Color

Sjoerd Stuit

Color vision has been selected for because it offers many evolutionary advantages. For primates the optical assessment of
edibility of fruits is of key importance.

hurlert 2004
 Today
❖ Why do we have color vision?

Color vision has been selected for

because it offers many
evolutionary advantages.

For primates the optical

assessment of edibility of fruits is
of key importance.

2
 Today

❖ Where does color ‘come’ from?

❖ Lights
❖ Cones
❖ Color-opponancy
❖ Color associations

3
 Color in light

❖ Early theories on vision:

❖ Eye send out light
❖ About 1000 years ago someone figured out light enters
the eye instead of the other way around
❖ but what is light?

it wasn’t untill 1000 year ago that someone figured out that ligth entered the eye for vision instead of emitting it. But they were
still not sure about what light was, how it behaved. Straigth lines or wave? Or what it even was.
 Color in light
Waves? Whatever

❖ +/- 1666: light behaves as a wave

and color is a combination of light
(which is red) and darkness
(which is blue)
❖ Newton: Let’s break some light

Newton didnt believe the wave theory because of how light is reflected. He also didn't believe that color was a combination of
light (red) and darkness (near blue) as Hooke proposed.
 Color in light

❖ Using a prism, newton

‘broke’ white light
❖ Also turned it back to
white

Newton ‘corrupted’ the light with a prism and than turned it back to white light using a second prism. This should not work if
light was a wave.
Newton concluded that light is composed of coloured particles which combine to appear white. He introduced the term 'colour
spectrum' and, although the spectrum appears continuous, with no distinct boundaries between the colours, he chose to divide
it into seven; red, orange, yellow, green, blue, indigo, and violet.
 Color in light

❖ Newton argued light consisted of particles

❖ Introduced the term ‘color spectrum’
❖ Separated 7 colors
❖ why seven?

Newton concluded that light is composed of coloured particles which combine to appear white. He introduced the term 'colour
spectrum' and, although the spectrum appears continuous, with no distinct boundaries between the colours, he chose to divide
it into seven; red, orange, yellow, green, blue, indigo, and violet.
 Color in light

Prism 1: white light Prism 2: individual

consists of light of colors (here: yellow)
many different cannot be further
colors subdivided
8

Newton wasn’t done. If color was indeed a particle, it should be able to be broken down even further. New experiment
Conclusion: yellow is ‘real’
 dafaq?
Color in light

Slits let through red Prism 2: red and

and green green make
yellow again

Mystery: if yellow=red+green, then why can

yellow not be split into red+green?
9
 Color in light
Under a monochromatic illuminant all
color discrimination vanishes.

Color discrimination depends upon

spectral composition differences.
10

for one…color appears to be in light…and of course this is true. but note that we have wavelengths and reflectance and we have
perception. One is not the other.
So Newton wasn’t completely wrong. We know now about wavelength. But this is surely not the whole story. Newton mixed up
color perception with light physics.
Color in light?

11
Color in light?

spectral colors

non-spectral
colors are very
common

12
 Color in …
❖ Goethe: Color is in the brain

“That I am the only person in this

century who has the right insight into the
difficult science of colors, that is what I
am rather proud of, and that is what
gives me the feeling that I have
outstripped many.”

13
 Color in cones
❖ Helmholtz, Young, Palmer:
Trichromacy

❖ Palmer argued that there were a

number of different retinal
"particles" at every point on the
retina to respond to light. Young
suggested that there might be
three such particles only, a view
later validated by science.
Helmholtz supported Young’s views.

14

here the discussion shift to the biological system that responds to light
Philosophical discussion, before the age of experimentation/microscopes
Various facts suggest it ‘must’ be like this.
Such philosophical proposals are often wrong
There was an extensive philosophy of vision, almost all invalidated by the arrival of experimentation
 Color in cones
❖ Cones response to different
wave lengths and thus decode
color…but is this color
perception?

15
 this one again!

16

Some more color adaptation…pay attention to the relation between adaptor color and aftereffect color
 Opponent colors
❖ Herings opponent colors

❖ Hering considered red, green, blue, and yellow to be elementary

color sensations; unique in that he could describe all other
experiences of color using these four, and they themselves could
not be described by any other colors.

❖ He proposed that red and green were opponent colors in that the
sensation of red and the sensation of green never appeared to co-
exist in the same color. Likewise, blue and yellow were mutually
exclusive (as were light and dark).
17
 Opponent colors

❖ Color space based on

Hering
❖ Helps describe color
effects

18
 Opponent colors

❖ Compressed
version

R/G B/Y (light / dark)

19

Black and white do not rely on the blue channel. Red+green is a good measure for ‘lightness’
Presence of blue with no other wavelength is a good blue measure.
Red+blue (lightness) WITHOUT blue means yellow.
Yellow is difficult because we don’t have a receptor, but circuitry deals with that
 Opponent colors

❖ Opponent color
schemes in the ganglion
cells of the retina

20

Note that this is about comparing wavelength quantities

 Light, cones & opponency
❖ Metamerism

21

Metamers are stimuli that are physically different yet result in the same percept.

