VueScan & Nikon LS scanner exposure control

[Don]

However
I would also suggest that any user who is prepared to dabble with
individual channel exposure *probably* isn't happy with the resultant
balance in any case

I already made this very point but it remains unanswered, probably,
because such rigid design decisions are indefensible. Failing to even
acknowledge that automatic color balance may not always work is (for
me, at least) a very troublesome sign.

As I suggested, a pop-up dialog warning of the
change of colour balance would be sensible and, IMO, exactly what most
users who wanted to do this would expect.

That's just one of the many possibilities... Any of them would make
the design flexible and user friendly. Of course, in the case of an
alert box the user *must* have the ability to turn it off because its
constant "ping" is bound to get old pretty quick. Case in point, boxes
of (dark) Kodachromes.

True, but there is always that 1%. ;-)

And a pretty large "1%" at that! Some of them quite vocal, too! ;o)

Don.

 

[Don]

And yes, it was my fault that I did not
expose those Kodachrome properly, but I really need to salvage those
images.

No, it's not your fault. As far as Nikon scanners are concerned every
Kodachrome is a "dark" Kodachrome.

I think this is good information to know. For now, I will stick with
NikonScan and do my own analog adjustment because VueScan can't quite
reproduce the original color of the scene in my Kodachrome even though
I select Kodachrome as my slide type.

I've been wrestling with this literally for months. After a few rounds
with (incompetent) Nikon's so-called "support", bothering and annoying
people around these parts, plus tons of tests, I've settled on the
following procedure.

First, a statement of intent (in no particular order):
1. Get the most (e.g. dynamic range) out of the scanner.
2. Leave final adjustments to post-processing (e.g. in Photoshop)
3. Eliminate the most unreliable part (me!) from the process, i.e.
automate and don't rely on subjective judgement.

In the hope it may help others this is what I do given a Nikon scanner
and NikonScan, although in theory it applies to scanning in general:

1. Turn off "everything" in NikonScan (color management, auto
exposure, etc). This is in order to get a reliable baseline.

2. Don't use any image manipulation tools (curves, etc). Same reason
as above. I leave all post-processing to Photoshop (PS) afterwards. If
you don't have PS or equivalent, you may do curves at the end before
the final scan(s).

3. Do an initial scan with AG at 0. Examine the histogram (I do this
in PS).

4. Adjust individual Analog Gain (AG) until all 3 channels bump up
against the right histogram edge. This may require repeated scans and
AG adjustments.

5. Examine the dark, left, side of the histogram. If all 3 channels
are away from the left edge, you have established optimum exposure.

At this point you may turn on ICE before doing the final scan. Do note
that ICE tends to shift the histogram a few points to the left.

However, if there is clipping at the dark, right, side of the
histogram I do "contrast masking" so another scan is needed:

6. Maintaining the same ratio of individual RGB channels boost
(Master) AG until all channels move away from the right, dark, edge of
the histogram. This may also require repeated scans and tweaks.

7. Perform two final scans, one with AG settings established in point
5 (the highlights scan) and one with settings in point 6 (the shadows
scan).

8. "Contrast mask" the two scans to get the best of both worlds.

There are many nuances to the above procedure but I omitted them for
brevity. If you wish I'll explain that in a separate message.

The above procedure may, in some cases, remove the cast but that's
only a side effect just like AutoLevels may remove the cast.

I always welcome constructive criticism and any further comments and
suggestions.

Don.

 

[Don]

Changing the ratio of the CCD exposure
times will change the result of this matrix multiplication if
the changed CCD exposure time isn't compensated for.

This assumes that the nominal CCD exposure is always flat (no cast).
What happens when the nominal CCD exposure has a cast?

Modifying individual Analog Gain could remove this cast but any
subsequent "compensation" would then essentially re-introduce it.

The main reason is that it's completely unnecessary to change the
ratio of the CCD exposure times, since for 99% of the images
people scan, the very slight gain in dynamic range that would
result wouldn't be visible.

I have boxes full of Kodachromes which all strongly disagree with both
the 99% claim and the "slight" gain in dynamic range. ;o)

Don.

 

[Ed Hamrick]

The reason I want to change the analog gain is that I have a bunch of
under-exposed Kodachrome. Currently, when I scan them in NikonScan, I
need to use a high positive gain in the master and red and a high
negative gain in the blue. And yes, it was my fault that I did not
expose those Kodachrome properly, but I really need to salvage those
images.

These are trivial to scan with VueScan. Just set "Input|Media type"
to "Image" (not "Slide film"), preview the image, click with the right
mouse button (or holding down the control key on Mac OS) on a neutral
color, then press the Scan button.

You never should use the analog gain controls to color balance an
image - this is just inefficient and completely unnecessary.

Regards,
Ed Hamrick

 

[Ed Hamrick]

I would also suggest that any user who is prepared to dabble with
individual channel exposure *probably* isn't happy with the resultant
balance in any case

Analog gain and color balance have nothing to do with each other.
Using analog gain to control color balance is a crude and
inefficient way to do this.

Regards,
Ed Hamrick

 

[Bart van der Wolf]

On Thu, 1 Apr 2004 17:32:19 +0200, "Bart van der Wolf"


Do you mean that when the exposure of a pixel with the value of, say,
5 is doubled it's "only" 10, but when a pixel with the value of 50 is
doubled it's a 100? (These are values out of thin air for
demonstration purposes only.) Therefore, the "dark" pixel would only
move 5 points while the "light" one would move 50.

Correct. Just like doubling the exposure, the dark end only increases by 5
and the light end increases by 50, but the relative difference stays a
constant luminance ratio of 10:1. On the histogram, the highlights would
race to the highlight clipping boundary, while the shadows only move slowly
up from the shadow end.

I realized that but I just thought the answer was more complicated.

For example, as Kennedy wrote, black is black, i.e. absence of
luminance, so no matter how much exposure is boosted it should stay
black.

Yes, zero stays zero no matter the exposure (which assumes correct
blackpoint calibration).

On the other hand, as I noted earlier, the black in slides is not
really true black (i.e. impervious to light). This can be easily
demonstrated by looking at an unexposed slide against a bright light.

Also correct, so we'll have to adjust for that by setting a Black point
after acquiring the scan. The main purpose (other than white balancing) for
increasing the exposure until close to saturating the highlights, is the
improvement of the Signal to Noise Ratio. The sensor noise that is constant,
will become a smaller percentage of the signal.

That should in theory accelerate the movement of dark pixels. It's
because of things like this I thought the answer was more complex.

You were right, it usually is ;-)
In fact, the CRT will reduce the shadow contrast again with the reciprocal
of the gamma applied to the Raw data, but you could just look at the Raw
(linear gamma) histogram to see what that distribution looks like (the top
histograms in VueScan).

Bart

 

[Bart van der Wolf]

Analog gain and color balance have nothing to do with each other.
Using analog gain to control color balance is a crude and
inefficient way to do this.

It has more to do with restoring a neutral balance where the film base has a
perceived off-white cast.
Color balancing has to do with (selectively) stretching/aligning individual
channels, as is correctly done with your implementation of the
Right-Mouse-Button click. But the RMB does presume a Neutral Blackpoint
which can be tweaked by locking the image color.

Maybe a suggestion for future enhancement could be a CTRL+RMB average when
clicking on several points, or even linking the Luminance of the clicked
point to the selective shift of the neutral point in a Color Balance look-up
table.

Bart

 

[Kennedy McEwen]

Analog gain and color balance have nothing to do with each other.
Using analog gain to control color balance is a crude and
inefficient way to do this.

Whilst I agree it is a crude control for this, I do not agree it has
nothing to do with colour balance - that is merely an interpretation
that you have of what analogue gain *should* do. You interpret it as
changing the scanner's characteristics in a way that the interface
should cope with. I interpret it as a change of the characteristics
which the user is forcing for specific reasons, generally including
colour balance, for which the interface should not attempt to
compensate.

 

[Don]

You were right, it usually is ;-)
In fact, the CRT will reduce the shadow contrast again with the reciprocal
of the gamma applied to the Raw data, but you could just look at the Raw
(linear gamma) histogram to see what that distribution looks like (the top
histograms in VueScan).

Like the saying goes: Nothing is as simple as it seems...

Don.

 

[Don]

You never should use the analog gain controls to color balance an
image - this is just inefficient and completely unnecessary.

Forget color balance, how about using Analog Gain to extend dynamic
range?

Given an image with the following histogram highlights (after
adjusting Master AG):

Red - around 120
Green - around 125
Blue - around 250

Setting the neutral point post-scan will *not* recover the missing R &
G range (it will just cause banding). Using individual Analog Gain
will.

Like all uncomfortable questions, I expect this to go unanswered as
well...

Don.

 

[Markus Plail]

Forget color balance, how about using Analog Gain to extend dynamic
range?

Given an image with the following histogram highlights (after
adjusting Master AG):

Red - around 120
Green - around 125
Blue - around 250

Setting the neutral point post-scan will *not* recover the missing R &
G range (it will just cause banding). Using individual Analog Gain
will.

Erm... how probably is that please? I can imagine a picture with
highlights of that color, but I can't imagine a picture with that
highlights that I want to correct.

regards
Markus

 

[Don]

Erm... how probably is that please? I can imagine a picture with
highlights of that color, but I can't imagine a picture with that
highlights that I want to correct.

I'm not quite sure I understand what you mean, Markus.

If you mean the above values are unlikely or the slide is bad, that's
not correct. The histogram is from a real scan and I've got quite a
few of them. This has nothing to do with the original slide, which is
fine, but with (older) Nikon scanner's inability to handle Kodachromes
resulting in scans which are dark and with a heavy blue cast.

And I'm not alone either. If you check the archives you'll find many
others.

