Sensation & Perception 6th Edition 5

Chapter
Courtesy of Lutz Baar
Lutz Baar, French Abstract, 2014

The Perception of Color
Questions to Contemplate
Think about the following questions as you read this chapter.
By the chapter’s end, you should be able to answer and discuss them.
� What is color for?
� Does everyone see the same colors?
� Suppose you have an orange on your plate for lunch. It looks orange. If you
send a photo of that orange, the photo will look orange too. The physical bases
of those two orange experiences are very different. How does that work?
� Suppose that orange looks a bit green. Will it taste different?
I
f you were forced to give up one dimension of your visual experience, color vi-
sion would be a sad but sensible choice. Vision works quite well without color.
Television thrived in black and white, for example. You would be much more
impaired if you were forced to give up orientation perception or motion percep-
tion (see Chapter 8). That said, an ability to use color has evolved multiple times in
multiple ways in the animal kingdom, and it seems very central to your perceptual
life. You have probably been asked about your favorite color. When was the last
time anyone asked you to name your favorite orientation? The goal of this chapter
is to explain how we see color and to speculate about why.
5.1 Basic Principles of Color Perception

Although color itself is not a physical property of things in the world, it is relat-
ed to a physical property. As discussed in Chapter 2, humans see a narrow range
of the electromagnetic spectrum between the wavelengths of about 400 and 700
nanometers (nm). Recall that 1 nm is 10-9 or 0.000000001 meters. The apparent
color of a piece of the visible world is correlated with the wavelengths of the light
rays reaching the eye from that piece of the world.
Most of the light that we see is reflected light. Typical light sources, like the sun
or a lightbulb, emit a broad spectrum of wavelengths that hit surfaces in the world
around us. Some wavelengths are absorbed by the surfaces they hit. The more light
a surface absorbs, the darker it appears. Other wavelengths are reflected, and some
of that reflected light reaches the eyes. The color of a surface depends on the mix of
wavelengths that reach the eye from the surface (and, as we will see later, from the
other surfaces and lights that are present in the scene). In the case of red roses, more
of the longer-wavelength light (>600 nm) is reflected into the eyes of the observer.
Even though (almost) all humans would declare this collection of wavelengths to
appear red, it would be a big mistake to think of specific wavelengths of light as
being specific colors. As Steven Shevell (2003) puts it, “There is no red in a 700 nm
light, just as there is no pain in the hooves of a kicking horse.” Like pain, color is the
result of the interaction of a physical stimulus with a particular nervous system.
132 Chapter 5
S-cone A cone that is preferentially Three Steps to Color Perception

sensitive to short wavelengths, colloqui- Several problems must be solved in order to go from physical wavelengths to the
ally (but not entirely accurately) known
as a “blue cone.”
perception of color. We will organize our discussion around three steps (Stockman
and Brainard, 2010):
M-cone A cone that is preferentially
sensitive to middle wavelengths, col- 1. Detection. Wavelengths must be detected. For this we need photoreceptors
loquially (but not entirely accurately)
known as a “green cone.” to convert light into signals in the nervous system.
L-cone A cone that is preferentially 2. Discrimination. We must be able to tell the difference between one wave-
sensitive to long wavelengths, colloqui- length (or mixture of wavelengths) and another. For this, we need neurons
ally (but not entirely accurately) known that compare inputs from different kinds of photoreceptors.
as a “red cone.”
3. Appearance. We want to assign perceived colors to lights and surfaces
spectral sensitivity The sensitivity of a
cell or a device to different wavelengths in the world. Moreover, we want those perceived colors to go with the
on the electromagnetic spectrum. object (that rose looks red) and not to change dramatically as the viewing
photopic Referring to light intensities conditions change (that rose should remain red in sun and shadow, for
that are bright enough to stimulate the example). For this, we need some very clever processing that we do not
cone receptors and bright enough to fully understand yet.
“saturate” the rod receptors, that is,
drive them to their maximum responses.
scotopic Referring to light intensities 5.2 Step 1: Color Detection
that are bright enough to stimulate the
rod receptors but too dim to stimulate Detection was largely covered in Chapter 2. To briefly review, we have three types
the cone receptors. of cone photoreceptors. These cones differ in the photopigment they carry, and
as a result, they differ in their sensitivity to light of different wavelengths. FIGURE
5.1 shows those sensitivities as a function of wavelength. Each cone type is named
for the location of the peak of its sensitivity on the spectrum: The cones that have
a peak at about 420 nm are known as short-wavelength cones, or S-cones. The
medium-wavelength cones, or M-cones, peak at about 535 nm. Long-wavelength
cones, or L-cones, peak at about 565 nm. As you can see from Figure 5.1, there is
a lot of overlap between the sensitivities of different cones. That is, even though
the L-cone is maximally sensitive at about 565 nm, the M-cone can detect that
wavelength as well. Their spectral sensitivities overlap. As mentioned in Chapter
2, S-cones are relatively rare, and they are less sensitive than M- and L-cones. The
combination of sensitivities of the three types of cones gives us our overall ability to
detect wavelengths from about 400 nm to about 700 nm (see Figure 5.1). Remember
also that cones work at photopic (daylight) light levels. We have one type of rod
photoreceptor; it works in scotopic (dimmer) light and has a somewhat different
sensitivity profile, peaking at about 500 nm.
420 498 535 565

100
Normalized absorbance
50
S Rod M L
FIGURE 5.1 Photoreceptor types The retina contains
four types of photoreceptors. These differ in their sensitivity
to the wavelengths of light. Three cone types are maximally 0
sensitive at short (S), medium (M), and long (L) wavelengths.
The single type of rod photoreceptor has its peak sensitivity 400 500 600 700
between those of the S- and M-cones. Wavelength (nm)
The Perception of Color 133
5.3 Step 2: Color Discrimination

We can detect wavelengths between 400 and 700 nm, but how do we distinguish,
for example, between lights of 450, 550, and 625 nm? To see how discrimina-
Receptor response
tion differs from detection, let’s examine the response of a single photorecep-
tor to a single wavelength of light. FIGURE 5.2 shows how one kind of human
photoreceptor responds to light of a specific wavelength while the intensity of
the light is held constant. Because of the properties of the photopigment in the
photoreceptor cell, 400 nm light produces only a small response in each cell
of this type, 500 nm light produces a greater response, and 550 nm light even
more. However, 600 nm light produces less than the maximal response, and 650
nm light produces a minimal response. Light of 625 nm produces a response of 400 500 600 700
moderate strength. Wavelength (nm)
FIGURE 5.2 Responding to one

The Principle of Univariance wavelength A single photoreceptor
So far, so good. We know that different wavelengths of light give rise to different shows different responses to lights of
experiences of color, and the varying responses of this photoreceptor to different different wavelengths but the same
intensity. A 625 nm light of this inten-
wavelengths could provide a basis for color vision. But there is a problem, as il- Wolfe
sity, indicated by the arrow, produces a
lustrated in FIGURE 5.3. Suppose we change the wavelength from 625 nm to 450 Sensation & Perception 6E
response midway between the maximum
OUP/Sinauer Associates
nm. Figure 5.3 shows that an equal amount of 450 nm light will produce the same and minimum responses.
response from this photoreceptor that 625 nm light does. If we were looking at the Wolfe6e_05.02.ai 8.25.2020
output of the photoreceptor, we would have no way of distinguishing between the
two lights. But when we look with a normal human color vision system, the 625
nm light looks orange and the 450 nm light looks bluish.
Actually, the problem is much worse. Remember that Figures 5.2 and 5.3 rep-
resent the photoreceptor’s response rate when all wavelengths are presented at the principle of univariance The fact
same intensity. Under these conditions, as the graphs indicate, light at either 450 that an infinite set of different
or 625 nm produces a response lower than the peak response obtained at about wavelength-intensity combinations
535 nm. If we had a 535 nm light, we could reduce its intensity until it produced can elicit exactly the same response
from a single type of photoreceptor.
exactly the same level of response from our photoreceptor as the 450 or 625 nm One photoreceptor type cannot
light did at the higher intensity. Indeed, we could take a “white light” or any mix of make color discriminations based
wavelengths and, by properly adjusting the intensity, get exactly the same response on wavelength.
out of the photoreceptor.
Thus, when it comes to seeing color, the output of a single photoreceptor is
completely ambiguous. An infinite set of different combinations of wavelength and
intensity can elicit exactly the same response, so the output of a single photore-
ceptor cannot by itself tell us anything about the wavelengths stimulating it. This
Receptor response
constraint is known as the principle of univariance (Rushton, 1972). (See Activity

5.1: The Principle of Univariance.)
Obviously, the human visual system has solved the problem, but not under
all circumstances. Univariance explains the lack of color in dimly lit scenes.
Remember that there is only one type of rod photoreceptor. All rods contain the
same type of photopigment molecule: rhodopsin. Thus, they all have the same
sensitivity to wavelength. As a consequence, although it is possible to tell light from
dark under scotopic conditions, the problem of univariance makes it impossible
to discriminate colors. Our nighttime color blindness is one hint that color is 400 500 600 700
Wavelength (nm)
psychophysical and not physical. The world seen under a bright moon (FIGURE
5.4) has not been physically drained of color. The same mix of wavelengths that FIGURE 5.3 Univariance Lights of
produces color perception during the day remains present on that moonlit night, 450 and 625 nm both elicit the same
response from the photoreceptor whose
but we fail to see colors under dim illuminants like moonlight, because dim light
responses are shown here and in Figure
stimulates only the rods, and the output of that single variety of photoreceptor Wolfe
5.2. This situation illustrates the principle
does not permit color vision. Sensation & Perception 6E
of univariance.
Wolfe6e_05.03.ai 8.25.2020
134 Chapter 5
FIGURE 5.4 Rod vision The moonlit world appears

drained of color because we have only one type of rod
photoreceptor transducing light under these scotopic
conditions. With just one type of photoreceptor, we
cannot make discriminations based on wavelength,
so we cannot see color.
Courtesy of John Bortniak, NOAA

The Trichromatic Solution
We can detect differences between wavelengths or mixtures of wavelengths precisely
because we have more than one kind of cone photoreceptor. FIGURE 5.5 shows
how, with our three cone types, we can tell the difference between lights of differ-
ent wavelengths. Look at the three cones’ responses to the two wavelengths—450
and 625 nm—that produced the same response from the single cone in Figure 5.3.
The two wavelengths of light still produce the same response from that type of
cone, now revealed to be the M-cone. However, these two wavelengths produce
different outputs from the L-cones and S-cones. In fact, as you will see if you try
Activity 5.2: Trichromacy, any wavelength from about 420 to 660 nm produces a
unique set of three responses from the three cone types. This combined signal, a
triplet of numbers for each “pixel” in the visual field, can be used as the basis for
color vision. (We can see from about 400 to 700 nm, but the very long and very
short wavelengths each stimulate only one type of cone.)
In our discussion of the univariance problem, we noted that we can make any
wavelength produce the same response as any other from a single cone type by
adjusting the intensity of the light. That is not a problem in the three-cone world
S M L
S-cone response
to 450 nm is big.
S
L-cone response to
L 625 nm is big.
M-cone response to M-cone response to

450 nm is moderate. M M 625 nm is moderate.
L-cone response to L
450 nm is smaller.
S-cone response to
FIGURE 5.5 Solving univariance The two wave- S
625 nm is absent.
lengths that produce the same response from one type
of cone (M) produce different patterns of responses 400 500 600 700
across the three types of cones (S, M, and L). Wavelength (nm)
Wolfe
of human color vision. A specific light produces a specific set of three responses trichromacy or trichromatic theory
from the three cone types. Suppose that the light produces twice as much M re- of color vision The theory that the
color of any light is defined in our visual
sponse as S response and twice as much S response as L response. If we increase system by the relationships of three
the intensity of the light, the response sizes will change but the relationships will numbers—the outputs of three receptor
not. There will still be twice as much M response as S response and twice as much types now known to be the three cones.
S response as L response, and those relationships will define our response to the Also called the Young-Helmholtz theory.
light and, eventually, the color that we see. (Think what might happen if the lights
were really bright.) The idea that ability to discriminate one light from another
is defined in our visual system by the relationships among three numbers is the
heart of trichromacy, or more elaborately, the trichromatic theory of color vision.
Metamers
The examples presented thus far involve the responses of the visual system to
single wavelengths. However, you may have noticed that we have been referring
to “wavelengths or mixtures of wavelengths.” That’s because we are not typically
exposed to single wavelengths. Almost every light and every surface that we see
is emitting or reflecting a wide range of wavelengths. A laser pointer would emit
a very narrow range of wavelengths, but a more normal situation is shown in FIG-
URE 5.6. It shows the relative amounts of light reflected from groups of Granny
Smith apples exposed to different amounts of sunlight during growth. How can
we discriminate the injured from the uninjured apples? When we’re studying color
vision, this real-world concern gets reduced to a different question: How do our
cones respond to combinations of wavelengths of light?
To answer this question, consider what happens if we mix just two wavelengths.
For the sake of this example, we will oversimplify by ignoring the S-cones and re-
drawing the M- and L-cones to make the numbers simpler. Imagine that we shine a
wavelength that looks red and a wavelength that looks green onto a white piece of
paper so that a mixture of both is reflected back to the eyes (FIGURE 5.7A). Suppose
that the light that looks green produces 80 units of activity in the M-cones and
40 in the L-cones (remember, we are ignoring the S-cones for now). In addition,
Class 0 Class 1 Class 2 Class 3 Class 4

(Shaded) (Exposed) (Mild) (Moderate) (Severe)
60
50
40
Class 4
Reflectance (%)
Class 3
30
Class 1
Class 2
20
FIGURE 5.6 Real world reflectance
Class 0 Class 0 functions Objects in the real world
10 Class 1 reflect light across the spectrum in different
Class 2 amounts. This graph plots the reflectances
0 Class 3 of sun injury development in Granny Smith
Class 4 apples. You can see how the appearance
changes as the amount of reflected
500 550 600 650 long-wavelength light changes. (After C. A.
Wavelength (nm) Torres et al. 2016. Sci Hortic 209: 165–172.)
136 Chapter 5
(A) What happens if you add this light suppose that the light that looks red produces 40 units of activity in the M-cones
that looks red to one that looks green? and 80 in the L-cones. If we assume that we can add the cone responses together,
then summing the “red” and “green” lights produces a response of 120 units in each
M-cone L-cone cone. The absolute value is not important, because it could change if the intensity
of the light changed. What is important is that these two lights, mixed together,
Response
produce a mixture that excites the L- and M-cones equally.

L-cone M-cone The key point is that the rest of the nervous system knows only what the cones
tell it. If the mixture of lights that look red and green produces the same cone
output as the single wavelength of light that looks yellow (FIGURE 5.7B), then
the mixture and the single wavelength must look identical. Mixtures of different
G R
wavelengths that look identical are called metamers. The single wavelength that
Wavelength
produces equal M- and L-cone activity will look yellow, and the correct mixture of
longer- and shorter-wavelength lights will also look yellow. Two quick warnings:
(B) This light that looks yellow produces 1. Mixing wavelengths does not change the physical wavelengths. If we mix 500
equal L- and M-cone responses.
and 600 nm lights, the physical stimulus contains wavelengths of 500 and 600
nm. It does not contain the average (550 nm). It does not contain the sum
M-cone L-cone
(1100 nm) (which we would not be able to see anyway). Color mixture is a
mental event, not a change in the physics of light.
Response
2. For a mixture of a red and green to look perfectly yellow, we would have to
have just the right red and just the right green. Other mixes might look a bit
reddish or a bit greenish.
This example generalizes to any mixture of lights. All the light reaching the retina
from one patch in the visual field will be converted into three numbers by the three
Y cone types. If those numbers are sufficiently different from the numbers in another
Wavelength
patch, you will be able to discriminate those patches. If not, those patches will be
FIGURE 5.7 Metamers In (A), the metamers: they will look identical, even if the wavelengths are physically different.
long-wavelength light that looks red and
the shorter-wavelength light that looks The History of Trichromatic Theory
green mix together to produce the same
response from the cones as does the From what you’ve read so far in this book, you would be forgiven for supposing that
Wolfe
medium-wavelength light that looks yel- clever anatomists and physiologists identified the three cone types and built the
Sensation two sets 6E
low in (B).&IfPerception of lights produce the trichromatic theory of color vision from there. Indeed, there have been beautiful
OUP/Sinauer
same responses, Associates
they are metamers and
experiments of this sort: For instance, Schnapf, Kraft, and Baylor (1987) managed
must look identical, so the red plus the
Wolfe6e_05.07.ai
green will look yellow. 2.18.2020 to record the activity of single photoreceptors. Nathans, Thomas, and Hogness
(1986) found the genes that code for the different photopigments. David Williams
and his students even developed a method for photographing and identifying dif-
ferent cone types in the living human eye (see Figure 2.18).
Such research has cemented our understanding of the physical basis of trichro-
macy, but the basic theory was established by psychophysical experimentation. The
theorizing started with Isaac Newton’s great discovery that a prism would break up
sunlight into the spectrum of hues, and a second prism would put the spectrum
back together into light that looked white. In 1666, Newton understood that “the
rays to speak properly are not coloured” (from Opticks, published originally in
1704). Newton knew that color is a mental event.
The three-dimensional nature of the experience of color was worked out in
the nineteenth century by Thomas Young (1773–1829) and subsequently by Her-
mann von Helmholtz (1821–1894). In their honor, trichromatic theory is often
called the Young-Helmholtz theory. James Clerk Maxwell (1831–1879) developed
a color-matching technique that was central to Helmholtz’s work on this topic.
metamers Different mixtures of (Somehow Maxwell missed having his name attached, or we would have the
wavelengths that look identical, or
more generally, any pair of stimuli
Young-Maxwell-Helmholtz theory.) Maxwell’s technique is illustrated in FIGURE
that are perceived as identical in 5.8. The observer in a modern version of Maxwell’s experiments would try to use
spite of physical differences. different amounts of “primary” colored lights (e.g., the lights looking red, green,
Bluish Blue Green Red FIGURE 5.8 Maxwell’s color-matching experiment

A color is presented on the left. On the right, the observer
adjusts a mixture of the three lights to match the color
on the left.
and blue on the right side of the figure) to exactly match another reference color additive color mixture A mixture of
(e.g., the light looking cyan, or bluish, on the left side). lights. If light A and light B are both
reflected from a surface to the eye, in
The central observation from these experiments was that only three mixing the perception of color the effects of
lights
Wolfe are needed to match any reference light. Two primaries are not enough, and
those two lights add together.
Sensation & Perception 6E
four are more than are needed. Long before physiology could prove it, these results
OUP/Sinauer Associates subtractive color mixture A mixture
led Young and Helmholtz to deduce that three different color mechanisms must of pigments. If pigments A and B mix,
Wolfe6e_05.08.ai
limit the human experience8.18.2020
of color. some of the light shining on the surface
will be subtracted by A, and some by
A Brief Digression into Lights, Filters, and B. Only the remainder will contribute to
the perception of color.
Finger Paints
The ubiquity of video screens in the twenty-first century
may make color mixing and metamers reasonably intui-
tive. If you’ve never done so, find a magnifying glass and 1. Take “white” light
that contains a broad “White” light
take a very close look at a yellow patch on your computer mixture of wavelengths.
screen. You’ll find that the patch is actually composed
of thousands of intermixed red and green dots; though,
with new monitors, it is getting harder and harder to
see the tiny pixels. The “red + green = yellow” formula
is an example of additive color mixture because we are 2. Pass it through a filter
taking one wavelength or set of wavelengths and adding that absorbs shorter
wavelengths. The result “Yellow” filter
it to another. will look yellowish.
For most of us, color mixture begins in kindergarten
or earlier, with paints. In that world, red plus green
doesn’t make yellow; that mixture typically looks brown.
A finger paint, or any other pigment, looks a particular
color because it absorbs some wavelengths, subtract- 3. Pass that through a
ing them from the broadband (“white”) light falling on bluish filter that absorbs
“Blue” filter
all but a middle range
a surface covered with the pigment. When a toddler of wavelengths.
smears together red and green, almost all wavelengths
are absorbed by one pigment or the other, so we perceive
the subtractive color mixture as a dark color like brown.
Actually, finger paint mixtures are rather complicated,
with some particles of different pigments sitting next to 4. The wavelengths that
each other and effectively adding their different reflected make it through both “Green” remainder
filters will be a mix
lights to the result that you see. Other particles occlude that looks greenish.
each other, and others are engaged in still other complex
interactions. Colored filters, like those you might put over FIGURE 5.9 Subtractive color mixture In this example, “white”—
stage lights, are a cleaner example of subtractive color broadband—light is passed through two filters. The first one absorbs
mixture. FIGURE 5.9 shows how a subtractive mixture (“subtracts”) shorter wavelengths, transmitting a mix of wavelengths that
looks yellow. The second absorbs longer wavelengths and the shortest
of yellow and blue filters would subtract wavelengths,
wavelengths, transmitting a mix that looks blue. The wavelengths that
leaving only wavelengths in a middle range, which appear can pass through both filters without being subtracted are a middle
green. An additive mixture of lights that look blue and range of wavelengths that appear green.
Wolfe
Wolfe6e_05.09.ai 2.18.2020
138 Chapter 5
Blue White Yellow yellow will look white (if you have exactly the right blue and yellow) (FIGURE 5.10)
because that combination produces a mix of wavelengths that stimulate the three
cone types roughly equally. (See Activity 5.3: Color Mixing.)
Additive color mixture with paints is possible. Georges Seurat (1859–1891)
and other Postimpressionist artists of the late nineteenth century experimented
with Pointillism, a style of painting that involved creating many hues by placing
small spots of just a few colors in different textures (FIGURE 5.11). Viewing the
painting up close, as illustrated in the figure, we can see each individual dot of
color. Like the red, green, and blue phosphor dots on a computer monitor, the
FIGURE 5.10 Additive color mixture dots in the painting combine additively to produce a wide range of colors. Thus,
If we shine a light that looks blue and from a distance the domes look gray even if, up close, they are composed of spots
a light that looks yellow on the same of blue, green, and white.
patch of paper, the wavelengths will add,
producing an additive color mixture.
Remember that the light that looks yellow
From Retina to Brain: Repackaging the Information
Wolfe
is equivalent
Sensation to a mix6E
& Perception of a long wavelength The cones in the retina are the neural substrate for detection of lights. What is the
and a mediumAssociates
OUP/Sinauer wavelength, so blue plus neural basis for discriminating between lights with different wavelength composi-
yellow results in a mix of short, medium,
Wolfe6e_05.10.ai
and long wavelengths. The mixture 2.18.2020
looks
tion? To tell the difference between different lights, the nervous system will look at
white (or gray, if it is not the brightest differences in the activities of the three cone types. This work begins in the retina.
patch in view). We could send separate L, M, and S signals to the brain, but that approach would
be less useful than one might think. For example, the L- and M-cones have very
similar sensitivities (see Figure 5.1), so most of the time they are in close agree-
ment: L says, “Lots of light coming from location X.” “Yes, lots of light coming from
location X,” M agrees.
Computing differences between cone responses turns out to be a much more
useful way to transmit information to the brain. The nervous system computes two
differences: L – M and (L + M) – S. Why these two? Comparisons across species
suggest that the comparison between S-cones and an LM-cone happened first,
Paul Signac, Grand Canal Venice, 1905. Oil on canvas
FIGURE 5.11 Pointillism Instead of

the usual illustration with a picture by
Seurat, here is a painting by Paul Signac
(1863–1935) using additive color mixture in
the manner of a modern computer moni-
tor. Thus, from a distance the domes look
grayish, even if, up close, they are com-
posed of spots of blue, green, and white.
perhaps 500 million years ago (Mollon, 1989). Then, about 40 million years ago, the lateral geniculate nucleus (LGN)
LM-cone split into very similar L- and M-cones, and the difference between those A structure in the thalamus, part of the
midbrain, that receives input from the
cone types turned out to be useful. We don’t know exactly why the L-M comparison
retinal ganglion cells and has input and
became important in color vision, but there are theories. For example, blushing output connections to the visual cortex.
and turning pale are useful signals to observe, and our specific photopigments may cone-opponent cell A cell type—
have evolved to help us see those signals (Changizi, Zhang, and Shimojo, 2006) by found in the retina, lateral geniculate
making it possible to discriminate different amounts of blood in skin. That is not nucleus, and visual cortex—that, in
the only possibility. The L – M difference may also be very useful if you want to effect, subtracts one type of cone input
from another.
tell the difference between fruit and leaves, different shades of green in the foliage,
or the ripeness of a berry (Regan et al., 2001). koniocellular Referring to cells in the
koniocellular layer of the lateral genicu-
In addition to L – M, we could create L – S and M – S signals. However, because late nucleus of the thalamus. Konio
L and M are so similar, a single comparison between S and (L + M) can capture from the Greek for “dust” refers to the
almost the same information that would be found in (L – S) and (M – S) signals. appearance of the cells.
Finally, combining L and M signals is a pretty good measure of the intensity of the parvocellular Referring to cells in the
light (S-cones make a rather small contribution to our perception of brightness). parvocellular layers of the lateral genicu-
late nucleus of the thalamus. Parvo from
Thus, on theoretical grounds, it might be wise to convert the three cone signals into
the Greek for “small” refers to the size
three new signals—L – M, (L + M) – S, and L + M (Buchsbaum and Gottschalk, of the cells.
1983; Zaidi, 1997)—and the visual system does something reasonably close to this.
Cone-Opponent Cells in the Retina and LGN

