Human Color Vision: Cones, Opponent Signals, and the Brain

Color vision is not an RGB camera. The retina samples a spectrum with overlapping L, M, and S cone fundamentals, adapts those cone signals, subtracts them into opponent channels, and ships parallel pathway signals toward cortex. This bench lets you move those stages together.

Color vision is a pipeline. Light enters the eye, gets focused on the retina, triggers a chemical reaction in a cone outer segment, propagates through several layers of retinal neurons that already begin processing it, travels down the optic nerve to the lateral geniculate nucleus (LGN) of the thalamus, fans out to primary visual cortex (V1), and continues through specialized areas - most famously V4 - that bind it together with motion, shape, depth, and memory into the unified experience you call "seeing a red apple."

Six features of this pipeline keep returning across every subsection below. They are worth naming early.

The retina is often described as "the back of the eye where light hits photoreceptors," but that drastically undersells it. The retina is a piece of brain tissue that developed - embryologically - as an outpouching of the diencephalon. It contains five principal cell classes arranged in three distinct nuclear layers, plus two layers of synaptic connections. The signal moves backward through this structure: light passes through every other layer before reaching the photoreceptors at the very back.

Inside each photoreceptor's outer segment sits a stack of thousands of membrane discs. Embedded in those discs are tens of millions of opsin molecules - the light-sensitive proteins. Each opsin is bound to 11-cis retinal, a vitamin-A derivative that twists shape when it absorbs a photon. The shape change kicks off a multi-step amplification cascade that closes sodium channels and hyperpolarizes the cell - photoreceptors are unusual in that they respond to light by going more negative, not more positive.

Three details matter for understanding everything that follows. First, the cascade amplifies enormously: a single photon can trigger the closure of hundreds of channels via the G-protein chain. Second, photoreceptors are tonically active in darkness - they constantly release glutamate when no light is present, and light reduces that release. Third, the response is graded, not all-or-nothing: more photons produce stronger hyperpolarization, encoding intensity as analog voltage.

The cones in your retina are not arranged in a tidy grid. The fovea contains a roughly hexagonal packing of L and M cones, with S cones essentially absent at the very center and appearing in a sparse irregular distribution beyond. The relative numbers of L : M : S vary substantially between individuals - typical ratios are around 64 : 32 : 4 - yet people across this range report extremely similar color experiences. The brain compensates for whatever mosaic it has.

A neuron's receptive field is the region of visual space where light affects its firing rate. Retinal ganglion cells - the only cells whose axons leave the eye - have famously circular, center-surround receptive fields. The center is excited by light; the surround is inhibited. This means the ganglion cell fires hardest not for uniform light, but for contrast: an edge or a spot against a different background.

Color enters this story when the center and surround draw from different cone types. An L-center / M-surround ganglion is excited by red light in the center and inhibited by green light around it. It fires hardest for red spots on green backgrounds - exactly the kind of opponent red-green discrimination Hering predicted in 1872. The retina does not just report cone signals; it computes opponent differences before the data leaves the eye.

The retina sends three principal streams to the brain, each carried by a distinct ganglion-cell class. Together they account for the vast majority of fibers in the optic nerve - though more than twenty additional minor classes have been identified, many with specialized roles like circadian rhythm or melanopsin-based light detection.

Roughly one million ganglion axons exit each eye through the optic disc, producing the famous blind spot. They cross at the optic chiasm - where fibers from the nasal half of each retina cross to the opposite hemisphere, so each side of the brain sees the contralateral visual world - and arrive at the lateral geniculate nucleus (LGN) of the thalamus.

The LGN is a six-layered structure, plus thin koniocellular sublayers between them. The two ventral magnocellular layers receive parasol input (luminance, motion); the four dorsal parvocellular layers receive midget input (red-green, detail); the koniocellular sublayers receive bistratified input (blue-yellow). The layers are organized retinotopically - neighboring locations in visual space map to neighboring cells - and each LGN cell receives input from only one eye.

The LGN does more than relay. It modulates incoming signals with attention-related feedback from cortex, gain control, and binocular interactions. But its overall role is to maintain the parallel streams set up in the retina: by the time signals leave the LGN for V1, color, motion, and detail are still travelling on distinct pathways.

Primary visual cortex (V1, also called striate cortex or Brodmann area 17) sits at the back of the brain in the calcarine sulcus. About a third of the cerebral cortex's surface area is devoted to vision in primates, and V1 is the first cortical stage. It is famously organized in layered orientation columns that respond to edges of particular tilts - the discovery for which Hubel and Wiesel received the 1981 Nobel Prize.

Color processing in V1 happens largely in patches called blobs - concentrations of cytochrome-oxidase-rich cells visible with the right histological stain. The blobs contain double-opponent cells whose receptive fields are organized like a color version of the retinal center-surround: a red-on/green-off center plus a green-on/red-off surround. This cell type, first described in detail by Conway in 2001 in macaque V1, is the cortex's basic unit for representing the color of a region rather than the color of a point.

V1's color processing is also where orientation and color first interact. Cells outside blobs respond to oriented edges with weak color tuning; cells inside blobs respond to color contrast with weak orientation tuning. The two forms of information remain partially segregated, and only begin to merge in downstream areas.

Signals leave V1 along two large streams. The dorsal stream heads upward through V2 and V3 into parietal cortex - the "where" pathway for spatial attention and motion. The ventral stream moves downward through V2 and V4 into inferior temporal cortex - the "what" pathway for object recognition, including color. V4, first emphasized by Semir Zeki in the 1970s, is a major hub on the ventral path with strong color selectivity.

Damage to V4 - usually from stroke - can produce cerebral achromatopsia: patients lose color vision while keeping intact form, motion, and brightness perception. The world becomes a movie shot in black and white. Crucially, this is different from congenital colour blindness (cone-pigment defects); it is a loss of the cortical experience of color while the receptor stage may still be intact.

A clean cone signal is not the end of the story. Cones, ganglion cells, LGN cells, and cortical neurons all adapt to the prevailing input - they rescale their sensitivity so that whatever the dominant signal is becomes the new "neutral." Walk from a tungsten-lit room outside into noon sun, and within seconds the world stops looking blue. The reverse happens going back inside, faster than you can notice.

Human vision encodes color and luminance at different spatial resolutions. Luminance contrast supports very fine spatial detail - we can resolve gratings finer than 30 cycles per degree of visual angle. Chromatic contrast - especially blue-yellow - resolves much coarser detail, perhaps a tenth as fine. Engineering uses this all the time: JPEG, MPEG, and video broadcasting throw away most of the chrominance bandwidth while keeping luminance intact (the famous 4:2:0 subsampling).

Six questions on the biology of color vision. Wrong answers come with explanations and a pointer back to the relevant section.