LightingExplainer

White lighting and grey values: why colour temperature is not enough

Two 6500 K white lights share one colour temperature yet can set grey values 10% apart; the spectral maths, and how to specify white vision lighting.

LumaSpec Vision10 min read Share on LinkedIn

On this page 9 sections
  1. 01Why do two 6500 K white lights give different grey values?
  2. 02What does colour temperature in kelvin actually measure?
  3. 03How does a monochrome camera turn a spectrum into one grey value?
  4. 04How large is the grey-value shift between two whites of the same kelvin?
  5. 05LED, fluorescent or halogen: which white belongs on a vision camera?
  6. 06Can white balance on a colour camera correct the difference?
  7. 07How do you specify white lighting so grey values stay repeatable?
  8. 08When should a monochrome camera drop white light altogether?
  9. 09Questions

Two bar lights arrive in one shipment. Both are labelled white, both rated 6500 K, and both pass incoming inspection. Mounted in turn over the same part at the same exposure, they set the mean grey value fifteen levels apart — far enough to flip a tuned threshold. Nothing is defective: colour temperature was never a complete specification of a light. If you integrate white lighting on colour or monochrome cameras, this page gives you the spectral reasoning behind grey-value shifts, a worked three-band example you can rerun with your own sensor data, and the specification habits that keep grey values stable from batch to batch.

Why do two 6500 K white lights give different grey values?

Because colour temperature records only the colour impression a source makes, not the spectrum behind it: two lamps that both match a 6500 K reference can distribute their power in completely different wavelength patterns, and your sensor weighs those patterns differently.

Inside a white LED, a narrow blue spike leaves the die and a broad yellow-orange band comes from the phosphor that converts part of it. A fluorescent tube emits discrete mercury lines, strongest in the green near 546 nm, softened by the tube's own phosphor coating. A halogen filament glows with a smooth curve that keeps climbing through red and into the infrared. All three can be graded 6500 K, because the grade asks one question only: does the overall colour impression match a heated reference at that temperature? It never asks where the energy sits.

Your camera does not perceive the impression. It integrates energy, wavelength by wavelength, and on a monochrome sensor that integration collapses into one grey number per pixel — a number that moves with the spectrum even when the label does not.

What does colour temperature in kelvin actually measure?

It is the temperature of a heated black body whose colour looks the same as the lamp's: around 2700-3000 K reads warm like an incandescent lamp, 5000-6500 K reads like daylight, and the number says nothing about what happens between the wavelengths.

The reference is Planck's black body, an ideal absorber and emitter whose spectrum depends only on temperature. A lamp's correlated colour temperature is the temperature of the black-body curve whose chromaticity best matches the lamp's. Two sources can land at the same point on the colour diagram while their spectra differ completely — one continuous, one built from lines and bands.

The usual companion number, CRI, scores how the source renders standard test colours for a human observer. Typical industrial white LEDs fall anywhere from the 70s to above 90, and both CRI and colour temperature are eye-weighted averages, so neither reveals the wavelength-by-wavelength detail your sensor responds to.

How does a monochrome camera turn a spectrum into one grey value?

It multiplies the light's spectrum by the part's reflectance and the sensor's sensitivity, wavelength by wavelength, and sums the result: the grey value is that integral, so any lamp that changes the spectrum changes the grey even at identical colour temperature.

grey value ∝ Σ SPD(λ) · R(λ) · QE(λ) · Δλ

SPD is the lamp's spectral power distribution — its power in each wavelength bin. R is how much the part reflects there, and QE is the sensor's quantum efficiency. A typical silicon machine-vision sensor responds from roughly 400 nm out past 900 nm, with sensitivity peaking in the 500-600 nm band; the exact curve belongs to your sensor's data sheet.

The part brings its own curve. Most grey-looking plastics and coatings reflect blue differently from red, so the lamp's weighting matters even on a part your eye calls neutral. Collapse the spectrum to one narrow wavelength — a single-colour LED — and the sum becomes a single product, which is why monochromatic light produces the most repeatable grey values.

How large is the grey-value shift between two whites of the same kelvin?

In the three-band worked example below, a blue-peak LED white and a green-weighted fluorescent white, both labelled 6500 K, set the same warm-grey part to grey 148 and grey 163 at 8 bits: a 10% shift, enough to break a threshold parked between them.

The shares, reflectances and sensitivities are illustrative values shaped like the real curves — LED blue peak plus phosphor band, fluorescent green weighting, sensor peaking near 550 nm — and the method is the point: run your own three-band sum with your lamp and sensor data.

Band LED white share Fluorescent share Part reflectance Sensor QE
450 nm (blue) 0.35 0.10 0.20 0.40
550 nm (green) 0.45 0.55 0.60 0.55
630 nm (red) 0.20 0.25 0.45 0.35

LED white: 0.35 × 0.20 × 0.40 = 0.0280, plus 0.45 × 0.60 × 0.55 = 0.1485, plus 0.20 × 0.45 × 0.35 = 0.0315. Sum: 0.208.

Fluorescent white: 0.10 × 0.20 × 0.40 = 0.0080, plus 0.55 × 0.60 × 0.55 = 0.1815, plus 0.25 × 0.45 × 0.35 = 0.0394. Sum: 0.2289.

The ratio is 0.2289 / 0.208 = 1.10: the fluorescent white drives the sensor 10% harder on this part at the same output rating. With exposure fixed so the LED white lands the part at grey 148 of 255, the fluorescent lands it at 148 × 1.10 ≈ 163. A threshold at 155 passes the part under one lamp and fails it under the other. Your own number will differ; the mechanism will not.

LED, fluorescent or halogen: which white belongs on a vision camera?

Each technology earns its keep somewhere: LED whites dominate machine vision for efficiency, shape and strobing; fluorescent whites still anchor colour-matching booths; halogen's smooth continuous spectrum keeps it alive in hyperspectral work.

White source Spectrum shape What it means on the line
White LED narrow blue peak near 450 nm plus a broad phosphor band through 500-700 nm efficient, compact, instant strobing; the exact spectrum follows the phosphor recipe and the bin
Fluorescent mercury lines near 436 and 546 nm plus phosphor bands, green-weighted large, even illumination; drive it at high frequency or it beats with your exposure
Halogen continuous black-body curve near 3000 K, strong red and near-infrared tail the smooth spectrum hyperspectral work wants; runs hot, ages redder, needs infrared management

Mains-driven fluorescents interact with shutter timing — the camera flicker guide shows the diagnosis. If your task borders on classifying materials by spectrum, halogen's continuum is why a spectral imaging validation plan still starts from a broadband source.

In an industrial cell, LED white usually arrives as a linear bar light or ring chosen for shape first and spectrum second:

LightingBar lights

Angled or low-angle light, any length

1,403 models · from $89.00

Can white balance on a colour camera correct the difference?

Partly: white balance rescales each colour channel to make a white target sit near neutral again, which repairs hue, but it cannot restore spectral detail the sensor never captured, and a monochrome camera has no per-channel gains to adjust at all.

Gains act on broad red, green and blue channels. If a lamp is deficient in a narrow band — a missing mercury line, a weak phosphor shoulder — no gain combination restores what one channel never separated. Colour cameras also expect an infrared-cut filter; without one, a red-heavy lamp pushes every channel at once and no balance recovers the colour.

On a monochrome camera the fix is recalibration, not balance: re-shoot a grey reference after any lamp change and re-set the threshold. Hold the camera still while you do it — lock the gains with the method in locking auto white balance, and take every auto mode off exposure as well, as in fixing brightness drift.

How do you specify white lighting so grey values stay repeatable?

Buy the white light by series and part number, drive it from a constant-current controller, shield it from daylight, and qualify every incoming batch against a grey reference target: then the grey values follow the part, not the shipment.

