Luminance vs. Brightness: The Meter and the Eye

Luminance is a physical quantity: the light a surface sends toward your eye, per unit area, per unit solid angle, with each wavelength weighted by the CIE V(λ) curve, the standardized sensitivity of daylight vision peaking at 555 nm in the green. The unit is the candela per square metre (cd/m², informally the nit; cinema still speaks in foot-lamberts, 1 fL = 3.426 cd/m²). That V(λ) weighting is the whole difference between photometry and radiometry: a watt of deep red carries far less luminance than a watt of green, because eyes are not flat-response detectors. Past that one standardized curve, though, luminance is fully objective: a spectroradiometer aimed at a patch returns the same number for any observer, in any room, at any hour, and it adds linearly: double the photons, double the luminance. Everything a calibrator reports (black level, peak white, EOTF tracking) is luminance.

Brightness is not a quantity; it is a sensation. In the CIE's own vocabulary it is the attribute of visual perception by which an area appears to emit more or less light. Its relative sibling, lightness, is brightness judged against a similarly illuminated white: "how gray is this?" rather than "how much light is this?". No instrument measures either. They exist only inside an observer, assembled from luminance plus adaptation state, surround, recent visual history, and unconscious assumptions about the illumination. And brightness is neither linear nor additive. The full moon, at roughly 2,500 cd/m², looks brilliant in a night sky; the same 2,500 cd/m² is an unremarkable patch of daylit cloud. Nothing about the moon changed: your adaptation did. Luminance is what the display does; brightness is what you experience.

The compression has been measured for 190 years. Weber (1834): the just-noticeable difference is a constant fraction of the stimulus, about 1–2% for luminance in good light. Fechner turned that into a logarithmic law; Stevens' mid-century scaling experiments settled on a power law with an exponent near ⅓. The CIE froze the working version in 1976 as lightness: L* = 116·(Y/Yn)^⅓ − 16, relative to the adapted white Yn. The consequences run every display decision: a patch at 18% of white's luminance reads as half as bright (L* = 50 at Y = 18.4%, the photographer's mid-gray). Doubling the light makes a patch only ~26% brighter (∛2 = 1.26). To look twice as bright, a surface needs roughly eight times the luminance. The meter and the eye genuinely disagree, and both are right, about different things.

The cube root is only the start; brightness is also computed relationally. Simultaneous contrast: an identical patch looks brighter on a dark surround than a light one, as the photo pair on this page demonstrates with your own eyes and a checkable number. Adaptation: the visual system slides a ~13-stop operating window across a ~24-stop total range (the dynamic-range module is built on exactly this), so the same luminance lands on different parts of the response depending on what you looked at last: clicking between the two rooms above is a scale model of that slide. And surround changes apparent contrast wholesale: Bartleson and Breneman showed in 1967 that images viewed in a dark surround look flatter, which is why cinema masters to a ~2.6 gamma, living-room video to ~2.4, and bright-office sRGB to ~2.2. Same pixels, three "correct" renderings, because brightness, not luminance, is what the audience receives.

The terms form a ladder from physics to perception, and each rung answers a different question: RADIANCEW·sr⁻¹·m⁻² · raw energy, no observer LUMINANCEcd/m² · energy × V(λ); a standard eye, still objective BRIGHTNESSpercept · absolute "how much light does this seem" LIGHTNESSpercept · relative to white; L* approximates it Two habitual confusions to retire: illuminance (lux) is light arriving at a surface, luminance is light leaving it toward you, and meters for one cannot stand in for the other. And luma Y′ is not luminance Y: it is a weighted sum taken after gamma, a signal-domain fossil covered in the Y′CbCr module. Even the monitor's "brightness" knob is historically mislabeled: it moved black level, while "contrast" moved white.

Suppose codes were spaced evenly in luminance. The step ramp above shows the result: nearly half the codes would describe whites the eye can barely tell apart, while the shadows, where a 1% luminance change is visible, would band grotesquely. Eight bits spaced linearly in light are not enough for a clean gradient; eight bits spaced perceptually are. That is the entire reason transfer functions exist: legacy gamma is, by historical accident, a rough inverse of the eye's cube root, and PQ was engineered directly from the Barten contrast-sensitivity model so every code step sits just below one JND from 0.0001 to 10,000 cd/m². The transfer-function module lets you quantize a real scene both ways and watch the linear encoding fall apart in the shadows.

Spec sheets sell luminance; viewers buy brightness, and the exchange rate is the cube root. A 1,000-nit display against a 100-nit reference is 10× the light but only about 2.15× the apparent brightness (∛10). Going from 1,000 to 4,000 nits (a heroic engineering leap) buys roughly ×1.6 more. This is why HDR's impact does not come from the whole picture getting brighter: full-screen peaks would just drag your adaptation up (and trip the panel's ABL current limiter anyway). It comes from small, brief highlights (speculars, flames, the Strip at night) sitting far above an average picture level that stays SDR-ish, so your adaptation stays low and the highlight lands on the steep part of the response. HDR done well spends its nits where brightness is cheapest: against a dark adapted eye.

Calibration targets are always luminance, because luminance is the only thing two people can agree on: 100 cd/m² reference white for SDR grading, 48 cd/m² (14 fL) on a cinema screen, absolute PQ code-to-nits tracking for HDR. But the viewer receives brightness, so the other half of the job is controlling the context that manufactures it: a D65 bias light at ~10% of peak white behind the monitor to anchor adaptation, neutral dim walls, no stray glare re-lighting the blacks. Skip that and a perfectly measured monitor still reads wrong: blacks look milky in a bright room, contrast looks crushed in a black one. If a client says "it looks dim," check the room before the display: the meter may already be right. Book a calibration →