Re: Printer profiles and testcharts
Re: Printer profiles and testcharts
- Subject: Re: Printer profiles and testcharts
- From: email@hidden
- Date: Tue, 3 Apr 2001 08:43:42 EDT
In a message dated 4/3/01 7:15:46 AM, email@hidden writes:
>
Has anybody tested the difference with profiles made with different
>
size standard CMYK-testcharts on different printers?
>
<<SNIP>>
>
>
If somebody has already done a comprehensive and serious test of
>
this, please let me (and all of us) know - one test less to do !
Lots of us have done lots of tests... and the results are variable, depending
on the questions you ask and the software you use.
First Principle: more patches offer more data, but not necessarily smoother
profiles. Added patches may actually introduce banding into images in some
situations.
Second Principle: 2d prints of patches for 3d tables gets quickly out of
hand. If you want to add data points to a 3d LUT, it makes for a massive
increase in the number of patches involved.
Limited Deduction: the optimal number of patches varies with the methods used
by the software, but seems to be between 700 patches and 1500 patches.
Smarter techniques require a bit less brute force in this area. Diminishing
returns set in quickly beyond 1200. Below 700 you reach a point where
non-liniarities are not accounted for sufficiently. Example:
ColorBlindMatchBox will make a profile that matches the default settings
profiles built with ColorBlind PRO (default only, because it has no CMYK
controls like CBPro does) for a liniar device (say a press) but falls behind
on a nonllniar device like a dye-sub.
First Limited Corollary: I'd prefer to print fewer patches (within reason)
when possible, read fewer patches when possible and use smart software with
it, rather than (as they say in the acoustical world) "throw drivers at it".
Second Limited Corollary: Liniarization processes in advance of profile LUTs
do not improve the accuracy of a profile, and may actually damage it.
Building a curve on a very limited number of patches and applying that in
advance of your LUT data makes "reliniarization" at a later date from the
same limited patch target possible; but this process assumes that the
relationship of your data points is similar even when the major points have
shifted, in otherwords that nonliniarites, spikes, crossovers, and reversals
follow the identical pattern, even if the major color densities have shifted.
If you can build a new profile from a reasonable number of patches instead of
reliniarizing, then you are not depending on this type of cooperation from
your variable device... and if you are profiling a stable device, then this
feature is not necessary.
I suspect other opinions will follow... <G>
C. David Tobie
Design Cooperative
email@hidden