Don.

 

[Markus Plail]

I'm not quite sure I understand what you mean, Markus.

If you mean the above values are unlikely or the slide is bad, that's
not correct. The histogram is from a real scan and I've got quite a
few of them. This has nothing to do with the original slide, which is
fine, but with (older) Nikon scanner's inability to handle Kodachromes
resulting in scans which are dark and with a heavy blue cast.

I don't own a Nikon scanner nor have I ever scanned Kodachrome. I heard
that they have a blue cast, but I really didn't think it is THAT bad.

regards
Markus

 

[Kennedy McEwen]

I don't own a Nikon scanner nor have I ever scanned Kodachrome. I heard
that they have a blue cast, but I really didn't think it is THAT bad.

Bad enough for Nikon to make substantial changes in their later scanners
to overcome the problem.

 

[Don]

I don't own a Nikon scanner nor have I ever scanned Kodachrome. I heard
that they have a blue cast, but I really didn't think it is THAT bad.

At least as far as 80s Kodachromes on an LS-30 go, it is! :-(

Don.

 

[Don]

Bad enough for Nikon to make substantial changes in their later scanners
to overcome the problem.

Oh, one quick thing, Kennedy, if you don't mind...

Why does Nikon limit AG boost to 4 (master + individual RGB AG)? I
sometimes (often?) would like to boost more than that.

I wonder whether this "4" was an arbitrary number or is there some
theoretical reason why boosting AG > 4 doesn't make sense (for
example, all you get is noise, or some such...).

In other words, should I go back to my disassembly of NikonScan and
try to tweak this limitation? ;o)

Thanks as always!

Don.

 

[Kennedy McEwen]

Oh, one quick thing, Kennedy, if you don't mind...

Why does Nikon limit AG boost to 4 (master + individual RGB AG)? I
sometimes (often?) would like to boost more than that.

No reason I can think of - they have to put a limit somewhere though,
and "64KBytes ought to be enough for everyone". ;-)

I wonder whether this "4" was an arbitrary number or is there some
theoretical reason why boosting AG > 4 doesn't make sense (for
example, all you get is noise, or some such...).

Certainly not random noise as the whole point of increasing the exposure
is to cope with increased density with the same noise levels as
otherwise. Ultimately, the dark current inherent in the CCD, which is
removed by the calibration procedure, would put an upper limit on the
practical exposure, but I would be surprised if this was significant
enough to limit exposure to 4x nominal. Usually it is only becomes an
issue with exposures of several seconds, which would be several thousand
times the nominal exposure.

In other words, should I go back to my disassembly of NikonScan and
try to tweak this limitation? ;o)

You can try, but it seems to me that not only do you need to find and
change the limit in the software, but you also need to modify how that
is displayed and adjusted in the GUI, which is probably a much more
difficult task.

 

[Don]

No reason I can think of - they have to put a limit somewhere though,
and "64KBytes ought to be enough for everyone". ;-)

Yes, just like "nobody needs individual RGB Analog Gain"... ;o)

Certainly not random noise as the whole point of increasing the exposure
is to cope with increased density with the same noise levels as
otherwise. Ultimately, the dark current inherent in the CCD, which is
removed by the calibration procedure, would put an upper limit on the
practical exposure, but I would be surprised if this was significant
enough to limit exposure to 4x nominal. Usually it is only becomes an
issue with exposures of several seconds, which would be several thousand
times the nominal exposure.

The reason I asked is that with some images I find that the histogram
(in particular red) stubbornly refuses to move away from the right
edge. I'm not setting the black point, nor clipping - leaving all that
for Photoshop later. The right edge looks sort of full of random
spikes (this is all at +4 AG) so I just thought I may be seeing noise.

It's no big deal, but I was just curious.

You can try, but it seems to me that not only do you need to find and
change the limit in the software, but you also need to modify how that
is displayed and adjusted in the GUI, which is probably a much more
difficult task.

Indeed, which is why I wouldn't worry about "pretty pictures". That
would be a hack just for personal use. As long as the numerical
display is correct and the exposure is extended I'd be happy.

Don.

 

[Kennedy McEwen]

The reason I asked is that with some images I find that the histogram
(in particular red) stubbornly refuses to move away from the right
edge. I'm not setting the black point, nor clipping - leaving all that
for Photoshop later. The right edge looks sort of full of random
spikes (this is all at +4 AG) so I just thought I may be seeing noise.

Err, if its stuck against the right hand side of the histogram then I am
not surprised that increasing the AG to +4 doesn't help - you need to be
using negative values to get off the right edge.

Presumably you meant "left" instead of "right". ;-)

 

[Don]

Err, if its stuck against the right hand side of the histogram then I am
not surprised that increasing the AG to +4 doesn't help - you need to be
using negative values to get off the right edge.

Presumably you meant "left" instead of "right". ;-)

Yes, that's right! I mean, "left". No, I mean "left" is right. Argh...
You know what I mean... ;o)

Don.