[argyllcms] Re: printer profiling best practice

  • From: Anders Torger <torger@xxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 02 Feb 2015 12:35:35 +0100

I've now done lots of experiments and I will publish my "recipes" 
soon, it may be useful to others.

I have a question though. One opinion I've come across more than once 
when reading all over the 'net is that profiles made with higher patch 
counts produce "better highlight detail". Not in relation to argyll 
specifically but X-rite profiling software.

I was assuming that a low patch count did not impact detail, but only 
accuracy.

A theory I've come up with how that could be true though is that with 
more patches you sample more of the gamut surface and get a more 
precise definition of it's boundary, and thus you could get a tiny bit 
extra detail towards gamut clipping. Is this true?

Making a render test of the standard timage file (the RGB cube and gray 
gradient) I could see that the low patch count profiles would not make 
the CMY to whitepoint edges visible as far into clipping as the 800 and 
1600 patch profiles did. That could be some sort of other side effect 
though.

Any info on this? Does higher patch count provide any other advantage 
than higher accuracy (ie lower Delta E)?

If it is true that it is an advantage to sample the gamut surface more 
precisely it would be nice with a weighting parameter to targen to 
increase density of patches on the gamut surface.

/Anders

On 01/28/2015 09:08:39 AM, Anders Torger wrote:
> On 01/28/2015 04:23:24 AM, Graeme Gill wrote:
> > Anders Torger wrote:
> > 
> > Hi,
> > 
> > > I'm thinking about a workflow with 500 - 700 patches, single-pass
> > (ie 
> > > no preconditioning). My rationale is that as the printers are 
> more
> 
> > > linear today and I don't need the highest end in terms of hue
> > accuracy 
> > > it should be enough, and I'm also thinking that with fewer 
> patches
> > it's 
> > > more likely that the resulting profile will render gradients
> > smoothly.
> > 
> > Start simple and work up in complexity as you feel it is needed.
> > 
> > It's pretty simple to figure out if a separate
> > linearization/calibration
> > step is warranted on a printer - if you print a standard image in
> > a standard colorspace on the printer, does it look roughly
> > right in terms of its tone ? Running into a raw printer with
> > high dot gain, the answer will be "no, it looks way too dark".
> > So you would calibrate it first. Most printers that provide a
> > fake RGB print path should look roughly right, unless you've
> > changed one of the printing control parameters, or something
> > about the paper and/or inks is very different from what it is
> > expecting.
> 
> Thanks! I think I've surfed up some good parameter settings, and the 
> remaining question seems to be how many patches do you really need. 
> As
> 
> it will be a bit a matter of taste there's no right answer to that,
> but 
> I'll try to find my own "right answer". When I started out it felt
> like 
> I had to test trough an infinite number of parameters, but now I 
> think
> 
> I'm down at only the patch count. I'll test 200 - 1600 patches and 
> compare resulting ICC profile performance both by eye, and by 
> scanning
> 
> test images and see how much they differ. Hopefully the difference 
> becomes ~0 before I reach 1600 patches...
> 
> The Pixma Pro-1 is indeed one of those that looks roughly right when 
> going through the bundled printer driver and selecting a matching
> media 
> type.
> 
> /Anders
> 
> 



Other related posts: