[argyllcms] Re: 0.60 CMYK profile misshaped
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Thu, 07 Dec 2006 01:36:28 +1100
Burckhart Seifert wrote:
After analyzing the result I have decided to profile the new measure
file again with Argyll CMS 0.53 before delivering the profile. The
reasons are:
The preview of 0.60 looks very misshaped in ProfileEditor, in VRML it
looks very different to the preview of the same file calculated with
0.53 and I have some results out of tolerance in some patches of the
MediaWedge.
The resulting profiles from the data you provided don't look
that different in VRML. They are a little different in behaviour,
but not grossly so.
Here's the results I get putting 10000 random device values through
the two profiles A2B tables, between V53 and V60:
Verify results:
Total errors (CIEDE2000): peak = 5.193954, avg = 1.144390
Worst 10% errors (CIEDE2000): peak = 5.193954, avg = 2.585372
Best 90% errors (CIEDE2000): peak = 2.057085, avg = 0.984281
The v53 result is a bit suspect in any case, because the smoothing
level was grossly low, and the profile fitting was stopping
prematurely. It did mean that it better accommodated strange
behaviour from a device, at the cost of poorer smoothness.
Given that, the results with V60 do have quite a large self
fit error (peak 10.5, average 1.93 delta E).
Most of the profile fit errors seem to be near cyan == 100%.
Something funny seems to be going on in that channel at least, as
it approaches 100%. Is there a calibration system on the RIP ?
How are the calibration curves targeted ?
The chart is very regularly arranged, which is often not very
efficient (that's why Argyll generate charts are different).
Although it's a bit of a pain with a strip instrument, more patches
may help accuracy, particularly if proofing is the aim. I generally
like about 3000 for a CMYK device.
Graeme Gill.
Other related posts:
