filmscanners: Nikon Scan Dmin and Negs

["Jawed Ashraf" <Jawed@cupidity.force9.co.uk>]

[Started a new thread - going off-topic somewhat]

Rafe wrote:
> [Jawed wrote:]
> > May I just jump in here briefly and make an observation about
> the way Nikon
> > Scan 3.x works when scanning negatives?:
> >
> > Negatives plainly have quite a high DMin.  The curious thing is
> that Nikon
> > Scan doesn't tweak the black point at all when doing a default
> scan (ROC/GEM
> > off) so that the black point ends-up at around 18-25, depending
> on film, it
> > seems.  I'm guessing that NS is just giving you the DMin, un-adjusted.
> > Seems like a faulty approach if you ask me.  Especially as NS insists on
> > clipping the white point at the slightest excuse.
>
>
> I have seen both of these effects on
> my 8000, with NS 3.1
>
> The black-point effect is a minor
> annoyance, IMHO.  Maybe Nikon set
> things that way because of the
> common perception that Nikon
> scanners have "poor shadow detail."

Yeah, except that poor shadow detail affects highlights in neg scanning.

>
> In any case, it's easy enough to
> deal with, either in NS or later n
> Photoshop.

Ah there's a gotcha here.  Assuming the scan has a black point of 21, I have
found that if I set the black point to, say, 0, I also need to set gamma to
1.5-ish (1.3-1.6 is the normal range, 1.5 is the figure that keeps on
cropping-up).  NS "reads" the neg as though it was a slide, and the gamma it
applies is fine if you leave the black point at 21.  But if you want the
black point at 0 then the gamma gets thrown by your adjustment.

Now for the gotcha: I find that if I apply a 1.5 gamma in Photoshop, I get
heavy posterisation, especially in "noisy" shadow detail (for obvious
reasons!).  Conversely, if I apply 1.5 gamma in NS using its curves tool, I
get no posterisation.  This is despite the fact that I've scanned at 12-bits
(LS40) so that PS is working with 16-bit data.  I FIND THIS REMARKABLE.

Worse, the ***SHAPE*** of the gamma curves that NS and PS apply, at the same
gamma setting, appears to be different - the results look different, anyway.

Finally, one can elect to "pre-gamma" the data in NS without setting the
black point (i.e. you set the black point when editing in PS).  Trouble is,
a gamma of 1.5 after setting the black point requires you to use a
completely different pre-gamma.  I reckon one needs to use a pre-gamma of 2
in NS so that setting a black-point of 21-ish in PS produces a sensible
result.

I have a theory:

In PS it seems to me that the Levels dialog first applies levels settings
(i.e. black point, white point) and then applies gamma.  You can show this
is the case by splitting your edit into two operations of the Levels dialog:
1. levels only; 2. gamma only.  The end-result comparing one-stop Levels
dialog and two-stops Levels dialog should look the same.  If you reverse the
order, they won't look the same.

In NS I can't tell you the order in which levels and gamma are applied
(because I haven't experimented yet - I don't quite know how to do the
experiment).  I do know that NS applies levels and gamma to each channel
(i.e. R, G, and B) before it applies levels or gamma to R-G-B combined.  I
know this because I've been experimenting with individual channel curves
very heavily, followed by a global gamma change.

My theory is that NS applies levels and gamma concurrently in some way,
"leading to less arithmetic errors, hence less posterisation" (guess!).  I
don't know how - but there has to be some explanation for the excessive
posterisation that I'm seeing in PS versus NS when using lots of gamma to
correct for the silly black point.

>
> The white-point clipping is much
> more subtle.  But yeah, you wonder
> why the behavior isn't symmetrical
> at the two ends of the histogram.

As I said before, if Nikon Scan was a camera, the exposure control it
displays when scanning negs would be rejected as useless.  Luckily by the
time one is scanning a neg, a huge amount of scene dynamic range has been
thrown away, consequently NS's errors have far less impact than they would
have if they were in a camera.

There is an answer within NS to the black point and gamma problems.  It's
ROC.  Try setting ROC to 1.  Remarkable.  In fact I'd go so far as to say
that NS works the way it does in order that ROC is mandatory for all neg
scans.  Except that ROC is really bad at under-exposed frames or night
photographs.  REALLY BAD.

I intend to write some more on this...

Jawed