Seeing colors¶

Color is not 'in' light. Light is one wave (E + B oscillating, ~380–740 nm visible to humans); 'color' is what 3 cone types report after the retina projects that continuous spectrum onto a 3-D space (trichromacy). Different people sample the spectrum slightly differently (8 % of men are red-green colorblind; ~0.1 % of women are tetrachromats and may see a 4th channel). Mantis shrimps have ~12 cone types but discriminate worse than us — more receptors ≠ more colors. Most of the electromagnetic spectrum is invisible (visible band is one 0.0035 % slice of EM that earth's atmosphere happens to pass and chlorophyll happens to reflect). Vision is filtering all the way down: filter wavelength → filter via three cone sensitivity curves → filter via opponent-process encoding → filter via top-down expectation. You cannot fully backtrack a percept to a physical spectrum (metamerism: many spectra give one color). Animals have senses we don't (electroreception, magnetoreception, polarization, IR); humans have ~5 textbook senses but functionally 10–20 (proprioception, interoception, equilibrioception, nociception, thermoception, time). Mastery follows the same training curve as any skill: minutes for the obvious channels, decades for the subtle ones.

🌱 seedling tended 2026-05-17 research senses vision color optics perception

flowchart LR
  em[EM spectrum · single wave family] --> vis[visible slice · 380–740 nm]
  vis --> eye[cornea · lens · iris]
  eye --> ret[retina · rods + 3 cone types]
  ret --> opp[opponent channels · L–M, S–LM, L+M]
  opp --> lgn[LGN · thalamus]
  lgn --> v1[V1 · primary visual cortex]
  v1 --> v4[V4 · color area]
  v4 --> exp[top-down expectation · prior]
  exp --> percept[reported color]

L0 — TL;DR (≤5 lines)¶

Light is a single wave family (the electromagnetic field). The eye samples a tiny slice (~380–740 nm) through 3 cone types (L, M, S) — color is a 3-D projection of an infinite-D spectrum, so many physical spectra collapse to one perceived color (metamerism). Some people sample the slice differently (~8 % of men are red-green colorblind; ~0.1 % of women may be functional tetrachromats). Mantis shrimps have 12+ cone types yet discriminate fewer colors than humans — extra receptors are used for fast classification, not finer hue. Vision is filtering at every stage (atmosphere → optics → cones → opponent channels → top-down prior). You cannot fully back-track a percept to a physical spectrum.

L1 — Overview¶

Core question¶

What is color, why do different observers see different ones, how much of the world is actually visible at all, what gets filtered out at each stage, and how many distinct senses can a person learn to use?

Why it matters¶

Color is a model, not a measurement. Two screens, two paints, and a rainbow can all evoke "red" while emitting completely different spectra. Designers, scientists, and clinicians who treat color as ground truth get fooled.
Most of the spectrum is invisible. Radio, microwave, infrared, UV, X-ray, gamma — all the same wave. The human window is a 0.0035 % slice. Other animals see other slices, and instruments see all of them.
Vision is filtering, not recording. Each stage discards information. You can't reconstruct the input from the output. Knowing where the filters are tells you where the illusions live.
Sense count is contested. The "five senses" model is folk taxonomy. Modern neuroscience names 10–20 receptor systems; some animals add more (electroreception, magnetoreception, polarization). Some are trainable; many are not consciously accessible at all.

Mermaid map (L1)¶

flowchart LR
  sun[solar / source spectrum] --> atm[atmosphere · absorbs UV, IR bands]
  atm --> obj[object · reflects subset of incident]
  obj --> eye[cornea + lens + iris]
  eye --> rod[rods · ~120 M · scotopic]
  eye --> cone[cones · ~6 M · 3 types L M S]
  cone --> opp[opponent encoding · R-G, B-Y, light-dark]
  rod --> opp
  opp --> lgn[LGN · thalamus relay]
  lgn --> v1[V1]
  v1 --> v4[V4 color area]
  v4 --> prior[top-down expectation · context · memory]
  prior --> exp[experienced color]

Skeleton sub-claims¶

Light is one wave (E + B oscillating). "Color" is the receiver's report.
Human vision samples ~380–740 nm with 3 cone types: L (~564 nm), M (~534 nm), S (~420 nm).
3 cones → 3-D color space → metamerism (different spectra, same perceived color).
Individual variation is real: cone shifts, density, density of macular pigment, lens yellowing with age.
~8 % of men, ~0.5 % of women are color-deficient (X-linked); a few women carry a 4th cone.
More cones ≠ more colors. Mantis shrimp has 12+ photoreceptor types but coarse discrimination.
The visible band is a tiny window on the EM spectrum; the rest needs instruments.
Color is computed, not received: opponent-process encoding starts at the retina.
Top-down priors close the loop (the dress; lightness constancy; chromatic adaptation).
Animals have senses humans lack (electroreception, magnetoreception, polarization-vision, IR pit, lateral line, echolocation).
Humans have many more than 5 senses by any modern count; mastery varies.

L2 — Deep dive¶

1. Light is one wave¶

Visible light, radio, infrared, ultraviolet, X-ray, and gamma are the same phenomenon — an oscillating electric + magnetic field — at different frequencies. The single equation is Maxwell's; the only thing that changes across the spectrum is wavelength (and therefore photon energy E = hν).

Band	Wavelength	Photon energy	Detector
Radio	km – 1 m	~10⁻⁹ eV	antenna
Microwave	30 cm – 1 mm	10⁻⁶ – 10⁻³ eV	antenna / bolometer
Far IR	1 mm – 30 µm	~0.05 eV	bolometer
Near IR	1.4 – 0.74 µm	~1 eV	InGaAs sensor
Visible	740 – 380 nm	~1.7 – 3.3 eV	cones / rods / CCD / CMOS
Near UV	380 – 200 nm	~3 – 6 eV	photomultiplier / silicon
X-ray	10 nm – 10 pm	~100 eV – 100 keV	scintillator
Gamma	< 10 pm	> 100 keV	scintillator / semiconductor

The visible band is ~0.0035 % of the spectrum by log frequency. The reason this slice is "visible" is not accidental:

Atmospheric window: Earth's atmosphere is transparent at ~300 nm – ~1 µm and again at radio. UV, mid-IR, X-ray, gamma are largely absorbed.
Solar peak: the Sun's blackbody peak (5778 K) is ~500 nm — green-yellow, right in the window.
Water transparency: liquid water absorbs ~10⁻⁴ in the visible and orders more elsewhere. Eyes are made of water.
Photon energy fits chemistry: visible photons (1.7–3.3 eV) match outer-electron transitions in organic pigments. Lower energy can't drive isomerization; higher tears bonds.

So vision evolved into the slice where the source had energy, the medium passed it, and the chemistry could absorb it without damage. Different ecologies tuned the slice slightly (insects shift bluer into UV; pit vipers add a separate IR organ; deep-sea fish keep only the blues).

2. From photons to cones — the periphery¶

Light passes through:

flowchart LR
  ph[incoming photon] --> co[cornea · most refraction]
  co --> aq[aqueous humor]
  aq --> ir[iris · aperture · pupil 2–8 mm]
  ir --> le[lens · accommodation · UV-yellowing with age]
  le --> vi[vitreous]
  vi --> ret[retina · upside down · rods + cones]
  ret --> rpe[RPE · pigment epithelium · absorbs leftover · recycles 11-cis retinal]

In the retina:

Cell	Count	Sensitivity	Role
Rod	~120 million	peak 498 nm; one pigment (rhodopsin)	scotopic (dim light), achromatic, peripheral, motion
S cone	~6 % of cones	peak ~420 nm (blue)	one of three color channels; sparse, not in fovea center
M cone	~33 %	peak ~534 nm (green)	red-green channel input
L cone	~64 %	peak ~564 nm (yellow-green)	red-green channel input

Three observations:

"Red" cones aren't red-tuned. L peaks at 564 nm (yellow-green). Red percept comes from L responding more than M — opponent comparison, not raw tuning.
The fovea has no S cones in its center. Tiny fine-blue detail is invisible; the visual system fills it in.
Cone-to-ganglion ratio is ~1:1 in the fovea, ~100:1 in the periphery. Foveal vision is high-resolution; peripheral is sensitive to motion and dim light at the cost of detail and color.

A photon striking a cone isomerizes 11-cis retinal to all-trans inside an opsin (a GPCR — same superfamily as olfactory receptors; see OLFACTORY-SENSES §1). That triggers a transducin cascade closing a cGMP-gated Na⁺ channel and hyperpolarizing the cone. Photoreceptors are unusual in that light reduces their firing.

3. From cones to color — opponent encoding¶

Cone outputs do not go to the brain as (L, M, S). They are immediately re-combined into opponent channels at the retinal ganglion cells (Hering 1892; Hurvich & Jameson 1957):

Channel	Formula	Percept axis
Luminance	L + M	black ↔ white
Red-green	L – M	red ↔ green
Blue-yellow	S – (L + M)	blue ↔ yellow

This is why:

You cannot perceive "reddish-green" or "yellowish-blue" — the axes are mutually exclusive at the encoding level. (You can see "yellowish-red" = orange, or "bluish-red" = purple, because those mix orthogonal axes.)
A flag stared at for 30 s leaves an after-image in opponent colors.
Color and edges share the same channel mathematics; this is why edge contrast distorts color (Cornsweet illusion).

The compression is severe: ~100 million photoreceptors → ~1 million ganglion-cell fibers in the optic nerve. The retina is performing the first 4–5 stages of visual processing locally, sending only the projection — not the raw spectrum.

4. Why color is not in the world — metamerism¶

A cone integrates ∫ S(λ) R(λ) dλ over its spectral sensitivity R(λ). Three cones produce three numbers. An infinite-dimensional spectrum S(λ) collapses to a point in 3-D LMS space.

So infinitely many physical spectra produce the same (L, M, S) triple — metamers. A pure 589 nm yellow laser and a mixture of 530 nm green + 670 nm red can both look the same yellow. Your screen exploits this: it emits only three narrow peaks (R, G, B) and can fool the eye into any color in its gamut.

Consequences:

You cannot back-track perception to physics. Given a color, you cannot determine the spectrum that produced it. The forward map (spectrum → color) is well-defined; the inverse is one-to-many.
Hyperspectral instruments are needed to recover the spectrum. Humans cannot do it; the information was thrown away at the retina.
Pigment matching vs light matching diverges. Two paint chips that match under sunlight may differ under fluorescent light (illuminant metamerism).
Camera ≠ eye. Even a 3-channel camera with different cone curves produces a different LMS that the brain has to color-correct.

This is the central fact: color is a lossy projection, and the loss is irreversible at the eye.

5. Individual variation — do different people see different colors?¶

Yes, in several measurable ways.

Source of variation	Effect	Prevalence
OPN1LW / OPN1MW polymorphism	shifts L and M cone peaks by up to ~5 nm	~50 % of population carries one variant
Anomalous trichromacy (protan, deutan, tritan)	one cone shifted toward another → poor R-G or B-Y discrimination	~6 % of men, ~0.4 % of women
Dichromacy (one cone missing)	true colorblindness	~2 % of men
Monochromacy / achromatopsia	no cones or one cone only	~1 in 30 000
Functional tetrachromacy	4-cone female carriers of color-deficient sons; an extra cone with shifted peak; some show 4-D discrimination	~0.01 % – ~0.1 % of women (Jordan 2010 identified one confirmed cDa29)
Macular pigment density	yellow filter in fovea; varies 5× between individuals	continuous; affects blue discrimination
Lens yellowing with age	absorbs short wavelengths; "blues fade"	progressive after ~40
Cataract surgery removing UV filter	some people then see ~360 nm UV ("bee purple")	post-surgical
Synesthesia (grapheme-color)	letters perceived as colored	~2–4 % of population
Linguistic / cultural priors	named-color boundaries shift discrimination (Berlin & Kay; Russian "blue" split)	weak but measurable

So "do we see the same red" has a literal answer: no, but mostly close enough that we agree on names. The 6 % of men with red-green deficiencies live in a measurably reduced color world; the rare tetrachromats may live in one with a 4th axis we cannot share.

The famous dress (#thedress, 2015) is an inferential difference, not a receptor one: same retinal input, different priors about illumination. People who unconsciously assumed warm light "saw" blue+black; people who assumed cool light "saw" white+gold.

6. Which animals see more colors?¶

The naive ranking by cone-type count is misleading.

Animal	Photoreceptor classes	Sees	Note
Most placental mammals	2	dichromat	dog, cat, mouse — sees blue + yellow-ish; cannot distinguish R from G
Old World primates / humans	3	trichromat	our 3-cone system re-evolved via gene duplication ~30 Mya
Some New World monkeys (female heterozygotes)	3 (females) / 2 (males)	dichromat or trichromat	the trick that may have triggered our own
Marsupials, rodents	often 2	dichromat	+ some UV sensitivity in many rodents
Birds	4	tetrachromat	+ oil-droplet filters narrow each cone's bandwidth
Reptiles	4–5	tetra / penta	same trick as birds
Honeybee	3	trichromat UV-blue-green	sees UV; nectar guides on flowers are UV patterns invisible to us
Butterfly (some)	6–9	super-tetrachromat?	discrimination not always tested
Mantis shrimp (Odontodactylus scyllarus)	12–16 + polarization	does not discriminate better than us	uses receptors as fast spectral classifiers, not for fine hue (Thoen et al. 2014)
Goldfish	4	tetrachromat	UV-sensitive
Deep-sea fish	often 1	monochromat	blue-tuned; no point in color where only one wavelength reaches
Pit viper / boa	IR organ separate from eyes	adds thermal imaging	not part of "color" but expands the sensed band
Snake (some)	2	dichromat	hunting by IR + chemical, not color

The mantis shrimp result is the single most counter-intuitive fact: 12 receptor types but worse discrimination than 3. The reason is that mantis shrimps appear to use their receptors as a lookup table (this band on, that band off → known prey class), not as inputs to a fine comparison. More dimensions can mean faster classification at coarser resolution, not finer.

The general rule: number of receptors sets the dimensionality of the color space; downstream wiring decides how finely it is sliced. Bees do better than humans at UV; humans do better than bees at red-green discrimination because the L-M opponent channel is finely tuned.

7. How much can be seen at all?¶

Hard bounds:

Spectral: ~380–740 nm, ~0.0035 % of EM by log frequency. Outside needs instruments (radio antenna, IR camera, X-ray detector).
Spatial: ~60 cycles per degree at the fovea — about a hair at arm's length. Periphery is ~10× worse.
Temporal: ~60 Hz flicker fusion in bright light, ~15 Hz in dim. Faster events are blurred or missed (the wagon-wheel effect).
Dynamic range: ~10⁹:1 across full dark adaptation (rods take 30 min to adapt); ~10³:1 instantaneously.
Field of view: ~210° horizontal (both eyes), ~135° vertical.
Color discrimination: ~1 million colors in optimal conditions (trained observer, side-by-side comparison); ~10 named categories in free naming.

Soft bounds — what gets filtered:

Atmosphere removes UV-C, much UV-B, mid-IR, X-ray, gamma.
Optics (cornea, lens) absorbs UV below ~340 nm, IR above ~1.4 µm. Diffraction-limits resolution at the pupil; aberrations distort.
Iris caps aperture (2–8 mm).
Retinal pigment absorbs anything that survived.
Cone sampling discards spectral detail (metamerism).
Opponent encoding compresses 6M cones → 1M fibers, projecting onto 3 axes.
Saccadic suppression blanks vision during each ~30 ms saccade (~3 per second). You don't see motion blur during eye movement.
Blind spot at optic disc (~15° temporal) — filled in by prior.
Inattentional blindness — unattended objects are not encoded (gorilla-on-the-basketball-court experiment, Simons & Chabris 1999).
Top-down priors override the input under ambiguity (the dress, the shadow-checkerboard illusion).

You see ~10⁷ bits/s entering the retina, ~10⁶ bits/s entering V1, ~10² bits/s into awareness. The funnel from photon to "what you can report" is roughly 10⁵×.

8. Is it mainly filtering?¶

Yes, and at multiple cascaded stages — but with constructive steps mixed in (filling-in, lateral inhibition, edge enhancement, opponent-color formation, motion energy, depth from disparity).

A useful frame: vision is a generative model that uses the image to constrain a hypothesis, not a recording. The hypothesis is mostly prior; the image is mostly evidence selecting among priors.

flowchart LR
  in[photons] --> filt[filter cascade]
  filt --> ev[evidence]
  prior[strong prior on scene] --> hyp[hypothesis space]
  ev --> hyp
  hyp --> percept[reported percept]

This explains:

Why illusions exist (the prior dominates when evidence is ambiguous).
Why color constancy works (the inferred illuminant is divided out).
Why training changes perception (priors update).
Why two people can see the same scene differently (different priors).

9. Can you back-track the percept?¶

No, not fully.

Forward (each step loses information):

spectrum S(λ) ─→ LMS triple ─→ opponent triple ─→ ganglion spikes ─→ V1 features ─→ V4 color ─→ verbal report

Stage	What is lost
Spectrum → LMS	infinite → 3 dimensions; metamerism
LMS → opponent	linear remix; reversible in principle
Opponent → spikes	rate code; ~6 bits per ganglion per second; lossy
V1 → V4	non-linear feature pooling; not invertible
V4 → report	constrained by language; ~10 base colors named freely

You can run the forward direction with full physics (and the screen industry does — ICC profiles, sRGB / Display P3 / Rec.2020 gamuts). You cannot run the inverse cleanly. To "back-track" you need either multi-spectral instruments (skip the cone bottleneck) or assumptions (known illuminant + known surface reflectance class) — which is exactly what color-constancy algorithms add.

This irreversibility is the same shape as compression in general (see UNIVERSE-EVOLUTION-AS-COMPRESSION). Color is a 3-D lossy code of an infinite-D signal; you cannot recover the signal from the code alone.

10. Do animals have more senses?¶

Yes — several whole modalities humans don't have.

Sense	Mechanism	Animal	Notes
Electroreception	ampullae of Lorenzini detect bio-electric fields	sharks, rays, platypus, electric fish	< 1 µV/cm sensitivity
Magnetoreception	cryptochromes in retina; magnetite in beak	birds, sea turtles, monarchs, possibly cows	navigation; debated mechanism (quantum radical pair)
Polarization vision	photoreceptors aligned to E-field axis	bees, octopus, mantis shrimp, many insects	reads sky polarization for navigation
Infrared (separate organ)	pit organ — thermal radiation, not photons	pit vipers, boas, some pythons	~0.003 °C resolution
Echolocation	active sonar — emit click, time return	bats, dolphins, some shrews; trained humans (Kish, Underwood)	spatial map from sound
Lateral line	hair cells in canals detect water flow	fish, amphibians	hydrodynamic "vision"
Geomagnetic + electric + chemical fusion	combined	salmon (home migration)	multi-modal beat anything single
UV vision	extra cone	bees, birds, reindeer (see polar-bear urine on snow)	also seen by humans post-aphakia

Plus mechanisms with sharper resolution than the human version (elephants' infrasound, dogs' olfaction, bats' echolocation).

11. How many senses do humans have?¶

The "five senses" (sight, hearing, touch, taste, smell) is Aristotle. Modern neuroscience names many more receptor systems. A working list:

Sense	Receptor	Pathway	Note
Vision	rods + cones	retina → LGN → V1	this page
Hearing	hair cells (cochlea)	auditory nerve → MGN → A1	20 Hz – 20 kHz; declines with age
Olfaction	~400 ORs	OB → piriform → OFC	OLFACTORY-SENSES
Gustation	taste buds (5 modalities)	VII, IX, X → solitary nucleus	sweet, sour, salty, bitter, umami
Touch (mechano)	Meissner, Pacinian, Merkel, Ruffini	DCML / spinothalamic → S1	pressure, vibration, stretch
Proprioception	muscle spindles, GTOs	DCML → S1 / cerebellum	limb position without looking
Equilibrioception	vestibular (otoliths + canals)	VIII → brainstem / cerebellum	head orientation, acceleration
Thermoception	TRPV1 (heat), TRPM8 (cold)	spinothalamic	also activated by capsaicin / menthol
Nociception	A-δ + C fibers	spinothalamic	"pain" distinct from touch
Interoception	vagal afferents, baroreceptors, chemoreceptors	NTS → insula	hunger, thirst, heartbeat, breath
Time	suprachiasmatic + striatal timing nets	many	circadian + interval (~ms to days)
Magnetoreception (vestigial?)	cryptochromes?	unclear	weak, contested in humans
Pheromones	VNO; mostly non-functional in humans	accessory olfactory bulb?	contested

A common modern textbook count is 9–12; with finer splitting (itch as separate from pain, separate cold/warm channels, proprioception split into joint/muscle/tendon) you get 20+.

12. How many senses can you master?¶

Mastery here means: become aware of the channel and use its information to guide behavior reliably.

Trainable to expert level in months to years:

Vision sub-skills: edge / motion / color discrimination, depth, faces.
Hearing sub-skills: pitch, timbre, location, language phonemes, echolocation (Kish, Underwood: months of practice).
Olfaction: perfumer / sommelier reach ~5000 named odors after ~5 years (Hummel; see OLFACTORY-SENSES §10).
Proprioception: athletes, dancers, surgeons train sub-mm precision for years.
Vestibular: pilots, gymnasts learn to dissociate from visual cues.
Interoception: meditators measurably improve heartbeat detection accuracy after ~8 weeks (Khalsa 2008).
Thermoception: cold-exposure adepts (Wim Hof) shift autonomic set-points.
Time perception: musicians, athletes hit ~10 ms timing in trained ranges.

Hard limits:

You cannot grow more cone types (gene therapy aside — animal trials exist for protan correction).
You cannot extend your visible band by training. UV / IR vision requires instruments.
You cannot grow electroreception, magnetoreception, or echolocation organs; you can substitute (sensory-substitution devices: BrainPort, vOICe — slow but real). Eagleman 2020 shows haptic vests can encode novel channels (stock-market dynamics, drone telemetry) reliably after weeks.

A reasonable estimate: a focused individual can reach competent discrimination in 3–5 senses in their lifetime, expert in 1–2. A polymath's edge is often moving information between modalities (seeing music, hearing geometry) — not raw within-channel expertise.

13. Practical takeaways¶

Trust your color naming for communication, not for measurement. Use a spectrophotometer if it matters (paint, screen calibration, art conservation).
Calibrate displays. sRGB / Display P3 / Rec.2020 are not the same 3-D space. Photo work without calibration is unreliable.
Test for color deficiency in children — Ishihara takes 1 minute. Diagnosis early changes course choices (electrician color codes, pilot certifications).
Treat illusions as data, not failures. Each illusion identifies a prior the visual system is using.
Lighting matters for everything color-related: paint chips, food appearance, dating-app photos, medical assessment of skin.
For low-light tasks, use red light to preserve rod adaptation.
Blue light suppresses melatonin (intrinsically photosensitive retinal ganglion cells, ipRGCs — a 4th non-image-forming photoreceptor in humans, ~2002 discovery). Late-evening screen use shifts circadian phase. See BODY-AS-ENGINE.
Polarized sunglasses cut sky/water glare by filtering one polarization axis — a sense bees use natively and humans access via technology.
Cataract surgery removes the UV-blocking lens. Many post-op patients report seeing UV ("an extra purple"). The receptors were always there; the filter was upstream.

Open questions¶

How common is functional tetrachromacy? Jordan 2010 identified exactly one (cDa29) of 25 carriers with measurable 4-D discrimination. Larger studies (Bosten 2020+) suggest the trait is rarer than carrier status alone implies. What does the 4th dimension actually feel like?
Mantis-shrimp color use: if 12 receptors aren't for discrimination, what are they optimal for? Fast prey ID? Object categorization?
Human magnetoreception: Wang & Kirschvink 2019 reported EEG responses to rotated magnetic fields in humans. Replication is sparse. Real or noise?
ipRGCs and mood: blue-blockers / chronotherapy effect sizes are small and contested. Bright light for SAD works (~70 % response); evening-screen impact is real but smaller than popular discourse claims.
How many distinguishable colors really? Pointer's gamut estimates the surface-color gamut at ~3 million colors; trained side-by-side discrimination reaches ~10 million; free naming bottoms at ~10–20.
Sensory substitution ceiling: BrainPort gives blind users ~20×20 tactile resolution from camera. Is there a cortical reorganization limit, or is it engineering?
Why exactly this visible band? The atmosphere / water / chemistry explanation is necessary but is it sufficient? Could a different evolutionary path settle on a different window?

References¶

Hecht, S. (1924). The visual discrimination of intensity and the Weber-Fechner law. J. Gen. Physiol.
Hering, E. (1892). Outlines of a Theory of the Light Sense. (English trans. Hurvich & Jameson, 1964.)
Hurvich, L. M., & Jameson, D. (1957). An opponent-process theory of color vision. Psychological Review.
Jordan, G., Deeb, S. S., Bosten, J. M., & Mollon, J. D. (2010). The dimensionality of color vision in carriers of anomalous trichromacy. J. Vision.
Thoen, H. H., How, M. J., Chiou, T.-H., & Marshall, J. (2014). A different form of color vision in mantis shrimp. Science.
Berlin, B., & Kay, P. (1969). Basic Color Terms: Their Universality and Evolution. (And the long debate that followed.)
Berson, D. M., Dunn, F. A., & Takao, M. (2002). Phototransduction by retinal ganglion cells that set the circadian clock. Science. (Discovery of ipRGCs.)
Macpherson, F., ed. (2011). The Senses: Classic and Contemporary Philosophical Perspectives. (On how to count senses.)
Simons, D. J., & Chabris, C. F. (1999). Gorillas in our midst: sustained inattentional blindness for dynamic events. Perception.
Eagleman, D., & Perrotta, M. (2020). The future of sensory substitution, addition, and augmentation. Frontiers in Systems Neuroscience.
Wang, C. X., et al. (Kirschvink) (2019). Transduction of the geomagnetic field as evidenced from alpha-band activity in the human brain. eNeuro.
Pointer, M. R. (1980). The gamut of real surface colours. Color Research & Application.
Bosten, J. M. (2020+). Modern psychophysics of tetrachromacy carriers.

Inspiration sources¶

Hermann von Helmholtz — Handbuch der physiologischen Optik (1867): the foundation. Trichromacy as quantitative theory.
Ewald Hering — opponent-process; the right idea before the receptors were known.
Edwin Land — retinex theory; color constancy as a computation.
David Hubel & Torsten Wiesel — V1 organization; Nobel 1981.
Margaret Livingstone — Vision and Art (2002): how painters reverse-engineer the visual system.
Oliver Sacks — The Island of the Colorblind (1997): living without cones.
David Eagleman — Livewired (2020): sensory substitution and the trainability of new channels.
Justin Marshall — mantis-shrimp psychophysics; the receptor-count myth-buster.