Mike Strickler wrote:
there is
inevitably a point at which increasing an instrument's accuracy
becomes
statistically meaningless. (To give a crude example, you don't need a
micrometer to frame a house.) Perhaps this can be approached
empirically: Can anyone demonstrate a noticeable and objectionable
variability in printed color that can be traced to the performance of
any recent model of spectrophotometer that has passed its
manufacturer's
certification process?
Of course one can: Take two new spectros (different models), create
two
profiles for the same printer and compare the results. In all
probability an expert eye will notice /considerable/ differences
between
the results at first go. Repeat the procedure with one of the
spectros.
In all probability no one will notice differences between the two
profiles measured with the same instrument. That's what Terry was
talking about: Repeatability is not the Problem, but inter-instrument
agreement is. An I guess it's playing in the same order of magnitude
than inter-observer differences.
The example conforms with the results of several
inter-instrument-agreement-tests, e.g. performed by the University of
Wuppertal/Germany.
Their conclusion: inter-instrument differences are a significant
Factor
in colormanagement-based process control, see e.g.
<http://www.digitalproof-forum.de/rueckblick/ergebnisse04.php>
(German).
Klaus
