Delta-E is the number quoted when a display is described as colour-accurate. It is also one of the easiest numbers to quote misleadingly, because the result depends on which formula was used, which target was chosen, and what content was measured.

This article covers what the figure means, how a calibration workflow produces it, and where colour accuracy stops being the answer to a display problem.

What delta-E measures and which formula applies

Delta-E expresses the perceptual difference between a measured colour and a reference colour as a single number. The measurement itself is a set of tristimulus values or spectral data; the formula converts that into a difference.

Several formulas are in use, and they do not agree. Older formulas treat differences in lightness and chroma similarly across the colour space; newer ones correct for the fact that the eye is more sensitive to some differences than others. Two laboratories can therefore produce different delta-E values from the same measurement if they use different formulas.

The practical rule is to state the formula with the number. A statement such as “average delta-E below 2” is incomplete without it, and comparisons between suppliers are unreliable unless the formula matches.

Choosing a reference target and gamut

Delta-E needs a reference. In display work that reference is usually a colour space with a defined white point, gamma and primaries – sRGB, Adobe RGB or DCI-P3, for example.

The choice changes the result substantially. A panel measured against sRGB may report a large error on saturated colours that fall outside the sRGB gamut, even though the panel reproduces them as well as it can. The same panel measured against its own native gamut may report a small error. Neither number is wrong; they answer different questions.

For an industrial product, the reference should follow the content: an interface defined in sRGB should be evaluated against sRGB, while a display used to judge imagery should be evaluated against the space that content is authored in.

Instrument and environment requirements

Colour work starts from the parameters on the specification sheet, and the definitions in reading an LCD datasheet are worth settling before calibration begins.

Calibration targets also differ by application. A display used to judge imagery is usually calibrated to a standardised colour space with a defined white point and gamma, while an interface display is often calibrated only to a consistent appearance across units. Choosing the lighter target where it is sufficient keeps both the measurement effort and the risk of over-specifying the panel in proportion.

Colour measurement needs a colorimeter or spectroradiometer. A colorimeter is faster and adequate for calibration against a known display type; a spectroradiometer is slower and more accurate across different panel technologies, which matters when verifying several suppliers.

The environment matters as much as the instrument. Ambient light reaching the screen changes the measured colour, and light reaching the instrument changes the reading. Measure in controlled conditions, with the instrument in contact with or at a fixed distance from the screen, and record the ambient condition.

Calibration status is part of the evidence. A measurement taken with an instrument whose calibration has lapsed is not comparable with one from a calibrated instrument, and the difference can exceed the tolerance being tested.

Warm-up and stabilisation

Colour drifts as a display warms up, and the drift is not the same for every colour. White point moves, and the primaries move slightly differently, so a calibration performed too early produces a profile that is correct only for the first few minutes.

The practical approach is to warm the display with the content it will actually show, then measure repeatedly until the values stabilise. Record the warm-up time, because a later verification performed after a different warm-up period will not reproduce the same numbers.

Calibration workflow step by step

Two practical details determine whether the workflow produces a usable result. The first is bit depth: corrections applied through an eight-bit pipeline can introduce banding in smooth gradients, which is why calibration on an eight-bit path usually limits how much correction can be applied without visible artefacts. The second is where the correction is stored. A profile held on the host is lost when the host is replaced; a correction applied in the display’s own processing survives, but is harder to verify.

Set the display’s own controls first. Brightness, contrast and colour temperature presets on the panel or the host should be fixed at the values the product will ship with, because calibration adjusts the output pipeline rather than the panel’s internal settings.

Then characterise: measure the primaries, the white point and a set of greys. Build the correction – a lookup table for greys and a matrix for colour – and apply it. Re-measure to establish how the correction behaves rather than introducing banding at the extremes.

Where the device supports it, store the result as a profile on the host. Where it does not, the correction may need to be applied by the graphics pipeline or at the application level, which is a design decision rather than a measurement one.

Recording calibration proof for later verification

Where the screen is used to judge images rather than to display an interface, the panel technology itself sets part of the limit; the comparison of IPS and TN panels covers colour depth and viewing behaviour. Verification is a separate activity from calibration and should use different content. Calibration optimises against a set of patches; verification checks the result on colours the calibration did not target, including skin tones, greys and the interface’s own palette.

Report the result as a distribution rather than a single figure: average error, maximum error and where the maximum occurs. A display with an average below 2 and a maximum of 8 on one primary is a different product from one with a uniform error of 2, even though the averages match.

Matching multiple displays in one system

Where several displays appear side by side – a control desk, a multi-screen instrument, a video wall – matching matters more than absolute accuracy. The eye is very good at detecting a difference between adjacent screens and poor at judging absolute colour.

Two approaches are used: calibrate each display to the same target, or calibrate one and match the others to it. The second is usually faster and produces a better visual match, because it corrects the differences between units rather than their absolute errors.

Drift and recalibration intervals

Set a trigger as well as an interval. Beyond the calendar, three events should prompt a check: a change of panel supplier or batch, a repair or replacement of the display, and any change to the ambient conditions in which the product operates. A recalibration interval without triggers usually means the check happens after the problem has been noticed rather than before.

Displays drift with age, and the drift is accelerated by heat and by high brightness settings. In a fixed installation, annual verification is a reasonable starting point; in a product that runs continuously at high brightness, more frequent checks are warranted.

Set the interval from the application’s tolerance rather than from a generic rule. A display used for colour judgement needs a tighter interval than one used to show status text, where a slow white-point shift will never be noticed.

When colour is not the real problem

Several problems are described as colour issues and are not. A low-contrast image in a bright room is a reflection problem, not a delta-E problem. Uneven colour across a large screen is a uniformity problem. Colours that differ between two screens showing identical content are often a gamma or profile mismatch rather than a panel error. And a screen that looks wrong only at an angle is a viewing-angle issue.

Measuring colour before identifying which of these is present produces a technically correct number that does not solve the user’s problem. For the uniformity case, the article on mura and luminance uniformity covers the diagnostic path.

Workflow checklist

Step	What to record
Reference target	Colour space, white point, gamma and formula used
Instrument	Model, calibration status and measurement geometry
Environment	Ambient light and the display’s own settings
Warm-up	Duration and the content displayed during it
Calibration	The correction applied and where it is stored
Verification	Average and maximum error, and the content used
Interval	Recalibration period and the trigger for an early check

If a colour problem is proving hard to pin down, describe the content, the ambient light and whether the issue appears on one screen or across several – those three answers separate a colour error from a uniformity, reflection or viewing-angle problem.

Frequently asked questions

What delta-E is good enough?

It depends on the content. Average values below 2 are often quoted as visually accurate for general work, but a single saturated colour with a large error can matter more than the average in a status-coded interface.

Do we need a spectroradiometer?

Not always. A colorimeter is adequate for calibrating against a known display type; a spectroradiometer is valuable when verifying several panel technologies or when the highest accuracy is needed.

Why do two calibrated displays still look different?

Usually because they were calibrated to different targets, with different gamma settings, or at different brightness levels. Matching adjacent screens to one reference usually resolves it.

Calibration also decays. A display that is accurate on the day it is set up drifts with temperature and with backlight ageing, so the useful record is not only the measured value but the interval at which it is worth repeating.