[argyllcms] Re: Best Practices

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Thu, 03 Jul 2014 16:59:45 +1000

Martin Lederhilger wrote:

> Why, or should I do a calibration? Isn't the behaviour already captured
> in the profile itself if using a non calibrated workflow? Is calibration
> done to reach a nearly linear response, so that afterwards for profiling
> the interpolation error in CLUT lookup is reduced?


Calibration is optional - it benefits print devices that are being
operated "raw", where dot gain and other non-linearities are
a significant factor. It can also be used as a mechanism for
resetting the device to a known state without a full re-profile.

As mentioned in the documentation, because calibration is a per
channel characterization it can explore the device behaviour in
much more detail than a general 3 or 4 dimensional profile sampling.
Since most useful color output devices have a degree of additivity,
linearizing the primary channels in fine detail can help
to make it behave in a more reasonable manner, benefiting profiling accuracy.

> Another question: When chartread has read all strips successfully, then
> is the read information for each patch assumed to be accurate (I mean
> that no patch is an outlier)?

The profile creation has a degree of regularization so as to
be able to "fill in" between measured points, and this has the
effect of smoothing or filtering the characteristic to some degree.
Outliers will show up as self fit errors.

> Should I average multiple reads, or move
> the device slower over the strips, or is the increase in accuracy
> negligible?

Using slower scans can improve the measurement accuracy slightly,
but in general if you are going to expend more resources on
printing and measuring, increase your test chart size. This
helps with improving the smoothing and exploring the device
characteristic in more detail.

Graeme Gill.

Other related posts: