3.6 Animal eyes🔗⧉

A bee can distinguish ultraviolet patterns that we cannot see, while a raptor can resolve fine detail at a distance. Other eyes emphasize sensitivity, wide fields of view, or polarization. Comparing them helps separate the information available in light from the particular measurements made by human vision.

3.6.1 The optics: many ways to form an image🔗⧉

Forming an image means getting light from each direction in the scene onto a different receptor, the same plenoptic-sampling problem a camera solves with a lens. Nature has found several radically different answers (Figure 3.6.2⛶ fullscreen).

A patch of photoreceptors can detect illumination; shading part of it can add directional information; a cup and a restricted opening can improve directional selectivity; a lens can gather more light while forming an image. This sequence illustrates plausible functional steps Land & Nilsson, Animal Eyes (Figure 3.6.1⛶ fullscreen). It is not a universal historical ladder followed by every eye lineage.

Even the bottom rung has a subtler cousin. The single-celled Euglena has no eye at all, only a light-sensitive spot shaded by an adjacent speck of orange pigment, its stigma. As the cell corkscrews through the water the pigment speck sweeps past the receptor and periodically shadows it, so the receptor reads the direction of the light from the rhythm of its own shadow, and the organism steers toward it. The trick is the exact optical inverse of a pinhole: where a pinhole is a small hole that lets a pencil of light through, this is a pinspeck, a small opaque speck that blocks one and casts a shadow, and a shadow carries directional information just as an image does. It is the same anti-pinhole principle we later exploit to form accidental images from ordinary occluders (Chapter 38.1, Accidental cameras).

Eye design depends on body size, habitat, and the visual tasks an animal performs. Larger apertures collect more light, finer angular sampling requires more receptors, and temporal integration improves sensitivity while reducing response to rapid changes. Shared ancestry also constrains which designs evolve.

• Camera-type eyes, including vertebrate and cephalopod eyes, form an image with a shared optical system. Their receptor arrangements differ despite similar overall imaging geometry. “Single-lens” does not mean the cornea contributes no refraction.
• Compound eyes contain repeated units called ommatidia. In apposition eyes, units sample different directions; in superposition eyes, optics can combine light from several facets. Facet spacing, optical acceptance angles, diffraction, and neural processing jointly limit angular resolution. They are not universally arrays of independent one-pixel cameras.
• The nautilus uses a lensless pinhole eye. Scallops use a concave mirror as a major focusing element, together with other ocular structures. The existence of reflective focusing does not imply that no refraction occurs.
• Many nocturnal eyes return unabsorbed light through receptors with a reflective tapetum, producing eyeshine. Other adaptations include larger apertures, receptor pooling, and longer temporal integration.

3.6.2 The pupil: aperture shape as a clue to ecology🔗⧉

Pupil shape changes both light collection and defocus blur. Banks and colleagues found associations between pupil shape, eye height, and ecological niche in terrestrial animals, and analyzed possible optical advantages (Banks et al. 2015). These associations are useful hypotheses, not a reliable way to identify every animal's behavior from its pupil.

Round pupils give approximately orientation-independent defocus in a rotationally symmetric eye. They occur across many ecological groups, including humans and large cats.

A vertical slit is narrow horizontally and wide vertically. Its defocus kernel therefore spreads a horizontal edge more than a vertical edge: vertical contours remain comparatively sharp, supporting horizontal stereoscopic disparity estimates, while blur of horizontal contours can supply another distance cue. This is the orientation dependence proposed for low-to-the-ground ambush predators. Slit pupils can also vary their area substantially between bright and dim conditions: Banks and colleagues report that a domestic cat's pupil area changes about 135-fold, against about 15-fold for a human's round pupil Banks et al. 2015.

Horizontal pupils occur in many grazing animals with laterally placed eyes. Their optics favor light collection and image quality along the ground plane. Goats and some other grazers rotate their eyes as the head lowers, helping maintain pupil orientation relative to the horizon. The wide field of view also depends on eye placement and the rest of the optics, not pupil shape alone.

Cuttlefish can have a W-shaped pupil when constricted. Proposed functions include controlling light from different parts of the field. Chromatic-aberration-based spectral discrimination has also been proposed for cephalopods, but should be presented as a hypothesis rather than established ordinary color vision.

Some gecko pupils constrict to several small apertures. This limits light entry and changes the defocus pattern. Multiple apertures can carry focus information in principle, but a claim that an animal uses their relative images for distance requires behavioral evidence.

Aperture shape affects directional blur and the range of admitted light. Its function should be evaluated together with eye geometry and behavior.

3.6.3 Amphibious vision: focusing in air and in water🔗⧉

A human eye loses much of its focusing power underwater because the refractive-index step at the front cornea becomes smaller: water is near 1.33, versus about 1.38 for corneal tissue and 1.00 for air. The resulting image is strongly defocused. A diving mask restores an air space in front of the cornea. Aquatic and amphibious eyes use different combinations of corneal power, lens power, accommodation, and pupil restriction. In a human eye the cornea supplies about two thirds of the eye's roughly 60 diopters, most of which disappears underwater. Diving birds such as cormorants instead squeeze an unusually soft lens through the iris, with reported gains of tens of diopters, to replace what the flooded cornea lost.

Evolution has answered this several ways (Figure 3.6.6⛶ fullscreen), and you can trace each one through a schematic eye (Figure 3.6.5⛶ fullscreen).

A pinhole can form an image geometrically without a powered lens, which reduces dependence on refractive focusing. It still has a brightness–blur tradeoff, and diffraction and transmission depend on wavelength and medium. The fully aquatic nautilus is an example of pinhole imaging, not evidence of identical vision in air and water.

Fish rely heavily on a strong lens because a submerged cornea contributes little power. Pinnipeds combine strong lenses with corneal regions and pupil shapes that help support aerial and underwater vision. Anatomy varies by species; a uniformly flat cornea and a sphere are teaching simplifications, not a complete seal-eye prescription.

The surface-dwelling fish Anableps receives light through aerial and submerged portions of each eye. Its divided pupil and specialized corneal and lens geometry support these different optical paths. It has two eyes, each sampling both environments.

Diving birds can accommodate strongly by changing lens shape. In other animals, accommodation may involve lens displacement as well as deformation. A generic statement that amphibians and turtles all squeeze their lenses would hide those differences.

Mudskippers use aerial vision while their eyes project above the head. Their ocular anatomy supports life at the water–air boundary; raising the eyes alone would not solve the refractive mismatch of an arbitrary fish eye.

The air–water comparison is an optical-design problem: changing the surrounding refractive index changes surface power. Figure 3.6.5⛶ fullscreen should be read as a schematic comparison unless its geometry and material properties are calibrated to particular species.

Stopping down provides a partial response to lost focusing power. A narrow aperture admits a smaller bundle of rays from each object point, so those rays make a smaller blur spot even when the focal plane is misplaced. It does not restore the missing corneal refraction. At the pinhole limit, image formation relies mainly on selecting ray directions rather than bringing rays together with a lens; the cost is little light, with diffraction limiting how far the aperture can be reduced. This separates two solutions an amphibious eye can use: change the focusing power, or tolerate the focus error by restricting the aperture.

3.6.4 The sensor: a retina that is not uniform🔗⧉

Many camera sensors use a regular pixel grid, while retinas have pronounced variation in receptor density, type, and connectivity. This allows an eye to emphasize particular directions without resolving its entire field equally.

Human foveal sampling supports fine detail near gaze, while peripheral vision contributes broader spatial information and motion sensitivity. Saccades bring selected features to the fovea. This motivates foveated imaging, but should not be interpreted as a blank periphery that is merely filled in.

Some retinas concentrate high sampling density in a fovea, others have an area of higher density or a visual streak. Cats have an area centralis rather than a primate-like foveal pit. Many raptors have two foveae with different viewing directions; this is not universal among birds. Grazers may combine a horizontal visual streak with lateral eyes to monitor a broad horizon.

Forward-facing eyes increase binocular overlap, but “predator” and “prey” are not anatomical categories: primates, birds, and other groups provide exceptions to a simple front-versus-side rule. Compare the actual field of view and retinal sampling rather than inferring them from diet alone.

3.6.5 Seeing in the dark: sensitivity and its tricks🔗⧉

Low-light vision uses the same tradeoffs as imaging instruments. A larger aperture admits more light. Spatial pooling combines signals from more receptors, sacrificing detail; longer temporal integration improves detection of slowly varying signals, sacrificing speed. A tapetum gives some unabsorbed light another pass through the receptors, with an effect that depends on the anatomy and spectrum. Deep-sea fish push the aperture route furthest. Many have tubular eyes: a large lens on a short cylinder of retina, usually pointing upward at the faint light from the surface, which buys aperture at the cost of a narrow field of view. Some stack several layers of rods in a multibank retina, so that a photon missed by one layer can be absorbed by the next. Owls use a tubular eye shape too, fitting a large lens and a long focal length into a skull without room for a spherical eye that large; the tubes barely move in their sockets, which is why owls turn their heads instead.

Rods are not the only route to night vision. Nocturnal helmet geckos have demonstrated cone-based color discrimination at illumination where human color discrimination is poor (Roth and Kelber). Dung beetles can orient using the broad Milky Way pattern without resolving its individual stars (Dacke et al.). The elephant hawkmoth discriminates colors in dim starlight, where our own color vision has long since shut off, by pooling signals over space and time Kelber, Balkenius & Warrant 2002. These are different tasks, so their thresholds should not be ranked on one universal “times better than humans” scale.

A receptor's response to one absorbed photon is also different from reliable behavioral detection. The distinction and the role of photon noise are developed in Chapter 3.3, Adaptation and constancy and Chapter 2.8, Noise, signal-to-noise ratio and dynamic range.

The alternatives can be compared by the resource they use:

The gecko and dung-beetle examples answer different questions: distinguishing colors and maintaining a heading. Pooling a broad sky pattern can support navigation without resolving individual stars. This is the same distinction a camera designer makes between detecting a large faint object and recovering its fine texture.

3.6.6 Acuity: the finest detail, and how to sharpen it🔗⧉

Acuity measures fine spatial discrimination under specified conditions. Grating acuity is expressed in cycles per degree; an eye-chart result uses a different task. A 20/20 optotype has a critical stroke width of one arcminute. Alternating one-arcminute light and dark bars form a two-arcminute cycle, or 30 cycles per degree, not 60. Favorable grating measurements can reach higher frequencies.

Receptor spacing, focal length, optical blur, contrast, and neural processing all matter. Large raptor eyes and dense retinal sampling support high acuity; the proposed extra magnification from a deep foveal pit is not a settled explanation for every species. Compound-eye resolution depends on interommatidial angle and optical acceptance angle, and varies substantially across insects. Jumping spiders concentrate fine sampling into a narrow field of their principal eyes.

Measured grating acuities give the scale. The values below are approximate, gathered by Caves and colleagues from studies that used different methods Caves, Brandley & Johnsen 2018:

A camera sampling calculation is useful if it is kept separate from behavioral acuity. Near the optical axis of a rectilinear camera, pixel pitch $p$ and focal length $f$ give an angular sample spacing approximately $p/f$ radians. The local Nyquist frequency is therefore

For a 60-megapixel, 3:2 full-frame sensor, about 9,487 samples span 36 mm, so $p\approx3.8\,\mu\mathrm m$. A 50 mm lens gives a central sampling limit near 115 cycles/degree. This is a sampling ceiling, not resolved detail: lens aberrations, diffraction, the color-filter mosaic, noise, and processing reduce usable contrast. Sampling also varies with position in a wide-angle image.

Comparing an eye with a camera therefore requires a common stimulus, angular scale, contrast criterion, and viewing condition. Megapixels or a nominal zoom multiplier alone cannot establish that a phone out-resolves an eagle.

We can also solve the sampling calculation backward. For the same sensor, the focal length needed for a chosen central sampling limit is $f\approx360p\nu_N/\pi$.

These are sampling targets, not measured camera acuities or a ranking against particular animals. Increasing focal length puts more samples on each degree by narrowing the field of view. It does not create more total pixels. An animal with a fovea allocates detail nonuniformly; a photographer can instead narrow the whole camera's field. This is why an eye-to-camera comparison needs both angular detail and field coverage, and why focal length belongs in the calculation alongside pixel count.

3.6.7 Color vision: opsins remixed🔗⧉

Species differ in their photopigments and in how they compare receptor outputs (Figure 3.6.10⛶ fullscreen). A honeybee has ultraviolet-, blue-, and green-sensitive receptors; a dog has two main cone classes; many birds have four. Receptor count does not by itself specify behavioral color discrimination. Optical filtering, adaptation, and neural processing also matter.

• Many birds, reptiles, and fish have four cone classes, although ultraviolet sensitivity and behavioral tetrachromacy must be established for the species. Many mammals have two main cone classes. In humans and other catarrhine primates, duplication and divergence of an ancestral long-wave opsin produced distinct L and M pigments. Different primate lineages have different histories.

• Some mantis shrimp have twelve spectral receptor classes. Thoen and colleagues found relatively coarse wavelength discrimination in the species and task tested Thoen et al. 2014. This motivates alternative coding hypotheses; it does not prove a universal winner-take-all neural mechanism or poor performance on every color task.

Colored oil droplets in many bird and reptile cones filter light before absorption by the visual pigment. Different droplet types can narrow and shift effective spectral sensitivities; some droplets are nearly transparent. Filtering reduces overlap but also discards photons, creating a discrimination–sensitivity tradeoff. The effect must be computed from both pigment absorption and droplet transmission, not a pigment peak alone (Figure 3.6.14⛶ fullscreen).

The vertebrate visual-opsin family includes RH1, SWS1, SWS2, RH2, and LWS classes; duplication and loss differ across lineages (Figure 3.6.12⛶ fullscreen). This is not a tree of all animal photopigments: insect and cephalopod visual systems use other opsin lineages. The species explorer should distinguish pigment-template curves from measured whole-receptor sensitivities (Figure 3.6.13⛶ fullscreen) Govardovskii et al. 2000 Vision, "Evolution of Eyes".

The oil-droplet example separates two ways to design a receptor. Changing the photopigment changes which wavelengths it absorbs. Placing a filter in front of an existing pigment changes which wavelengths reach it. The effective sensitivity is the product of filter transmission and pigment sensitivity. A filter that suppresses the short-wavelength flank of a broad sensitivity can separate it better from a neighboring channel, but the rejected photons no longer contribute to detection. The same calculation is needed for a camera dye over a photodiode. A sharper spectral response is not a free increase in information under low light (Figure 3.6.14⛶ fullscreen).

Different sensitivities also mean different metamers. Two lights that match for us can differ to an animal with another set of receptors, while two lights that we distinguish can match for it. Adding a receptor class can preserve another independent spectral measurement, but the nervous system still has to use it. That is the importance of the mantis-shrimp result: a receptor inventory and a behavioral discrimination experiment measure different things.

3.6.8 Polarization: the axis we are blind to🔗⧉

Light carries more at each direction than an intensity and a spectrum. Its electric field oscillates in a particular orientation, and that orientation, the polarization, is a dimension our own eyes throw away almost entirely (recall polarization from Light and optics). Sunlight leaves the sun unpolarized, but the world polarizes it: Rayleigh scattering in the sky, reflection off water and leaves and cuticle, and scattering underwater all stamp a partial polarization onto the light that a suitable eye can read. Many animals read it, and put it to uses we can only approach with a filter screwed onto a lens.

A compass in the sky. Rayleigh scattering lays a spatial pattern of polarization across the blue sky, bands that wheel around the sun and move with it through the day. Insects read the pattern as a compass. A foraging honeybee or a desert ant holds its heading by the sky's polarization even when the sun itself is hidden behind a cloud or a dune, using a specialized strip of the eye, the dorsal rim area, whose receptors are tuned to the angle of polarization. It is celestial navigation with a visible sky pattern, run off a patch of blue.

Finding water, and cutting glare. Light reflected from a horizontal surface comes back strongly polarized in the horizontal direction, the very fact a photographer uses when a polarizing filter kills the glare on a pond or a windshield. Water beetles, dragonflies, and mayflies find water by exactly that horizontal polarization, diving toward any surface that reflects it, which is why they are now regularly fooled into laying eggs on car roofs, asphalt, and solar panels. The animal carries a permanent polarizer and a brain wired to what it reveals.

A private, high-contrast channel underwater. Below the surface, where color washes out with depth but polarization persists, the cephalopods excel. Octopus, cuttlefish, and squid are colorblind, and many species have polarization sensitivity, and they lean on it where color fails: to lift a silvery, transparent, or countershaded animal out of a flat background, since the camouflage that beats a brightness-and-color eye still disturbs the polarization of the light passing through or bouncing off it. Cuttlefish may even signal to one another with polarized patterns on their skin, a potential signaling channel; color deficiency alone does not imply polarization blindness. The mantis shrimp goes furthest, an animal with demonstrated circular as well as linear polarization sensitivity, perhaps for just such a private band (Figure 3.6.15⛶ fullscreen).

Aligned microvilli can make rhabdomeric receptors sensitive to polarization orientation. Neural comparisons can extract polarization information from differently oriented receptors. Vertebrate photoreceptors have different structures, but the ability to sense polarization is not exclusively invertebrate. Humans have weak polarization sensitivity, visible under suitable conditions as Haidinger's brush; it is not equivalent to a camera with a complete polarization analyzer.

A division-of-focal-plane polarization sensor places analyzers at several orientations over different pixels. Four linear-analyzer orientations estimate linear polarization components under assumptions about neighboring samples. They do not measure circular polarization without additional optics. This capture can help separate reflections or constrain surface shape, but it trades spatial sampling and light throughput for polarization information.

3.6.9 Temporal vision: flicker, and motion first🔗⧉

Animal temporal sensitivity varies with species, adaptation, stimulus area, contrast, and measurement method. Critical flicker-fusion frequency measures when a specified modulation becomes indistinguishable from a steady light. It is not a frame rate or a direct measure of how slowly time feels to the animal. Nor does a high threshold mean that a sample-and-hold television looks like a slideshow.

Fast visual responses can aid flight and prey tracking, while longer integration can improve detection in dim light. Compare temporal contrast-sensitivity curves under reported conditions rather than ranking species by one unqualified frequency. With that caveat, measured flicker-fusion thresholds span a wide range: roughly 50 to 60 Hz for a human under bright conditions, above 100 Hz for many small birds, and near 250 Hz or more for fast-flying flies, while nocturnal and deep-sea animals sit far lower. Across species, the threshold rises with metabolic rate and falls with body mass Healy et al. 2013. Chapter 3.4, Spatial and temporal vision distinguishes flicker, apparent motion, and display refresh.

Motion selectivity can begin early in visual processing. Classic frog-retina recordings revealed cells selective for particular small moving stimuli, but do not establish that frogs cannot perceive stationary scenes or inevitably starve beside motionless prey. Correlation-based models such as the Hassenstein–Reichardt detector compare neighboring signals with a temporal delay. They illustrate motion-sensitive computation; they are not identical to a general optical-flow estimator.

Event cameras provide a related engineering example: pixels report thresholded changes in log intensity asynchronously. They can offer high temporal resolution without conventional frame-exposure blur, but still have bandwidth limits, noise, thresholds, and finite latency. A stationary camera can emit events under flickering illumination even when nothing moves. Biological motion processing and event sensing are useful comparisons, not identical mechanisms.

Animal eyes show several ways to allocate sensitivity, spatial resolution, temporal response, and spectral information. The useful comparison is the task each system supports under its operating conditions, rather than a single ranking of better and worse eyes.

Consider two neighboring light sensors along the direction of travel. A bright feature moving from the first to the second produces similar temporal signals, but the second arrives later. Delay the first signal and multiply it by the second: at a suitable speed the responses overlap strongly. Reverse the delay-and-multiply arrangement and compare the two products. Motion in the opposite direction favors the other product. This explains the directional response of a correlation detector without asking it to identify the object. Its response depends on contrast, spatial pattern, and speed, so it is not a general displacement measurement. The comparison with optical flow is the use of spatial and temporal structure to infer motion, not an identity of algorithms.

An event camera offers an engineering comparison. Instead of reading the whole array at fixed frame times, it reports local brightness changes asynchronously, usually when a change in log intensity crosses a threshold. A static, noiseless scene produces little output; a moving edge generates events along its path. This can resolve rapid changes without the exposure integration of an ordinary frame, but thresholding, noise, finite bandwidth, and latency still limit the measurement. It is a useful comparison with change-sensitive biological vision, not a claim that an animal retina and an event sensor implement the same computation.