The earliest work on the combination of cone signals was done with fish (Svae-
tichin and Macnichol, 1959). By the 1960s, Russell De Valois and others had begun
to show that these sorts of signals actually exist in the lateral geniculate nucleus
(LGN) of macaque monkeys (De Valois, Abramov, and Jacobs, 1966). As described
in Chapter 3, many ganglion cells in the retina and the LGN of the thalamus are
maximally stimulated by spots of light. These cells have receptive fields with a
characteristic center-surround organization. For example, some cells are excited
when a light turns on in the central part of their receptive fields and inhibited when
a light turns on in the surround (see Figure 2.20).
A similar antagonistic relationship characterizes color. Some of these retinal
and LGN ganglion cells are excited by the L-cone onset in their center and inhib-
ited by M-cone onsets in their surround. These L – M cells are one type of cone-
opponent cell, so named because different sources of chromatic information
are pitted against each other. There are also M – L, (M + L) – S, and S – (M + L)
cells—just the sorts of cells we would like to have to support the repackaging of All red
cone signals, as described in the previous section. The cells that were excited by
With permission from AAAS/CC BY-NC 4.0

light onset could be thought of as L + M cells. Thus, we have the three signals that
e1600797. DOI: 10.1126/sciadv.1600797.

From R. Sabesan et al. 2016. Sci Adv 2:
we wanted on theoretical grounds. The actual physiology is quite complicated. As
Almost
mentioned in Chapters 2 and 3, for example, the S-cone signals go through the all green
koniocellular layers in the LGN, while the M- and L-cone opponent signals are
mostly found in the parvocellular layers (Xiao, 2014). Mostly
Suppose you could stimulate a single cone. What color would you see? Any spot white
of light you could put up on a screen would cover many photoreceptors. However,
recall from Figure 2.12 that Roorda and Williams (1999) figured out how to use
some very sophisticated optics to allow them to see and classify individual cones
FIGURE 5.12 The retinal mosaic
in the retina of a human observer. Now Roorda and his colleagues (Sabesan et al., A piece of the retina with the L-, M-, and
2016) have managed to focus tiny spots of light onto individual cones and ask those S-cones colored red, green, and blue.
observers what they see (FIGURE 5.12). Often, as you might think, L-cones produce On top of many cones are white rings
red responses and M-cones produce green regardless of the wavelength of stimulus of varying degrees of completion. Each
Wolfe
shows the responses of that cone to tiny
(remember univariance, discussed earlier). However, interestingly, much of the Sensation 6Ewhite ring, for exam-
spots &ofPerception
light. A full
time, the spots look white. Amazingly, when the same cone is tested in sessions OUP/Sinauer Associates
ple, means the observer always said the
many days apart, the pattern of responses can stay quite stable. Why do we have all spot looked white.
Wolfe6e_05.12.ai 7.30.2020
140 Chapter 5
these “white” cones? Cones could be contributing to many circuits. For example,
an L-cone could be part of an L – M circuit that responds to color and an L + M
circuit that responds to brightness. Maybe the L + M brightness response is just
stronger, so the spot looks white even though that cone might also contribute to the
apparent color of a more normal patch of light. Alternatively, it could be that the
response looks white because some cones just don’t contribute to color sensation.
This might help explain why spatial resolution (acuity, contrast sensitivity; see
Chapters 2 and 3) is quite bad if you use equiluminant stimuli—stimuli that vary
in color but not in luminance (Mullen, 1985). It may be that many cones just don’t
contribute to perception of such stimuli.
5.4 Step 3: Color Appearance

Returning to the conscious perception of the visual world, we have seen that the
three cones detect a range of wavelengths. Rods make a small, important con-
tribution to color vision, but only in fairly dim light (Stabell and Stabell, 2002).
Interestingly, the three cone types and the rods are all active in a middle range of
light intensities, called the mesopic range, but the visual system never developed
“tetrachromacy,” color perception based on four photoreceptor types (Webster,
2017) (but we will return to this topic in Section 5.5). The retina and LGN contain
cells that have repackaged the three cone signals into cone-opponent difference
signals that constrain our ability to see differences between regions. Now is the
time to think about what colors will be perceived.
Three Numbers, Many Colors

To begin, it is useful to have a system for talking about all the colors we can see.
Because we have exactly three different types of cone photoreceptors, the light
reaching any part of the retina will be translated into three responses, one for
each local population of cones. After that translation, the rest of the nervous sys-
tem cannot glean anything more about the physical wavelengths of the light. If
the light rays reflecting off two isolated surfaces produce the same set of cone re-
sponses, the two surfaces must and will appear to be exactly the same color. They
will be metamers, even if their physical characteristics are quite different. Thus, it
is possible to produce bloodred color on the page without mimicking the physical
properties of blood.
Working with just three numbers might not sound very promising, but it has
been estimated that, with this system, we can discriminate the surfaces of more
than 2 million different colors (Pointer and Attridge, 1998; Linhares, Pinto, and
Nascimento, 2008). A lot of these colors are lightness variations of what we would
colloquially consider the same color: bloodred in dim light, bloodred in brighter
light, and so on. But even if we ignore lightness, we can still distinguish about
26,000 colors (Foster, 2011). Unless you’re in paint sales, you don’t know this many
distinctive color names, but you could tell two different colors apart even if they
equiluminant Referring to stimuli were both named “pea green” or “sky blue.”
that vary in color but not in luminance. We need some way to talk about all those colors in an orderly way. We can
mesopic Referring to the middle describe each of the colors in the spectrum with a single number—for example,
range of light intensities. the light’s wavelength. Going beyond the spectrum, we have a three-dimensional
color space The three-dimensional color space analogous to a three-dimensional physical space. Of course, it doesn’t
space, established because color
perception is based on the outputs
make much sense to talk about height, length, and width in color space, but the
of three cone types, that describes space that contains the set of all perceptible colors has three dimensions based on
the set of all colors. the three numbers that come from the outputs of the three cone types.
● Sensation & Perception in Everyday Life

Picking Colors
Representing the three-dimensional color space be- blue elements in each pixel, and using RGB values is
comes a practical problem when you are using your a good way to specify the signal to send to each of
computer.
NOTE: thisManygoes
applications
in S&P let in you pick theLife
Everyday colors
Box those elements. Thus, the green shown in the figure
of fonts, objects, and so on, and the app developers is what you get from 62 units (out of 255) delivered by
needed to devise an interface that would make that the red elements, 180 from the green, and 60 from
practical. One solution has been simply to present a the blue. These RGB sliders define a three-dimension-
fixed set of options (e.g., the crayon-box color-picking al space that can be represented by the cube shown
tool in many applications). However, if the developers in Figure 5.13C. Notice that white is in the upper cor-
want to provide access to all the possible colors, then ner closest to you on the cube. Black is hidden in the
they can present one or more color pickers that rep- back corner, farthest from you.
resent a three-dimensional color space. For example, In Figure 5.13B, we see the same shade of green de-
look at FIGURE 5.13. It shows two different ways of fined by a different set of three numbers. This time, the
using three “sliders” to define colors (here, borrowed variables are hue (H), saturation (S), and brightness (B).
from PowerPoint). In Figure 5.13A are three sliders for Part of the resulting HSB color space is shown in Fig-
red, green, and blue. These RGB sliders are useful ure 5.13D. These variables are very different concepts
because computer monitors have red, green, and from red, green, and blue, but notice that again the
system is three-dimensional.
(A) (C) Look at the circular, top surface
Red of the HSB space. Hue is the
Colors
chromatic aspect of a light. It
changes as you move around
the circumference of the circle.
Each point on the spectrum
RGB Sliders
defines a different hue. As you
can see from the HSB sliders,
Red
62 the whole family of colors with a
Green hue of 120 (out of 360 degrees
180 around the circle) will look like
Green Blue
Blue some version of green. The sat-
60 uration dimension corresponds
to the amount of hue present
in a light. In the HSB space
(B) (D)
(Figure 5.13D), it is shown as
Hue
the distance from the center of
the circle. At the center, where
0 saturation is 0, the resulting
100 Satu color would appear somewhere
HSB Sliders ratio 100
n on the achromatic axis from
Hue black to white. Here, 67 (out of
120
100) describes a fairly saturat-
Saturation
Br
ed green. Finally, brightness

ig
67 %
ht
describes that black-to-white

ne
Brightness
s s
71 % achromatic axis. The black

bottom of the HSB cone has
0
a brightness of 0 (out of 100).
FIGURE 5.13 Computer color pickers may offer several ways to specify a color The space tapers to a point
in a three-dimensional color space These “sliders” from Microsoft ® PowerPoint ® because when brightness is 0,
are based on an RGB (red, green, blue) set of primaries in (A and C) and are based on
HSB (hue, saturation, brightness) in (B and D). See the text for details. (Continued )
Wolfe
Wolfe6e_05.13.ai 2.18.2020
142 Chapter 5
Sensation & Perception in Everyday Life (continued )

hue and saturation no longer have meaning. The full many applications will also give you CMYK. This is
three-dimensional space, in this case, includes a cone, a four-dimensional system, with sliders for cyan (C),
rising from a black point to a multicolored surface at magenta (M), yellow (Y), and black (K because B might
a brightness of 100 and then climbing to a white peak be thought to stand for blue). You might notice that
(not shown in the figure). Brightness is the perceptual these are the colors of the ink cartridges in your color
consequence of the physical intensity of a light. For printer. Color printing is a subtractive process, and
instance, the physically intense light of the sun looks CMYK is designed for printing. The black slider is a
brighter than the less intense moon. convenience because using black ink is a better way
Thus, when you use your color picker to find just to get to the black point in your color space than mix-
the right color, you are illustrating the fundamen- ing cyan, magenta, and yellow inks. We will leave it
tal nature of trichromacy. There is one interesting as an exercise for the reader to figure out why CMYK
exception. Along with RGB and HSB color pickers, uses C, M, and Y rather than R, G, and B inks.
The Limits of the Rainbow

The spectrum that you can see in the hue slider in Figure 5.13B is like the rainbow/
wavelength spectrum that you can see at the bottom of FIGURE 5.14A , but it is
not identical. There are hues that you can see that do not exist on the wavelength
spectrum. Figure 5.14 illustrates how this comes to be. For example, suppose
we combined a pure 420 nm light with a pure 680 nm light. This combination
would strongly stimulate the L- and S-cones and produce minimal stimulation in
the M-cones. No single wavelength of light could do that, but such a mixture is
visible and must look like something. In fact, such mixtures will produce purplish
magentas that seem to us to lie naturally between red and blue on a color circle
(FIGURE 5.14B). When we wrap the wavelength spectrum into a color circle,
we join the red and blue ends with a set of colors that are called nonspectral
(A) The spectrum of wavelengths (B) The color circle
S M L
600
550
650
500
nm
700
450
400
400 450 500 550 600 650 700

Wavelength (nm)
FIGURE 5.14 Nonspectral colors
If you combine lights that look red and No single wavelength Nonspectral
blue, the result will look purple, but there on the spectrum will hues
is no purple on the spectrum (A). Purples have the hue of this
are nonspectral colors that join the ends mixture of a long and
of the spectrum into a color circle (B). a short wavelength.
hues—hues that can arise only from mixtures of wavelengths. While we are on opponent color theory The theory
the topic, it is worth noting that there are other commonly perceived colors that perception of color is based on
the output of three mechanisms, each
that are not included in the spectrum’s “all the colors of the rainbow.” Brown of them resulting from an opponency
is one such color. There are no brown lights. Brown is seen when a mixture of between two colors: red-green,
wavelengths that would look yellow, greenish yellow, or orange is seen in the blue-yellow, and black-white.
company of other, brighter patches of color. You cannot see an isolated brown unique hue In the context of opponent
light in the dark (Buck, 2015). color theory, any of four colors that can
be described with only a single color
Opponent Colors term: red, yellow, green, blue. Other
colors (e.g., purple or orange) can also
In the nineteenth century, Ewald Hering (1834–1918) described a curious feature be described as compounds (reddish
of color vision: some combinations of colors seem to be perceptually “illegal.” We blue, reddish yellow).
can have a bluish green, a reddish yellow (which we would call “orange”), or a bluish
red (which we would call “purple”), but reddish green and bluish yellow don’t exist.
Red and green are, in some fashion, opposed to each other, as are blue and yellow
(Hering, 1878). Young and Helmholtz described a trichromatic theory with three
basic colors (red, green, and blue); Hering’s opponent color theory had four basic
colors in two opponent pairs: red versus green, and blue versus yellow. A black-
versus-white component formed a third opponent pair.
FIGURE 5.15 illustrates this idea. The center ring shows the hue dimension of
HSB space, wrapped into the color circle as in Figure 5.14B. The outer ring offers
a cartoon of four color mechanisms in two pairs. The inner patch on the center
left looks yellow, in opponent color theory, because it stimulates the yellow pole of
the yellow-blue opponency and does not stimulate either red or green. Move a bit
counterclockwise, and the orange patches add increasing red to the decreasing yellow.
Leo Hurvich and Dorothea Jameson (1957) revived Hering’s ideas and developed
one way to quantify this opponency. The method, called
hue cancellation, is shown in FIGURE 5.16. Let’s start with
the blue-yellow mechanism. For every wavelength, we Green
will find out how much blue or yellow we need to cancel
the yellow or blue in a light of that color. So, in Example
1, we start with a wavelength, say, 430nm, that looks
blue. Those yellow circles indicate that we are adding
increasing amounts of yellow and asking if the mixture
still looks at all blueish. It takes quite a lot of yellow to
cancel that blue, leaving a pale green. This tells us that
this wavelength strongly stimulates the blue part of the
blue-yellow mechanism. The blue in the bluish-green Yellow Blue
of Example 2 can be cancelled with less yellow. For the
orange color in Example 3, we would need to add blue
to cancel the yellow component in orange. A color that
is neither blue nor yellow occurs when the blue-yellow
function crosses zero (see graph in Figure 5.16A). The
color is “unique green”. No one wavelength is “unique
red”. Everything above ~650 nm will look red.
The orange spot in Example 4 has both yellow and
red in it. If we cancel the red with green we are plotting Red
the red-green mechanism (see Figure 5.16B). The spots FIGURE 5.15 Opponent colors Ewald Hering noted that all the
where the function crosses zero are the wavelengths that colors on the “color circle” (the center ring) could be represented by
are “unique blue” and “unique yellow”. We could think of two pairs of opposing colors: blue versus yellow, and red versus green
Wolfe
(shown in&the outer ring).
6E Thus, a color could be a reddish yellow or
gray as one more unique hue (or antihue). It is what you Sensation Perception
a bluish green,
OUP/Sinauer but not a reddish green or a bluish yellow. (From
Associates
get if you cancel both red-green and blue-yellow opponent A. Stockman and D. H. Brainard. 2010. In OSA Handbook of Optics,
processes (Webster, 2017). Wolfe6e_05.15.ai 2.18.2020
3rd ed., M. Bass [Ed.], pp. 11.11–11.104. McGraw-Hill: New York.)
144 Chapter 5
(A)
Example 1. Start with a Example 2. Start with a Example 3. Start with a
blue patch blue-green patch reddish-yellow (orange) patch
+ = Much too little.

Looks blue
+ = Too little.
Looks blue
+ = Much too little.
Looks orange
+ = Too little.
Looks blue
+ = Too little.
Still a bit blue
+ = Too little.
Still orange
+ = + = + =
Just right
Too little. Looks green OK, reddish
Still bluish No blue or yellow No blue or yellow
+ = + = + =
OK, it is a
Too much. Now it is a
grayish green
Looks yellowish bluish-red
+ = + = + =
No blue or yellow
Too much. Far too much. Too much.

Looks yellowish Looks very yellow Looks blue
Point 1 on graph Point 2 on graph Point 3 on graph
Unique green
Values above the

line mean that you 3
are adding blue to
cancel yellowness.
Values below the

line mean that you
are adding yellow 2
to cancel blueness.
1
400 500 600 700

Wavelength (nm)
(B)
Example 4. Start with a Unique hues
reddish-yellow (orange) patch
+ =
Blue Yellow
Much too little.
Looks orange Values above the 4
+ =
line mean that you
are adding green
OK, here there is to cancel redness.
No red or green
+ =
Values below the
Now it is line mean that you
yellowish green are adding red to
+ =
cancel greenness.
Now it is
greener
+ = And now it is
much too green
400 500 600 700
Wavelength (nm)
Point 4 on graph
FIGURE 5.16 Color cancellation method Using color cancellation to measure opponent
processes and unique hues in a blue-yellow (A) and a red-green (B) process. The locations
where the hue cancellations cross the neutral midpoint are the locations of the unique green
(A) and blue and yellow (B) hues—for example, the green hue with no hint of blue or yellow
in it. (Graphs after L. Hurvich and D. Jameson. 1957. Psychol Rev 64: 384–404.)
Wolfe Sensation and Perception 6e:

This opponent-process, color-appearance story sounds rather similar to the L – M

and (L + M) – S story. However, L – M is not the same as red versus green, and (L +
M) – S is not the same as yellow versus blue. If this were the case, an (L + M) – S cell
should be a yellow-blue cell: a cell that would be maximally excited by unique yellow
and maximally inhibited by unique blue. However, that (L + M) – S cell would actually
be stimulated most strongly by a yellowish-greenish hue and least by a purplish hue.
The L – M cells aren’t in quite the right place either. The L-cone end of the axis is
near perceptual red, but the M-cone end is a bluish green (Eskew, 2008). Krauskopf,
Williams, and Heeley (1982) call these endpoints the “cardinal directions” in color
space, but they are not perceptual red, green, yellow, and blue. We need at least three
steps to get to color appearance. These are shown in FIGURE 5.17. First, three cones
detect light in different ranges of wavelengths. Then, opponent processes measure the
differences in activity between cone types. Finally, some further transformations are
needed to create the color opponency described by Hering (Stockman and Brainard,
2010). (See Essay 5.1: More About Opponent Processing in Color Vision.)
(A) Step 1: Detection (cones)
1.0
Relative sensitivity
to this wavelength
0.8
0.6
S M L
0.4
0.2
0.0
400 500 600 700
(B) Step 2: Discrimination

1.5 1.5
cone-opponent mechanisms
Relative responses of
1.0 1.0
L–M (L + M) – S
0.5 0.5
0.0 0.0
–0.5 –0.5
M–L S – (L + M)
–1.0 –1.0
549 nm 486 nm
–1.5 –1.5
400 500 600 700 400 500 600 700
(C) Step 3: Appearance (opponent colors)

FIGURE 5.17 Three steps to color
1.5 perception (A) Detection: light is differ-
color-opponent mechanisms
2 entially absorbed by three photopigments

Relative responses of
1.0
G Y in the cones. (B) Discrimination: differences
0.5 1
are taken between cone types, creating
R R B cone-opponent mechanisms, important for
0.0 0
wavelength discriminations. (C) Appear-
–0.5 –1 ance: further recombination of the signals
creates color-opponent processes that
–1.0
477 nm 580 nm –2 504 nm 510 nm support the color-opponent nature of color
–1.5 appearance. (From A. Stockman and
D. H. Brainard. 2010. In OSA Handbook
400 500 600 700 400 500 600 700 of Optics, 3rd ed., M. Bass [Ed.],
Wavelength (nm) pp. 11.11–11.104. McGraw-Hill: New York.)
146 Chapter 5
Color in the Visual Cortex

We know that transformations that produce perceived color take place in the vi-
sual cortex; it is not clear how the physiology gives rise to perception (Webster,
2017). Many cells in cortex are interested in color but do not seem to linearly add
and subtract inputs from different cone types. There is some evidence that they
add and subtract in a nonlinear manner. For instance, a cell might be responding
to something more like (L2 – M2 + S2), though exactly how this would produce the
colors we see remains a topic for future work (Horwitz and Hass, 2012; Wandell
and Chichilnisky, 2012). For now, we may have to agree with Conway (2014, p. 201):
“There is still no good neural explanation for Hering’s psychologically important
colors … or … the unique hues often associated with them.”
Is there a specific area or are there areas in the brains of monkeys and humans
specialized for color vision? In the 1980s, evidence favored a separate pathway
for color. If you recall the maps of the visual areas in Chapter 4 (Figures 4.2–4.4),
in the 1980’s researchers found “blobs” in V1 where cells did not seem interested
in orientation but seemed very interested in color. The blobs sent output to “thin
stripe” regions in V2 (Livingstone and Hubel, 1988) and from there to V4, an
area that Semir Zeki argued was specialized for color, with cells that responded
not to wavelength but to perceived color (Zeki, 1983a, 1983b). More recently,
functional imaging in monkeys has been used to uncover color hot spots in visual
cortex. These have been named “globs” (really!) (Conway, Moeller, and Tsao,
2007), and cells in these regions also seem more interested in color than do cells
outside. However, although these anatomical pathways are there (Federer et al.,
2009), it has become less clear that we can separate color processing from other
perceptual processes in cortical anatomy (Shapley and Hawken, 2011). Indeed, you
may remember from Chapter 4 that V4 is a popular candidate for a “shape” area.
This doesn’t mean that it is uninterested in color, but merely that it is probably
not only interested in color (Roe et al., 2012).
FURTHER DISCUSSION of the elegantly named “blobs” can be found in

Section 3.6.
Modern imaging studies show some areas of the human visual cortex that
seem particularly interested in color (Grill-Spector and Malach, 2004), but we can’t
record from single cells in humans under most circumstances. Perhaps the best
evidence for specialized brain areas for color in humans comes from certain cases
of achromatopsia, a loss of color vision after brain damage (Zeki, 1990). People
with achromatopsia may be able to find the boundaries between regions of different
colors, but they cannot report what those colors might be. In these individuals,
vision is largely intact, while the experience of color seems specifically impaired.
5.5 Individual Differences in Color Perception

Thus far, with a little caution, we have been talking about color as if we all see
colors the same way, but do we? This is one of those questions that everyone asks
at one point or another, often as a child. Like many “childish” questions, this has
been the topic of much decidedly unchildlike philosophical work. (See Essay 5.2:
The Philosophical Problem of “Inverted Qualia”.)
Language and Color

achromatopsia An inability to perceive
colors that is caused by damage to the Putting aside the finer points of philosophy, suppose you are in a clothing store
central nervous system. and you find a shirt that you like but you want a different color. You might go to
the clerk and ask, “Do you have this in blue?” You simply assume that you and that basic color terms Color words that
clerk agree about the meaning of blue. You can discriminate on the order of millions are single words (like blue, not sky blue),
are used with high frequency, and have
of different colors, but you don’t have a separate word for each of these. There is a meanings that are agreed upon by
vast range of color words, but in looking for that shirt, you would not typically ask speakers of a language.
for “azuline” or “cerulean” (both, varieties of blue). You would use a word like blue cultural relativism In sensation and
that almost every speaker of your language would use quickly and consistently in perception, the idea that basic percep-
naming colors. These color names are the basic color terms of the language. What tual experiences (e.g., color perception)
makes a color term basic? Berlin and Kay (1969) asserted that it must be common may be determined in part by the cul-
tural environment.
(like red and not like beige), not an object or substance name (excluding bronze and
olive), and not a compound word (no blue green or light purple). This classification
is a little subjective (is beige that uncommon?), but Berlin and Kay argued that, in
English, these rules yield a list of 11 terms: red, green, blue, yellow, black, white,
gray, orange, purple, brown, and pink.
Interestingly, the numbers of “basic” color terms differ dramatically across
cultures, down to as few as 2 or 3. At one time it was thought that the differing
numbers of basic color terms in different languages meant that color categorization
was arbitrary. This notion was called cultural relativism, meaning that each group
Everyone
was free to create its own linguistic map of color space. Berlin and Kay’s important
discovery was that the various maps used in different cultures are actually rather Black
similar (Lindsey and Brown, 2006). After surveying many languages, they found White
Red
that the 11 basic color terms in English are about as many as any group possesses.
11 Basic colors
Yellow
Of course, the words themselves differ. Red becomes rouge in French or adom in Green
Hebrew. Moreover, languages do not select randomly among the possible color Blue
terms. If a language has only two basic color terms, speakers of the language divide Brown
Pink
colors into “light” and “dark.” If a language has three color terms (one chromatic Orange
term beyond light and dark), what do you think the third usually is? If you guessed Purple
red, you are correct. Typically, the fourth color term would be yellow, then green Gray
Peach
and blue. This ordering is not absolute, but you won’t find a language with, for
Teal
example, just purple, green, and gray as its basic color terms. Lavender
How do new basic color terms emerge? Berlin and Kay argued that a big color Maroon
term is partitioned into two smaller terms. Levinson (2000) suggested that new Tan
Gold
basic terms tend to emerge at the boundary between two existing color terms, in Turquoise
the area where neither existing term works well. In fact, both processes may be Burgundy
at work. Lindsey and Brown (2014) looked at the use of color words in American Aqua
Violet
English by asking 51 Americans to name the color of each of 330 color patches (like Salmon
paint chips). These observers were told to use a single word. That word had to be Magenta
a word that could be used for anything of that color (you can’t have a “blond” car, Olive
can you?). Lindsey and Brown were looking for the sort of word you might use in Fuchsia
Lime
everyday speech to name the color of a car or a shirt. Those 51 Americans used Periwinkle
122 terms for 330 colors. Everyone used the basic 11, but there was evidence that Lilac
American English might be moving beyond the 11 basics. The color term teal may Mustard
Beige
become basic. It emerges in the no man’s land of colors that are neither blue nor
Brick
green. A term like purple may eventually be partitioned, with a term like lavender Flesh
or lilac taking over some of the purple real estate in color space (FIGURE 5.18). Forest
If our set of basic color terms increased, would that change the way we see color? Chartreuse
Coral
If a language has only two or three basic color terms, do its speakers see colors Rust
Rose
Mauve
FIGURE 5.18 Basic color names When Lindsey and Brown (2014) asked Americans to Navy
name color patches, everyone used the 11 “basic” color names. The other color terms were Chocolate
used by fewer and fewer people, going toward the bottom of the graph. A few of these Wine
other terms, like peach and teal, may be considered basic color names over time. (From Sky
D. T. Lindsey and A. M. Brown. 2014. J Vis 14: 17, 1–25. DOI: https://doi.org/10.1167/14.2.17.) Goldenrod
Wolfe
148 Chapter 5
FIGURE 5.19 Color category boundaries (A) (B)

