Is it the same as white balance?

The ideas are related, but white balance is a broader imaging operation. Its illuminant estimation and channel processing need not implement a particular von Kries transform.

Why are three separate gains used?

They represent separate adjustments of the three response channels in the model's chosen adaptation space.

Do all von Kries-type transforms give identical results?

No. The selected coordinate transformation, white points and adaptation assumptions influence the result.

AI Glossary / Computer vision and multimodal AI / Intermediate

Von Kries adaptation

Definition

Von Kries adaptation is a model of chromatic adaptation in which the three cone-response channels are adjusted by separate multiplicative gains. In color-processing applications, a von Kries-type transform maps colors between adaptation conditions by scaling responses in a chosen cone or cone-like space, usually using source and destination white points.

What it means

A sheet of white paper can continue to look approximately white under different lighting even though the light entering the eye changes. Chromatic adaptation contributes to this relative stability of color appearance.

The von Kries idea describes part of that adjustment through three independent channel gains. It is useful because the computation is simple and can be incorporated into color-management systems. It remains an approximation rather than a complete account of color perception.

The coordinate system matters. A diagonal scaling in cone-response space generally becomes a full matrix transform when expressed in another color space. Multiplying the three numbers of an arbitrary RGB file is therefore not automatically the same operation.

How it works

A typical computational procedure begins with color values and source and destination white points expressed consistently, often in CIE XYZ. A matrix converts them into the selected adaptation space.

For a full white-point adaptation, each channel gain is the destination white response divided by the source white response in that space. The gains are applied to the stimulus channels, and an inverse matrix converts the result back.

The complete transform has the form: convert into the adaptation space, apply diagonal gains, then convert out. Different choices of adaptation space can produce different corresponding-color predictions.

This white-point calculation assumes a specified adaptation situation. Partial adaptation, surround, luminance and other appearance effects can require additional modeling. It is also distinct from estimating the scene's illuminant: a transform can use a supplied white point without explaining how that white point was obtained.

Three practical examples

A color-management developer applies a documented adaptation transform when converting color values between reference whites.
An imaging researcher compares predicted corresponding colors under two lighting conditions and tests how well different adaptation models match observations.
A photography workflow uses a calibrated color-processing pipeline to reduce an unwanted lighting cast, while distinguishing illuminant estimation from the subsequent color transform.

Think of it this way

Think of adjusting three measuring channels so that a reference sample produces the intended reference readings. Other samples are then transformed using the same channel adjustments.

The analogy is limited because color appearance involves more than three independent controls. The chosen channels are also not necessarily the displayed red, green and blue values. The adjustment's meaning depends on the representation and viewing assumptions.

Worked example

Calculating channel gains

Use an invented cone-response example with source white L, M and S values of 100, 80 and 20. The target white in the same coordinate system is 100, 100 and 100.

The three gains are 100 / 100 = 1, 100 / 80 = 1.25 and 100 / 20 = 5. Applying them maps the source white exactly to the target white: (100, 80, 20) becomes (100, 100, 100).

Now consider a stimulus with responses (20, 20, 10). Applying the same gains gives (20, 25, 50). These are transformed responses in the chosen space, not display-ready RGB values or a claim that the color has become white.

If the source S-channel white estimate were incorrectly entered as 25, the third gain would be 4, giving a stimulus S response of 40 instead of 50. The calculation would be internally consistent with the wrong input, yet answer a different adaptation question.

A real workflow must also perform the appropriate coordinate conversions and maintain compatible scales. The example isolates the diagonal-gain step so that its assumptions and sensitivity are visible.

Illustrative case study

Illustrative case study - fictional

A fictional catalog studio sees color mismatches after combining images from two capture setups. An engineer notices that one processing path applies a white-point correction directly to encoded image channels while another uses the intended color-space conversions.

The team standardizes the pipeline, documents its white points and adaptation transform, and evaluates reference patches under controlled viewing conditions. It also discovers mixed lighting in some photographs, for which one global white-point adjustment is insufficient.

The response is to address both the processing mismatch and the capture conditions. The case does not claim that a single transform restores every object's true color or removes the need for a viewing test.

Common misconceptions and mistakes

Von Kries adaptation is not a complete model of human color appearance.
Cone-response or cone-like channels should not be confused with arbitrary image RGB channels.
Mapping the source white to the target white does not guarantee perfect appearance matches for every color.
Knowing how to apply an adaptation transform does not establish the correct illuminant estimate.
One global correction may be unsuitable when different parts of a scene are illuminated differently.

Business use

The business value lies in consistent, documented color reproduction across workflows. A small implementation mismatch can affect product images, design proofs or comparison studies even when each individual image looks plausible.

Record the input color space, encoding, white points, adaptation model and output representation. Check reference samples rather than judging only a neutral patch. Separate errors in capture, illuminant estimation, adaptation and display. For color-critical work, assess the final viewing conditions and recognize that a mathematical white-point match alone is not a complete quality test.

Check your understanding

A transform maps a source white to the desired target white. Does that prove every color will look identical under both viewing conditions?

Answer: No. The white-point match is one constraint, while color appearance and the model's approximation involve additional factors.

Related terms

chromatic adaptation
colour constancy
white balance
LMS colour space
white point
Bradford transform

Learn next

Repeat the worked example with a different source-white estimate and compare the resulting stimulus responses. Then examine a documented XYZ-to-adaptation-space conversion to see why channel gains cannot be transferred unchanged to arbitrary RGB data.

Sources

Sources: Colour developers, von Kries adaptation implementation and documentation: https://colour.readthedocs.io/en/develop/_modules/colour/adaptation/vonkries.html ; Sharma and Trussell, Digital Color Imaging, chromatic-adaptation discussion: https://www.hajim.rochester.edu/ece/sites/gsharma/papers/dciip97.pdf . Relevant channel-scaling implementation, white-point inputs and model limitations were reviewed. The illustrative response values are invented.

Previous: Voiced sound Next: Von Mises-Fisher distribution