Re: I need some clearing up ...
Re: I need some clearing up ...
- Subject: Re: I need some clearing up ...
- From: Graeme Gill <email@hidden>
- Date: Mon, 23 Sep 2002 12:08:46 +1000
Daniel Lowicki wrote:
>
a) When the relcol table is written the whitepoint scaled
>
to 100,0,0 but the blackpoint is not scaled so it remains
>
absolute colorimetric ? Or is there a blackpoint tag that
>
is thorougly supported and the relcol table is also blackpoint
>
relative ?
All these things depend somewhat on the profile software writers
interpretation of the ICC spec. In general, I think I'm
right in saying that most profiling packages don't scale the
data in any way in relation to the black point. (This was
confusing because at one stage, one spot the ICC spec.
implied that it should be, but this was contradicted at other
spots in the same ICC spec.).
By definition, the white point is used to scale the absolute
instrument data, making it relative colorimetric, not absolute
colorimetric. (Note that I'm using the words "absolute" and "relative"
in the ICC sense, not the normal colorimetry sense).
>
b) When I do a perceptual profile conversion from scanner to printer
>
and I look at it as a two step conversion scanner->Lab
>
and Lab->printer there should be absolutely no compression
>
in step one and the values should be passed to Lab in absolute
>
colorimetry (no matter which intent is selected).
>
Then the second step should do the perceptual compression.
The ICC spec. doesn't mandate that the scanner->Lab not perform
any sort of compression, but once again, I think that the majority of
profiling packages will do what you expect, and not compress.
The data will most likely be relative colorimetric (ie. normalized
to the white point of the scanner calibration media), not
absolute colorimetric. Depending on the color system you are
using, you might be able to select between relative and absolute
colorimetric intent in the scanner->Lab conversion.
The second steps intent will depend on which intent you select.
This in turn (should) select between the three tables in an
output devices profile, and the use of relative table combined
with the white point to give the absolute intent.
Most ICC based systems let you select intents for the two
device profiles used in a device to device conversion
independently.
>
After all the Lab space should be able to hold all the scanner values.
>
True?
True, but note that an Lab base lut scanner profile, will clip any
values that are whiter than the white point of the media used to
calibrate the scanner.
>
c) The compression is hard coded into the profile tables. So in theory
>
the pcs to device table of a printer profile (say perceptual) has
>
to have an entry for each Lab value that could occur and define target
>
values for the whole Lab space. This would mean that even if the source
>
only covers a small portion of the Lab space and is still bigger than
>
the target space the compression would be huge. True ?
Yes. This is a fundamental limitation (I'd call it a flaw) of the ICC
way of doing things. All the PCS->device tables are indexed in a
coordinate space that covers all reasonable (and many unreasonable)
PCS values, meaning that the table has full control over how
out of gamut colors will be dealt with.
Since the output tables are pre-computed, they have to deal with
an unknown source gamut. This means that it is likely that they
compress the gamut by an inappropriate amount (too much or too little),
for any particular source gamut.
There are ways around this limitation, but they involve using ICC
profiles in un-conventional ways, and aren't supported in
most popular color management systems.
>
Question: Is there a weighting in the icc math that controls the compression
>
by how much of the Lab gamut a source space uses ?
Nothing in the ICC spec dictates how gamut compression is performed.
This is all up to the software writer to invent.
>
Question: To my understanding a perceptual conversion is only useful if
>
the image uses portions of Lab outside the target gamut and if visible clipping
>
occurs. Is there some math involved that checks for overlaps and passes the
>
conversion through the colorimetric tables rather than through the perceptual
>
one if there are no overlaps?
No. This sort of thing is way beyond the current capability of ICC profiles,
when used in a conventional fashion.
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.