It is easier to remember which of two colors you
have seen if the choices are categorically differ-
ent. For example, suppose you had to remember
the “blue” patch shown at the top of each part of
the figure. Picking between two “blues,” as in (A),
would be rather hard. The task would be easier
if one of the choices were “blue” and the other
“green,” as in (B), even if the distances in color
space were the same in the two cases. Harder choice Easier choice
? ? ? ?
differently than we do with our 11 basic terms? Eleanor Rosch (Heider, 1972) studied
this question among the Dani of New Guinea, a tribe whose language has only two
basic color terms: mola for light-warm colors and mili for dark-cool colors. Now, it
Wolfe
is hard enough
Sensationto&ask your neighbor
Perception 6E to define the experience of “blue” and then to
ask if that is the same as your experience of “blue.” It is much more difficult to ask
these questions across a great cultural divide. But there are tricks, as the experiment
Wolfe6e_05.19.ai 2.18.2020
illustrated in FIGURE 5.19 demonstrates. Suppose you are shown a bluish color
chip and asked to remember it. Then you are shown two test chips and asked to
pick the color you saw before. Obviously, the less similar the two test colors are,
the easier this task is. But more important, you will do better if the wrong choice
is on the other side of a color categorical boundary. Color boundaries are sharper
than you might think. If you show people a collection of colors and ask, “Which
are blue and which are green?” people do the task without much difficulty. If you
have to remember a color, as in the task shown in Figure 5.19, you are likely to give
it a label like “green” or “blue.” If the next color has the same label, you are more
likely to be confused than if it has a different label.
Rosch found that the Dani’s performance on such tasks reflected the same color
boundaries, even when their language did not recognize the distinction between
the two colors (a Dani might use the term mili for all the colors in Figure 5.19 but
would still do better with the task in Figure 5.19B). This finding leads to the con-
clusion that color perception is not especially influenced by culture and language;
blue and green are seen as categorically different, even if one’s language does not
employ color terms to express this difference. If you want some quite convincing
evidence that color categorization isn’t just a function of language, consider a study
by Caves et al. (2018). They did a version of the experiment illustrated in Figure
5.19 with songbirds as the observers and obtained evidence that they, too, seem
to group colors into categories.
On the other hand, there is some evidence that language and color naming can
influence color discrimination. In the late 1990s, Debi Roberson went up the Sepik
River in New Guinea to study the Berinmo, whose language, like the Dani’s, has a
limited set of basic color terms. Unlike previously studied groups, the Berinmo have
terms that form novel boundaries in color space that we do not have. For example,
their nol/wor distinction lies in the middle of colors we categorize as green. Nol
and wor may roughly distinguish live from dead or dying foliage. When Roberson
asked the Berinmo to do the color memory task, they performed better across their
nol-wor boundary than across the blue-green boundary. English speakers showed
the opposite result (Davidoff, Davies, and Roberson, 1999).
Slowest FIGURE 5.20 A color category

760 experiment Observers see four color
patches and simply locate the one patch
740
that is different from the other three.
Response time (ms)
720 Responses are faster and more accu-

Slower rate when the two colors cross a color
700
boundary—in this case, between red
680
Fastest and brown. (After C. Witzel and K. R.
660 Gegenfurtner. 2016. J Exp Psychol 42:
640 540–570. Published by APA; reprinted
with permission.)
620
600
Red-red Red-brown Brown-brown
Color pairs
Red Red Brown Brown
You might think that this was just about the role of language in memory. Sup-
pose you call a spot “green” when you see it first. Then, when you see one spot that
looks green and another that looks blue, you know which one you have seen. If you
see two different green spots, the word green won’t help. Of course, that would not
Wolfe what the Caves et al. (2018) birds were doing. Moreover, color boundaries
explain
have an influence,
OUP/Sinauer even if the task doesn’t have a memory component. As you can
Associates
see in FIGURE 5.20, Witzel and Gegenfurtner (2016) measured how long it took to
Wolfe6e_05.20.ai
say which one of four patches 2.19.2020
was of a different color. They used four different pairs
of colors selected from red and brown patches. Each patch was separated from the
next patch by two just noticeable differences. (If you don’t remember just noticeable
differences, revisit Section 1.2). They found that people were faster to identify the
different one and more accurate when the patches were separated by the border
between red and brown than when all patches were within the “red” or “brown”
category. So the color names matter, even when you compare two colors side by side.
Genetic Differences in Color Vision

The individual differences described in the previous two sections are small. Un-
der most circumstances, if you declare two lights to be metamerically matched,
those around you will generally agree, even if we can’t make definitive statements
about the qualia (singular quale) that each individual experiences. Qualia is the
term that refers to the conscious experience (in this case of a color). We can talk
about your qualia—your experiences, but only you can actually experience them.
Even with properties we can measure, there will be some variation between indi-
viduals. For example, unique green can vary between observers from at least 495
to 530 nm (Nerger, Volbrecht, and Ayde, 1995). Some of these differences will be
due to factors like age, which turns the lens of the eye yellow (Werner, Peterzell,
and Scheetz, 1990). Still, to a first approximation, your performance on standard
measures of color vision will be the same as others’.
However, there is a significant exception to this universality of color matching.
Some 8% of the male population and 0.5% of the female population have a form of
color vision deficiency commonly known as “color blindness,” in which there is a
malfunction in one or more of the genes coding the three cone photopigments. It’s
a “guy thing” because the genes that code for the M- and L-cone photopigments are
on the X chromosome (Nathans, 1986). Males have only one copy of the X chromo- qualia In philosophy, private conscious
some, so if one is defective, the male in question will have a problem. Females have experiences of sensation or perception.
150 Chapter 5
tetrachromatic Referring to the rare two copies and can have normal color vision even if one copy is abnormal. In fact,
situation (in humans, at least) where the some women can end up with four different cone pigments, and in very rare cases,
color of any light is defined by the rela-
tionships of four numbers—the outputs
that produces tetrachromatic color vision—color vision based on four numbers
of those four receptor types. per patch of light (Jordan et al., 2010). Such individuals may actually see colors that
deuteranope An individual who suf- trichromats cannot see. The S-cone photopigment is coded elsewhere, so everyone
fers from color blindness that is due to has two copies, and therefore S-cone color deficiencies are rare (Alpern, Kitahara,
the absence of M-cones. and Krantz, 1983). (See Essay 5.3: Experiencing Color Blindness.)
protanope An individual who suffers There are several different types of color blindness. One determining factor is the
from color blindness that is due to the type of cone affected. A second factor is the type of defect; either the photopigment
absence of L-cones.
for that cone type is anomalous (different from the norm) or the cone type is missing
tritanope An individual who suffers altogether. Although we call people who are missing one cone type “color-blind,” it is
from color blindness that is due to the
absence of S-cones. a mistake to think that this means they cannot see colors at all. As you will recall, if
you have all three cones with their standard photopigments, you need three primary
color-anomalous A better term for the
commonly used term color-blind. Most colors to make a metameric match with an arbitrary patch of color. If you have two
“color-blind” individuals can still make cone types rather than three, the normally three-dimensional color space becomes
discriminations based on wavelength. a two-dimensional space. The world will still be seen in color, but you will have a
Those discriminations are different from
“flatter” color experience, different from that of people with normal color vision.
the norm—that is, anomalous.
Because M- and L-cone defects are the most common, most color-blind individ-
cone monochromat An individual with
only one cone type. Cone monochro-
uals have difficulty discriminating lights in the middle-to-long-wavelength range.
mats are truly color-blind. For example, consider the wavelengths 560 and 610 nm. Neither of these lights
rod monochromat An individual with activates S-cones very much, and the L-cones fire at about the same rate for both.
no cones of any type. In addition to But most of us can distinguish the lights on the basis of the M-cone outputs they
being truly color-blind, rod monochro- elicit, which will be higher for the 560 nm light than for the 610 nm light (you can
mats are badly visually impaired confirm these assertions by consulting Figure 5.5). English-speaking trichromats
in bright light.
would label the colors of these two lights as “green” and “reddish orange,” respectively.
Now consider a deuteranope, someone who has no M-cones. His photorecep-
tor output to these two lights will be identical. Following our maxim that the rest
of the visual system knows only what the photoreceptors tell it, 560 and 610 nm
lights must and will be classified as the same color by our deuteranopic individual.
A protanope—someone who has no L-cones—will have a different set of color
matches based on the outputs of his two cone types (M and S). And a tritanope—with
no S-cones—will be different again. Genetic factors can also make people color-
anomalous. Color-anomalous individuals typically have three cone photopigments,
but two of them are so similar that these individuals experience the world in much
the same way as individuals with only two cone types.
We actually have some notion of exactly what the world looks like to color-de-
ficient individuals, because there are a few, very rare cases of individuals who are
color-blind in only one eye. They can compare what they see through the color-blind
eye with what they see through the normal eye, enabling us to reconstruct the
appearance of the color-blind world (MacLeod and Lennie, 1976).
True color blindness does occur, but it is very unusual. It is possible to be a cone
monochromat, with only one type of cone in the retina. Cone monochromats (who
also have rods) live in a one-dimensional color space, seeing the world only in shades
of gray. Even more visually impaired are rod monochromats, who are missing cones
altogether. Because the rods work well only in dim light and are generally absent
in the fovea, these individuals not only fail to discriminate colors, but also have
very poor acuity and serious difficulties seeing under normal daylight conditions.
There have been various clever efforts to improve the color vision of individ-
uals with abnormal color vision. For example, if you have L and M cones that are
anomalously similar to each other, the right filter, placed in front of the eyes, can
more effectively separate the two cone types. This will change what you would see
and, to judge by testimonials, some “color-blind” individuals experience a richer
world of color. Unfortunately, these glasses do not give people normal color vision agnosia A failure to recognize
(Gómez-Robledo et al., 2018; Martínez-Domingo et al., 2019). objects in spite of the ability to see
them. Agnosia is typically due to brain
We already mentioned one other very interesting class of color blindness, coming damage.
not from photoreceptor problems but from damage to the visual cortex. Lesions of
anomia An inability to name objects in
specific parts of the visual cortex beyond primary visual cortex can cause achroma- spite of the ability to see and recognize
topsia. An achromatopsic individual sees the world as drained of color, even while them (as shown by usage). Anomia is
showing evidence that wavelength information is processed at earlier stages in the typically due to brain damage.
visual pathway. Brain lesions can also produce various forms of color agnosia or anomia synesthesia The perceptual experi-
(Oxbury, Oxbury, and Humphrey, 1969). In agnosia, the patient can see something ence (e.g., a color) elicited by a stimulus
(e.g., a letter) that does not typically
but fails to know what it is. Anomia is an inability to name—in this case, an inability produce that experience, while the stim-
to name colors. A patient with anomia might be able to pick the banana that “looks ulus (e.g., wavelength information) that
right” but unable to report that the banana is or should be yellow. does normally produce the experience
is absent.
Does Everyone See the Same Colors?
The Special Case of Synesthesia
In 1885, the artist Van Gogh was taking piano lessons. He would tell his music
teacher that, for him, the sounds made by different keys on the piano were asso-
ciated with different colors. The music teacher found this strange enough that he
decided that Van Gogh was insane, and that ended the lessons (Voskuil, 2013).
Now, Van Gogh might not have been the most emotionally stable of men. (There
is that time where he cut off an ear.) However, in associating specific colors with
specific sounds or even in seeing colors in response to sounds, he was not insane
or even particularly unusual. He was experiencing synesthesia. Synesthesia is a
phenomenon where one stimulus evokes the experience of a different stimulus. The
prevalence of synesthesia in the general population is estimated at around 4–5%
(Safran and Sanda, 2015). It is hard to know exactly, because experiences like Van
Gogh’s, even today, discourage people from reporting the phenomenon (Safran
and Sanda, 2015). The most common form is so-called grapheme-color synesthe-
sia, experienced by about two-thirds of synesthetes. For these individuals letters,
numbers, or sometimes words have specific and idiosyncratic colors. So, the letter
A might always look red. Next comes colored time units (e.g., Blue Monday) and
then colored music. We are talking about this phenomenon in this chapter, which
is about color, but many other types of synesthesia have been described (sounds
with specific smells, for instance; Zellner, 2013).
We are all synesthetes to some extent. If you were asked about the color of a
slow, sad piece of music, you would probably have an answer that would be dif-
ferent than if we asked you about a fast, happy dance tune. These colors might be
associations that you have between emotions and colors (Curwen, 2018). (“I feel
blue” is a nice multisensory statement.) Most of us would not say that we actually
see those colors. They simply come to mind. A “projective” synesthete would see
those colors out in the world.
How do we know this is a real phenomenon, not just a story that a would-be
synesthete is telling us? Of course, we can’t directly interrogate someone else’s
conscious experience (qualia again), but there are methods. For example, a graph-
eme-color synesthete will give a specific color name to each of the 26 letters of the
alphabet. OK, anyone can do that, but the synesthete will give the same answer
next week—something a nonsynesthete could do only with practice by memorizing
the pairs. Synesthetes can do this because they just need to look at the letter and
report what they always see (Rich, Bradshaw, and Mattingley, 2005).
What is going on here? We don’t really know. One hypothesis is that syn-
esthetes have different, more extensive connections between sensory areas of the
brain (Ramachandran and Hubbard, 2001), but neural evidence for this is rather
152 Chapter 5
color contrast A color perception thin. Others would argue that there is a continuum between those who project
effect in which the color of one synesthetic colors into the world and those who internally associate one stimulus
region induces the opponent color
in a neighboring region.
with another, but mapping a specific color to every letter seems qualitatively different
from associating sadness with blue. We do know that synesthesia is acquired, not
entirely innate, because people have to learn letters (for example) before they can
develop color-letter associations. Perhaps the most important point to be made is
that, whatever Van Gogh’s piano teacher thought, synesthesia is an interesting and
normal multisensory experience. �
5.6 From the Color of Lights to a World of Color

