[argyllcms] Re: Purplish blues with Perceptual in Argyll 1.1.0_RC1

  • From: Gerhard Fuernkranz <nospam456@xxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Fri, 27 Nov 2009 00:49:59 +0100

Elena [service address] wrote:
> 3) A pity, Argyll profile still has "bumps",or "waves", expecially noticeable 
> when printing out fully saturated RGB test fades, HUE circles and such 
> things. If you don't understand what I'm meaning, it's the same problem 
> discussed in a past topic
>    //www.freelists.org/post/argyllcms/profile-black-generation,8
>    Yes, they are those waves whose frequency increases if you increase the 
> LUT resolution. I noticed no major improvement in this problem from 1.0.4 to 
> 1.1.0. Btw, ProfileMaker too has this defect, perhaps a bit less noticeable, 
> but it's there.

IMO these artifacts result from trilinear interpolation of 3D LUTs. If
you resample for instance a matrix/TRC-based CIELAB to sRGB
transformation at the N^3 vertices of a regular grid in order to obtain
a 3D LUT, and then perform a transformation sRGB -> exact
matrix/TRC-based transformation -> CIELAB -> trilinear interpolation of
the CIELAB-to-sRGB 3D LUT -> resulting_sRGB, then you'll notice the same
kind of "ripples" in the "resulting_sRGB" too, even though the the LUT
grid points have been precisely resampled from the original function. So
this is not necessarily a defect of the profile (ok, there still can be
problems in some profiles too, of course), but rather happens in the
engine which applies the profile and interpolates the LUTs in the
profile. And in order that the ripple in the above example decreases to
a hardly noticeable level, a pretty high LUT resolution (say N=65 or
even N=129) seems to be necessary if one sticks with trilinear
interpolation (for comparison: colprof seems to use N=17 for "-qm", and
N=33 for "-qh" (for 3D LUTs)).


Other related posts: