A camera made of water and protein
The front of the eye has one job: bend light so that everything from one point in the world lands on one point at the back. Get it wrong by a millimetre and the world goes soft.
Simplified: the diagram lumps the cornea and lens into one bending surface and exaggerates distances inside the eye so the focus error is visible. The letter shows how blurry the image on the retina would be.
The cornea does most of the work. The clear dome over your iris provides about 43 of your eye's roughly 60 dioptres of focusing power, because the jump from air into the watery cornea is where light bends most. That is why swimming underwater blurs everything: water cancels the cornea.
The lens does the fine-tuning. A ring of muscle lets the lens bulge to focus up close. Drag the object toward the eye and watch the lens fatten. The extra power is called accommodation, measured in dioptres (1 ÷ distance in metres).
The lens stiffens with age. A child can add about 15 dioptres, enough to focus 7 cm from their face. By 45 most people have 3 or 4 left, and by 60 almost none. Set the age to 55 and bring the object in to 25 cm: that is why reading glasses arrive in middle age. This is called presbyopia.
The iris is an aperture. Your pupil ranges from about 2 mm in sunlight to 8 mm in the dark, a 16× change in light. A small pupil also narrows the cone of light, so a focus error makes a smaller blur. Shrink the pupil in the simulator. This is why squinting helps you read a distant sign.
Your eye cannot focus all colours at once. Short wavelengths bend more than long ones, so green comes to focus slightly in front of red. Optometrists use this. Look at the letters below, with glasses if you normally wear them.
Red side sharper? Your focus lands a little in front of the retina, a slight lean toward myopia. Green sharper: a slight lean the other way. Equal: your focus is balanced.
Two cameras sharing one sensor
The retina is a sheet about a quarter of a millimetre thick, lining the back of the eye. It holds two kinds of light catcher, arranged very unevenly.
Rods (about 92 million) are so sensitive they can respond to a single photon, but they only report brightness. They run your night vision. Their pigment, rhodopsin, is sometimes called visual purple, which is why rods are purple on this page.
Cones (about 4.6 million) need much more light but come in three types, tuned to short, medium and long wavelengths. Comparing the three gives you colour. Cones also give you fine detail.
Almost all the cones crowd into the fovea, a pit about 1.5 mm across directly behind the lens. Its centre has no rods at all. Everywhere else, rods dominate. Hover the chart, or drag the slider, to see the mosaic at different distances from the centre of your gaze.
The retina is built backwards. Light must pass through layers of nerve cells and blood vessels before it reaches the photoreceptors, which face away from the light. The wires have to leave somewhere, so they bundle together and exit through one hole: the optic disc. About 1.2 million nerve fibres carry the output of those 97 million receptors, so the retina has already compressed and processed the image before it leaves the eye.
- ~150,000
- cones per mm² at the centre of the fovea
- 0
- rods in the central 0.35 mm, the foveola
- 0
- receptors of any kind at the optic disc
Fovea: only cones, packed tight. Long (orange), medium (green) and short (blue) cones, with no short cones in the very centre.
Data table
Approximate values after Osterberg (1935) and Curcio et al. (1990), in thousands per mm². Mosaic is illustrative.
There is a hole in your vision right now
Where the optic nerve leaves the eye there are no receptors, so each eye has a blind patch about 5° wide, around 15° to the side of wherever you look. That is about the size of a fist at arm's length. You have never noticed it.
Cover or close your left eye. With your right eye, stare at the +. Don't look at the dot. Start about 50 cm from the screen and slowly lean in. At some distance the dot vanishes completely. It has landed on your optic disc.
What replaces the dot is the interesting part. You don't see a black hole. You see the background colour, and in the other two tests your brain continues the line and the pattern across the gap. It is inventing what should be there from the surroundings. This is called filling-in.
Sharp vision is the size of your thumbnail
Only the central 2° of your view, about a thumbnail at arm's length, is truly sharp. Your eyes make 3 or 4 jumps (saccades) every second to aim that tiny patch at whatever matters, and your brain stitches the results together.
Stare at the centre dot. The letters get bigger the farther out they are, and at the right viewing distance they all look about equally readable. This chart (after Stuart Anstis, 1974) shows how steeply acuity falls: a letter 20° out must be roughly ten times larger than one near the centre.
Crowding: seeing that something is there, not what it is
Fix your gaze on the red + below. The single letter on the left is easy to name. The same letter on the right, squeezed between neighbours, becomes a jumble, even though it is exactly as far from the centre.
Your periphery is not just blurry. It pools features across a region, so nearby shapes merge. This is why you can't read a line of text out of the corner of your eye even when the letters are large.
You also don't notice any of this. During each saccade your brain suppresses the smeared image, and it covers the gap with the view you land on. Try it in a mirror: look from one of your eyes to the other. You will never see your own eyes move, though someone watching you will.
- 2°
- of the field is seen at full acuity
- ~50%
- of primary visual cortex is devoted to roughly the central 10°
- 3–4 / s
- saccades, each lasting 20 to 200 milliseconds
Colour is three numbers
A cone doesn't know which wavelength hit it. It only reports how strongly it responded. Your sense of colour comes from comparing three such reports, and that shortcut has strange consequences.
Data table
Curves are smoothed approximations of human cone sensitivities, each scaled to a peak of 1.
Look at how much the M and L curves overlap. They evolved from one gene that duplicated in our primate ancestors around 30 to 40 million years ago. Most mammals, including dogs, have only two cone types.
The genes for M and L sit on the X chromosome. If one is missing or shifted, the red–green comparison breaks. About 1 in 12 men and 1 in 200 women of Northern European descent have some form of red–green colour deficiency. You can simulate it in the lab below.
The left patch is pure 580 nm light. Mix the two lights on the right, 630 nm red and 540 nm green, until your cones can't tell them apart. Watch the bars: when all three match, the two lights are identical to you, even though they are physically nothing alike. Such pairs are called metamers.
This is how every screen works. There is no yellow light coming out of your display. The yellow patch above is red and green subpixels, tuned to produce the same three cone responses as real yellow light.
Tired cells paint in colour
After the cones, the retina recodes colour into opposing pairs: red against green, blue against yellow, light against dark. Stare at one side of a pair long enough and the cells tire. Look away and the opposite side wins for a few seconds.
The picture has had its colours flipped. Press start and hold your gaze on the black dot, without moving your eyes, until the timer ends. The image will switch to plain black and white. Keep staring at the dot.
For a few seconds you should see the scene in its correct colours: a blue sky, red apples, green leaves. None of that colour is on the screen. Your red–green and blue–yellow channels adapted to the flipped colours, and the grey image now reads as their opposites.
The same fatigue explains why a white page looks bluish after a spell under warm lamplight, and why the green spot hangs in front of you after a camera flash.
This opponent coding is also why no colour looks reddish-green or yellowish-blue, while reddish-yellow (orange) and bluish-red (purple) are easy to imagine. Red and green sit on opposite ends of one channel.
- 10¹⁰
- range of light levels you can see in, from starlight to snow in sunshine
- ~30 min
- for rods to fully adapt to darkness, after about 10 minutes for cones
You see a best guess, not the light
The brain doesn't measure the light reaching your eye. It works out what surfaces are out there and what they are probably like, and shows you that. Usually this helps. Sometimes it shows.
The checker shadow
Square A looks dark and square B looks light. They are exactly the same grey, pixel for pixel. Press the button to remove their surroundings.
Your brain knows B sits in a shadow, so it discounts the shadow and reports what the surface is probably made of: a light square in shade. It is answering "what colour is the paint?" rather than "how much light is this?". That is almost always the question you need answered.
Fading into the background
Stare at the + and hold still. Within 20 seconds or so the soft blobs begin to fade into the grey. Unchanging input with no sharp edges is treated as uninteresting and dropped. This is Troxler fading, first described in 1804. Blink or move your eyes and they come back.
Take the eye apart
Remove any part of the system and see what that eye would see. The ring follows your pointer: that is where the eye is looking, so the fovea is aimed there. Each change here corresponds to a real condition, which is named in the notes.