The Ultimate Scanner Software vs. Photoshop Question

[Howard]

Below is a question that BEGS for a response from Wayne Fulton,
Kennedy McEwen, C.D. Tobie, or others at their level of understanding.
It derives from a Yahoo Groups discussion, but for which no
definitive answer has yet been forthcoming.

Assume a good quality, flatbed scanner, and that one scans an image
using the manufacturer's or a third-party's scanning software through
which some adjustments, such as moving the white-point and black-point
in the "Levels" function, and boosting the gamma using the "Curves"
function, are made.

Question 1: When physical scanning occurs, do the settings in the
scanner software affect how the image is scanned so as to produce the
desired output, OR does the scanning occur independent of any settings
in the scanning software, but then the scanned image file is "handed
off" to the scanning software which then processes the scanned file so
as to make the indicated "Levels" and "Curves" or other adjustments?


Question 2: (First some background, then the question.)

Background

When I first learned digital imaging, Photoshop (5.5) and how to scan,
the theory was: Because image manipulations in Photoshop --
especially adjusting the white and black points -- inherently
degrade/damage the image, which usually can be seen in the shape of
Photoshop's histogram, do as much image processing in the scanner
software as possible so that Photoshop begins with the best image
possible. Because I do photo restorations, where the unadjusted image
data is almost always scrunched toward the middle of the histogram
thereby needing aggressive black- and white-point adjustments, I would
spend A LOT of time scanning and rescanning, going back and forth
between the scanning software and Photoshop so as to obtain an
excellent scan with a nice histogram prior to any adjustments in
Photoshop.

My scanner was a UMAX PowerLook III, with its MagicScan software. I
always scanned in hi-bit (16/48-bit) mode. In Photoshop 5.5 through
7, use of Adjustment Layers of course required lowering the mode to 8
bit. But we have Photoshop CS, where the important image-adjustment
functions, including Adjustment Layers, are available in 16-bit mode.
(And I have a new Epson 4870 scanner.)

The Argument:
If it is correct that scanning occurs independent of any settings in
the scanning software, and then the scanned image file is "handed off"
to the scanning software which then processes the scanned file so as
to make the indicated "Levels" and "Curves" or other adjustments, and
if both the scanner and Photoshop are set to high-bit mode, then there
should be no difference "where" the image adjustments occur – in the
scanning software or in Photoshop. And if so, and inasmuch as
Photoshop's tools tend to be more precise than scanning software, why
not save a LOT of time by scanning "wide open", e.g., white and black
points at "0" and "255", and do the Levels adjustment in Photoshop at
16 bit?


Argument Against:
First, if it is true that, in the past, the ONLY reason that one
should execute major image manipulations in the scanner software was
that scanners were capable of high-bit image manipulations, whereas
most Photoshop functions were limited to 8-bit, then setting the black
and white points in the scanner software while in low-bit mode should
have shown a jagged histogram when imported into Photoshop. But that
was not the case. So long as performed in the scanner software, even
in low-bit mode, the histogram in Photoshop was nice and smooth
(unless and until later manipulated in Photoshop).

Second, I question the fundamental assumption that scanning occurs
entirely independent of the adjustments in the scanning software. One
of the reasons given for the workflow of (i) determining the image's
final output resolution and then (ii) scanning at the next higher dpi
level which is an even integer of the scanner's maximum optical
resolution, was so that there was even spacing among the (CCD)
scanning elements, which thereby necessitated far less image
processing (read: interpolation) by the scanner software. If true,
then the scanning software does have an affect upon the actual,
physical, scan.

Third, if scanned in high-bit mode and "Photoshopped" in high bit mode
yield the same result in terms of limited degradation to the image,
and since Photoshop's tools tend to be better than those in most
scanning software – and certainly no worse – why would anyone pay
extra for Silverfast, VueScan, etc. Instead, simply use the
manufacturer's bundles software to scan "wide open" (in high-bit mode)
and do all manipulations in Photoshop. NOTE: let's not get into the
issue of scanning software's bells, whistles, Digital ICE and other
features. Assume that the corrections being made are through
"Levels", "Curves", Saturation and Color Balance – equally available
in the scanning software and in Photoshop.

