[argyllcms] Re: Very poor results with 1000+ patch target.
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Mon, 06 Oct 2014 10:21:45 +1100
Michael Gallagher wrote:
> After reading in all the patches, I grabbed my local AdobeRGB1998.icc from
> my ColorSync folder (as I've read you're supposed to do), copied it to my
> working directory and used this command to generate the profile:
>
> colprof -v -qh -S AdobeRGB1998.icc -dpp -D "Arches 300gsm Canon Pro 100"
> 1050patch_Arches
One of the things emphasized in the documentation is to check
the profile self fit before proceeding onwards. What were the values
returned in your case ?
> In case anyone wants to look at both ICC profiles on their computer, here's
> a link to a zip file with both new and old scans included:
> https://www.dropbox.com/s/9c51et2luunvuap/1050_Arches.zip?dl=0
What's of more interest is the .ti3 file, rather than the resulting ICC
profile, but by default it is included in the profile, so if I extract it
from the v2 file and run a simple sanity check:
colprof -v datafile.ti3
I get
Profile check complete, peak err = 37.499292, avg err = 0.889221, RMS =
2.756460
That's a pretty obvious sign that the chart hasn't been read in properly.
You can identify the patches that are very wrong by looking at the output
of profcheck
profcheck -v2 datafile.ti3 datafile.icm
It appears that a number of patches in strip AD were not read accurately.
Deleting the worst 26 patches and re-running the profile yeilds:
colprof -v datafile_P5.00.ti3
Profile check complete, peak err = 2.177640, avg err = 0.342337, RMS = 0.401542
which would be a good result. Correcting the bad reads may not end up quite as
tight
a fit as this, but a peak of 5 dE or less would be quite reasonable.
I'm not sure why you are getting such consistent measurement errors, although
given the types of problems you are seeing, I would re-print the target
without the 60% scaling before trying again.
A last resort is to read problematic strips patch by patch. You can
do this by quitting chartread part way through in one mode, and then
resuming in a different mode (ie. chartread -p and -r flags).
Graeme Gill.
Other related posts:
