More on 'posterization again'
2002-12-01 by Benjamin C. Pierce
Following up my own query from a couple of days ago about posterization
effects from the MIS sepia inks...
I've continued experimenting and have been able to produce some better
results, though without a systematic understanding of how to do it
reliably. Some observations I've made in the process:
* I discovered I had two settings that were not identical to the "canonical
recommendations" that various people posted.
1) I was using the PQIJ paper setting, not Heavyweight Matte.
2) My color profile was still set to the PS default (which is
suitable for web graphics and apparently has a small gamut)
rather than Adobe RGB 1998.
My sense is that the first setting doesn't matter much, but that the
second does -- at least, the on-screen image after applying the Rourk
[not Rourke, as I wrote in my last message: sorry Paul] curves looks
significantly more colorful and with smoother gradations now.
* After some poking around (using the Info tool) in one particular image
where the posterization was especially bad, it seemed as though the
worst-affected areas were maxed-out (value=255) in the green channel.
So a hypothesis for where the posterization effect could be coming from
might be that the Epson driver is somehow trying too hard to make those
areas "really green" and ignoring the information from the other two
channels.
I don't know anything about what the Epson driver is actually doing
internally, so it's hard to evaluate this hypothesis. But it seems
consistent with the observation (posted by several people) that
overlaying an "un-Roark'd" (B/W) copy or two of the image on top of the
one with the curves applied will sometimes mitigate the posterization
effect, since this extra layer will dampen the bright green to
something less than the max value.
* I also noticed that the images giving me trouble tended to be
less-than-ideal scans, which had required some non-trivial levels
adjustments to make them look good on the screen. Obviously, the
levels adjustment will have introduced some posterization in the file
itself, and this seems to be somehow getting picked up and enormously
magnified by the final stages of the workflow (either the curves or the
Epson driver).
* The posterization I was observing does *not* seem to have anything do
with inks bleeding once they are on the paper (as I thought it might):
if I crop a small region of the image and then print it at the same
size as the original, the posterized regions just scale.
All in all, I must say I've found the experience of working with the MIS
inks and partitioned workflow rather frustrating so far -- there seem to
be a *lot* of hidden variables. I'm inclined to give up for the moment
and have a try with the piezo driver and inks, in hopes of getting a bit
more direct control over the process. Fortunately, I do have a few
really gorgeous prints to show for my efforts. (In particular, some
nudes where, ironically, a little bit of posterization has a very
interesting effect, changing the look to something closer to marble than
flesh.)
-- Benjamin