the chromatic and
achromatic components of the color, bounded by neutral
- J0 420 40 500 540 580 620 660 700 740
(0%o)and 100%. They then judged the hue of the chro-
matic component by choosing relative percentages of red,
WAVELENGTH (nm)
green, yellow,and blue (constrained to sum to 100%o).
Fig. 1. Reh itive spectral power of the phosphors of the RGB
monitor. Opponent-Color Plots. The judgments were analyzed
and plotted in a conventional opponent-color graph. The
subjects never responded with more than two hue compo-
lb . 2b nents, one from each opponent pair (red-green, yellow-
I
I
I
I ,, 4.2 9 blue), in agreement with most previous experiments.
I I
I
I
I
I
9
I
I
9 The mean hue was plotted on the angular coordinate, with
9 -9
the ratio of the two hue components determining the loca-
tion within the quadrant. The mean ratio of chromatic to
achromatic content was plotted as the radial coordinate.
Training. In a training session before the main experi-
la 2a .41 ment, simulated Munsell squares were presented one at a
100W 50W/50R time on the monitor. Each of 18 colors appeared once in
each of 6 blocks, in pseudorandom sequence. The sub-
13.3 32.0
4.2' jects gave numerical descriptions of the color appearance
3a 4a of each patch by the method described above.
50W/50R 1OOR
5.5 13.3
Both subjects' judgments corresponded closely with the
Munsell designations of the patches, as shown by the op-
ponent analysis. The coordinates of the mean judgments
were computed as described above. The coordinates of
--- J---- 9~~~~~~r-
-- - -
the Munsell designations were computed as follows. The
*3b ,4 Munsell hue circle was distorted to align pure red (5R),
green (5G), yellow (5Y), and blue (5B) with the red, green,
9 1
yellow, and blue opponent poles, respectively. Munsell
purple (5P) was assumed to lie at -45°, midway between
Fig. 2. Contiguous (solid lines) and separated (dashed lines) spa- the red and blue poles. All other Munsell hue designations
tial arrangements of the four color squares. The balance of red were assigned angles that were in linear proportion to their
and white primaries and the luminance (in units of candelas per positions relative to those five pure Munsell hues. The
square meter) in the high mean luminance condition is indicated
in each square.
Table 1. Stimulus Specifications
lines) all four patches (lb, 2b, 3b, 4b) were simultaneously High Low
present, with the separations indicated. (0.0 ND) (2.5 ND)
Scotopic Scotopic
Each of the spatial configurations was studied at two
Square x y Trolands Trolands
overall luminance levels. The chromaticity coordinates
and the photopic and scotopic luminances for each square 1 0.313 0.329 27.3 0.086
at each luminance level are shown in Table 1. In the low- 2 0.467 0.329 39.1 0.124
luminance case the subject viewed the monitor through a 3 0.467 0.329 6.6 0.021
goggle with a 2.5 ND Kodak Wratten Number 96 gelatin 4 0.615 0.329 5.0 0.016
filter fitted to the right eyepiece and the left eyepiece Photopic Photopic
blocked. In the high-luminance condition the filter was Trolands Trolands
removed. The monitor screen was 45 in. (1.14 m) from
the chinrest. Stimuli were presented continuously. 1 376.0 1.19
2 904.8 2.86
The two subjects (the authors) were experienced psycho-
3 155.5 0.49
physical observers with normal (Farnsworth-Munsell 100- 1.19
4 376.0
Hue Test) color vision.
 32 J. Opt. Soc. Am. A/Vol. 9, No. 1/January 1992 J. Skon and L. Arend

saturations of Munsell chroma /12 patches were always of them did see a definite yellow in similar patterns pro-
judged to be 100% chromatic, so Munsell chromas /0-/12 duced with different equipment on a separate occasion.
were linearly scaled into the radial dimension, 0-100%. Another observer reported a peculiar transparency ap-
For subject LA the mean hue and saturation judgments pearance, a slightly bluish-red square seen through a
differed from the Munsell hue and chroma designations strong yellow illumination or transparency. The observ-
by 1.340 and 2.26%o,respectively. For subject JS the cor- ers here (the authors) are puzzled by these reports, since
responding mean differences were 4.67° and 0.99%. For both always see an apparently normal yellow square under
LA the standard deviations of the hue and saturation a wide variety of conditions. LA reported saturated yel-
judgments were 5.65° and 4.99%, respectively. For JS the low with 10 sec of dark adaptation or 10 min and over a
standard deviations of the hue and saturation judgments mean luminance range extending at least 0.5 log unit
were 4.070 and 3.94%o,respectively. above and below that described above. Our research so far
has given us no good hypotheses regarding this unusual
degree of variability.
RESULTS
Subjects' averaged color judgments for the four red-and-
white squares presented contiguously, alone, and spatially DISCUSSION
separated at two luminance levels are plotted in Fig. 3. In Land's red-and-white pictures of still-life natural
Both subjects concurred that the three types of arrange- scenes yellows and blues occurred even at relatively high
ment-contiguous, alone, and spatially separated-gave luminances. Those scenes contain much finer gradations
less variation in perceived color for the high-intensity case. of red/white ratio than our stimuli.
The most prominent feature of the high-intensity-low- For the red-and-white design with contiguous suares
intensity shift was the transition in color of square 2a presented at the lower luminance level, both subjects
from pink at high intensity to a pale, warm yellow at low reported a set of hues roughly symmetrical about the
intensity. This square was judged as desaturating almost neutral point-red, blue-green, violet, and yellow,qualita-
completely when it was viewed alone on a black back- tively agreeing with the results of Pearson et al.4 Cur-
ground. Square 3a was described by both observers as rent models seem to provide no obvious explanation for
changing from a red-violet at high intensity to a slightly these effects. Perhaps depressed signal-to-noise ratio in
bluer violet at low intensity. Finally, square la, which the visual system with the high- to low-intensity change
was bluish cyan at high intensity, appeared to have a enhances color induction effects owing to contrast at
higher green content at low intensity for subject JS. borders. At present there are too few data regarding
The yellow appearance of square 2a deserves further color appearance at low luminances to develop a more
comment. Experimental conditions for yellow appear- definite model.
ance seem to be particularly critical, especially for some Our main finding was that a particular balance of
observers. In informal demonstrations approximately a red and white light produced a more yellow appearance
third of observers report no yellow. However, at least one with lowered intensity in simple patterns. Boynton and