So, there's the challenge. Probably far more words have been expended
to describe the issue than will be needed to provide answers.

Who dares to take up the challenge?

Howard

 

[Wayne Fulton]

Photoshop has excellent tools. Not all scanners do. Therefore using
Photoshop may seem a very good plan.

Not all photo editors have excellent tools. Some scanners do.
Therefore use of the scanner tools may seem a very good plan.

When the situations are more equal, it really only depends on a users
preference of work method. Certainly all users are not equal. Some will
vote one way, some the other, but no one is wrong, it just may not match
your preference. Lots of others dont want ANY adjustments at all, they
just want the picture. From time to time, I may have any one of those
opinions at the moment.

If you want to argue that outputting 16 bits into Photoshop is superior,
then that's great, enjoy. I normally do that when scanning film with
VueScan. There have been many debates about 16 bits, with no obvious
winner yet, it has been impossible to show that it matters, at least
with real world photo images. I do it because I think I get better
color results from film with a wide image from VueScan, followed by
Photoshop Levels. 16 bits is easy today, slow, but no longer a
struggle.

If you want to argue that other ways are better or more convenient,
that's great too, enjoy. I normally use the scanners 8 bit tools when
scanning prints or documents, it is indeed convenient.

All that is important is to find a good solution for your case, which
gives good results and a preferred way of working. I find it very
reasonable that one user has a best way of working, and another has a
very different best way of working. We dont have to agree.

 

[Kennedy McEwen]

Below is a question that BEGS for a response from Wayne Fulton,
Kennedy McEwen, C.D. Tobie, or others at their level of understanding.
It derives from a Yahoo Groups discussion, but for which no
definitive answer has yet been forthcoming.

Assume a good quality, flatbed scanner, and that one scans an image
using the manufacturer's or a third-party's scanning software through
which some adjustments, such as moving the white-point and black-point
in the "Levels" function, and boosting the gamma using the "Curves"
function, are made.

Question 1: When physical scanning occurs, do the settings in the
scanner software affect how the image is scanned so as to produce the
desired output, OR does the scanning occur independent of any settings
in the scanning software, but then the scanned image file is "handed
off" to the scanning software which then processes the scanned file so
as to make the indicated "Levels" and "Curves" or other adjustments?

Answer 1: It depends on the scanner and the software. ;-)
"In the old days", it was common for scanners only to output 8-bits per
channel, even high quality scanners which had an internal ADC of 10 or
12 bits. Consequently, they implemented the black/white point and gamma
correction in a hardware lookup table which the software programmed up
before the scan. Even so, the scan was usually the same, but the
software adjustments were implemented in scanner firmware, rather than
the software itself.

Modern scanners usually pass full bit depth data back to the PC for the
software to implement the adjustments after the scan, so the software
doesn't actually influence the scanner configuration at all.

The only adjustment which is actually implemented in the scanner
hardware itself is, if it is available, exposure control. I don't know
any flatbed scanners that have real exposure control, but someone will
pipe up if they have one.

Question 2: (First some background, then the question.)

Background

When I first learned digital imaging, Photoshop (5.5) and how to scan,
the theory was: Because image manipulations in Photoshop --
especially adjusting the white and black points -- inherently
degrade/damage the image, which usually can be seen in the shape of
Photoshop's histogram, do as much image processing in the scanner
software as possible so that Photoshop begins with the best image
possible. Because I do photo restorations, where the unadjusted image
data is almost always scrunched toward the middle of the histogram
thereby needing aggressive black- and white-point adjustments, I would
spend A LOT of time scanning and rescanning, going back and forth
between the scanning software and Photoshop so as to obtain an
excellent scan with a nice histogram prior to any adjustments in
Photoshop.

My scanner was a UMAX PowerLook III, with its MagicScan software. I
always scanned in hi-bit (16/48-bit) mode. In Photoshop 5.5 through
7, use of Adjustment Layers of course required lowering the mode to 8
bit. But we have Photoshop CS, where the important image-adjustment
functions, including Adjustment Layers, are available in 16-bit mode.
(And I have a new Epson 4870 scanner.)

The Argument:
If it is correct that scanning occurs independent of any settings in
the scanning software, and then the scanned image file is "handed off"
to the scanning software which then processes the scanned file so as
to make the indicated "Levels" and "Curves" or other adjustments, and
if both the scanner and Photoshop are set to high-bit mode, then there
should be no difference "where" the image adjustments occur – in the
scanning software or in Photoshop. And if so, and inasmuch as
Photoshop's tools tend to be more precise than scanning software, why
not save a LOT of time by scanning "wide open", e.g., white and black
points at "0" and "255", and do the Levels adjustment in Photoshop at
16 bit?


Argument Against:
First, if it is true that, in the past, the ONLY reason that one
should execute major image manipulations in the scanner software was
that scanners were capable of high-bit image manipulations, whereas
most Photoshop functions were limited to 8-bit, then setting the black
and white points in the scanner software while in low-bit mode should
have shown a jagged histogram when imported into Photoshop. But that
was not the case. So long as performed in the scanner software, even
in low-bit mode, the histogram in Photoshop was nice and smooth
(unless and until later manipulated in Photoshop).

Second, I question the fundamental assumption that scanning occurs
entirely independent of the adjustments in the scanning software. One
of the reasons given for the workflow of (i) determining the image's
final output resolution and then (ii) scanning at the next higher dpi
level which is an even integer of the scanner's maximum optical
resolution, was so that there was even spacing among the (CCD)
scanning elements, which thereby necessitated far less image
processing (read: interpolation) by the scanner software. If true,
then the scanning software does have an affect upon the actual,
physical, scan.

Third, if scanned in high-bit mode and "Photoshopped" in high bit mode
yield the same result in terms of limited degradation to the image,
and since Photoshop's tools tend to be better than those in most
scanning software – and certainly no worse – why would anyone pay
extra for Silverfast, VueScan, etc. Instead, simply use the
manufacturer's bundles software to scan "wide open" (in high-bit mode)
and do all manipulations in Photoshop. NOTE: let's not get into the
issue of scanning software's bells, whistles, Digital ICE and other
features. Assume that the corrections being made are through
"Levels", "Curves", Saturation and Color Balance – equally available
in the scanning software and in Photoshop.

So, there's the challenge. Probably far more words have been expended
to describe the issue than will be needed to provide answers.

Who dares to take up the challenge?

OK, I got the background, the argument for and three arguments against
the workflow presented in the background, but I seem to have missed the
question... ;-)

A lot of this was actually raised in a recent thread which will drive
regulars crazy if it degenerates into that again. ;-) With modern
flatbed scanners and software capable of high bit depth processing there
is little (note not no!) benefit to adjusting in scanner software before
handing off to Photoshop. However, you make a few erroneous assumptions
in your arguments above, and the difference between "little" and "no"
difference is a consequence of some of those, so that may influence what
you conclude.

In your first argument against the workflow, you argue that processing
in the scanner software resulted in a smooth histogram even when working
in 8-bit mode. The reason is that the scanner software didn't work in
8-bit mode *internally*, it only output 8-bit data. So all of the
black/white point adjustment, gamma correction, saturation and hue
adjustments were implemented to at least the precision of the hardware
and the final result was then truncated or rounded to 8-bit precision -
resulting in a smooth histogram which the alternative of processing in
Photoshop from only an 8-bit source file could not match.

With hi-bit capability in Photoshop, then that is less of an issue - but
not negligible for the reasons of a misassumption in your third argument
against the workflow. Here you assume that Photoshop tools are "better
and certainly no worse" than those in the scanner software, which is not
necessarily the case. In particular, Photoshop's authors are on record
as stating that its high bit capability covers 32769 discrete levels
which is, near as makes no difference, only a 15-bit capability, not
true 16-bits. I don't know if this is still applicable to PS-CS
although I expect it is, but it certainly is in PS7.1 which is what I
use and have available. Now, if your scanner has a 16-bit ADC and
software capable of exploiting the scanner's capabilities (which would
usually be the case) then Photoshop's tools are actually inferior to
those in the intrinsic scanner software. Once again, I don't know of
any flatbed scanners which fall into this category yet but, in the
previously mentioned thread, I demonstrated that PS7.1 is inferior to
NikonScan in this respect and there may well be other flatbed scanner
software which is equally capable of true 16-bit performance. So for
those *film* scanners, there is still an argument for processing in the
scanner software as much as possible using the available tools there
before handing off to Photoshop for further adjustments. It may not be
a significant advantage, but it does result in less corruption and as
scanner technology improves and 18 or 20-bit scanners hit the streets
(and it will happen) this will become even more significant than just
the single bit difference there is at present. We are now in the
"twilight zone" (do-da-do-do do-da-do-do!) where the improvements in
Photoshop have made it almost capable of matching the best scanners, but
pretty soon we will be back to the status quo ante with the original
workflow being just as valid.

Finally, in your second argument against the workflow, you refer to the
resampling process and the use of an even integer division of the
intrinsic scanner resolution. I don't see why you restrict that to only
even integers. Odd integers will give exactly the same effect, which is
that every new output sample is centred on an original scanner sample.

Usually the resolution is decimated simply by dropping the unused
non-central samples from the CCD output, so no interpolation is
implemented in the software. Thus each individual output sample is a
CCD sample itself, which has an intrinsically high resolution but is now
heavily undersampled, resulting in vastly increased aliasing. However,
in higher quality scanners, the decimation is implemented by a more
accurate downsampling algorithm, which makes use of those intervening
samples, thus reducing aliasing even at low resolutions. Furthermore,
both of these types of decimation process can be implemented quite
separately in Photoshop after a full resolution scan, either using
nearest neighbour interpolation for the first method or usually bilinear
interpolation for the second. Most scanner software does not offer
bicubic interpolation, which is available in Photoshop and results in an
even better decimation process in terms of an optimisation between
residual resolution and the amount of aliasing that occurs. So it does
not follow from your second argument that the scanner hardware is
implementing the scan any differently than a full resolution scan - it
might, it might not, and, generally, the better the scanner the less
likely that it would. However, in cases where it does make a difference,
in that only certain CCD outputs are processed and transferred across
the bus and the scanner makes larger motor steps, the results are likely
to be the poorer for it.

So, that may have risen to your challenge and addressed your arguments
but, not having asked a question, I don't claim to have provided you
with an answer. ;-)

 

[Howard]

Thank you Wayne and Kennedy.

Kennedy, after taking a deep breadth, I briefly followup your
response.

1. As to the primary focus of the original inquiry:
In a nutshell (not a 55-gallon drum), it seems that your position is:
if Photoshop CS were truly, fully 16 bit (instead of the bug-induced
15-bit maximum), and if the scanner software were operated in full
(ignore potential future technology) 16/48-bit mode, and ignoring all
the bells and whistles offered by some scanner software so that we are
considering only the effects of using the “Levelsâ€, “Curves†and
“Color Balance†functions either in the scanner software or in
high-bit mode Photoshop CS, then it makes no difference whether (i)
such functions are implemented through 16-bit scanner software (and
touched-up in Photoshop) or (ii) the image is scanned “wide open†in
the 16-bit scanner software and ALL adjustments are performed in
16-bit mode Photoshop. (I accept as true all of the qualifiers stated
in you original reply; but, as I hope is clear, here I just want to
understand the basic issue — all else being equal.)

2. As to why, in earlier times (mostly 8-bit Photoshop), 8-bit scans
produced a nice, smooth histogram when imported into Photoshop, you
have explained something that I did not previously know. If, as you
wrote, the scanner software was making the Levels, Curves and Color
Balance adjustments in high-bit mode but EXPORTING the data to
Photoshop in 8-bit mode, then that would indeed explain the
significant advantage of making such edits in the scanner software,
and why the initial histogram in Photoshop would be nice and smooth.

3. Finally, as to the better/preferred scanning technique:
This was mentioned in the earlier post only as potential corroborating
evidence that the scanner software is actively “doing things†even if
the software adjustment functions are left “wide open.†But since you
went into a little (!) detail, I respond.

I go back to the days of the first “epson-inkjet†user group. Back
then, and still to some, lesser, extent today, one of the most
frequent subject strings of posts was/is what method is best for
scanning? Responses generally came in three categories (with
variations):

a) Always scan at the highest optical resolution of the scanner;
resize in Photoshop to the desired final size; then downsample to the
printer’s “preferred†resolution – which in the case of Epson was 360
or 720 dpi (240 at minimum). (I believe the new HPs are 300/600; I
have no clue regarding Canon printers.)

b) Do your scan resolution/print size calculation first. Then choose
a scanning resolution which will give you exactly the final image size
you want. For example, to obtain a final print size of 8x10 at 300
dpi, one might have to scan a smaller original at 421 (or 421.638)
ppi.

c) To a post that suggested method “b†immediately above (if I recall
(from more than three years ago), the poster was a Mr. Franklin), you
posted the response: “NO, NO, NO! Don’t do that!†Of course, you
then provided a DETAILED explanation. You explained that oddball
scanning resolutions force significant interpolations by the scanning
software, most of which was not as good as Photoshop’s bicubic
interpolation. As to method “aâ€, I believe you noted that scanning at
the maximum optical resolution of the scanner not only often produces
a far greater file size than is necessary, but the later downsampling
may be quite severe — and this is exacerbated when Photoshop is set to
force a final print resolution of, say, exactly 360 dpi. You reported
that, from your tests, one will usually obtain the best compromise by
scanning at the next highest ppi level which (i) is an even divisor,
i.e., an integer divisor, into the scanner's maximum optical
resolution (that’s what I meant to say in my original post), and (ii)
will provide enough pixels for the desired print size. And what this
necessitates, as observed in Photoshop’s “Image Size†dialog, is that
the final print resolution should be allowed to “fall where it mayâ€
(your actual words) -- but always being at least the minimum for the
printer to produce a good image (240 ppi, preferably 300 - 360, for
Epsons).

So that’s where I got that part of my original post; although, as
stated, it was really meant only to support the assumption that
scanner software is doing something even if image adjustment functions
are left untouched.

Howard

 

[Kennedy McEwen]

Thank you Wayne and Kennedy.

Kennedy, after taking a deep breadth, I briefly followup your
response.

1. As to the primary focus of the original inquiry:
In a nutshell (not a 55-gallon drum), it seems that your position is:
if Photoshop CS were truly, fully 16 bit (instead of the bug-induced
15-bit maximum), and if the scanner software were operated in full
(ignore potential future technology) 16/48-bit mode, and ignoring all
the bells and whistles offered by some scanner software so that we are
considering only the effects of using the “Levelsâ€, “Curves†and
“Color Balance†functions either in the scanner software or in
high-bit mode Photoshop CS, then it makes no difference whether (i)
such functions are implemented through 16-bit scanner software (and
touched-up in Photoshop) or (ii) the image is scanned “wide open†in
the 16-bit scanner software and ALL adjustments are performed in
16-bit mode Photoshop. (I accept as true all of the qualifiers stated
in you original reply; but, as I hope is clear, here I just want to
understand the basic issue — all else being equal.)

Pretty much, except that the PS limit is "by design" rather than bug
induced. ;-)

Also, make sure that you use the scanner software to optimise the
exposure or analogue gain as part of the production of the "wide open"
scan.

3. Finally, as to the better/preferred scanning technique:
This was mentioned in the earlier post only as potential corroborating
evidence that the scanner software is actively “doing things†even if
the software adjustment functions are left “wide open.†But since you
went into a little (!) detail, I respond.

I go back to the days of the first “epson-inkjet†user group. Back
then, and still to some, lesser, extent today, one of the most
frequent subject strings of posts was/is what method is best for
scanning? Responses generally came in three categories (with
variations):

a) Always scan at the highest optical resolution of the scanner;
resize in Photoshop to the desired final size; then downsample to the
printer’s “preferred†resolution – which in the case of Epson was 360
or 720 dpi (240 at minimum). (I believe the new HPs are 300/600; I
have no clue regarding Canon printers.)

b) Do your scan resolution/print size calculation first. Then choose
a scanning resolution which will give you exactly the final image size
you want. For example, to obtain a final print size of 8x10 at 300
dpi, one might have to scan a smaller original at 421 (or 421.638)
ppi.

c) To a post that suggested method “b†immediately above (if I recall
(from more than three years ago), the poster was a Mr. Franklin),

I remember the discussion well - and as I recall you had some input to
that thread as well! Unfortunately my archive CDs from that period
aren't convenient to access (I would need to reinstall old news software
to read them) otherwise I would look up the exact details. Since the
three Leben lists closed in April 2002 though it probably was more than
3 years ago. ;-) I did keep in touch with Austin for a while
afterwards, but I haven't heard from him for a year or so and don't know
if he is still active in scanning or photography.

you
posted the response: “NO, NO, NO! Don’t do that!†Of course, you
then provided a DETAILED explanation. You explained that oddball
scanning resolutions force significant interpolations by the scanning
software, most of which was not as good as Photoshop’s bicubic
interpolation. As to method “aâ€, I believe you noted that scanning at
the maximum optical resolution of the scanner not only often produces
a far greater file size than is necessary, but the later downsampling
may be quite severe — and this is exacerbated when Photoshop is set to
force a final print resolution of, say, exactly 360 dpi. You reported
that, from your tests, one will usually obtain the best compromise by
scanning at the next highest ppi level which (i) is an even divisor,
i.e., an integer divisor, into the scanner's maximum optical
resolution (that’s what I meant to say in my original post), and (ii)
will provide enough pixels for the desired print size.

And that is still what I would recommend, however I don't think I used
the term "even integers" - or if I did then it was not what I meant!
Sorry :-(

Even integers are 2, 4, 6... etc., but all integer, ie. 2, 3, 4, 5...
etc., divisions of the optical resolution are suitable. So for a
3200ppi scanner that means 1600 (3200/2), 1066.667 (3200/3), 800
(3200/4), 640 (3200/5), 533.333ppi (3200/6) etc.

Time has moved on of course and what was a huge file 3 or 4 years ago
that very few systems could handle now commonly fits within the RAM
complement of most machines, so the optimum compromise has moved towards
making best use of the information in the larger file.

 

[Don]

1. As to the primary focus of the original inquiry:
In a nutshell (not a 55-gallon drum), it seems that your position is:
if Photoshop CS were truly, fully 16 bit (instead of the bug-induced
15-bit maximum), and if the scanner software were operated in full
(ignore potential future technology) 16/48-bit mode, and ignoring all
the bells and whistles offered by some scanner software so that we are
considering only the effects of using the “Levelsâ€, “Curves†and
“Color Balance†functions either in the scanner software or in
high-bit mode Photoshop CS, then it makes no difference whether (i)
such functions are implemented through 16-bit scanner software (and
touched-up in Photoshop) or (ii) the image is scanned “wide open†in
the 16-bit scanner software and ALL adjustments are performed in
16-bit mode Photoshop. (I accept as true all of the qualifiers stated
in you original reply; but, as I hope is clear, here I just want to
understand the basic issue — all else being equal.)

A few of things to keep in mind...

Even if you have a true 16-bit scanner IMHO it's still better to scan
raw and do all image editing in Photoshop.

Namely, when you view things in context, the loss of 1 bit in
Photoshop is more than offset by vastly larger toolbox and general
usability of Photoshop vs. scanner software. In particular:

1. Scanner software generally offers only a very limited subset of
editing tools and utilities e.g. there is no threshold, you can't
truly examine the histogram, etc. etc. In other words, a good edit in
15-bit (i.e. Photoshop) will be superior to a bad edit in 16-bit (i.e.
scanner software).

2. Scanner software doesn't allow you to view the image at full
resolution (100%) - unless you have a 6000 x 4000 pixel monitor... So
any editing decisions made based on this "keyhole" view of the image
will be very questionable (e.g. examining noise in dark areas which
often needs 300% enlargement, if not more).

3. If, after scanning and examining the image in Photoshop, you decide
the scan was not satisfactory you will have to repeat it and make
another guess. However, by scanning raw, you can edit this raw image
in Photoshop with full complement of tools and utilities. If you don't
like the result, simply reload the raw file and start again. This is
not only considerably quicker than re-scanning but it's also much
easier on the scanner as it cuts down on unnecessary use.

2. As to why, in earlier times (mostly 8-bit Photoshop), 8-bit scans
produced a nice, smooth histogram when imported into Photoshop, you
have explained something that I did not previously know. If, as you
wrote, the scanner software was making the Levels, Curves and Color
Balance adjustments in high-bit mode but EXPORTING the data to
Photoshop in 8-bit mode, then that would indeed explain the
significant advantage of making such edits in the scanner software,
and why the initial histogram in Photoshop would be nice and smooth.

Again, a few things to keep in mind. The histogram is an absolutely
essential tool, but a histogram must be viewed in context, otherwise
the conclusions will be in error.

Namely, a smooth histogram is not always a "good thing". It depends on
how it was generated. A smooth histogram of a 16-bit scan reduced to
8-bits is "good" but an 8-bit scan which was modified by scanner
software to produce a smooth histogram is "bad" because this
smoothness was achieved by corrupting the data instead of a true
representation of what came from the scanner.

For example, if you scale to 99% and scan at 8-bit you will get a
smooth histogram, but this will be the results of scanner software
taking raw scan data and then interpolating this to 99%. A simple
visual comparison of a 100% 8-bit scan (gappy histogram) to a 99%
8-bit scan (smooth histogram) will show that the former is
considerably sharper than the latter.

3. Finally, as to the better/preferred scanning technique:

I personally don't print so I'll let other people comment on that.
However, as a rule of thumb, you should always scan at target
resolution. Any change of image size - even in Photoshop - will always
result is some loss of quality, in particular the image will be less
sharp.

The only thing I would harp on is to scan as raw as possible i.e. not
use any of the scanner editing tools. The only thing I personally use
in scanner software is exposure i.e. Analog Gain and focus.

I look at it this way: Using scanner software editing tools is
comparable to taking a Polaroid picture, while scanning raw is
comparable to shooting negatives. In the former you'll do all your
"editing" and decision making at the time of taking the picture and
won't be able to meaningfully improve afterwards. Using the latter
will give you the opportunity to improve afterwards and repeat this
until you get satisfactory results.

All usual caveats apply: My 2c worth... Your mileage may vary... etc.
;o)

Don.

 

[false_dmitrii]

<snip>

I meant to catch you and Don during your earlier debate, but this
thread looks close enough. I'm curious whether you'd recommend
scanning RAW instead of "corrected" when the following factors are
considered:

1) The goal is to archive as much of the original image data as
faithfully as possible for my own and perhaps others' unforeseen
future use. In other words, I worry that making even basic
adjustments to the original might somehow reduce a future user's
flexibility. Particularly relevant for faded older photos, where even
the slightest differences in HSL could be useful to someone. I'm not
sure how the initial difference between the original and the scanner's
perception of it factors into this.

2) I'm still too inexperienced to trust my own judgment when
"correcting" color, contrast, etc. My fear is that at a later date I
would think a photo had been adjusted toward the wrong direction. For
instance, I boosted saturation in the scanner software for a number of
early scans. Now I'm not so sure, especially since it often caused
apparent clipping at the high end of the red channel. Scanning RAW
would allow me to approach an image differently without rescanning the
original.

3) At present, I'm most interested in archiving my images. Current
output quality is a secondary concern. One can expect full 16-bit
editing to become available in consumer graphics software at some
point in the (distant?) future; from your snipped post I gather that
the editor-adjusted RAW file could then fully match the point-of-scan
adjusted output. (Can anyone reading this comment on the 16bpc
post-scan adjustment quality of Vuescan and Silverfast HDR, either of
which already has the potential to replace Photoshop CS for RAW image
adjustments?)

