ICC profiles are usually classified in different ways and, if these levels are not separated, it is easy to confuse the function that a profile fulfills within a color management workflow with the mathematical structure used to build it. This article focuses on the latter issue, especially on the options available in ArgyllCMS and RoughProfiler, but it is advisable to start by distinguishing both planes.
A profile can belong to a certain class (input, monitor, output, or one of the processing classes defined by the ICC) and, at the same time, be constructed using a matrix and tone curves, multidimensional lookup tables, or a combination of both strategies. None of these decisions alone guarantees that the profile is “better”: its quality depends on the characterization, measurement conditions, the device, and the intended use.
ICC Profile Classes
The three basic classes of device profiles are input, monitor, and output profiles. Input profiles describe how a capture device, such as a camera or scanner, converts device signals to a profile connection space or PCS. Monitor profiles describe the response of a display. Output profiles characterize printing devices and allow conversion from the PCS to the ink or dye combinations required by the printer.
In addition to these basic classes, the ICC specification defines DeviceLink profiles, color space conversion profiles, abstract profiles, and named color profiles. DeviceLink profiles store a pre-calculated transformation between specific spaces or devices; abstract profiles perform adjustments within the PCS; color space profiles describe conversions between spaces not associated with a physical device; and named color profiles manage color libraries with specific denominations.
Soft proof is not a profile class. It is an on-screen simulation of the appearance an original would have after being reproduced on an output device under certain conditions. At least the monitor profile, the output profile, and the conversion options chosen in the software are involved in performing it.
Matrix-Based Profiles
Matrix profiles are especially common in RGB devices. Their basic structure relates the three device values to the PCS, usually XYZ, through a 3×3 matrix. Each column of that matrix expresses the contribution of one of the device primaries to the XYZ coordinates.
Before applying the matrix, tone response curves or TRCs may intervene. These curves describe the behavior of each channel and allow approximating the relationship between the device’s numerical value and its colorimetric response. A matrix profile does not directly model complex interactions between channels or behaviors of a device with more than three colorants, but its simplicity provides smooth, compact, and easy-to-interpret transformations.
In ArgyllCMS, matrix profiles contain a colorimetric transformation; they do not incorporate the perceptual or saturation tables that may be present in CLUT-based profiles. Therefore, it is not necessary to say that they “only support relative colorimetric,” but rather to explain that their capacity to encode transformations dependent on the conversion intent is more limited.
Matrix Only

The Matrix Only option uses only the 3×3 matrix. It can be useful when the device response sufficiently approximates the chosen model and a simple transformation is sought. It should not be interpreted as a universal recommendation for “perfectly linear” sensors: the relevant linearity is that of the data entering the profiling process, after considering black levels, white balance, encoding, and any prior processing.
Gamma + Matrix and Single Gamma + Matrix

In these profiles, a gamma-type tone curve is added before the matrix. The Single Gamma variant uses a single curve for all channels; the ordinary variant allows an independent curve for each channel. A single curve may be sufficient in very balanced devices, but if midtone dominants appear or if channels have differentiated responses, it is advisable to use separate curves or a more flexible model.
The gamma curve does not assume that the sensor is “perfectly logarithmic” or “perfectly linear.” It is a functional approximation of the tonal response; its usefulness should be verified through profile validation and not by a general rule based solely on the device type.
Shaper + Matrix and Single Shaper + Matrix

A shaper curve allows a shape more flexible than a simple gamma and can better adjust complex tonal responses. This does not mean it is always more accurate: the quality of the result depends on the characterization chart, the quantity and distribution of patches, measurement repeatability, noise, and device stability.
Shaper curves mainly correct the tonal response of each channel. If the device presents significant nonlinear chromatic interactions, a matrix and tone curves may not be sufficient. In that case, it may be appropriate to use a multidimensional table, provided there are adequate measurement data to support it.
Table-Based or CLUT Profiles
Profiles based on lookup tables, or CLUTs, use a multidimensional table to describe the transformation between device values and the PCS. This structure allows representing more complex relationships than a 3×3 matrix, as well as devices with more than three channels, for example CMYK printers or systems with additional inks.
A CLUT is not bidirectional by definition. A profile may contain tables for device-to-PCS transformation, inverse transformation, or both. The possibility of using a profile in a specific direction depends on the tables present and the workflow needs. In ArgyllCMS, for example, B2A tables are not mandatory for an input profile, while they are relevant in output profiles and certain linking or conversion processes.
The greater complexity of a CLUT does not automatically equate to greater accuracy. A camera profile created with a very detailed table may reproduce very well the conditions under which the chart was measured but respond worse when the illuminant, exposure, development, or scene changes. The problem is not that the table is “too precise,” but that the profile may be describing a capture condition that is too specific or an insufficient sample of the camera’s actual response.
Similarly, a CLUT may show problems in highly saturated colors, highlights, or areas with few measurement patches if the table resolution or interpolation is inadequate. Validation with independent charts and real scenes remains necessary.
Lab CLUT, XYZ CLUT, and PCS
The PCS of an ICC profile can be expressed in XYZ or Lab. Neither of these options belongs exclusively to additive or subtractive devices. The choice should respond to the workflow, application compatibility, and characterization needs, not to whether the device is a camera, monitor, or printer.
In ArgyllCMS, the usual option for a CLUT is Lab; XYZ is an available alternative. This choice may influence numerical representation and interpolation behavior but does not allow a general statement that XYZ is always preferable for cameras and monitors, nor that Lab is necessarily better for printers.
Before establishing a preference, it is advisable to generate comparable profiles, validate them with independent data, and check the result in the final workflow. The difference between Lab and XYZ may be smaller than that introduced by poor measurement, an insufficient chart, or unstable capture conditions.
Practical Selection Criteria
For cameras, the choice between matrix and CLUT should consider the file type, profiling illuminant, and development workflow stability. A matrix profile can offer a smooth and robust transformation; a CLUT can model more detail when sufficient measurements are available and usage conditions are well defined. Neither structure eliminates the camera’s response dependence on the illuminant.
Scanners usually operate under more stable illumination and geometry than cameras, so it may be reasonable to use more detailed characterizations. In monitors, profile construction depends on panel technology, calibration, and the application that will use it. In printing, CLUTs are especially useful because they allow describing multichannel devices and building output transformations with black generation rules and ink limits.
In all cases, the profile should be chosen based on validation. File size, mathematical complexity, or number of tables do not replace verification with independent data from the chart used to characterize the device.
Conclusion
Matrix profiles and CLUT-based profiles are not hierarchical alternatives but models with different capabilities and limits. The former are compact, smooth, and suitable for describing relatively simple RGB responses. CLUTs allow modeling more complex behaviors and devices with multiple channels but require quality data and careful validation to avoid unreliable extrapolations.
To choose an option in ArgyllCMS or RoughProfiler, it is advisable to separate three questions: what class of profile the workflow needs, which structure best describes the device behavior, and under what conditions the profile will be used. Only the joint answer to these three questions allows deciding between a matrix, a gamma curve, a shaper, or a CLUT.
References
- International Color Consortium. ICC.1 v4 Specification.
- ArgyllCMS. Documentation of
colprof.
