Ben Goren wrote:
Just to make sure I've been doing this right--
I use Bruce Lindbloom's BetaRGB as my working space in Photoshop, etc. I convert images to that space early, and only convert them to something else for a specific reason (such as to sRGB as the last step in making something to post on the 'Net). Images (etc.) that I print (that I care about) are always in BetaRGB.
I feed the BetaRGB profile to -S. This is correct, no?
That's a straightforward way of approaching it, and may work well enough in most cases.
You may find that it is not optimal though. Using a colorspace to define a gamut is really a shorthand for convenience. Many (perhaps most ?) colorspaces we come across at the moment are what is known as "output referred" colorspaces. An output referred space represents an output medium (ie. printer, display etc.), and will have real world constraints on the size of its gamut.
Generally, when we get an image that is in such a colorspace, it has been "rendered" to that colorspaces gamut. By rendered, we mean that it has been optimized to best make use of the output devices gamut. If this is the case, then the colorspace indeed represents the gamut of the images encoded in that space, and it's pretty convenient to use that space to setup our gamut mapping to another output referred space. The gamut mapping can be computed once, and used efficiently many times on images of that colorspace, and the output will be consistent among themselves.
There is another sort of colorspace though, and that is an "input referred" or "scene referred" colorspace. Such a colorspace's gamut doesn't have to have any relationship to a practical device, since its purpose is to represent colors as they are, without rendering. In this situation, the colorspace an image is encoded in tells you little or nothing about the gamut the image actually occupies.
So the idea of converting images into a large gamut working space before output, works in a direction that breaks the convenient shorthand we tend to use, of assuming that the colorspace an image is encoded in represents its gamut as well.
To get good looking output, someone or something has to optimize the image for the output medium. If an image is delivered in a state in which it has been optimized for one medium (say the colorspace it's encoded in), then it's possible to do a fair job of re-rendering it to another output medium automatically. That's the type of thing that happens when converting though an ICC profile pair, where the output gamut mapping has been setup correctly for the two colorspaces involved, and a perceptual or saturation intent is used.
If an image is delivered in a state in which it has not been optimized for a particular medium, then a different workflow is needed. Converting images into a large gamut working space, effectively converts images into this state.
One possible workflow is to manually manipulate the images so that they fit within a target gamut. The tool used would have to encode the images during manipulation in a non-gamut constrained way, and have a way of representing the target output gamut constraints to the user. Something like photoshop probably allows this, as the images can be converted into something like a large gamut RGB space, or even a non-gamut constrained space like L*a*b*, and the "gamut alarm" feature can be used as a (crude) guide as to how best to manipulate the image to fit the targeted output gamut. After manual adjustment, the image would normally be saved into the target colorspace without further manipulation. If the image was saved in a working space, then an external conversion to the target output space would be done in a "colorimetric" way, since no further manipulation of the gamut is desirable, it having all been done by the user.
A more automated workflow would be one it which using the colorspace the image encoded in as shorthand for its gamut is abandoned. In this sort of workflow, the gamut of each image is measured, and then used to setup the gamut mapping. A different gamut mapping (hence different ICC profile or device link) would be needed for each image in this workflow, and the output images would not be consistent amongst themselves.
Argyll has the tools to try this type of workflow out :- See <http://www.argyllcms.com/doc/tiffgamut.html>, the -g flag of profile <http://www.argyllcms.com/doc/profile.html#g>, the -G flag of icclink for the device link workflow <http://www.argyllcms.com/doc/profile.html#g>.
Graeme Gill.