The material presented so far in this chapter about color is quite complex, but
the fact is that it has only addressed the relatively simple problems related to the
detection, discrimination, and appearance of isolated lights. We pointed to the
limits of this approach to color when we noted that there are no “brown” lights. A
surface looks brown when there are other, typically brighter surfaces in the neigh-
borhood. A nonspectral color like brown just scratches the surface of the puzzles
that must be solved if we want to understand color in the world. Think about this.
FIGURE 5.21 shows a black-and-white zebra on grass. Let us assume that you are
looking at this in print (the story would be different on a screen). The black, white,
and green are products of reflection from paper, covered with specific inks. Let
us suppose you are reading this deep in the bowels of the library, by the light of a
yellowish incandescent lightbulb. Now, you take the book outside and continue to
read in sunlight. The number of photons coming from the page to your eye is now
thousands of times greater than it was inside. Outside, a patch of “black” stripe is
now sending much more energy to your photoreceptors than did a patch of “white”
stripe inside. Moreover, the mix of wavelengths reflected from the grass will be
very different if the light source is the sun rather than a dim bulb. Nevertheless,
the grass looks green and the zebra looks black-and-white in both locations. How
(and why) do you do that?
Let’s start by just putting colors next to each other. We live in a world where
regions of one color abut regions of another, and this proximity changes the appear-
ance of colors. FIGURE 5.22 shows color contrast effects. The color of one region
© Karel Miragaya/123RF
FIGURE 5.21 Color constancy This zebra looks like

a black-and-white beast in a green field whether you
are looking at this book under a dim yellow bulb inside,
under the bright sky outside, or for that matter, on your
computer screen. How does that work?
FIGURE 5.22 Color contrast The central square takes on

chromatic attributes that are opposite those of the surround.
For instance, the green central square in column 2, looks
greener on the red background than on the green background.
(From A. Stockman and D. H. Brainard. 2010. In OSA Handbook
of Optics, 3rd ed., M. Bass [Ed.], pp. 11.11–11.104. McGraw-Hill:
New York.)
induces the opponent color in a neighboring region. Thus, the yellow surround color assimilation A color perception
weakens the yellow of a central square and strengthens the blue. FIGURE 5.23 effect in which two colors bleed into
each other, each taking on some of the
shows a vivid version of a color assimilation effect. Those spheres are all the same
chromatic quality of the other.
color underneath the red, green, or blue stripes. In color assimilation effects, two
unrelated color A color that can be
colors bleed into each other, each taking on some of the chromatic quality of the experienced in isolation.
Wolfe
other (Shevell
Sensation and6E
& Perception Kingdom, 2008).
related color A color, such as brown or
Not onlyAssociates
OUP/Sinauer can other colors in the scene alter the color of a target region, but gray, that is seen only in relation to other
scenes can contain colors that cannot be experienced in isolation. Though it may colors. For example, a “gray” patch in
Wolfe6e_05.22.ai 2.18.2020 complete darkness appears white.
be hard to believe unless you try it, you cannot sit in complete darkness and see a
gray light, all by itself. That light will look white if seen as an isolated or unrelated
color. To be seen as gray, it must be seen in relation to other patches of color.
Thus, it is a related color. Brown is another related color. We can distinguish a few
thousand unrelated colors. Allowing for context effects is what boosts the number
of distinguishable colors to the millions (Shevell, 2003).
Adaptation and Afterimages

Color contrast effects show how the spatial relations between colors can influence
color appearance. Temporal relations matter too: what you saw before has an influ-
ence on the color you see now. You already know this from the discussion of light
Courtesy of David Novick
FIGURE 5.23 Color assimilation Here, colors blend together locally. The apparent
color of each sphere comes from stripes that lie on top of it in the group on the left.
Remove the stripes, as on the right, and the spheres are all the same color.
154 Chapter 5
negative afterimage An afterimage adaptation in Chapter 2. Adapting to a bright light makes a moderate light look
whose polarity is the opposite of the darker. Adapting to darkness would make that same moderate light appear brighter.
original stimulus. Light stimuli produce
dark negative afterimages. Colors are
complementary; for example, red pro- FURTHER DISCUSSION of the time course of dark adaptation can be found
duces green afterimages, and yellow in Section 2.3.
produces blue.
adapting stimulus A stimulus whose Now let’s extend that principle to color. Adaptation can be color-specific, as we
removal produces a change in visual see in the phenomenon of negative afterimages. If you look at one color for a few
perception or sensitivity.
seconds, a subsequently viewed achromatic region will appear to take on a color
opposite to the original color. We can call the first colored stimulus the adapting
stimulus. The illusory color that is seen afterward is the negative afterimage. (See
Activity 5.4: Afterimages.)
The principle is illustrated in FIGURE 5.24. Figure 5.24A consists of a circle of
gray spots. Now, stare at the black dot at the center of Figure 5.24B and consider
(A) (B)
(C) (D) (E)

All photos courtesy of Jeremy Wolfe
FIGURE 5.24 Negative afterimages Study the image in (A) and convince yourself that
the ring of circles is gray. Now stare at the black dot in (B). After 10 seconds or so, shift your
fixation to the black dot in (A). The circles should now look colored. This is a negative after-
image. Why does it happen? (If it didn’t happen, try fixating more rigorously. Really look at
the black dot.) Now try with a real scene. Fixate the star in (E), and then flick your eyes to
the same spot in (D). Looking back and forth helps. You should see a washed-out version of
the colors in (C).
Wolfe
what happens as you expose one bit of your retina and visual system to the red dot neutral point The point at which an
at the top of Figure 5.24B. The L-cones will be more stimulated than the M- or opponent color mechanism is gen-
erating no signal. If red-green and
S-cones. L+/M– opponent processes will be stimulated. You will see “red.” When blue-yellow mechanisms are at their
you move your eyes back to fixate on the black dot at the center of Figure 5.24A, neutral points, a stimulus will appear
the red is withdrawn from that area of the visual field. The L-cones will be more achromatic. (The black-white process
adapted than M- or S-cones, as will the later processes in the retina and brain that has no neutral point.)
were more stimulated by the red spot. Adapted processes behave as though they color constancy The tendency of
are somewhat tired. They respond less vigorously than unadapted processes. The a surface to appear the same color
under a fairly wide range of illuminants.
result is a bit like what would happen if you held a pendulum up and released it.
illuminant The light that illuminates
The red-green opponent color mechanism swings back toward the neutral point, a surface.
overshoots this point, and slides over to the green side. As a consequence, the gray
spectral reflectance function The per-
spot appears greenish until the opponent mechanism settles back to the neutral point. centage of a particular wavelength that
If you look at the green dot at the bottom of Figure 5.24B and then look back is reflected from a surface.
at the gray image (Figure 5.24A), you will see the result of pushing the red-green spectral power distribution The phys-
mechanism in the other direction. Other colors will produce other results, which ical energy in a light as a function of
you should now be able to predict. In Figures 5.24C–E, you can try this with a real wavelength.
scene. If you stare at the star in Figure 5.24E and then quickly move your eyes to the
same spot in Figure 5.24D, you will see a pale, tinted version of the real photograph
(Figure 5.24C), though this effect will be more dramatic if you look at the version in
Activity 5.4: Afterimages. Notice that we are not attributing negative afterimages
to just the cones or just one set of cone- or color-opponent processes. Adaptation
occurs at multiple sites in the nervous system, though the primary generators are
in the retina (Zaidi et al., 2012).
Color Constancy
Let us return to the zebra. The fact that the picture looks black, white, and green
whether viewed inside or outside is an illustration of color constancy, the tenden-
cy for the colors of objects to appear relatively unchanged in spite of substantial
changes in the lighting conditions. All the color figures in this book will seem to
have more or less the same colors wherever you read the book (though there is an
entire research area hidden in what we might mean by “more or less”; Foster, 2011).
FIGURE 5.25 illustrates why color constancy is yet another difficult problem for
the visual system to solve. In fact, we don’t fully understand the apparent lightness
of a surface (Soranzo and Gilchrist, 2019). For instance, why does this paper look
white inside the library or outside in the sunlight? The heart of the problem is that
the illuminant, the light that illuminates a surface, is not constant. Lighting changes
as we go from indoors to outdoors or as the sun moves from the horizon to high in
the sky. Figure 5.25A shows the spectral reflectance function for a surface—the
percentage of each wavelength that is reflected from a particular surface. With its
preponderance of long and short wavelengths and that dip in the middle wave-
lengths, this surface probably looks purplish. Let’s call it “lilac.” Figure 5.25B shows
the spectral power distribution—the relative amount of light at different visible
wavelengths—of two different types of “daylight”: sunlight and skylight. Sunlight
is a yellowish light, richer in middle and long wavelengths; skylight is more bluish,
with more short-wavelength energy. Figure 5.25C shows that the light reflected to
our eyes is the product of the surface and the illumination. For example, a surface
might reflect 90% of 650 nm light, but no 650 nm light would reach the eye unless
there was some in the original illumination. Figure 5.25D and E show that those
two different products of surface and illumination are converted into two different
sets of three numbers by the L-, M-, and S-cones. (See Essay 5.4: Color Constancy
in the Lab and Activity 5.5: Color Constancy.)
156 Chapter 5
FIGURE 5.25 Color constancy (A) The surface The surface

The same surface (A) illuminated by two
different lights (B) will generate two differ-
Percentage A surface in the world
ent patterns of activity in the S-, M-, and
of light reflects different
L-cones (C–E). However, the surface will
reflected percentages of
appear to be the same color under both different wavelengths.
illuminants. This phenomenon is known
as color constancy. (After H. E. Smithson.
× ×
2005. Philos Trans R Soc Lond B Biol Soc
360: 1329–1346.)
(B) Illuminant 1 Illuminant 2
Yellowish sunlight
Energy and bluish skylight
of light are composed of
different mixtures
Sunlight Skylight of wavelengths.
(C) Surface × Surface ×

illuminant illuminant
Relative What reaches the

amount eye is the surface
of light reflectance multiplied
by the illuminant.
× ×
(D)
Cone The result is seen

S M L S M L by the three cones.
sensitivity
S M L S M L
(E) The surface The surface

This produces two
very different sets
Cone of three numbers from
responses the same surface. How
do we know what
color that surface is?
Here’s the problem: Even though the three numbers from the three cones are
different in the two conditions, that lilac-colored surface will look lilac under both
Wolfe
illuminants.
Sensation White6Epaper will look white. A banana will look yellow. This color
& Perception
constancy isAssociates
OUP/Sinauer beneficial because we want to know the color of the object. Under
normal circumstances, we don’t care about the spectral composition of the lights.
Wolfe6e_05.25.ai 08.11.2020
The Problem with the Illuminant
Let’s think about color constancy as a math problem. In simple terms, we have an
illuminant (call it I) and a surface (S). As shown in Figure 5.25C, what we can sense is
a result—the product of I × S—but what we want to know is S. We say that we want
to “recover” the surface, S, from the mixture, I × S. It is as if we were given a number
(say, 48), told that it is the product of two other numbers, and asked to guess what
those two numbers might be. The answer could be 12 and 4. Or it could be 16 and 3.
Or 6 and 8. Given just the number 48, we cannot solve the problem. Nevertheless,
given the result of I × S, the visual system does a pretty good job of figuring out S.
We sometimes talk about “discounting” the illuminant as if our whole goal were
to throw away the I term and just see the surface color. However, this is not quite
right. For instance, you can get different answers if you point to two patches under
two different illuminants and ask if the patches were “cut from the same cloth” or
if they are the “same color” (Arend and Reeves, 1986). If you just discarded the
illuminant information, these answers would be the same—but they are not. Simi-
larly, you can tell the difference between a scene lit by the morning sun and a scene
illuminated by the sun at high noon. Thus, not only can you recover the color of the
surface, but you also know something about the illuminant. How do you do this?
Physical Constraints Make Constancy Possible

As noted in the previous section, it is impossible to know which two integers are
multiplied to produce 48. However, if you are told that the first number is between
9 and 14, you’re saved. The first number must be 12, and the second, then, must
be 4. In an analogous way, color constancy must be based on some information or
assumptions that constrain the possible answers. There are many possible assump-
tions that could help. Suppose we assumed that, in a complex scene, the brightest
region was white (Land and McCann, 1971) or that the average color across the
whole scene was gray (Buchsbaum, 1980). We could scale the other colors relative
to these white or gray anchors. However, this can’t be entirely right. Think what
would happen if you were in a dark room with two spots of light on the wall: a red
one and a blue one. Under a simple version of a bright-is-white theory, the brighter
spot should look white and the other spot should change color. That can’t be right
(and theorists knew this, so the actual theories are more subtle, but they still do
not work perfectly).
There are other possible constraining pieces of information. Assumptions
can be made about the illuminant. For instance, natural light sources (and most
artificial ones, such as standard lightbulbs) are generally “broadband.” That is,
they contain many wavelengths, even if some wavelengths are not as intense as
others. Furthermore, their spectral composition curves (see Figure 5.25) are usually
smooth; spikes at particular wavelengths are uncommon, though this generalization
is violated by some artificial light sources. Highly unnatural (e.g., monochromatic)
light sources make the world look highly unnatural—a fact exploited in nightclubs
the world over. Indeed, with a monochromatic source, color vision is impossible,
though not much broadband light is needed to get color back.
Assumptions can be made about surfaces. Real surfaces also tend to be broadband
in their reflectances (recall the Granny Smith apple distributions of Figure 5.6). It
would be very unlikely, for example, to find a surface that reflected 100% of 535 nm
light, 0% of 538 nm light, and 100% again of 540 nm light. Even surfaces that look
like single wavelengths of light typically reflect a wide range of wavelengths. That
is, the red bars in Figure 5.25 may be roughly metameric with something like a 600
nm light, but that region is sending many other wavelengths to your eye. There are
other limits on the reflectance of real surfaces. The whitest surface rarely reflects
more than 95% of any wavelength, and the blackest rarely reflects less than 5%. The reflectance The percentage of light
hitting a surface that is reflected and
brightest thing in the visual field is likely to be white. A “specular reflection” (like not absorbed into the surface. Typically
the shiny spot on a billiard ball) has a wavelength composition very similar to that reflectance is given as a function of
of the illuminant. Any of these facts might help. wavelength.
158 Chapter 5
(A) FIGURE 5.26 What color is this dress? (A) The color of this dress may not look the
same to different people because each observer is making different assumptions about
the nature of the light shining on the dress. (B) Maybe the light is just broadband white
light. Then the dress appears to be blue and black. (C) Maybe there are a couple of light
sources: one bluish and the other more yellow. Then the dress could appear to be white
and gold. (B, C after S. L. Macknik et al. 2015. Sci Am Mind 26: 19–21.)
In 2015, a fairly ordinary picture of a dress swept the

By kind permission of Ecilua Bleasdale
internet. The picture that caused all the fuss is shown in

FIGURE 5.26. Some people assert that the stripes appear
to be black and blue. Others insist on white and gold.
If you saw this dress on someone out on the street, you
would likely agree with almost everyone that the dress
looks black and blue. Why don’t people agree? We are
not really sure, but it seems that people are making dif-
ferent assumptions about the light source in the original
(B) (C) photograph (Figure 5.26A). Some people assume that
the illuminant is white (Figure 5.26B). They would see
the dress as black and blue (the actual colors). Others
may be assuming a more complicated situation with one
diffuse blue light and a more direct, more yellow light
(Figure 5.26C). This could persuade those observers to
see the black as a sort of golden brown and the blue as
a bluish white. Then, when asked about the cloth, they
might report (as would the author of this bit of the text)
that the dress looks like a white-and-gold dress under
a bluish light. If you asked such an observer to match
the color of the lighter stripes, the patch that matched
the color might be quite blue. Remember, “Are these cut
from the same cloth?” and “Are these the same color?”
are different questions (Arend and Reeves, 1986).
The real mystery about the dress is that it looks dra-
matically different to different people. Many researchers
have something to say about the dress (Brainard and
Hurlbert, 2015; Gegenfurtner, Bloj, and Toscani, 2015;
Lafer-Sousa, Hermann, and Conway, 2015; Wallisch,
2017), but we still don’t have a clear explanation for the
dramatic differences between perceptions. Nevertheless,
we think it must be telling us something significant
about the mechanisms of color perception in complex
scenes. (See Activity 5.6: Illusions of Lighting.)
5.7 What Is Color Vision Good For?

Having introduced some of the basics and some of the complications of color vi-
sion, we will wrap up this chapter by returning to our opening question about the
usefulness of color vision. The ability to use wavelength information has evolved
several times in several ways during the course of time. Evolutionary theory tells
us that, for this to be true, color vision must provide an advantage that makes it
worth the trouble. Color is not an absolute requirement. If we could not make
wavelength discriminations, we could still identify a lion (FIGURE 5.27A), and we
(A) (B)
© Pakhnyushchy/Shutterstock.com
© Joerg Huettenhoelscher/123RF
FIGURE 5.27 Do humans need color vision? (A) A black-and-white lion is still a lion,
and (B) you could still find a path in the woods without color vision.
could still find our way through the forest (FIGURE 5.27B). Although color vision
might make the lion stand out a bit better from the background, we would be much
more impairedNOTE: if we lacked
Okay “orientation
to stack vision”
photos to save space,or “motion vision.” Across the an-
if needed
imal kingdom, however, there seem to be at least two realms of behavior where
Wolfe
color vision is especially useful: eating and sex. More generally, color vision seems
to be particularly
OUP/Sinauer helpful in visual search tasks (see Chapter 7).
Associates
Having color vision does seem to make it easier to find candidate foods and
Wolfe6e_05.27.ai
to discriminate good food2.18.2020
from bad food. Comparing the two versions of FIGURE
5.28 shows that finding berries is easier with color vision (Bompas, Kendall, and
Sumner, 2013). Notice also that it makes it easier to decide which of these berries
is ripe. You may recall that we mentioned this as a possible usefulness of the L- and
M-cone difference earlier in the chapter. Most diurnal animals (animals who are active
in daytime) have two photopigments: roughly an S-cone and an LM-cone. Some
primates have evolved separate L and M photopigments and the neural circuitry to
exploit the rather small differences between the responses of those cone types. There
has been considerable debate regarding whether this really conveys an advantage in
telling ripe from unripe fruit or distinguishing subtle differences among green leaves
Both photos © Vphoto/Shutterstock.com
FIGURE 5.28 Using color vision Finding a raspberry is easier if you have color vision,
as is deciding if that berry is ripe.
160 Chapter 5
FIGURE 5.29 Color and flavor The color of a food can change
its reported flavor. Here white wine is reported to taste like rosé
when it is colored pink. (After Q. J. Wang and C. Spence. 2019.
Food Res Int 126: 108678.)
(Troscianko et al., 2003). Some have argued that shape

and texture information, along with dichromatic color
vision, are good enough for these purposes. How can
we find out? In some monkey species (e.g., capuchins)
there is a mix of dichromats and trichromats (as in the
White wine White wine Real rosé wine human species). Do the trichromats have an advantage
dyed rosé color in hunting for food? Rather than test this in capuchins,
40 Citrus 40 40
White fruit Melin et al. (2013) carefully simulated six varieties of
Percentage of people saying they tasted…
35 Red fruit 35 35 capuchin color vision in human observers and had those
observers search for capuchin food in photographs of
30 30 30 those fruits as they would appear in the Costa Rican
forests where the monkeys live. The result was a clear
25 25 25
advantage for trichromatic vision.
20 20 20 Not only does color help you find something to
eat, it can have a significant impact on your ex-
15 15 15 perience of the flavor of a food. One nice example comes
from the world of wine tasting. Wang and Spence (2019)
10 10 10
had participants taste a white wine, a rosé wine, and
5 5 5 some more of that white wine that had been colored
to look like the rosé. For each wine, participants picked
0 0 0
Taste Taste Taste terms referring to smells and flavors that they detected
in each wine. As you can see in FIGURE 5.29, white
wine was judged to have “notes” of white fruits (like
pears), while rosé produced reports of red fruit (strawberries, raspberries, etc.).
Interestingly, the fake rosé was reported to have a flavor profile like the real rosé
even though it was the white wine in disguise.
Did the color change the flavor of the wine, or did it change the answers in a
difficult, subjective taste test? It would be interesting to know what the participants
would have reported if they had not seen the wine. Does the real rosé still taste of red
fruit if you can’t see the red? That condition was not part of this experiment. We do
know that there are limits on the effects of color on flavor. Shankar et al. (2010) did
an experiment where participants tasted a purple drink designed to inspire thoughts
of grape juice. If the real flavor was cranberry, they got grape reports. However, if
the real flavor was vanilla, mere color was not enough to move the reported flavor. �
In addition to searching for and assessing food, animals spend significant time and
effort searching for and assessing potential mates. Here, too, color plays a central role.
Colorful displays—from the dramatic patterns on tropical fish (FIGURE 5.30A) to the
tail of a peacock (FIGURE 5.30B) to the face of the mandrill (FIGURE 5.30C)—are all
sexual signals. Why is the male peacock with the most colorful tail the most desirable
mate for a female peacock? We can’t ask the female, of course, but a colorful tail might
somehow indicate that this male’s genes are better than his competitors’. A peahen
who could only see the world in black and white wouldn’t be able to perceive this
information and would be at an evolutionary disadvantage. In another example that
we mentioned earlier, the development of separate L- and M-cones may have made
primate color vision particularly well suited to detecting the amount of blood in a
blushing or blanched cheek (Changizi, Zhang, and Shimojo, 2006).
Wolfe Sensation and Perception 6e:
Wolfe6e_05.29
08/11/20
Dragonfly Media Group
(A) (B) (C)
© Dynamic Graphics Group/Creatas/Alamy Stock Photo

© Marcel Mooij/Shutterstock.com
© Vlad61/Shutterstock.com
FIGURE 5.30 Color and sex The colors of animals—from (A) tropical fish to
(B) peacocks to (C) mandrills—are often advertisements to potential mates.
Color vision is accomplished in different ways in different species. We are trichro-

mats, with three different types of cone photoreceptors. Dogs appear to be dichromats,
with two types of cone photoreceptors (Neitz, Geist, and Jacobs, 1989). Chickens,
surprisingly enough, turn out to be tetrachromats, with four (Okano, Fukada, and
Yoshizawa, 1995). There is not much gain in information if the number of cone
Wolfe
types is increased beyond 3 or 4 (Maloney, 1986), which probably explains why
Sensation & Perception 6E FIGURE 5.31 Extreme color vision
octachromats or dodecachromats—individuals with 8 or 12 types of cones—are
OUP/Sinauer Associates (A) The absorption spectra of the 38 (!)
very rare. Rare, but not nonexistent. If you have some time, look into the case of the different rod photopigments of the silver
Wolfe6e_05.30.ai
mantis shrimp, a wonderfully2.18.2020
eccentric beast with at least 12 types of photoreceptors spinyfin (Diretmus argenteus). Notice that
(Marshall and Arikawa, 2014). Amazingly, the current record for the greatest number these pigments have peak sensitivity in the
short wavelength (blue) end of the spec-
of photopigments for a vertebrate may lie with a deep-sea fish, the silver spinyfin trum. These fish and other occupants of
(Diretmus argenteus), which has two cone photopigments and 38 rod photopigments the deep sea emit their own light (a bit like
(Musilova et al., 2019). The spinyfin is just one of a group of frightening-looking fish fireflies), and that bioluminescence is in the
(FIGURE 5.31) that live where there is very little light and may have evolved a rich set short wavelengths. (B) The threadfin drag-
onfish (Echiostoma barbatum), another
of photopigments in an effort to catch every photon that they can. deep-sea fish with many photopigments.
(A after Z. Musilova et al. 2019. Science
364: 588–592.)
(A) Diretmus (B) Echiostoma barbatum

argenteus
1.0
Normalized absorbance
0.8
0.6
0.4
Courtesy of Zuzana Musilova
0.2
350 400 450 500 550 600 650

Wavelength (nm)
162 Chapter 5
FIGURE 5.32 Two ways to make (A) Different photopigments can tune photoreceptors
photoreceptors with different to different wavelengths.
spectral sensitivities (A) Our S-,
M-, and L-cones are different because
they contain different photopigments.
+
(B) Some animals have only one type of
+ =
photopigment. These animals can have S
M L
color vision because colored oil droplets
sitting on top of photoreceptors create
S M
groups of photoreceptors with different L
sensitivities to wavelength.
400 700 400 700 400 700 400 700
(B) Colored oil droplets can also tune photoreceptors

to different wavelengths.
Light
+ + =
S
M L
S
M L
400 700 400 700 400 700 400 700
Our S-, M-, and L-cones are different because they contain different photopig-
ments (FIGURE 5.32A). It is also possible to use a single photopigment to create
more than one functional type of cone. The trick is to put a different filter in front
Wolfe
of each type of cone so that some wavelengths are subtracted before light reaches
the photoreceptor (FIGURE 5.32B). A cone with a reddish oil droplet in front of
Wolfe6e_05.32.ai 8.11.2020
it will respond more vigorously to long-wavelength light than will a cone covered
by a greenish droplet. Chicks and other birds have these droplets, as do a variety
of reptiles (Govardovskii, 1983).
Even fireflies get into this act in a limited way. Fireflies signal each other with
bioluminescence: they make their own light. Different species have different lights,
and each species’ visual system appears to be tuned to its particular wavelength
signature. A combination of a photopigment and a colored filter makes signals
from conspecifics (members of the same species) appear brighter than the flashes
of from other fireflies in the vicinity (Cronin et al., 2000). With this sort of visual
system, the firefly will never appreciate the palette of colors in a sunset. But it will
be able to locate an appropriate mate, and that, after all, was the pressure shaping
the development of this limited sensitivity to wavelength.
● Scientists at Work
Filtering Colors
Question How well can you direct your attention to have no effect on ability to judge the centroid of one
one color? target color.
Hypothesis A “feature-based attention” mechanism Results When the different dots varied in hue (see
NOTE:allow
should This goes within “Scientists
an observer to selectat Work”
all thebox at end
items of of
onechapter Section 2.1), this task was quite easy. The researchers
color and treat them as a group, effectively ignoring could calculate the impact of each color dot on the
the other items. centroid, and in the hue condition, observers could
Test The researchers showed their observers images create an attentional filter that effectively removed
like those in the first column of FIGURE 5.33 (Sun et al., every dot except those of the correct hue (here green).
2016). (The colors were much more carefully chosen If the dots varied along a saturation axis as they do in
than we can print here.) Observers were asked to mark the second row, observers couldn’t easily select just
the centroid of the three dots of a particular color. If the dots of one saturation. (The actual experiment
you imagine the triangle formed by the three dots, the was a bit different from what is shown.)
centroid is that triangle’s center of mass. People can Conclusion We can use our attention to select a
do this easily enough with three dots (as in the three group of items on the basis of hue. Other dimensions
dots in the upper right panel of the figure). If filtering of color cannot be filtered as effectively. The authors
by color worked, the dots of different colors would speculate that “hue is a more reliable cue to material
identity than either lightness or saturation” (p. E6717),
so it might not be as useful to filter along
Find the “centroid” We must use It works for hue, those dimensions.
of the three disks attention to filter but not so well Future work Are there other features of
of exactly this color. the image. for saturation.
the visual world for which it might be useful
Hue to have this kind of filtering ability? Suppose
Attention filter
you were a product designer; might this

result influence the way that you designed
the bottles for a line of shampoo?
FIGURE 5.33 The Sun et al. attention filter

Only the correct experiment Column 1 shows sample stimuli.
Stimulus three dots matter. Observers would be asked to pay attention to just
three dots in each image, defined by their color. Col-
Saturation
umn 2 shows the “attention filter” that can be inferred
Attention filter
from the data. Column 3 shows the hypothetical

? strength of the dots in the mind of the observer when
the filter has been applied. When the dots differ in
hue (top row), the filter leaves only the target dots.
When the dots vary in saturation (bottom row), non-
target dots get past the filter and influence behavior.
Other dots leak in. (After P. Sun et al. 2016. Proc Natl Acad Sci USA 113:
Stimulus E6712–E6720. doi: 10.1073/pnas.1614062113.)
Wolfe
Wolfe6e_05.33.ai 2.18.2020
164 Chapter 5
Summary
1. Probably the most important fact to know about color vision is that lights
The Sensation & Perception and surfaces look colored because a particular distribution of wavelengths
digital resources include
of light is being analyzed by a particular visual system. Color is a mental
chapter overviews, activities,
essays, flashcards, and
phenomenon, not a physical phenomenon. Many animal species have some
other study tools. form of color vision. It seems to be important for identifying possible mates,
possible rivals, and good things to eat. Color vision has evolved several times
in several different ways in the animal kingdom.
2. Rod photoreceptors are sensitive to low (scotopic) light levels. Humans have
only one type of rod photoreceptor; it yields one “number” for each location in
the visual field. Our rods can support only a one-dimensional representation of
color, from dark to light. Thus, human scotopic vision is achromatic vision.
3. Humans have three types of cone photoreceptors, each having a different sen-
sitivity to the wavelengths of light. Cones operate at brighter light levels than
rods, producing three numbers at each location; the pattern of activity over the
different cone types defines the color. Some animals have many more types of
photoreceptors, but we know rather little about their color experience.
4. If two regions of an image produce the same response in the three cone
types, they will look identical; that is, they will be metamers. And they will
look identical even if the physical wavelengths coming from the two regions
are different.
5. In additive color mixture, two or more lights are mixed. Adding a light that
looks blue to a light that looks yellow will produce a light that looks white (if
we pick the right blue and yellow). In subtractive color mixture, the filters,
paints, or other pigments that absorb some wavelengths and reflect others
are mixed. Mixing a typical blue paint and a typical yellow paint will subtract
most long and short wavelengths from the light reflected by the mixture, and
the result will look green.
6. Color blindness is typically caused by the congenital absence or abnormality
of one cone type—usually the L- or M-cone, usually in males. Most color-
blind individuals are not completely blind to differences in wavelength.
Rather, their color perception is based on the outputs of two cone types
instead of the normal three.
7. A single type of cone cannot be used, by itself, to discriminate between
wavelengths of light. To enable discrimination, information from the three
cones is combined to form three cone-opponent processes. In the first,
cones sensitive to long wavelengths (L-cones) are pitted against medi-
um-wavelength (M) cones to create an L – M process that is roughly sensitive
to the redness or greenness of a region. In the second cone-opponent pro-
cess, L- and M-cones are pitted against short-wavelength (S) cones to create
an (L + M) – S process roughly sensitive to the blueness or yellowness of a
region. The third process is sensitive to the overall brightness of a region.
8. Color appearance is arranged around opponent colors: red versus green, and
blue versus yellow. This color opponency involves further reprocessing of the
cone signals from cone-opponent processes into color-opponent processes.
9. The visual system tries to disentangle the properties of surfaces in the world
(e.g., the “red” color of a strawberry) from the properties of the illuminants
(e.g., the “golden” light of evening), even though surface and illuminant
information are combined in the input to the eyes. Mechanisms of color con-
stancy use implicit knowledge about the world to correct for the influence of
different illuminants and to keep that strawberry looking red under a wide
range of conditions.

Sensation & Perception 6th Edition 5

Uploaded by

Copyright:

Available Formats

Sensation & Perception 6th Edition 5

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sensation & Perception 6th Edition 5

Uploaded by

Copyright:

Available Formats

Chapter

Courtesy of Lutz Baar

Lutz Baar, French Abstract, 2014

5.1 Basic Principles of Color Perception

S-cone A cone that is preferentially Three Steps to Color Perception

420 498 535 565

5.3 Step 2: Color Discrimination

FIGURE 5.2 Responding to one

constraint is known as the principle of univariance (Rushton, 1972). (See Activity

FIGURE 5.4 Rod vision The moonlit world appears

Courtesy of John Bortniak, NOAA

M-cone response to M-cone response to

Class 0 Class 1 Class 2 Class 3 Class 4

produce a mixture that excites the L- and M-cones equally.

Bluish Blue Green Red FIGURE 5.8 Maxwell’s color-matching experiment

FIGURE 5.11 Pointillism Instead of

Cone-Opponent Cells in the Retina and LGN

With permission from AAAS/CC BY-NC 4.0

e1600797. DOI: 10.1126/sciadv.1600797.

5.4 Step 3: Color Appearance

Three Numbers, Many Colors

● Sensation & Perception in Everyday Life

ed green. Finally, brightness

describes that black-to-white

71 % achromatic axis. The black

Sensation & Perception in Everyday Life (continued )

The Limits of the Rainbow

(A) The spectrum of wavelengths (B) The color circle

400 450 500 550 600 650 700

+ = Much too little.

Too much. Far too much. Too much.

Point 1 on graph Point 2 on graph Point 3 on graph

Values above the

Values below the

400 500 600 700

Wolfe Sensation and Perception 6e:

This opponent-process, color-appearance story sounds rather similar to the L – M

(A) Step 1: Detection (cones)

400 500 600 700

(B) Step 2: Discrimination

400 500 600 700 400 500 600 700

(C) Step 3: Appearance (opponent colors)

2 entially absorbed by three photopigments

Color in the Visual Cortex

FURTHER DISCUSSION of the elegantly named “blobs” can be found in

5.5 Individual Differences in Color Perception

Language and Color

FIGURE 5.19 Color category boundaries (A) (B)

Slowest FIGURE 5.20 A color category

720 Responses are faster and more accu-

Red Red Brown Brown

Genetic Differences in Color Vision

5.6 From the Color of Lights to a World of Color

FIGURE 5.21 Color constancy This zebra looks like

FIGURE 5.22 Color contrast The central square takes on

Adaptation and Afterimages

Courtesy of David Novick

(C) (D) (E)

FIGURE 5.25 Color constancy (A) The surface The surface

(C) Surface × Surface ×