[argyllcms] Re: Question regarding gamut mapping for photographic images

Greg Sullivan wrote: 
Thanks. I had only ever seen mention of perceptual contracting the gamut,
and it seems that this is the case.


That's the usual approach. The gamut is contracted where it has to
be (by the destination gamut being smaller than the source), while
in other areas the source gamut will be retained.

Now, for saturation intent, my understanding is that this *maintained*
saturation, without expanding it, so that's not what I want. For example,
sRGB can't reproduce the dark vivid cyan hues that print can. (right?) I'd
like a profile which can expand this area of the gamut out, and yes, I don't
mind sacrificing accuracy, as long as the result looks appealing, and not
silly. So it looks like the enhanced saturation mode is what I'd like to
experiment with, yes. I hope it does it smoothly, though. I.e, I've read
that saturation intent is commonly used for business graphics. :) 

Saturation can't really do much more than perceptual in regard to
gamut limitations. Because "business graphics" is the typical example,
it is important that fully saturated colors in the input become
fully saturated colors in the output. There is much variation
on how this is achieved. Some profilers (I believe) go to extremes,
and do things like map the "cusps" of each color space onto each
other, often grossly distorting the hues. Some may not maintain
good continuity of the lightness or saturation either, in an attempt
to maximize business graphics "punch".

Argyll isn't as aggressive as that, doesn't alter hues grossly,
maintains lightness and saturation continuity, but does make
sure that saturated colors stay saturated. I've actual
put in two pre-canned saturation intents, the default one
applying a mild "saturation enhancement" for better "business graphics"
type output, but either give very satisfactory reproduction of photographs
and images, and the differences between perceptual and saturation
intent are not all that big.

Regarding the speed, it seemed to slow down at the stage where it updates a
percentage complete figure, which I think is related to the perceptual
processing(?). 

Yes, that's computing the B2A tables (as I expected). When you want
gamut mapping, it increases the number of tables, and although
these are computed in parallel (to improve memory usage), it
still slows things down. Argyll uses a very accurate analytical
method of inverting the A2B table, which is perhaps not
as fast as other approaches.

> Without the perceptual table creation, a "high" quality
profile is quite quick to create. (feels like a few minutes, haven't timed
it) Adding the perceptual tables added between 1 and 2 *hours*, I think,
including perceptual saturation tables. FWIW, patches = 3000, and platform
is a 1GHz Pentium V, 1GB memory, and is running Windows XP. 

Hmm. That's more than I would have imagined. I haven't noticed such a
dramatic slowdown with the change you mention. Perhaps if you sent
me the .ti3 and the options you used, I can take a look at it.

Creating an "ultra" profile does take a very long time, even without the
perceptual tables. I perceive an incy wincy improvement in smoothness in
shadow flesh tones using ultra but I am not sure yet whether this is real.
(e.g, the dithering pattern may depend on the exact position that the image
is printed on the paper?)


"ultra" is really there for experimenting. I haven't found any valid use for
it.

Graeme Gill.

Other related posts: