Re: nothces on the L* axis
Re: nothces on the L* axis
- Subject: Re: nothces on the L* axis
- From: Klaus Karcher <email@hidden>
- Date: Fri, 14 Dec 2007 12:06:13 +0100
eugene appert wrote:
The disappointment I am trying to express with respect to the
apparent inability of colour management to distinguish perceivable
from nonperceivable variations in density goes to the heart of
whether it can ever replace the leg work of test strips and step
wedges.
A printer needs to know where tonal distinctions create lines and
where they don't. I just printed out a linearization chart on Museo
Silver Rag using a QTR grey profile and read it with an eye one. So
now I know that L* 11 prints out at L*18.1 and L* 12 prints out at
L*18.5. I could have just as easily converted the chart to the
output profile in Photoshop and then read the L* values with absolute
configured in my colour settings box. Even if the L* values had been
expressed in two decimal figures, what's my reference for knowing
whether or not the distinctions are important enough to produce a
line? Should I expect to see a line between L* 1.1 and L* 1.2 ?
between L* 16, 700, and 16, 720?
Don't forget that there is a long processing chain between Photoshop and
the actual output. It's beyond the scope of ICC colormanagement to
predict every possible output value with the precision you expect.
It starts with the way you measure your device values: in case of a 8bit
CMYK device, there are 2^8^4 = 4294967296 possible device values. Even
when we assume that there's only one "correct" separation (i.e. K is a
function of C, M an Y), there are still 16777216 possible device states
to be described. A typical test chart contains about 1600 distinct
patches -- that means that 99.990% through 99.99996% of your possible
output values have to be interpolated somewhere in the CMS.
When you render your test chart -- what's your input: a 8bit TIF or a PS
version with higher precission? What numbers are in the reference file
for the test chart? (in many cases rounded integer percentages for CMYK)
How does the profiling software interpret these values? As floating
point? As 8bit integer? As 16bit integer? How much noise does your
measurement device add to the result? What's the repeatability of your
output- and your measurement device? Are there linearization curves in
the output chain? With what precission?
When it comes to the output of "real" data, there are further questions:
what will be sent to the device when you define e.g. 50% in your vector
program 49.8, 50.0 or 50.2? Will your RIP treat vector- and bitmap data
in the same way? Do you use JPEG compression? Are there entities in your
workflow that add noise to the data to hide banding artifacts
graciously? Will that noise be added before or after the compression?
What interpolation methods will be used to scale the data to the output
resolution? Are there sharpening filters in the processing chain?
You see: there are many questions. And I am often amazed about how
predictable colormanaged system can be in spite of all these uncertainties.
Regards,
Klaus Karcher
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden