Changing the reference white
Why vision preserves a sense of white under changing illumination, how von Kries, Bradford, CAT02 and other CATs model that behaviour, and where adaptation actually lives in a colour pipeline.
Chromatic adaptation tries to answer a mathematical question: if the observer adapts to a different white, how should the coordinates of all other colours change so that they are perceived as corresponding colours? It is not a creative look and not simply a Tint control.
Quick reference
Reinterpretation of colour coordinates between two adaptation states.
The basic idea of independent gain on receptor-like channels.
Historically important in ICC/D50 workflows.
CIE reviews a family of models rather than one eternal universal CAT.
The interpretation of colours relative to a different adapted white.
Not a creative tint, not gamut mapping and not a substitute for identifying the source colour space.
Move from a warm room to daylight at a window and the spectrum reflected by a white object changes, yet after adaptation the brain again tends to perceive it as neutral. Colour constancy is not perfect, but the visual system partially compensates for the illuminant. Chromatic-adaptation models try to predict corresponding colours between different adaptation states.
Suppose one RGB space uses D65 and another system expects D50. We can convert RGB to XYZ, but XYZ describes the stimulus; it does not automatically change the observer's adaptation state. If the same XYZ is simply placed in a system with another neutral reference, neutrals and colour relationships may not appear as intended. A CAT inserts a reinterpretation step relative to a source white and a target white.
The classic von Kries hypothesis reduces adaptation to scaling three receptor-like channels relative to the source and target whites. In a modern matrix workflow this often looks like XYZ being transformed into an LMS-like space, each channel multiplied by its own white-balance ratio, and the result transformed back. The strength of the idea is simplicity; the limitation is that real vision is more complex than three independent gain controls.
The Bradford transform grew out of corresponding-colour experiments and became especially common in ICC and desktop-colour workflows. In practice it uses a matrix that maps XYZ into a special cone-like space, applies diagonal scaling from source to target white, then maps back. Bradford is not 'universally correct by definition', but historically became a very important interoperable choice.
CIE 160:2004 compares many chromatic-adaptation transforms and makes clear that the problem is not solved by one eternal matrix. CAT02, CMCCAT2000, CMCCAT97 and other models were built from different corresponding-colour datasets and appearance models. Resolve's menu of Methods is therefore not a collection of stylistic presets; it exposes different mathematical adaptation models.
XYZ Scaling simply scales the XYZ components according to source white → target white. It is useful as an understandable baseline model, but the XYZ axes are not direct physiological cone channels and this approach generally models colour appearance less well than specialised CATs. It is best understood as a simplified reference idea.
ICC v4 uses a D50 Profile Connection Space: data with another white chromaticity must be adapted to D50 inside the profile or transform. The user may therefore never see a separate Bradford button even though white-point adaptation is already happening. Likewise a managed video pipeline may include the necessary adaptation as part of a larger input/output transform.
Camera or RAW White Balance usually corrects the estimated illuminant of a particular scene. Chromatic adaptation in colourimetry translates colours between reference adaptation states. Gamut Mapping solves a different problem entirely: what to do with colours that do not fit the target gamut. All three may exist in one pipeline, but they are not interchangeable.
Source Illuminant defines the original neutral reference, Target Illuminant the destination, Method the CAT model, and Current Color Space/Gamma tell the effect what signal context the values are in. If CST, RCM, ACES or another transform has already performed the required adaptation, a separate FX can cause double processing. Build the pipeline logic first, then choose the tool — not the other way around.
The right question when choosing a CAT is not 'which one looks nicer?' but 'which source/target adaptation conditions am I modelling, and is another part of the pipeline already doing this?'
Technical definitions are anchored to standards bodies and primary specifications wherever that distinction matters.
CIE review of experimental data and multiple CAT models including CAT02/CMCCAT families.
A real-world architecture where chromatic adaptation is built into a D50 PCS workflow.
ICC describes Bradford-based correction when converting D65 data to the D50 PCS.