(a) y y y

G R G R G R

Js Js Js
Alone Together Exploded
B B B
(b) V V V

G R G R G R

LA LA LA
Alone Together Exploded
B B B
Fig. 3. Mean color judgments for the four patches in the three spatial arrangements, presented in opponent polar coordinates. Angular
coordinate: Mean hue, based on the ratio of the reported pair of hues (no third hue was ever reported for any patch). Radial coordi-
nate: Mean ratio of chromatic to achromatic content of the color appearance. Open symbols: High mean luminance condition (0.0
ND). Solid symbols: Low mean luminance condition (2.5 ND). Circles: Upper left patch (see Fig. 2). Triangles: Upper right patch.
Squares: Lower left patch. Diamonds: Lower right patch. (a) Subject JS, (b) subject LA.
J. Skon and L. Arend Vol. 9, No. 1/January 1992/J. Opt. Soc. Am. A 33

Gordon's 9 experiment on the Bezold-BrUcke effect showed "Visual processes in color photography," J. Opt. Soc. Am. 33,
579-614 (1943).
that monochromatic red stimuli shift away from yellow in 2. E. H. Land, "Color vision and the natural image," Part I,
appearance as luminance is decreased. In our experi- Proc. Nat. Acad. Sci. USA 45, 115-129 (1959).
ment, on the other hand, a patch that appeared red at high 3. J. Skon, "The measurement of perceived colors in red-and-
luminance shifted toward yellow at low luminance. white projections," Ph.D. dissertation (Harvard University,
Cambridge, Mass., 1981).
The role of rod responses in our experiment is not clear. 4. D. E. Pearson, C. B. Rubinstein, and G. J. Spivack, "Compari-
Our subjects fixated the 1.4° patches, but the surrounding son of perceived color in two-primary computer-generated ar-
patches fell outside the rod-free region, and our conditions tificial images with predictions based on the Helson-Judd
did not prevent eye movements from carrying the target formulation," J. Opt. Soc. Am. 59, 644-658 (1969).
patch into those regions. McCann and Benton' 0 and 5. L. Wheeler, "Color-naming responses to red light of varying
luminance and purity," J. Opt. Soc. Am. 52, 1058-1066
McCann" have reported that a variety of color sensations (1962).
are achieved under conditions in which only the rods and 6. L. Wheeler, "Color-matching responses to red light of vary-
the long-wave cones are responding. Trezona'2" 3 has pre- ing luminance and purity in complex and simple image,"
sented evidence, most convincingly from an experiment J. Opt. Soc. Am. 53, 978-993 (1963).
7. J. J. McCann and S. Auerbach, Polaroid Corporation, Cam-
by Stiles,' 4 that, with continual decrease in luminance, the bridge, Mass. 02139 (personal communications).
weighting of the rod component of visual response changes 8. The chromaticity coordinates of Land's two primaries are
the output of the blue-yellow opponent system. Less rod
saturation apparently permits stronger yellow and blue Xred = 0.693, Xwhite = 0.417,
response at lower luminances. Yred= 0.307, Ywhite= 0.397.
In conclusion, at a mesopic luminance we observed a Those of the monitor are
greater variety of hues in our red-white mixtures than at
high luminances. Xred'= 0.615, Xwhite'= 0.313,
Yred'= 0.329, ywhite'= 0.329.
ACKNOWLEDGMENTS 9. R. Boynton and J. Gordon, "Bezold-BrUcke hue shift mea-
sured by color-naming technique," J. Opt. Soc.Am. 55, 78-86
The authors wish to thank Charles Stromeyer and Rhea (1965).
Eskew for their editorial comments. This research was 10. J. J. McCann and J. L. Benton, "Interaction of the long-wave
supported by U.S. Air Force Office of Scientific Research cones and the rods to produce color sensations," J. Opt. Soc.
grants AFOSR 86-0128 and AFOSR 89-0377 to L. Arend. Am. 59, 103-107 (1969).
11. J. J.-McCann, "Rod-cone interactions: different color sen-
L. Arend is also with the Department of Ophthalmology, sations from identical stimuli," Science 176, 1255-1257
Harvard Medical School, Boston, Massachusetts 02114. (1972).
12. P. W Trezona, "Rod participation in the 'blue' mechanism and
its effect on color matching," Vision Res. 10, 317-332 (1970).
REFERENCES AND NOTES 13. P. W Trezona, "Is the retina trichromatic or tetrachromatic?"
Farbe 34, 211-219 (1987).
1. W F. Fox and W H. Hickey, "Improvements in kinemato- 14. W. S. Stiles, "The average colour-matching functions for a
graphic apparatus," British Patent No. 636, issued July 1914. large matching field," Nat. Phys. Lab. (UK) Symp. 1, 211-247
Evans did related work on colored shadows: R. M. Evans, (1958).