Is 50% Michelson contrast a 2:1 luminance ratio?

No. With a positive minimum, it corresponds to 3:1.

Does multiplying every luminance by two change it?

No, provided the scaling is uniform and nothing clips.

Can two equally contrasted patterns be differently visible?

Yes. Their spatial frequency, size, viewing conditions, and other properties can differ.

AI Glossary / Computer vision and multimodal AI / Beginner

Michelson contrast

Definition

Michelson contrast measures the difference between a pattern's maximum and minimum luminances relative to their sum: C = (Lmax − Lmin)/(Lmax + Lmin). It is often useful for repeating light-and-dark patterns, such as sinusoidal gratings, where both extremes are meaningful parts of the stimulus.

What it means

Two patterns can have different overall brightness yet the same relative modulation. Michelson contrast describes that modulation with a dimensionless number rather than a difference expressed in luminance units.

For nonnegative luminances and a positive denominator, the value ranges from zero to one. Zero means the two extremes are equal. One occurs when the minimum is zero and the maximum is positive. Multiplying by 100 expresses the result as a percentage.

The number is a physical measurement, not a complete prediction of how visible, attractive, or readable a pattern will be. Spatial scale, viewing conditions, display behavior, and the observer all matter. Do not confuse this measure with every software control named contrast or every accessibility contrast ratio.

How it works

Specify the pattern and measurement region, then determine its maximum and minimum luminances in consistent units. Subtract the minimum from the maximum, add the two values, and divide the difference by the sum.

For an ideal sinusoidal luminance pattern written as L(x) = Lmean[1 + C cos(2πfx)], its extremes are Lmean(1 + C) and Lmean(1 − C). Substitution into the formula returns C. Here f describes the spatial frequency and x the position along the pattern.

Use measured luminance or an appropriate linear-light representation when that is the quantity being claimed. Gamma-encoded pixel values are not automatically proportional to emitted luminance. Also state whether extrema are theoretical values, measured patch values, or extreme pixels from an image; noise and clipping can make these different measurements.

Three practical examples

1. Vision experiments: a researcher specifies the modulation of a grating while separately controlling its spatial frequency, mean luminance, size, and presentation time.

2. Camera testing: an imaging team measures the captured light - dark modulation of a repeating target under a defined setup and compares it with the target's known input modulation.

3. Display evaluation: a designer checks whether glare raises the darkest part of a displayed pattern enough to reduce its measured modulation, even when the digital image file has not changed.

Think of it this way

Think of the number as the size of a wave relative to its average water level: a larger rise and fall relative to the baseline corresponds to greater modulation.

The analogy is limited. Luminance cannot ordinarily be negative, the formula uses the two extrema, and a complicated image need not behave like a regular wave. The metric does not describe where all its details occur.

Worked example

Suppose a measured striped pattern has Lmax = 90 cd/m² and Lmin = 30 cd/m². Its Michelson contrast is (90 − 30)/(90 + 30) = 60/120 = 0.5, or 50%.

Now double both luminances: 180 and 60 cd/m². The result is 120/240 = 0.5 again. Uniform multiplicative scaling preserves this contrast if there is no clipping or other change to the relationship.

Instead, add 20 cd/m² to both original levels, representing an idealized uniform veiling contribution. The new extrema are 110 and 50. Contrast becomes 60/160 = 0.375, or 37.5%. The absolute difference remains 60, but the raised baseline reduces the normalized modulation.

For the original pattern, the maximum-to-minimum luminance ratio is 90/30 = 3:1. This is a different numerical description from 50% Michelson contrast; neither should be relabeled as the other. For a positive minimum, the relation is C = (r − 1)/(r + 1), where r is that ratio.

If both extrema are zero, the formula is 0/0 and undefined. A program should handle that case explicitly rather than presenting it as an ordinary computed contrast value.

Illustrative case study

Fictional case: an educational display team compares a striped demonstration shown in a dark room and near a bright window. The file's black and white code values remain unchanged, but measured luminances show that reflections have raised the darker stripes.

The team calculates Michelson contrast for the same region and records the ambient setup alongside the result. It also checks whether the stripe spacing and viewing distance allow the pattern to be resolved.

The report separates the measured modulation from a claim about readability or user performance. It uses the calculation to explain one effect of the environment, rather than treating a single number as a complete assessment of the teaching display. No measured outcome from a real organization is implied.

Common misconceptions and mistakes

Do not assume this contrast is unchanged by adding brightness: multiplicative scaling and additive offsets behave differently.

Avoid measuring an entire photograph by just its brightest and darkest pixels and calling that a comprehensive image-quality score. One bright reflection or dark outlier can dominate the extrema while saying little about useful detail elsewhere.

Do not compare results with different spatial frequencies or exposure processing as though only contrast changed. Sharpening, clipping, tone curves, and blur can alter measurements. When evaluating an imaging system, document the target, sampling, calibration, and analysis method.

Business use

Michelson contrast gives imaging and display teams a clear, reproducible description of modulation for suitable patterns. It helps turn vague comments such as the stripes look washed out into a measurable observation.

Use it within a defined test rather than as a universal purchasing score. A business decision about displays, cameras, or visual interfaces may also require resolution, noise, color, viewing-angle, and task-performance evidence. For AI image evaluation, match the metric to the actual feature the model must detect.

Check your understanding

A pattern has extrema 80 and 20 cd/m². What is its Michelson contrast?

Answer: (80 − 20)/(80 + 20) = 60/100 = 0.6, or 60%. These extrema alone do not tell you whether the pattern will be resolved at a particular viewing distance.

Related terms

Spatial frequency
Gamma encoding
Dynamic range
Luminance
Weber contrast
contrast ratio
modulation transfer function
contrast sensitivity

Learn next

Calculate Michelson contrast for a sinusoidal grating, then compare its purpose with Weber contrast for a small target against a background. Next explore how a modulation transfer function tracks changes in modulation across spatial frequency.

Sources

CIE e-ILV, luminance contrast: https://cie.co.at/eilvterm/17-22-091 MIT computational photography glossary, Michelson contrast: https://people.csail.mit.edu/fredo/comp-photo-book/glossary.html The primary educational and terminology pages were opened for the formula and measurement context. Calculations, caveats derived from the formula, and scenarios are original explanations.

Previous: Metropolis within Gibbs Next: Micro-thread (multi-core)