[argyllcms] Re: xicclu -g predictability issue

  • From: Elena [service address] <1007140@xxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sat, 29 Jan 2011 13:49:33 +0100

Hello Gerhard

On 29-Jan-2011, Gerhard Fuernkranz wrote:
> As Graeme suggested, I did not investigate just arbitrary slices through
> the whole color space, but a very small region around an actually
> visible discontinuity in the soft-proof of an image (source image
> created with "timage"), in order to get even more evidence that that
> these artifacts actually happen in those regions where the CMYK
> colorants make steep transitions (or are crossing steeply). And
> obviously, yes, this seems to be the case - so absolutely no
> contradiction to your findings, IMO.

No, ok read Yes... I had a bit headache these days :-) I meant,
perhaps my method (and the attached plots) was more radical in showing how
bad and abrupt the problem is at the orygin.

I don't know to what point checking the first derivative or using
smoothing approaches can radically fix the problem. Because analyzing
my slices I notice mainly two kind of discontinuities. Those where
channels don't swap (and which are usually not noticeable and don't
cause errors at interpolated boundaries) and those where channels
do swap, which are the most problematic. So in checking the derivatives
one should take not just K but other channels into account at the same time.

One approach would be empyrical, and that's why I asked kindly Grame for
the custom slices plot option in xicclu: this way one interactively checks
the effects of the chosen K parameters not just in the grey ramp but in
other color ramps aswell. Yet, maybe one finds there is no optimal solution, but
he's at least able to predict how severe will possible problems be, if any.

So, again, as for today I'm still convinced of the suggested approach.
-When computing B2A table, for every grid point keep a list of CMYK candidates
 sorted by their colorimetrical precision (high to low)
-When B2A is complete, check every intermediate point (all points interpolated
 at 50%) if their device space->absolute space value lay in between or not.
 If they don't, you're on the boundary of a discontinuity area.
 At this point other candidates should be tried iteratively or with some
 refinement, knowing that usually those areas ARE closed spaces (checked).

It's my feeling that an easier way would be to do such check while computing
the table, however. And starting from the white corner there're less chances
to start rigth inside a discontinue area (this is true for CMYK but may be
not the case for other colorants of course)


Other related posts: