How You See

Chapter 2 Photoreceptors and color

Rods and cones

Every cell at the back of your eye answers one question: how much light did I catch? Color exists only because three kinds of cone answer it differently.

10 min read · 5 interactive figures · 2 questions

The measured sensitivity of the three kinds of cone (S, M and L) and of the rods to light of each wavelength. Drag the light along the spectrum: the bars show how strongly each receptor responds. Notice that no single receptor can tell you the wavelength; only the pattern across the three cones can.

Rods and cones are the two kinds of light-sensitive cell, or photoreceptor, in the retina. Rods are extremely sensitive, work in dim light and cannot tell colors apart; a human retina has about 92 million of them. Cones need brighter light, give sharp detail and color, and come in three types; there are about 4.6 million, crowded most densely in the fovea at the center of gaze[1][2]. Color vision exists because the three kinds of cone respond to overlapping but different ranges of wavelength, and the brain compares them.

The figure at the top uses the measured sensitivity curves of all four receptor types. Move the light along the spectrum and watch the bars: that pattern of four numbers is all the retina knows about the light's wavelength. The rest of this chapter is about what follows from that.

Two kinds of photoreceptor

Both kinds of cell work the same way. An outer segment is packed with a light-absorbing pigment; when the pigment catches light, the cell's membrane potential changes, and so does the amount of neurotransmitter it releases onto the next cells in the retina[2]. They differ in shape (the long, stacked outer segment of the rod; the short, tapered one of the cone), in pigment, in number and in wiring. established

RodsCones
Number in one retinaabout 92 million (78–107 million)about 4.6 million (4.1–5.3 million)
Where they are densesta ring around the fovea at about the distance of the optic disc; none at the very centerthe fovea: about 199,000 per mm², varying from 100,000 to 324,000 between people
Pigmentsone, rhodopsinthree opsins, one per cone type (S, M, L)
Light leveldim light; overwhelmed in daylightdaylight; unresponsive in very dim light
Colornoneby comparison between the three types
Detaillow: many rods pool onto one ganglion cellhigh in the fovea, where one cone can have its own pathway

The numbers come from Christine Curcio and colleagues' 1990 counts of eight whole human retinas[1]; the wiring from OpenStax, which notes that toward the edge of the retina up to 50 photoreceptors converge, through bipolar cells, onto one ganglion cell, while in the fovea a cone can connect to a single one[2]. established Pooling makes rods good at catching scarce light and poor at detail. That trade-off explains most of what follows.

Two details of the layout are worth knowing. The very center of the fovea has no rods at all: Curcio's team measured a rod-free zone 0.35 mm across, about 1.25 degrees of visual angle[1]. And cone density falls tenfold within a millimeter of the foveal center, which is why only the few degrees around where you are looking are sharp. The rods reach their highest density in a ring at about the eccentricity of the optic disc[1]. established

One kind of cone cannot see color

Question 1 Prediction: can one cone type tell two colors apart?#

A photoreceptor counts the light its pigment absorbs. Light of some wavelengths is absorbed more readily than others, but once a photon is absorbed its effect is the same whatever its wavelength. So a single receptor's output is one number, and many different lights produce the same number: a dim light at the wavelength it likes best, or a bright light at one it barely notices. This is called the principle of univariance. established

Figure 1 Left: a 550 nm reference light. Right: a 610 nm test light at the same intensity. With One cone type (M) selected, each patch is shown as the gray level of the M-cone signal, and the test looks darker. Press the button to adjust the test light’s intensity until the M cones report the same number: the patches become identical. Then switch to Three cone types and try to make them match.#

With three cone types, no amount of turning the brightness up or down makes a 550 nm light and a 610 nm light identical, because the three cones change by different factors. The color you see is the brain's reading of the ratios between them. People and animals with only one working receptor type see the world in shades of a single dimension. That is exactly your situation at night, when only rods are working[2]. established

Three cones make color a comparison

The three cone types are usually called S, M and L, for short-, middle- and long-wavelength sensitive. They are sometimes called blue, green and red cones, but the names mislead: the L cone's sensitivity peaks at about 570 nm, in the yellow-green of the spectrum rather than the red, and the M and L curves overlap so much that almost any light stimulating one stimulates the other.

The curves in the figures here are the Stockman and Sharpe cone fundamentals, derived from color-matching measurements in observers whose cone genes were known[3], together with the CIE scotopic curve for rods, all taken from the Colour & Vision Research Laboratory database[4]. Measured at the cornea in energy units, they peak near 445 nm (S), 545 nm (M) and 570 nm (L), and the rod curve near 505 nm. established

Figure 2 The receptors’ response to 470 nm light, a blue: of the three cones the S cones respond most strongly, but the M cones respond too, and the L cones a little; the rods (dashed) respond strongly as well. Turn off the rods to see the cone pattern alone. Then drag toward 550 to 600 nm and notice that the S cones fall silent while the L and M responses change only in their ratio.#

Notice how the readout under the figure reports the ratio L : M. Between about 550 and 650 nm the S cones contribute almost nothing, and the only information about wavelength is that ratio. Yellow, orange and red are, as far as the retina is concerned, different values of one number. That is why mixtures can fool the eye: any two lights that produce the same three cone signals look identical, whatever their spectra. Such pairs are called metamers, and they are why a screen with three kinds of pixel can display the colors it does. established

Night vision: rods take over

In normal sunlight the rods' pigment, rhodopsin, is bleached as fast as it can regenerate, and the cones do the seeing. In a dark room there is not enough light for the cones, and vision depends on the rods alone, which do not distinguish colors: low-light vision is, in effect, grayscale[2]. established

The switch takes time. After you step from bright light into darkness, your sensitivity climbs in two stages. The cones recover first; the rods, whose pigment has to be rebuilt, take over later. In the classic measurements by Hecht, Haig and Chase, published in 1937, after the brightest pre-exposure the rods' contribution appeared only after 12 to 13 minutes in the dark[6]. established

Question 2 Prediction: which looks brighter at dusk, red or blue?#

Rods and cones also prefer different parts of the spectrum. Daytime brightness follows the CIE photopic curve, which peaks at 555 nm; night-time brightness follows the rods' curve, peaking near 505 nm[4]. A red flower and a blue one that look equally bright at noon will not look equally bright at dusk.

Figure 3 A 630 nm red light and a 470 nm blue light, set to equal brightness for daylight vision (equal photopic luminance). Switch to Night (rods): the patches are redrawn as gray levels in proportion to the rod signal each produces, and the readout gives the ratio. Move either wavelength to see how the effect depends on it.#

This change in relative brightness between day and night vision is called the Purkinje shift, after Jan Evangelista Purkyně, who described in 1825 how red flowers that were bright in afternoon sun looked dark at dawn[7][8]. The figure computes it from the two CIE curves alone. It models only the two extremes; at twilight both systems contribute at once, a regime called mesopic vision that the figure leaves out.

Color blindness

Not medical advice. This section and the simulator explain how color vision deficiency works. They cannot tell you whether you have it. Screens vary in their colors, and diagnosis uses standardized tests under controlled lighting. If you have questions about your color vision, see an optometrist or ophthalmologist.

Most color blindness is a change in one cone type. If the L cones are missing or altered, the condition is called protan; the M cones, deutan; the S cones, tritan. When a cone type is entirely absent the person is a dichromat (protanopia, deuteranopia, tritanopia), relying on two cone types. In the milder and more common forms the cone type is present but works abnormally, which is anomalous trichromacy (protanomaly, deuteranomaly, tritanomaly). The National Eye Institute lists deuteranomaly as the most common type, and red–green deficiency as the most common kind overall[9]. Because the M and L curves already overlap so much, losing or shifting either one mainly collapses the reds, oranges, yellows and greens toward one another: red–green color blindness. established

It is common. Large population surveys find inherited red–green deficiency in about 8% of men and about 0.4% of women of European descent, and in 4% to 6.5% of men of Chinese and Japanese descent[10]. established

Figure 4 A fruit stall, and the same picture simulated for deuteranopia (no working M cones) with the model of Machado, Oliveira and Fernandes. The red and green apples turn the same olive hue and differ only in lightness. Change the type and the severity; at severities below 1 the model shifts the affected cone’s sensitivity rather than removing it. Try the line chart: four series told apart only by color.#

The simulation uses the model published by Gustavo Machado, Manuel Oliveira and Leandro Fernandes in 2009, which is based on the stage theory of color vision and handles normal vision, anomalous trichromacy and dichromacy in one framework[11]. Their published matrices act on linear RGB values; this site applies them after undoing the sRGB encoding of the picture, as specified in CSS Color 4[12]. documented

The most useful thing the simulator shows is a design lesson. A chart whose series differ only in red and green, or a map that marks danger in red and safety in green, carries no information for roughly one man in twelve of European descent. Adding shape, labels, pattern or a difference in lightness fixes that for everyone. The full-size simulator, which also accepts your own pictures (processed in your browser and never uploaded), is on the color blindness simulator page.

From the retina onward

By the time a signal leaves the eye, the retina has already done a good deal of work: pooled rod signals, compared cone signals and sharpened edges. What it has not done is recover the third dimension. Each retina holds a flat pattern; seeing how far away things are is the next problem the brain has to solve, in the chapter on depth perception.