Don said:
Now, why didn't you say so at the beginning? ;o)
I did, numerous times. As I recall, it came up in our very first
discussion on this newsgroup some years ago, let alone in recent times.
This is exactly what metamerism is!
That's going to be tough because I mothballed my analog equipment,
although you read my mind as I was just about to ask: How can use my
scanner as a poor-man's densitometer?
Is there any place I can get these scanner curves? (Like Kodak
provides theirs.) I can't find anything on the Nikon site.
I know... I know... Now you'll probably tell me that's what the
scanner profile is for, right? ;o)
That's the one. ;-)
Yes, that's exactly the key part of my neverending agony. This
tug-of-war between all of the different regions. Since AG is linear,
using it to fix one part skews another.
Yes, but the toe region isn't the only thing that causes that. The
different slopes in the characteristic curves, whilst linear on a
log-log plot, are not linear when plotted linearly.
What you have on the characteristic curve for each colour is:
log(D) = m * log(E) + c
where log(D) is density (intrinsically measured as a log parameter),
E is exposure and m & c are unique parameters for the r/g/b curves
determining the slope and intercept respectively of the linear portion
of the curve.
Converting that to a linear scale, (where D becomes the inverse of what
is seen by the scanner) gives:
D = c*E^m
So what actually gets to the scanner after illumination is proportional
to 1/(c*E^m).
Remember c & m are the curve coefficients that you are trying to match.
However, if all you have at your disposal is exposure at the scanner
then all you can possibly correct for is that proportionality term, c.
This is the difference in offset on the original characteristic curves -
which is why exposure can correct for the huge difference in offset
between the curves in K-II, but not the difference in slope of KC.
That slope difference is a power term in the resulting signal, and
requires a gamma adjustment to fix it. Of course, you need something a
lot more complex than a simple gamma curve though, since this only
accounts for the linear portion of the curve and you need to correct for
the entire curve really, not just the linear part (although that would
be a big improvement to start with). It is called a profile. ;-)
If you try - as you are suggesting - to fix this power term using only
the exposure control then you can only achieve correct balance at one
density level on the film. Above that, there will be a colour cast
towards one colour, below it, towards its complement.
One way of considering this is that the scanner is simply transforming
the film curves into data. By adjusting the exposure, all you are doing
is moving the curves *up* and *down* as far as the scanner sees them.
You are *not* changing their slope. Changing slope requires gamma
correction - and, as you can see, KC requires a different slope for each
colour.
Erm... fx: Looks away uncomfortably and whispers: Noo... ;o)
Seriously though, to get specific, here's one of my main test slides.
The crop should have all the important bits. The black and white keys
for the extremes at both ends, while the midrange is represented by
the wall which is basically white (perhaps with a hint of beige - if
memory serves) but the shadow part should be neutral midrange.
The exposures (reflected in file names) are *absolute*, that is to say
I turned off AE and did everything manually. These are "raw" scans
i.e., no merging or any post-processing took place other than the PS
conversion from 16 to 8 bits and then to JPEG. Gamma = 2.2.
Nominal scan: KC mode on, AG Master=+2, R=0, G=0, B=0
http://members.aol.com/tempdon100164833/nikon/KC_2M_0R_0G_0B.jpg
"Adjusted" scan: KC mode on, AG Master=+2, R=+1, G=+0.5, B=0
http://members.aol.com/tempdon100164833/nikon/KC_2M_1R_0.5G_0B.jpg
I omitted remaining points for brevity, but with all of them firmly in
mind, I "measured" various regions in PS by selecting them and then
reading out mean/median values with the histogram.
The wall in the nominal KC scan is clearly too blue (and the most
visible source of my frustration). The second scan at +1R, +0.5G is
roughly (using visual assessment) what the slide really looks like
(and, approximately, my elusive goal). The mean/median read-outs
confirm this.
However, the black keys clearly have a red cast in the second scan,
even though that may be harder to see with a naked eye. In the nominal
scan, the black keys' mean/median values are fine i.e., neutral
(shouldn't they be blue?). Also, the whites (between the keys) are too
blue (?) in the nominal scan, even though they should be OK - if I
understood correctly - because the KC curves converge there.
Anyway, given the above examples, could you please reflect on all
this?
Well, if you have read what I wrote above, and have previously written
in this thread, you won't be surprised to learn that your results are
just what I would expect. You have made two raw scans based on some
nominal starting point (not an autoexposure reference) and consequently
the initial colour balance could be anywhere in your R=G=B=0 scan. As
it turns out, it is quite blue, but that is primarily irrelevant.
However, by adjusting only the exposure, you can only correct that blue
cast at one density. As it turns out, in your example above, this is
somewhere in the mid greys - the whites themselves are very slightly
biased away from red, to cyan. I would ignore the deep blacks, since
the toe region can make this misleading but, as you note, they do tend
to be biased to red as, more importantly, are some of the less dense
regions. The point is that you cannot correct this bias across even the
linear part of the KC response simply with an exposure adjustment. You
need to adjust the balance at all densities independently. For the
linear region that can be approximated as a gamma curve, but it requires
the black point to be correctly adjusted (which is *NOT* the same thing
as setting the black point to the darkest point in each channel) for
that to have the correct effect.
Quite simply, you cannot achieve matched colour balance across the
entire range with a this emulsion using only exposure control. That is
clear from a mathematical position, viewing the curves, and you have
demonstrated this in practice. This means that, as I pointed out
several posts back, you cannot achieve your aim with a raw scan - you
must apply channel specific post processing, either in the scanner
itself or in your preferred image processing software.
I can understand why you are having a hard time coming to terms with
this. The entire discussion reminds me of one I had about 30 years ago
when I was working as a photographer and first dabbling in my own colour
printing from negatives. My job was mainly B&W and colour slide
material but my boss, who had been in the game for decades at the time,
let me use some of the darkrooms to set up my own colour D&P kit since
he was interested in seeing what sort of results could be achieved from
"modern" kit.
Now *he* was the first person that I ever heard talking about "slope
control" - he wanted to know how you could ever achieve a matched colour
balance across the tonal range without what he termed "slope control". I
hadn't a clue what he was talking about, and tried to explain the
procedure was just down to getting the correct exposure, through colour
correction filters dialled up on the enlarger head, based on colour
analyser measurements and test strips. He kept telling me that you
could only get the colour balance at one point, and you needed some form
of "slope control" to get it at across the range.
It was several years later, long after I had given up working in that
field and lost touch with him, that I actually cottoned on to what he
was talking about. The "slope control" is achieved by Kodak matching
the response curves of the film and the paper at the specific process
temperatures - together with the orange mask on colour negatives. Change
film or paper manufacturer, or process temperature, and you did indeed
get a slightly different colour balance, albeit slight because the main
manufacturers were essentially working to the same baseline. Those that
weren't tended to produce less than ideal results, but at the time I
didn't know why.
Now, as you can see from looking at the KC curves, they have different
slopes. And just as my old boss said all those years ago, you need
"slope control" to match the colour balance across the tonal range.
These days, "slope control" is called gamma - because that is exactly
what gamma does, it adjusts the slope of the denisty-log(exposure)
curve.
So, for a nicely behaved film which only has second order curves (no
wiggly bits) you need black point adjustment to correct for the film toe
or density limit, gain adjustment to correct for the offset in the
curves and gamma adjustment to correct for the slope. Exposure
adjustment just controls the gain, which is only one of the minimum set
of controls required and that is why you cannot achieve what you are
attempting with exposure control alone.
