Gamut Mapping and Rendering Intents: What Happens When a Colour Does Not Fit

Every colour management problem eventually reduces to one question: what to do with a colour that the destination device cannot produce. A printer cannot make the saturated green of an emissive display; a display cannot make the deep blue of a print. The gamut is the set of colours a device can reproduce, a region in a perceptual colour space that is closed and irregular, and mapping into it is not a single operation but a family of them with different definitions of “best”. The four rendering intents named in the ICC specification are four consistent answers, and understanding what each preserves and destroys is the core of the topic.

The gamut boundary and why clipping is the naive answer

A gamut in a perceptual space such as Lab* is a closed volume whose boundary is a curved surface. A colour outside the boundary has to be moved somewhere inside, and the simplest rule is to move it to the nearest point on the boundary by some metric. In RGB terms that is a clamp: any channel above 1.0 or below 0.0 is set to the limit. A clamp is a projection in the wrong space, and its failures are structural.

The first failure is hue shift. Clamping the red channel of an orange that is too saturated leaves the green and blue unchanged and pins red to maximum, moving the colour toward yellow or white along a path determined by which channel clipped. A colour that clips in two channels moves toward the corner primary; a colour that clips in all three is white. The clamp’s behaviour is a function of the RGB basis, so the same out-of-gamut colour maps to different in-gamut colours in different RGB spaces, which is why two displays with different primaries show visibly different handling of the same saturated content.

The second failure is the loss of neighbourhood structure. A clamp is discontinuous in the derivative: two out-of-gamut colours very close to each other can map to the same boundary point if they differ only in the clipped channel, collapsing a smooth gradient in a saturated region into a flat patch. The same gradient, mapped by a method that preserves the ratio of the channels, would keep its structure and merely reduce in saturation.

The four intents and their definitions

The ICC specification names four intents, and each is a different answer to what to preserve.

Perceptual. Compress the whole image, not just the out-of-gamut colours, so that the full range of the source maps into the full range of the destination and the relative relationships between colours survive. A perceptual transform moves in-gamut colours too, because it is fitting a larger gamut into a smaller one smoothly rather than clipping the edge. The result is a pleasing image whose colours are all shifted slightly, and the cost is that colours that did fit are no longer exact, so a spot colour reproduced perceptually is not metrically correct.

Relative colorimetric. Clip only the out-of-gamut colours to the boundary and leave everything else unchanged, after a white-point adjustment. The in-gamut colours are reproduced as accurately as the device allows, and the out-of-gamut colours are projected to the boundary by a fixed rule. The result is exact for most of the image and wrong at the edges, where clipping produces flat saturated patches and hue shifts.

Saturation. Maximize chroma, at the expense of accuracy, favouring vivid, saturated reproduction over a match. This is the intent chosen for business graphics, charts, and diagrams, where a saturated red must stay a vivid red even if it is not the same red. Saturation and perceptual transforms are related; saturation pushes chroma outward where perceptual compresses the whole gamut.

Absolute colorimetric. Reproduce the source colours as physically as possible without white-point adjustment, so a source white prints as the destination’s white patch even when that patch is not the paper’s white. The intent is used for proofing, where the goal is to simulate one device on another exactly, including the colour of the source’s white point.

The white-point adjustment is the defining difference between relative and absolute colorimetric. Relative adjusts so that the source white maps to the destination white; absolute does not, so the source white reproduces as whatever the destination produces for those coordinates. A proof of a newsprint target on a bright display must be absolute to show the paper’s cast; a photograph printed for viewing must be relative to make white paper read as white.

The mechanisms: projection versus compression

Two families of algorithm implement these intents, and they correspond to the clip-versus-compress distinction.

Projection methods move each out-of-gamut colour to the boundary. The choice of projection determines the trade-off: projecting toward a fixed point on the lightness axis preserves hue but shifts lightness; projecting toward the lightness of the colour itself preserves lightness but shifts hue and chroma; projecting toward a point on the grey axis chosen by the source colour’s lightness preserves a compromise. The Lab* geometry makes these different, and the “nearest point on the boundary” in a Euclidean metric is not hue-preserving, because the gamut boundary’s shape is not a simple cone.

Compression methods fit the whole source gamut into the destination by a smooth function, typically compressing lightness and chroma together so that the source’s extremes map to the destination’s extremes. The advantage is the absence of flat patches: gradients stay smooth, and the structure of the image survives. The disadvantage is a loss of accuracy everywhere, since every colour moves.

Intent	Moves in-gamut colours	Preserves	Typical use
Perceptual	yes	overall relationships	photographic reproduction
Relative colorimetric	no	in-gamut accuracy	digital printing, display
Saturation	yes	vividness	charts, graphics
Absolute colorimetric	no (white included)	physical coordinates	proofing

The role of the profile and the PCS

A profile is the data structure that carries the transforms, and the profile connection space (PCS) is the intermediate colour space through which conversions travel — in the ICC model, conversions go from the source space to the PCS and then from the PCS to the destination, never directly. The PCS is what makes the four intents well-defined: each intent is a different pair of transforms through the same neutral space, so the source and destination profiles are independent and can be combined with any matching pair of intents.

The profile’s gamut boundary is encoded as a lookup table, not a formula. A profile contains a set of tables that map the PCS into device space and back, and the boundary is implicit in those tables: the set of PCS values that map to in-range device values. Out-of-gamut detection is therefore a table lookup, and the precision of the boundary is the precision of the table’s grid, typically 33 points per axis or fewer. A colour that falls between grid points near the boundary can be classified in-gamut or out-of-gamut incorrectly, which is a source of the small mismatches between colour management engines that nominally implement the same specification.

This structure also explains why the two intents differ only at the edges in principle but differ everywhere in practice. A perceptual table is a separate table built by compressing the gamut, so its in-gamut entries are already shifted; the relative colorimetric table leaves them alone. The engine’s job is to select which pair of tables to compose, and the selection is the intent.

Failures and their signatures

Flat saturated patches are the signature of clipping. A photograph of a saturated flower that shows a uniform red-orange region where there should be a gradient has been clipped, and the fix is a perceptual or compression mapping rather than a colorimetric one.

A visible white border on a proof, or a colour cast on paper, is the signature of a wrong absolute-versus-relative choice. Proofing that adjusts the white point hides the paper’s cast and makes the proof look better than the print; proofing that fails to adjust it on a normal reproduction makes white look dirty.

A hue that changes between the source and a soft-proof on screen is the signature of a projection chosen for a different trade-off, or of a perceptual transform whose compression is not hue-linear. Hue stability across the mapping is the same concern raised in hue shifts in tone mapping, and the same perceptual-space machinery — the choice of Lab* versus a more hue-linear space — is what determines whether a gamut mapping is hue-stable.

The last failure is a mismatch between the gamut mapping and the tone mapping that precedes it. A pipeline that tone maps an HDR scene into display range and then applies a gamut map has made two decisions about the same colours, and the order changes the result. Tone mapping first changes where the colours sit relative to the gamut boundary, then gamut mapping moves the stragglers; gamut mapping first, then tone mapping, compresses colours that the tone map would have brought into range anyway. The two operations are separated in description and entangled in effect, which is why tone mapping operators and this topic must be read together, and why the PQ pipeline of ST 2084 defines the brightness range within which the gamut question is then asked.

Why there is no single correct mapping

The reason four intents exist is that “best” is not a property of a colour but of a job. A colour that must match a brand specification wants relative or absolute colorimetric; a photograph wants perceptual; a chart wants saturation. The same source and destination pair, the same engine, and the same profiles produce four different results, all defensible, because the operation has a free parameter that is the definition of acceptable loss. Knowing which loss a given task tolerates — accuracy at the edges versus smoothness everywhere versus vividness — is the substance of choosing an intent, and there is no way to derive the choice from the colour data alone.