<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> <html> <head> <title>About ICC profiles and Gamut Mapping</title> <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1"> </head> <body> <h2><u>About ICC profiles and Gamut Mapping</u></h2> <h3>How ICC profiles support different intents</h3> CLUT (Color Lookup Table) based ICC profiles support multiple <span style="font-weight: bold;">intents</span> by having a table for each intent. In a typical device CLUT profile, there are up to 6 CLUT's, three for input (AtoB tables, that convert from device space to PCS (Profile connection space)), and three for output (BtoA tables, that convert from PCS to device space). The tables allow the use of different color transforms, each transform being tailored for a different effect:<br> <br> AtoB0, BtoA0: Perceptual<br> AtoB1, BtoA1: Colorimetric<br> AtoB2, BtoA2: Saturation<br> <br> The colorimetric intent is meant to convey the exact device color behaviour, without any gamut mapping. Typically it is used to store the devices behaviour (characterization), and is also used where exact color reproduction is required, such as for proofing.<br> The Colorimetric tables double up for both relative colorimetric and absolute colorimetric with the application of a white point restoration.<br> <br> The Perceptual and Saturation tables are meant to contain gamut mapping combined with the device characterization. The allowance for this in both the AtoB direction, as well as the BtoA direction permits a profile to gamut map from the device gamut to some intermediate gamut, and then from the intermediate gamut to the device gamut.<br> <h3>ICC Version 2 behaviour<br> </h3> Apart from defining the general purpose of the different tables, the ICC Version 2 specification doesn't specify exactly how they are to achieve this, so it is up to the profile maker to make a choice in this regard. There is no common gamut boundary specified for the PCS, and such an approach limits the achievable intents in any case (see ICC Version 4 behaviour for an explanation why).<br> <br> What I've chosen to do with Argyll profiles, is to make all the AtoB tables the same as colorimetric. This means that the conversion used for the source profile is always colorimetric, and also means that the source gamut seen by the destination profile is the source colorspace gamut.<br> This means that the gamut mapping is done solely in the BtoA tables, and that their task is to map the source colorspace gamut to the destination colorspace gamut. So to construct the perceptual and saturation intent mapping tables, a source profile or source gamut needs to be specified, so that a gamut mapping can be constructed.<br> <br> The advantages of this approach is that the behaviour is precisely defined, a full range of gamut mapping options is available, and compatibility with matrix profiles (which do not have gamut mapping transforms) and other foreign profiles can be assured, by simply using such profiles as colorimetric sources. The main disadvantage is that the gamut mapping will only operate exactly as intended, when the profile is linked with the source profile it was setup for. This is really a fundamental limitation of the idea of having pre-computed gamut mapping color transforms, that the ICC profile format was intended to support.<br> <h3>ICC Version 4 behaviour</h3> (Note that Argyll does not currently support ICC V4)<br> <br> By default, ICC Version 4 profile operate exactly the same as the ICC V2 profile in regard to gamut mapping. A slight adjustment was made to the permitted tag contents, to allow things like Display profiles to contain the full range of AtoB and BtoA tables, so that they could also be gamut mapped. But an optional part of ICCV4, is to use the <span style="font-weight: bold;">Reference Medium Gamut</span> (RMG) as an intermediate gamut boundary between the source colorspace, and the destination colorspace. If this option is used, then an additional tag in the ICCV4 profile indicates that this is the case. This then solves the problem of the gamut mapping having to know the source and destination gamuts to operate. Instead, the gamut mapping is split into two parts, the first where the source gamut to RMG is done by the AtoB tables, and then the RMG to destination gamut is done by the BtoA tables. Profiles can therefore be mix and matches, while retaining true gamut mapping.<br> <br> This approach has a number of drawbacks though. One is that the colors get gamut mapped twice. Gamut mapping is sometimes not very precise, and<br> the geometry of the transforms may not cancel out, especially since different profile vendors may choose different algorithms in their gamut mapping. By "cancel out", I mean that even if you were linking the same source colorspace to the same destination colorspace, the gamut may be expanded (say) in the process of mapping to the RMG, and then compressed again in mapping from the RMG to the device space, and these expansions and compressions may not quite match.<br> <br> The chief drawback, is that only one (non colorimetric) intent can really be supported, that of saturation. <br> <br> The usual behaviour of colorimetric intent gamut mapping, is to compress any areas of the source gamut that lie outside the destination gamut, but for areas that fall within the destination gamut, to leave the colors unchanged. This preserves the source "look" as much as possible, while ensuring that out of gamut colors are smoothly brought within the destination gamut.<br> <br> Typical behaviour of a saturation intent, is (at least), to not only compress out of gamut source colors to fit within the destination, but to expand any source boundary that falls within the destination gamut, outwards match the destination gamut. By mapping the source gamut to the RMG, all information about what areas of the source gamut are inside or outside of the destination gamut are lost, so the destination gamut mapping can only map the RMG to match the destination gamut, thereby effecting a saturation intent.<br> <br> Once again, this is a fundamental limitation of using pre-computed gamut mappings. The only effective way of overcoming such limitations is to move to a more active color management architecture, in which gamut mappings are computed at link time.<br> <br> <br> <img alt="Illustration of perceptual and saturation gamut mapping." src="gamutmapping1.jpg" style="width: 665px; height: 215px;"><br> <br> <br> <br> <br> <br> </body> </html>
