Did you "reprint" the target?
Or did you simply "re-measure the same printed target"?
Whatever you did, you're on the right track!
Your perseverance paid off.
/ Roger
From: argyllcms-bounce@xxxxxxxxxxxxx <argyllcms-bounce@xxxxxxxxxxxxx> On
Behalf Of Yves Gauvreau
Sent: December 9, 2019 12:10 PM
To: argyllcms@xxxxxxxxxxxxx
Subject: [argyllcms] Re: Something is wrong
Still don't know why but I started from scratch and got much better results.
profcheck -k RR_PaloDuroSGR_PRO1000.ti3 RR_PaloDuroSGR_PRO1000.icm
Profile check complete, errors(CIEDE2000): max. = 1.504325, avg. = 0.417007,
RMS = 0.479644
That seems very good!
Thanks,
Yves
PS I'll try this idea of measuring multiple times to see if I can get even
better results.
On 12/9/2019 11:13 AM, Yves Gauvreau wrote:
Roger,
On 12/9/2019 7:22 AM, graxx@xxxxxxxxxxxx <mailto:graxx@xxxxxxxxxxxx> wrote:
Yves,
In general, to get the best results possible, should we just pruned out
these
outliers? Would taking multiple measures of the patches and use the
average
or the "median" be a good practice or just a waste of time?
You need to understand what you're doing. In the context that the device
(paper) is highly non-linear, *all* the patches are important to build a
model of the output.
Lut base profile are so sparse that it's impossible to capture the fine
details of this non-linearity you speak of. I don't know how the CMMs
compute their output when they need to interpolate between grid points but I
doubt it is in a non-linear fashion. I believe these Lut are in PCS space,
so gamma = 1.0 and I wouldn't be surprised if the interpolation was
basically linear as well.
Outliers are always a nuisance IMHO.
Is there a way you can start with a tiny patch set? Instead of the full
1000+?
That way, less chances of errors. I seem to remember, a long time ago, that
Argyll was able to make profiles from as low as 60 some patches. If that was
still the case, you could make all the measurements one at a time (if the
software allows manual reading) and analyze each patch. You need to better
understand what you're measuring to be able to tell if there is something
wrong with the generated profile. That's what I do. You have to understand
the underlying device behavior from the measurements.
/ Roger
Maybe I don't understand what you mean but I would believe a more robust
approach then a patch by patch one, would be to measure all the patches
several times and use "average(.exe)" to sort out which of these measure we
will actually use (average, median or a Geometric Median). This should
improve our chance of obtaining a measure that is closer to the real thing.
Your approach, sound nice, but choosing a measure just because it's closer
to the target isn't good practice IMHO, there is nothing to say that this
measure is the best one or the worst for that matter. Maybe the printer is
way off for this particular color or maybe it's an out of gamut color and
the "best" measure should be farther away for this or that particular
target.
Yves