For a fixed linear station, a diffused white bar light is the carrier to start from:

Bar lightsBar light, 372 x 28 mm, White (LS-BAR-1276)LS-BAR-1276
Size
200-400 mm
Type
Diffused
Voltage
24 V
$439.00View partBuild a system with it

The LS-BAR-1276 is a 372 x 28 mm white diffused bar light that runs from a 24 V supply. Hold its output steady with a constant-current controller:

Light controllersConstant current controller, 4 ch, 120 W, 89 x 147 mm, for 24 V lights (LS-CTL-0180)LS-CTL-0180
Output
Constant current
Channels
4 channels
Control
Digital
Feature
Constant current
$359.00View partBuild a system with it

The LS-CTL-0180 provides four constant-current channels rated 120 W for 24 V lights. Browse the light controllers family or the controller finder when you need more channels or a network option.

When should a monochrome camera drop white light altogether?

When the task is contrast or defect detection on a monochrome camera, skip white: a single-colour LED collapses the spectrum to one wavelength, so lamp-to-lamp spectral differences disappear and contrast is set by reflectance at that wavelength alone.

Pick the colour from the part and the defect — the colour contrast guide and the lighting colour wheel walk that selection. Red or infrared also slips past coloured surface contamination that scatters white light.

White keeps the job in three cases. Colour inspection on a colour camera needs a broad, neutral spectrum — the colour inspection guide shows what raw RGB can and cannot carry. Multi-camera cells that share one fixture benefit from one common spectrum. And hyperspectral work wants the smooth continuum only a heated filament provides.

So the decision per case: monochrome contrast, go single-colour; colour grading, specify white by part number and qualify batches; spectral classification, stay broadband. If your task keeps white light, start from the light finder and go straight to configuration:

Build the whole station

The configurator matches a camera, lens, light, light controller and cables to each other, step by step, and lists only parts that fit what you already picked.

Start with the light

More lighting guides sit on the lighting topic page, and the team answers selection questions through the contact page.

Questions

Does a high CRI white LED keep grey values closer to daylight?

CRI rates how faithfully a source renders colours for human vision, using eye-weighted test colours. A CRI above 90 tells you the spectrum covers the visible band fairly evenly, which helps colour cameras, but it is still not a sensor-weighted specification: two high-CRI whites can differ at 450 nm or 630 nm, exactly where your sensor's response is steep. Treat CRI as a screening filter, then compare measured spectra or a grey-target shot.

Do I need an IR-cut filter when a halogen light feeds a monochrome camera?

Yes in almost every case. A halogen filament at typical voltages radiates a large share of its power beyond 700 nm, and a silicon sensor keeps responding out to roughly 1000 nm. Without a filter that infrared reaches the sensor: grey values climb, and because the lens focuses different wavelengths at different depths, best focus shifts too. Fit an IR-cut filter, or accept that your greys include a band you cannot see.

How often should I re-check grey values as the light ages?

Tie the check to your change control rather than to a fixed calendar. Shoot a matte grey reference card at installation, save the mean grey value, and repeat the shot whenever the light, controller or exposure changes, plus on a schedule that suits the line, for example weekly. An LED white drifts bluer and dimmer as its phosphor ages; a halogen lamp reddens as the filament thins. A drift of more than a few grey levels at fixed exposure is your trigger to re-threshold or replace.

Can I combine a white bar light with coloured lights in one station?

Yes, and sequencing is often better than mixing. Drive each light from its own controller channel and fire them in sequence with the exposure, so no frame ever contains two spectra; grey values then stay attributable to one lamp. Mixing white and colour in the same exposure recreates the spectral weighting problem described above, only with a recipe you control.

Why do two orders of the same 6500 K bar light still give different greys?

LED manufacturing sorts finished dies and phosphor lots into bins, and the bins bound colour and flux, not identical output. Two production runs of one part number can therefore sit at opposite ends of the bin, which shows up in your image as a brightness or tint step. Qualify each delivery with a grey reference shot, order enough spares from one lot for the cell's life, or ask about matched-lot supply when the tolerance is tight.