We can trick the system by presenting specific color bands, instead of a range
These appear as the same color, and that is how we get intermediate colors from primary colors in a display
 Light, cones & opponency
Ratio the same,
so both look yellow
R/G B/Y (light / dark)

22
 dafaq?
Light, cones & opponency

Slits let through red Prism 2: red and

and green green make
yellow again

Mystery: if yellow=red+green, then why can

yellow not be split into red+green?
23

Gangion cells can’t note the difference between pure yellow (in wavelength) and the combination of red and green
newton was confused/tricked with metamerism
 Color constancy

24

The first three colours come from the banana under different lights. The next three come from the apple, the last three from the
bowl
Seeing these side-by-side emphasizes the differences, but we are normally very good an determining the actual colour of the
object regardless of lighting conditions
 Color constancy

25

The first three colours come from the banana under different lights. The next three come from the apple, the last three from the
bowl
Seeing these side-by-side emphasizes the differences, but we are normally very good an determining the actual colour of the
object regardless of lighting conditions
 Color constancy

26

The first three colours come from the banana under different lights. The next three come from the apple, the last three from the
bowl
Seeing these side-by-side emphasizes the differences, but we are normally very good an determining the actual colour of the
object regardless of lighting conditions
Color constancy

27
 Color constancy
To determine the color an object will be, we
need to know the spectrum of the lighting
and the spectrum of the object under even,
Average Daylight spectrum
white light
x The color of the object under the current
lighting is the product of these two spectra

But this works backward too: if we know the

Banana Reflectance Spectrum
lighting spectrum and the observer
= reflectance spectrum we see, we can
calculate the reflectance spectrum of the
object under white light: color constancy

Spectrum reflected by banana in average daylight

28

The role/problem of illumination

 Color constancy
❖ To achieve color constancy, an object's
spectral reflectance is the constant color
parameter that needs to be evaluated.

❖ Any information that helps characterize

an object's spectral reflectance is a cue to
color constancy.

29

We perceive colors mostly as they are, regardless of lighting

We can determine actual reflectance spectrum by knowing reflected color and lighting conditions
We can also perceive lighting
Both represented together
It’s actually hard to perceive the actual color reaching the retina due to contextual interactions
 Cues to color constancy

❖ Local color contrast

30

look at the square with the dot

Cues to color constancy

31
 Cues to color constancy

Relative M-cone contribution Relative M-cone contribution

Green disc appears less saturated 32
Green disc appears more saturated

less M in center compared to surround…less green

More M in center compared to surround…more green
 Color in Context

❖ Lateral interaction based on color

33

Each horizontal cell type has characteristic short distance and or long distance connections to particular photoreceptor types. No
need to learn all of these, remember that inhibition interactions are both spatial and spectral
 Cues to color constancy

❖ Local color contrast

❖ Lateral interaction boots color contrasts
❖ Color adaptation
❖ Selective reduction in responses based on previous
exposer

34
35
36
 Cues to color constancy

❖ Local color contrast

❖ Color adaptation
❖ Changes relative
activity

37

note that we need color opponency to be able to explain this

 Cues to color constancy

❖ Local color contrast

❖ Color adaptation
❖ Changes relative
activity
R/G B/Y (light / dark)

38

note that we need color opponency to be able to explain this

 Cues to color constancy

❖ Local color contrast

❖ Color adaptation
❖ Global color contrast
❖ global spectral changes generally represent changes

in lighting conditions; take this into account to solve

local color content

39
 Cues to color constancy

❖ Looks like different

lighting conditions
❖ Use this spectrum to
make inferences about
the objects colors

40
 Cues to color constancy

❖ Looks like different

lighting conditions
❖ Use this spectrum to
make inferences about
the objects colors

41
Cues to color constancy

For the Non-Believers...

42
Cues to color constancy

For the Non-Believers...

43
Cues to color constancy

For the Non-Believers...

44
 Cues to color constancy
❖ Local Color Contrast

❖ Color Adaptation

❖ Global Color Contrast

❖ Luminance Highlights ?
❖ Mutual Reflections ?
❖ Range of Reflected ?
Spectrum
45

Local colour contrast and adaptation happen in the retina, LGN, and V1
Global colour contrast requires larger receptive fields, so mostly occurs in V4.
By V4, colour constant responses are well established
Some other cues we use depend on higher-level processing, probably in later cortical areas, and are poorly understood.
It seems likely these cue are processed later and then send feedback signals to earlier areas to influence their activity.
 Cues to color constancy: incomplete
Kraft and Brainard
(1999) demonstrated that
even with a full set of
cues, color constancy
was not perfect (83%
accuracy), and that the
most powerful cue to
constancy was local
color contrast.

46

http://www.pnas.org/content/96/1/307.full.pdf
ms act on a local subjects adjusted the fruits in the direction away from their typical
hereas others act setting. We then normalized the memory color index by dividing
ver larger image through the distance between the control settings and the fruit’s typical
is that they are Cues to color constancy
na.
jects also affects a 90 b
s have a typical

(L + M) – S (% cone contrast)
or. This knowl- 45
the process of ❖ Hansen et al., 2006:
minant. Previous ❖ Make this banana grey!
0
be distinct from
more saturated
2005). However, –45
nd, because the
he colors of the –90
47

–10 –5 0 5 10
e effects were of
L – M (% cone contrast)

uit objects on a Figure 1 The chromatic adjustment method. (a) The distribution of
bjects could be chromaticities in the original photograph of the banana (yellow) in the
bananas of any isoluminant plane of a color space spanned by an L – M axis and an (L + M)
hat does a blue – S axis. Subjects could adjust the color of the stimulus in two dimensions.
This was achieved by rotating and scaling the whole distribution of
which we started chromaticities (for example, toward magenta). The black cross indicates the
ubject to inter- mean of the distribution and how it changes when it is rotated by 901 and its
s in color space amplitude scaled to 70%. (b) Stimuli corresponding to the two chromatic
hod satisfies the distributions shown in a.

sen, Otto-Behaghel-Str. 10F, 35394 Giessen, Germany. Correspondence should be addressed to K.R.G.

ne 15 October 2006; doi:10.1038/nn1794

NOVEMBER 2006 1367

 B R I E F C O M M U N I C AT I O N S
Cues to color constancy
short-wavelength energy in the morning a
a b energy of longer wavelengths during the co
90 18
objects are yellowish rather than bluish; th

(L + M) – S (% cone contrast)
45 9
grayish in the morning, as color constancy
even in realistic situations11. The appearanc
contribute toward color constancy under t
0 0
Our results show a high-level cognitive ef
mechanisms. Our knowledge of the world
–45 –9
much in the way that Bayesian modelin
settings1. Just as we apply prior knowle
–90
–10 –5 0 5 10
–18
–2 –1 0 1 2 comes from above12, we apply prior kno
L – M (% cone contrast) L – M (% cone contrast) color of fruit objects. This knowledge is use
and global information about the scene to d
Figure 2 Color settings. (a) The achromatic
48 (gray) and typical (yellow) This allows the visual system to function
settings for the banana stimulus. Each data point indicates the position of
reduced conditions when only single ob
the mean chromaticity averaged across all pixels within the banana shape, as
adjusted by each observer, averaged over five trials. The large symbols depict unknown illuminant, as in our exper
the average settings for all 14 subjects. The black symbol indicates the that determine color appearance act throu
average setting for the control stimuli. (b) The achromatic settings for the of the visual system, from the retina to vis
seven fruits studied (colored symbols) and for the two control conditions memory. Modulatory feedback is a can
where a uniform disc (indicated by the black circle) or a noise patch underlying the integration of bottom-up
(indicated by the black square) was adjusted. Each data point is the average
down expectations13,14.
for 14 observers.
Note: Supplementary information is available on the N

setting (Supplementary Methods). The indices varied between 4– 13% ACKNOWLEDGMENTS

for the different fruit objects (mean ¼ 8.23%, t13 ¼ 6.195, P o 0.001). We thank L. Kaim and W. Kirchner for technical a
This amounts to an effect that is approximately three to five times L. Maloney for helpful discussions. This work was
above the threshold of discrimination. The effect was highly reliable Forschungsgemeinschaft (Ge 879/5-2) and by a Gi
fellowship to S.W.
even for individual observers. All 14 subjects had positive memory
color indices, indicating that they all adjusted the fruits in the direction COMPETING INTERESTS STATEMENT
opposite to their typical settings. The authors declare that they have no competing fi
In previous studies on memory color, subjects had to either view an
Published online at http://www.nature.com/naturene
object or recall its color from memory, and then pick a matching color Reprints and permissions information is available
from a set of colors. In our study, observers could actually manipulate reprintsandpermissions/
“Our results show a high-level cognitive effect on low-level perceptual mechanisms. Our knowledge of the world affects our
the color of the object while viewing it, without the need for a cognitive
perception, very much in the wayretrieval
that Bayesian
of memorymodeling hasadjustment
during the predictedprocedure.
in many settings1. Just as 1.
Rather, memory weKnill,
apply prior knowledge that
D.C. & Richards, W. Perception as Bayesian
the light usually comes from above12,
seemedwe toapply
have aprior
directknowledge about
top-down effect thatthe natural color
continuously of fruit objects. This Massachusetts,
modulated Cambridge, knowledge1996). is used
2. Smithson, H.E. Philos. Trans. R. Soc. Lond.
the incoming sensory data and changed basic color sensations.
together with other local and global information about the scene to determine color appearance. “ (2005).
Our results show that natural fruit objects tend to be perceived in 3. Duncker, K. Am. J. Psychol. 52, 255–265 (1939).
their typical color. Objects with a typical color might seem to constitute 4. Bruner, J.S., Postman, L. & Rodrigues, J. Am. J. Ps
5. Fisher, S.C., Hull, C. & Holtz, P. Am. J. Psychol. 69
only a small percentage of the objects surrounding us, but one has to 6. Bolles, R.C., Hulicka, I.M. & Hanly, B. Can. J. Psyc
keep in mind that our visual system evolved in a world that was not 7. Bartleson, C.J. J. Opt. Soc. Am. 50, 73–77 (1960)
filled with man-made objects of arbitrary color. Another question 8. Delk, J.L. & Fillenbaum, S. Am. J. Psychol. 78, 290
9. Jin, E.W. & Shevell, S.K. J. Opt. Soc. Am. A 13, 19
posed by our results is whether they are applicable to the type of 10. Taylor, A.H. & Kerr, G.P. J. Opt. Soc. Am. 31, 3–8 (
lighting conditions occurring naturally. Notably, natural illumination 11. Kraft, J.M. & Brainard, D.H. Proc. Natl. Acad. Sci.
12. Ramachandran, V.S. Nature 331, 163–166 (1988
during the course of the day varies mostly along the blue- 13. Grossberg, S. Psychol. Rev. 87, 1–51 (1980).
10
yellow dimension . Natural daylight contains a high proportion of 14. Hupé, J.M. et al. Nature 394, 784–787 (1998).

1368 VOLUME 9 [ NUMBER 11 [ NOVEMBER 200

 Color preference

49

Everyone prefers blue… so why the pink toys for girls?

 horizontal bar indicates only components (see Supple

Mean proportio
approximately the tested
0.4 data). These account for
hues and is not an accu-
rate reproduction of the of the population varianc
spectrum. sub-populations by sex a

Color preference
0.2 nationality, the fixed com
Current Biology Vol 17 No 16 account for between 64%
R624
0 females) and 72% (British
1 2 3 4 5 6 of the variance.
Hue angle (radians) Each individual hue pre
Figure 1. Mean hue prefer- S–(L+M) (‘blue–yellow’) and L–M curve is thereby reduced
A 1.0 B 1.0
ence curves.
UK Female (n=92)
(‘red–green’) neuronal mechanisms
China Female (n=18) two physiologically mean
(A) British subjects. (B) Chi-
UK Male (n=79) which encode
China Male (n=19)colors. We therefore
weights. While the ‘blue–y
nese subjects (±s.e.m.).
0.8 Hue0.8values are obtained decomposed the hue preference
contrast component acco

Mean proportion preferred

Mean proportion preferred
from CIE-LUV coordinates, curves in terms of fixed basis
the greatest variance acro
using the background color functions which explicitly match
0.6 as 0.6
population (44.5% S−(L+M
reference white. The the two cone-opponent contrast
horizontal bar indicates only L−M), the ‘red–green’ con
components (see Supplemental
approximately the tested
data). These account for 70% component accounts for
0.4 hues0.4and is not an accu-
of the population variance. For greater variance within th
rate reproduction of the population alone (41% L−
spectrum. sub-populations by sex and
0.2 0.2 nationality, the fixed components S–(L+M)). Only the ‘red–g
account for between 64% (Chinese weights show a consisten
females) and 72% (British females) difference across all popu
0
1 2 3
1
4
2
5
3
6 of4 the variance.
5 6
0 On average, all males give
Hue angle (radians) Each individual hue preference
Hue angle (radians) negative weight to the L−
B 1.0 curve is thereby
Current reduced
Biology to whereas all females weigh
China Female (n=18) two physiologically meaningful slightly positively (sex diff
❖ China Male (n=19) Hurlbert & Ling (2007) weights. While the ‘blue–yellow’ p < 0.00001). That is, fem
0.8 contrast
were British Caucasian (79 male). A component
The average accounts
female forpreference prefer colors with ‘reddish
Mean proportion preferred

sub-population (37) were mainland the greatestrises variance

steeply across
to athesustained peak against the background, w
0.6
50Han Chinese (19 male), the majority population (44.5%
in the S−(L+M);
reddish-purple 25.5% region, and males prefer the opposite
having left China for the UK within L−M), the ‘red–green’
falls rapidly contrast
in the greenish-yellow average, all subjects give
the past year (range 0.5–3 years). component region, accounts for the the male
whereas weight to the S–(L+M) con
0.4 greater variance within the male towards
Observers were tested in three preference is shifted component (‘bluish’ contr
different experiments, each of population alone
blue-green L−M;
(41% and less28%pronounced. with British females weigh
0.2 which included the same pair-wise S–(L+M)). Only the ‘red–green’
Although there is a significant significantly higher than B
comparisons for a standard weights show maina consistent
effect of hue sex for both sexes males (p < 0.00001) (Figur
difference across all populations.
group of eight colors (varying independently (p < 0.000001 Although male reaction
0 On average, all males give large
1 2 3 4 5 6 hue; saturation 0.5; lightness 80). males; p < 0.000001 females) are significantly faster on
Hue angle (radians) negative weight to the L−M axis,
A subgroup of 90 subjects (28 and together (p < 0.000001), the (1.26 seconds) than fema
Current Biology whereas all females weight it
British females, 25 British males, variance in preference over all hues seconds) (p < 0.00001), b
slightly positively (sex difference
and the Chinese sub-population), p < 0.00001). isThat
significantly
is, females greater for females females and males respo
were British Caucasian (79 male). A The average performed the standard
female preference prefer colorsversus males (pcontrast
with ‘reddish’ < 0.00001). to ‘bluish’ versus ‘yellowi
sub-population (37) were mainland rises steeplyexperiment
to a sustainedtwice,peak
with a two-week Individual female
against the background, whereas preference contrasts (reaction times
Han Chinese (19 male), the majority interval. Here
in the reddish-purple we report
region, and results for prefercurves
males are alsoOn
the opposite. more stable over negatively with S-cone-co
having left China for the UK within falls rapidlythe
in standard
the greenish- color group common
yellow average, all time, subjectsfor the givesubgroup
positive of 90 increments of the preferre
the past year (range 0.5–3 years). to all experiments.
region, whereas the male weight to the subjects
S–(L+M)tested contrasttwice (p < 0.002). female r = −0.1061, p < 0.
Observers were tested in three preference isWe obtained
shifted towardshue preference component (‘bluish’ The predictability
contrasts), of the male r = −0.0348, p < 0.01
different experiments, each of blue-greencurves
and less by pronounced.
plotting for each ofwith the British individual
females weighting hue preference
it curves Thus, while both males
which included the same pair-wise eight is
Although there standard hues the proportion
a significant significantlyprompted
higher than usBritish
to seek more concise females share a natural p
comparisons for a standard main effect ofoftrials
hue on for which it was preferred
both sexes descriptors.
males (p < 0.00001) (FigurePrincipal
2). component for ‘bluish’ contrasts, the
group of eight colors (varying independently(Figure We found that hue Although analysis
(p <1).0.000001 male reaction revealstimes
that three factors preference for ‘reddish’ c
hue; saturation 0.5; lightness 80). males; p <preference
0.000001 females) curves do not vary are significantly faster on average
alone explain 79% of the variance further shifts her peak tow
A subgroup of 90 subjects (28 and together (p < 0.000001),
significantly the
for different (1.26 seconds)
lightness thanthe
across female
entire(1.33
population. the reddish region of the
British females, 25 British males, variance inand preference
saturation over all hues
levels seconds)
(Figure S1 in (pThe< 0.00001),
first twoboth factors strongly circle: girls’ preference fo
and the Chinese sub-population), is significantly greater for females
the Supplemental females andresemble
data). The mean males respond the cone-fasteropponent may have evolved on top
performed the standard versus males hue(p < 0.00001).curves for males
preference to ‘bluish’ versuscontrast ‘yellowish’
components of the natural, universal preferen
experiment twice, with a two-week Individual and
female preference
females differ significantly. contrasts (reaction times correlate
stimuli — the fundamental for blue. We speculate th
interval. Here we report results for curves are also more stable over negatively with S-cone-contrast
the standard color group common time, for the subgroup of 90 increments of the preferred hue;
Bias towards reddish colors for girls.
to all experiments. subjects tested twice (p < 0.002). female r = −0.1061, p < 0.00001;
We obtained hue preference The predictability of the male r = −0.0348, p < 0.01).
From hunter/gatherer times. curves by plotting for each of
eight standard hues the proportion
the individual hue preference
prompted us to seek more concise
curves Thus, while both males and
females share a natural preference
The degree of preference for redof correlated with femininitydescriptors.
trials on which it was preferred
(Figure 1). We found that hue
index Principal component
analysis reveals that three factors
for ‘bluish’ contrasts, the female
preference for ‘reddish’ contrasts
further shifts her peak towards
Chine: Red good luck, so ppl likepreference
red more…so curves do not vary
why bring evolution
significantly for different lightness
alone explain 79% of the variance
into this??
across the entire population. the reddish region of the hue
and saturation levels (Figure S1 in The first two factors strongly circle: girls’ preference for pink
the Supplemental data). The mean resemble the cone-opponent may have evolved on top of a
hue preference curves for males contrast components of the natural, universal preference
and females differ significantly. stimuli — the fundamental for blue. We speculate that this
 Color preference
❖ Palmer & Schloss (2010):
❖ Color preference task
❖ Object-association task
❖ Object-valence task
❖ Color-object matching task
❖ All with different participants!

