Dynamic range / Analog Gain

  • Thread starter Thread starter Don
  • Start date Start date
D

Don

Given a 10-bit film scanner its available dynamic range is 3.01 - from
log10(2^10). Correct?

The dynamic range of a slide (Kodachrome) is, allegedly, 3.4. Correct?

So, I'm missing about 0.4 of dynamic range or about 1.3 bits. I'm
compensating for this by scanning twice and contrast masking.

Question: What is the Analog Gain equivalent to 0.4 of dynamic range?

In other words, even given a slide with the maximum possible dynamic
range what is the theoretical maximum difference between shadows and
highlights scan expressed as Analog Gain so I don't go beyond that
because I'd be just wasting time scanning noise?

Are there any catches (there usually are...) e.g. removing 1.5 bits
for noise from 3.01 figure above, thus making it plummet to 2.56, etc?

Don.
 
Don said:
Given a 10-bit film scanner its available dynamic range is 3.01 - from
log10(2^10). Correct?

The dynamic range of a slide (Kodachrome) is, allegedly, 3.4. Correct?
Click to expand...

At it's best.
So, I'm missing about 0.4 of dynamic range or about 1.3 bits. I'm
compensating for this by scanning twice and contrast masking.

Question: What is the Analog Gain equivalent to 0.4 of dynamic range?
Click to expand...

Um, 1.3 bits = 2 ^ 1.3 = 10 ^ .4 = 2.5
In other words, even given a slide with the maximum possible dynamic
range what is the theoretical maximum difference between shadows and
highlights scan expressed as Analog Gain so I don't go beyond that
because I'd be just wasting time scanning noise?

Are there any catches (there usually are...) e.g. removing 1.5 bits
for noise from 3.01 figure above, thus making it plummet to 2.56, etc?

Don.
Click to expand...

Yes, there are gotchas. The biggest one is that if you actually scanned
with the equivalent of 11 or 12 bits, the density steps between binary
counts would be huge at the highest densities, i.e. in the shadow areas.
Granted, the film granularity effect is also big there, but it's the reason
most high-performance scanners have a 16 bit A/D converter. It's not to be
able to scan densities of 4.8, but to be able to scan densities of 3.5 or so
with respectable density step sizes.

I don't think it's as big a deal as it sounds, however. Unless you are
trying to enhance the detail in the shadow area of a slide, it's unlikely
that the increased noise/posterization there would be noticable under
reasonable viewing conditions.

BTW, you *are* storing those scans in a 16-bit format, aren't you?

Also Don
 
Don said:
Given a 10-bit film scanner its available dynamic range is 3.01 - from
log10(2^10). Correct?
Click to expand...

Yes, if we disregard scanner induced noise for a moment.
The dynamic range of a slide (Kodachrome) is, allegedly, 3.4. Correct?
Click to expand...

Yes, but the different film layers have a different D-max though so it
ranges from 3.1 to 3.4 roughly.
So, I'm missing about 0.4 of dynamic range or about 1.3 bits. I'm
compensating for this by scanning twice and contrast masking.

Question: What is the Analog Gain equivalent to 0.4 of dynamic range?
2.5119

In other words, even given a slide with the maximum possible dynamic
range what is the theoretical maximum difference between shadows and
highlights scan expressed as Analog Gain so I don't go beyond that
because I'd be just wasting time scanning noise?
Click to expand...

You may still want to do that, because increasing exposure will improve the
signal to noise ratio.
Are there any catches (there usually are...) e.g. removing 1.5 bits
for noise from 3.01 figure above, thus making it plummet to 2.56, etc?
Click to expand...

Indeed, but if you, say, increase exposure for the shadows 8x you should be
able to compensate for the lack of scanner dynamic range AND avoid the noisy
bits. You may then be confronted with blooming of ajacent highlights into
shadows, if so try 4x. Then still in linear Raw, divide the shadow exposure
scan values by the amount you increased the exposure (e.g. with Photoshop
creating a layer filled with black and Apply Image|Add the shadow scan with
a percentage opacity).

Bart
 
Um, 1.3 bits = 2 ^ 1.3 = 10 ^ .4 = 2.5
Click to expand...

Ah, it's that simple! Thanks!
Yes, there are gotchas. The biggest one is that if you actually scanned
with the equivalent of 11 or 12 bits, the density steps between binary
counts would be huge at the highest densities, i.e. in the shadow areas.
Granted, the film granularity effect is also big there, but it's the reason
most high-performance scanners have a 16 bit A/D converter. It's not to be
able to scan densities of 4.8, but to be able to scan densities of 3.5 or so
with respectable density step sizes.

I don't think it's as big a deal as it sounds, however. Unless you are
trying to enhance the detail in the shadow area of a slide, it's unlikely
that the increased noise/posterization there would be noticable under
reasonable viewing conditions.
Click to expand...

Yes, the shadows scan of the contrast masking procedure is supposed to
bring out the detail in the shadows, so I am concerned with getting
the most out if it.
BTW, you *are* storing those scans in a 16-bit format, aren't you?
Click to expand...

No, because my scanner only delivers 8 bits... :-( Although, it
allegedly works with 10 bits internally.

Don.
 
Yes, but the different film layers have a different D-max though so it
ranges from 3.1 to 3.4 roughly.
Click to expand...

If it's lower, that's OK... :-)

Seriously though, do you happen to know which layers are peaking at
3.1 and which at 3.4?
2.5119
Click to expand...

OK, thanks! That's not as much as I feared.
You may still want to do that, because increasing exposure will improve the
signal to noise ratio.
Click to expand...

The problem is twofold. One channel (red!) is always very reluctant to
move away from the left edge of the histogram so boosting too much
trims the dynamic range of green and blue as they run away from the
left edge. On the other hand, if I change the RGB ratio of the shadows
scan to fix this, then I have trouble contrast masking because the two
scans have different RGB ratios and so introduce ugly color
imbalances.
Indeed, but if you, say, increase exposure for the shadows 8x you should be
able to compensate for the lack of scanner dynamic range AND avoid the noisy
bits.
Click to expand...

8x is equivalent to Analog Gain of 3. In my tests, so far, the
difficult slides do end up with a difference of between 2 and 3 AG
from highlights to shadows scan. The problem is when the highlights
scan is very dark to start with and I ran out of AG for the shadows
scan since NikonScan only lets me boost by a total of 4 AG.
You may then be confronted with blooming of ajacent highlights into
shadows, if so try 4x. Then still in linear Raw, divide the shadow exposure
scan values by the amount you increased the exposure (e.g. with Photoshop
creating a layer filled with black and Apply Image|Add the shadow scan with
a percentage opacity).
Click to expand...

The blooming should be cured by contrast masking but there I have
another problem. During contrast masking I'm supposed to apply some
Gaussian Blur to the layer mask.

Now then, there are basically two distinct cases during contrast
masking: one, when an image mostly comes through (highlights scan
supplies the highlights, shadows scan supplies the shadows), and two,
the "bit in the middle" where both layers contribute.

The problem is, case one (image comes through) wants minimum, read
*no*, Gaussian blur while case two (the mix) wants *lots* of Gaussian
blur.

Choosing the "happy medium" of Gaussian blur just makes both
unhappy... Case one develops "auras" while case two becomes washed
out.

I could edit the layer mask but that's far too time consuming and I'm
trying to streamline and automate this as much as possible.

Anyway...

Thanks as always, Bart!

Don.
 
Back
Top