[argyllcms] Re: Determining proper error value for -r

On 2006 May 23, at 8:11 AM, Graeme Gill wrote:

> Ben Goren wrote:
>
>> Anyway, it occurred  to me that the best way  to figure out the
>> proper error value to use for the -r option of profile would be
>> to  print out  a chart  with  a bunch  of patches  of the  same
>> colors,  measure it,  and  then figure  out  what the  standard
>> deviation is  (and convert that into  a percentage). So, that's
>> exactly what  I did: I created a  39-patch chart, did a  copy /
>> paste in a text editor (with some editing) to make eight copies
>> of the patch set, printed it, measured it, imported the results
>> into  an OpenOffice  spreadsheet,  and fiddled  around with  it
>> until  I came  up with  about  a third  of one  percent as  the
>> error--which is in line with my eyeball guesses. (I'll be doing
>> some more  testing later today  to further confirm  and explore
>> all this.)
>
> This is a pretty reasonable way to go about things. It's tedious
> though, and tells you about the level of error at only one point
> in the colorspace.

I  think one  of us  is  misunderstanding the  other. I got  eight
samples each from 39 different colors, the same colors that Argyll
generates from a 39-patch chart.

>> I'm also pretty sure that that  value would be much too low for
>> my parents' cheap  color laser printer, and probably  a bit too
>> high for  ``real'' paper  (I did  this on  plain paper  just to
>> figure out if it even made sense to do in the first place). I'm
>> also curious to learn just how much of an impact patch size has
>> on accuracy.
>
> The most sensitive  "real world" test I stumbled  across, was to
> simply make  up the profile,  and eyeball the gamut  surface.  I
> found quite  noticeable changes in  the smoothness of  the gamut
> surface, as  I varied the  -r factor in profile. Too  small, and
> the surface  was noticeably bumpy. As the  number increased, the
> surface got  visibly smoother,  and looked  more like  one would
> expect for  a well behaved  device. The self fit errors  rise as
> the -r  factor goes up  too, so I  stopped at a  suitable "knee"
> point.
>
> Now, whether this is a good way  of doing it, or the only way of
> doing it, I'm not sure.

It sure sounds like an easier way than what I just did.

Might I ask, what do you use  to look at the gamut surface? Do you
think  OS X's  ColorSync  utility  would be  sufficient,  or do  I
need something  with a  higher resolution? I've never  noticed any
irregularities when  looking at  the 3-D graph  of the  profile in
ColorSync, which is why I ask.

It could  also be that I  don't know what I'm  looking for. I just
enlarged the graph  in ColorSync, and my profiles  do look bumpier
than, for example, Canon's...but I  suspect that it's not the best
tool to examine profiles for this sort of thing.

>> There'd be  a new flag  to targen,  similar to the  existing -e
>> flag, that  tells how many  times to repeat /all/  patches, not
>> just the white ones. When profile  is fed a chart with repeated
>> patches,  it calculates  the appropriate  value for  -r itself,
>> uses that, then  creates a profile and  reports the recommended
>> value for  -r for  future use. You can  then use  the generated
>> profile with the -c option of targen or just toss it entirely.
>
> I'  not sure  such  a  thing is  worth  doing,  unless there  is
> an  indication  that such  a  calculated  number is  useful  and
> meaningful. You've already  expressed doubt that the  number you
> came up with for your printer seems realistic.

Again, I think there's a  misunderstanding here. For the paper and
printer I used,  it seems like it's probably  the right value. For
other papers and printers, it probably wouldn't be right.

> How does it  compare with the number you arrive  at in getting a
> smooth looking gamut surface ? Do the numbers agree ?

I'll get back to you as I figure out how to do that....

> What sort of color was your test patch ? It would be interesting
> to do the same test with a quite light, and quite dark color, to
> see if the errors have a different magnitude.

Like I said, I used 39 test patches from an Argyll 39-patch chart,
with eight copies of each patch.

If the number really is at least reasonably close, I'll repeat the
experiment with more patch samples  and on different papers to see
how reliable an indicator this is.

Cheers,

b&

Other related posts: