Re: Smart CMMs (Was: Who's right?)
Re: Smart CMMs (Was: Who's right?)
- Subject: Re: Smart CMMs (Was: Who's right?)
- From: Graeme Gill <email@hidden>
- Date: Thu, 10 Jun 2004 13:54:50 +1000
- Organization: Color Technology Solutions
Andre Sch|tzenhofer wrote:
what you say is that you / your software creates individual gamut
mapping from source to destination in your CMM. How exactly does this
happen - on the fly, based on a pre-defined algorithm, always
individual - and what about performance?
The way their packaged up at the moment, you would link the
profiles to create a device link, and then use the device
link to transform image data. There's nothing special
about this though, the whole thing could be hidden
behind machinery that creates the links automatically as
required, and then caches the device links so that common
or subsequent transformations of the same source/destination/intent
are fast.
Basically there are three modes of creating the links:
1) Simple mode, the same as most common CMMs, where the chosen
intent source A2B is concatenated with the chosen intent
destination B2A.
2) Gamut map mode, where the source and destination gamuts
are determined from the respective profiles, a mapping
is computed between them, and colorimetric A2B from the
source and the colorimetric B2A from the destination is used.
3) Inverse lookup mode, which is the same as 2), but rather
than using the B2A table for the destination profile,
the A2B table is used and inverted. This improves
accuracy, smoothness, and allows a new choice of ink limit
and black generation (including "through black" etc.).
The Gamut mappings can be chosen from a selection that
roughly covers the usual 4, but allows for more subtle variations.
The source gamut can be a particular images gamut, so that the
gamut mapping can be optimized for a particular sub gamut of the source colorspace.
The mapping can be carried out in a CAM (Color Appearance Model) colorspace.
In terms of speed, the first is very fast to setup, the second
typically takes a second or two of computation (depending on profile
resolution), while the third can be quite slow - several minutes to
several tens of minutes, depending on the resolution of the device link
requested, and the grunt of the CPU.
For most casual use, the first two seem perfectly practical in terms of
speed, and in particular, 2) could be speeded up if there was a standard
way of storing the gamut surface in the profile. (I was considering creating
a private ICC tag to do this at some stage). If 1) didn't need to be supported,
then all the non-colorimetric intent tables could be omitted saving some profile
space, and profile creation time.
The basic gamut algorithm has quite a few parameters to tune its behavior,
and the selection makes available a preset combination of parameter settings.
The "Smart Profile, Dumb CMM" currently in widespread use made a lot of sense
when CPU power and memory were more limited, and the consequent compromises in
color reproduction were not likely to worry the ordinary user. Now that compute
power is more readily available on the desk top, and with expert users in mind,
a "Dumb profile, smart CMM" seems to be more feasible and worthwhile.
Graeme Gill.
_______________________________________________
colorsync-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/colorsync-users
Do not post admin requests to the list. They will be ignored.