51

“In this article we articulate an ecological valence theory in which color preferences arise from people’s average affective
responses to color-associated objects. An empirical test provides strong support for this theory: People like colors strongly
associated with objects they like (e.g., blues with clear skies and clean water) and dislike colors strongly associ- ated with objects
they dislike (e.g., browns with feces and rotten food). Relative to alternative theories, the ecological valence theory both fits the
data better (even with fewer free parameters) and provides a more plausible, comprehensive causal explanation of color
preferences.”
 Color preference

Fig. 1. (A) The present sample of 32 chromatic colors as defined by eight hues,52
consisting of four approximately unique hues (Red, Green, Yellow, Blue) and
their approximate angle bisectors (Orange, cHartreuse, Cyan, Purple), at four “cuts” (saturation-lightness levels) in color-space: saturated (s, Upper Left), light
(l, Upper Right), dark (d, Lower Right), and muted (m, Lower Left). (B) The projections of these 32 colors onto an isoluminant plane in CIELAB color-space. (C)
Color preferences averaged over all 48 participants. Error bars show SEM. (D) WAVEs for the 32 chromatic colors estimated using data from independent
participants performing three different tasks.

adaptations across generations resulting in hardwired neural Results and Discussion

mechanisms), the EVT extends the range of potentially adaptive Each of 48 participants rated each of the 32 chromatic colors of
mechanisms to include individual organisms learning color pref- the Berkeley Color Project (BCP) (Fig. 1 A and B) in terms of
erences on an ontogenetic time scale. An analogy to taste pref- how much the participant liked the color using a line-mark rating
erences is apt: Taste preferences have both an evolutionary scale that was converted to numbers ranging from −100 to +100
component, because some genetic variations in taste are more with a neutral zero-point. Average preference ratings (Fig. 1C)
adaptive than others, and a learned component resulting from show that the saturated (s), light (l), and muted (m) colors
experiences that arise from eating various flavored foods that have produced approximately parallel functions with a broad peak at
affectively different outcomes (17). The connection of the EVT to blue and a narrow trough at chartreuse. The s colors were pre-
the emotion-based theory of Ou et al. (15, 16) is that the envi- ferred to the l and m colors [F(1,47) = 9.20, P < 0.01], which did
ronmental feedback required for a learning-based heuristic to not differ from each other (F < 1). Although the pattern of hue
work for color preferences is provided by the emotional outcomes preferences across s, m, and l cuts† did not differ [F(14, 658) =
of color-relevant experiences during a person’s lifetime. The more 1.66, P > 0.05], it did vary for the dark (d) cut relative to the
enjoyment and positive affect an individual receives from expe- other three [F(7,329) = 17.87, P < 0.001]. In particular, dark
riences with objects of a given color, the more the person will tend orange (brown) and dark yellow (olive) were significantly less
to like that color. preferred than other oranges and yellows [F(1,47) = 11.74,
In this article
Color preferences we test
found forthe EVT
their by determining how well it can
subjects 41.06, P < 0.001, respectively], whereas dark red and dark green
account for average preferences across individuals for a wide were more preferred than other reds and greens [F(1,47) =
gamut of colors. The EVT implies that the average preference for 15.41, 6.37, P < 0.001, 0.05, respectively].
any given color over a representative sample of people should be The central assumption of the EVT is that color preferences,
determined largely by their average affective responses to corre- averaged across people, are determined by the average affective
spondingly colored objects. Accordingly, people should be valence of people’s responses to objects that are strongly associ-
attracted to colors associated with salient objects that generally ated with each color. We tested this claim by measuring the
elicit positive affective reactions (e.g., blues and cyans with posi- weighted affective valence estimate (WAVE) for each of the 32
tively valued clear sky and clean water) and repulsed by colors chromatic BCP colors (Fig. 1D) and correlating the result with the
associated with salient objects that generally elicit negative reac- corresponding average color preferences (Fig. 1C). Calculating
tions (e.g., browns with negatively valued feces and rotting food). the WAVEs of the 32 BCP colors required collecting and analyzing
As reported here, we tested this central prediction of the EVT
and compared its fit with those of three other theories: the cone-
opponent contrast model, a color-appearance theory based on our †
We use the term “cut” (through color space) to refer to the four combinations of
observers’ ratings, and the color-emotion theory. lightness and saturation levels we used (Fig. 1 A and B).

8878 | www.pnas.org/cgi/doi/10.1073/pnas.0906172107 Palmer and Schloss

 Color preference

Fig.1.1. (A)
Fig. (A)The
The present
present sample
sample of 32 chromatic
chromaticcolors
colorsas
asdefined
definedby byeight
eighthues,
hues,consisting
consistingofof
four approximately
four approximately unique hues
unique (Red,
hues Green,
(Red, Yellow,
Green, Blue)Blue)
Yellow, and and
53
theirapproximate
their approximate angle
angle bisectors
bisectors (Orange,
(Orange, cHartreuse,
cHartreuse,Cyan,
Cyan,Purple),
Purple),atatfour
four“cuts”
“cuts”(saturation-lightness
(saturation-lightness levels) in color-space:
levels) saturated
in color-space: (s, Upper
saturated Left),
(s, Upper lightlight
Left),
(l,(l,Upper
UpperRight),
Right), dark
dark (d,
(d, Lower
Lower Right),
Right), and
and muted
muted(m,(m,Lower
LowerLeft).
Left).(B)
(B)The
Theprojections
projectionsofof
these
these3232
colors onto
colors an an
onto isoluminant plane
isoluminant in CIELAB
plane color-space.
in CIELAB (C) (C)
color-space.
Colorpreferences
Color preferences averaged
averaged over
over all
all 48
48 participants.
participants.Error
Errorbars
barsshow
showSEM. SEM.(D)
(D)WAVEs
WAVEs forfor
the
the3232
chromatic
chromatic colors estimated
colors using
estimated datadata
using from independent
from independent
participants performing
participants performing three
three different
different tasks.
tasks.

adaptations across
adaptations across generations
generations resulting
resulting in in hardwired
hardwired neuralneural Results Resultsand andDiscussion
Discussion
mechanisms), the EVT extends the range of potentially adaptive Each
mechanisms), the EVT extends the range of potentially adaptive Each of 48participants
of 48 participants rated
ratedeach
eachof of
thethe
32 32chromatic
chromatic colors of of
colors
mechanisms to
mechanisms to include
include individual
individual organisms
organisms learning
learningcolorcolorpref-pref- the Berkeley Color Project (BCP) (Fig. 1 A and B) in terms of of
erences on an ontogenetic time scale. An analogy to taste pref-
the Berkeley Color Project (BCP) (Fig. 1 A and B) in terms
erences on an ontogenetic time scale. An analogy to taste pref- how much the participant liked the color using a line-mark rating
erences is apt: Taste preferences have both an evolutionary how much the participant liked the color using a line-mark rating
erences is apt: Taste preferences have both an evolutionary scale that was converted to numbers ranging from −100 to +100
component, because some genetic variations in taste are more scale that was converted to numbers ranging from −100 to +100
component, because some genetic variations in taste are more with a neutral zero-point. Average preference ratings (Fig. 1C)
adaptive than others, and a learned component resulting from with a neutral zero-point. Average preference ratings (Fig. 1C)
adaptive than others, and a learned component resulting from show that the saturated (s), light (l), and muted (m) colors
show that the saturated (s), light (l), and muted (m) colors
experiences that arise from eating various flavored foods that have produced approximately parallel functions with a broad peak at
experiences that arise from eating (17).various flavored foods
of thethatEVThave produced approximately parallel functions
affectively different
affectively
outcomes
different outcomes
The connection
(17).etThe connection of the
to blue and a narrow trough at chartreuse. The swith a broad
colors peak at
were pre-
the emotion-based theory of Ou al. (15, 16) is that theEVTenvi-to ferred blue to andthea lnarrow
and m colors [F(1,47) = 9.20, P < 0.01], which didpre-
trough at chartreuse. The s colors were
the emotion-based
feedbacktheory of Oufor eta al. (15, 16) is that the envi- ferred
ronmental required learning-based heuristic to not differtofrom
the leach
and m colors
other (F <[F(1,47) = 9.20,the
1). Although P< 0.01], of
pattern which
hue did
ronmental feedback required for a learning-based
by the emotionalheuristic
outcomesto preferences not differ from
work for color preferences is provided acrosseachs, m,other
and l(F < †1).
cuts did Although
not differ the pattern
[F(14, 658) of=hue
work for color preferences is provided by the emotional outcomes preferences across
it dids, vary
m, and for lthe
cutsdarkdid(d)
notcut differ [F(14,
to 658)
the =
†
of color-relevant experiences during a person’s lifetime. The more 1.66, P > 0.05], relative
ofenjoyment
color-relevant experiences
and positive affectduring a person’s
an individual lifetime.
receives Theexpe-
from more 1.66, P > 0.05], it did vary for the dark (d) cut relative to the
other three [F(7,329) = 17.87, P < 0.001]. In particular, dark
enjoyment
riences withand positive
objects affectcolor,
of a given an individual
the more the receives
personfrom expe- orange
will tend other (brown)
three [F(7,329)
and dark= yellow 17.87, (olive)
P < 0.001]. In particular,
were significantly lessdark
riences
to like with objects of a given color, the more the person will tend preferred
that color. orange (brown)than other and oranges
dark yellow and (olive)
yellows were[F(1,47) significantly
= 11.74,less
to like thatarticle
In this color.we test the EVT by determining how well it can 41.06, preferred than respectively],
P < 0.001, other oranges and dark
whereas yellowsred [F(1,47)
and dark = green11.74,
In this on
modelaccount
based article we test
for different
average the EVT by
ppl.
preferences determining
across individuals howforwell it can were
a wide 41.06, P < preferred
more 0.001, respectively],
than otherwhereas reds and dark red and
greens dark green
[F(1,47) =
account
gamut offor average
colors. preferences
The EVT implies thatacross individuals
the average for a wide
preference for were 6.37,
15.41, moreP preferred
< 0.001, 0.05, thanrespectively].
other reds and greens [F(1,47) =
Different
gamut groups,
of colors.
any given different
color overThe aEVT models
implies thatsample
representative the average
of peoplepreference
should be for 15.41, The central
6.37, Passumption
< 0.001, 0.05, of the EVT is that color preferences,
respectively].
any
The model given
determined color
(right) overby
shows
largely awhat
representative
their the
average sample responses
preferences
affective of people
(left) should
should
to look
corre- be like based
averaged on object
The central
across associations.
assumption
people, of the EVT
are determined Fit by
the data
isthe
that verypreferences,
color
average well!
affective
spondingly largely
determined coloredbyobjects.
their averageAccordingly,
affectivepeople
responses should
to corre-be valence
averaged of people’s responses
across people, areto objects thatbyare
determined thestrongly
averageassoci-
affective
attracted to colored
spondingly colors associated
objects. with salient objects
Accordingly, peoplethat should
generallybe ated valencewith ofeach color.responses
people’s We tested to this claim
objects thatbyaremeasuring
strongly theassoci-
elicit positive
attracted affective
to colors reactions
associated (e.g.,
with bluesobjects
salient and cyansthatwith posi-
generally weighted affective valenceWe estimate
ated with each color. tested(WAVE)
this claim for byeach of the 32 the
measuring
tivelypositive
elicit valued affective
clear sky reactions
and clean(e.g., water) andand
blues repulsed
cyans bywithcolors
posi- chromatic
weightedBCP colorsvalence
affective (Fig. 1D) and correlating
estimate (WAVE) theforresult
eachwith
of the
the 32
associated
tively valued with salient
clear sky objects
and clean thatwater)
generallyandelicit negative
repulsed by reac-
colors corresponding average
chromatic BCP colorscolor preferences
(Fig. 1D) (Fig. 1C).
and correlating theCalculating
result with the
tions (e.g.,with
associated browns withobjects
salient negativelythatvalued feceselicit
generally and rotting
negative food).
reac- the WAVEs of theaverage 32 BCP colors required collecting and analyzing
As reported here, with
we tested this central corresponding color preferences (Fig. 1C). Calculating
tions (e.g., browns negatively valued prediction of the EVT
feces and rotting food). the WAVEs of the 32 BCP colors required collecting and analyzing
and compared its fit with those of three other
As reported here, we tested this central prediction of the EVT theories: the cone-
opponent
and compared contrast model,
its fit a color-appearance
with those of three othertheory based
theories: theoncone-
our †
We use the term “cut” (through color space) to refer to the four combinations of
observers’contrast
opponent ratings, model,
and theacolor-emotion
color-appearance theory.
theory based on our lightness and saturation levels we used (Fig. 1 A and B).
†
We use the term “cut” (through color space) to refer to the four combinations of
observers’ ratings, and the color-emotion theory. lightness and saturation levels we used (Fig. 1 A and B).
8878 | www.pnas.org/cgi/doi/10.1073/pnas.0906172107 Palmer and Schloss

8878 | www.pnas.org/cgi/doi/10.1073/pnas.0906172107 Palmer and Schloss

 Recap
❖ Different wavelengths
❖ Different cones
❖ Different ganglion cells with color opponent processes
❖ Perception depends heavily on response ratios, lighting
conditions and temporal and spatial context
❖ Learned association play a role in both color perception
and color preference

54