You and Don also awakened me to some basic issues surrounding exposure
and "proper" RAW scanning, but I'll post those separately.

Also, one of your comments left me with the impression that NikonScan
might be capable of "scanning" files. Is this the case?

Many thanks,
false_dmitrii

 

[RSD99]

"Howard" posted:
"...
In a nutshell (not a 55-gallon drum), it seems that your position is:
if Photoshop CS were truly, fully 16 bit (instead of the bug-induced
15-bit maximum), and if the scanner software were operated in full
(ignore potential future technology) 16/48-bit mode,
...."

Picture Window Pro is (or at least claims to be) a *full* 16-bit program.

http://www.dl-c.com/Temp/products/pwspecs.html

= = = = = Begin Quote = = = = =
"Full support for 16-bit Black and White and 48-bit Color Images

Picture Window Pro fully supports scanning, read, writing, and editing images with up to
16 bits per component. Increasingly, scanners are being introduced with the ability to
deliver 30-bit, 36-bit, or even 42-bit color data to your computer. This extended dynamic
range gives you more leeway to adjust the brightness and contrast of the image without
losing the smooth tonal range in the original image. If your scanner's TWAIN driver
supports this feature, Picture Window Pro not only lets you read 48-bit color images into
your computer, but all of its tools, transformations, and commands can be used to save,
load, and edit these images without first reducing them to 24 bits.

Picture Window Pro can read and write 16-bit black and white images in either TIFF or FITS
format, and it can read and write 48-bit color images in TIFF format."
= = = = = End Quote = = = = =


How does this figure into the discussion?

 

[Kennedy McEwen]

I'm curious whether you'd recommend
scanning RAW instead of "corrected" when the following factors are
considered:

1) The goal is to archive as much of the original image data as
faithfully as possible for my own and perhaps others' unforeseen
future use. In other words, I worry that making even basic
adjustments to the original might somehow reduce a future user's
flexibility. Particularly relevant for faded older photos, where even
the slightest differences in HSL could be useful to someone. I'm not
sure how the initial difference between the original and the scanner's
perception of it factors into this.

In principle, at some time in the future when Adobe come out with 24 or
32-bit functionality then you would get more precise results by saving
the data as a raw scanner output. However, that is merely precision,
not accuracy. Since you are limited to 16-bit output from the CCD then
you are not losing any significant accuracy in processing the data to
16-bit precision.

2) I'm still too inexperienced to trust my own judgment when
"correcting" color, contrast, etc. My fear is that at a later date I
would think a photo had been adjusted toward the wrong direction. For
instance, I boosted saturation in the scanner software for a number of
early scans. Now I'm not so sure, especially since it often caused
apparent clipping at the high end of the red channel. Scanning RAW
would allow me to approach an image differently without rescanning the
original.

It might, but remember that you can only see a certain range of tones -
just over 6-bits on a gamma corrected scale. That means that you need
to perform some pretty severe adjustments with 15 bit precision before
the limit of bit depth becomes visible. As you saw from the example I
used in the earlier thread, it took some extreme adjustments to make the
Photoshop limitations compared to Nikonscan become visible. Sometimes
such extreme adjustments are required and it is obvious when they are.
If you need major adjustments then do it in NikonScan or some other true
16-bit application first. But if you are looking at initial or
subsequent minor adjustments, even several minor adjustments in
sequence, then the PS limitation is pretty much irrelevant. I would
guess, from your comments, that this is the region you are concerned
about.

3) At present, I'm most interested in archiving my images. Current
output quality is a secondary concern. One can expect full 16-bit
editing to become available in consumer graphics software at some
point in the (distant?) future; from your snipped post I gather that
the editor-adjusted RAW file could then fully match the point-of-scan
adjusted output.

I would hope so, and I hope it won't be too long before Adobe wake up to
the fact that current imaging hardware has now surpassed the limitations
of their software. You might have seen Don't quote of Chris Cox's
comment referring to the 14-bit scanners that existed around the time it
was made - ignoring the fact that multiscanning made these same scanners
true 16-bit output machines. So there is some evidence that my hope is
in vain. :-(

You and Don also awakened me to some basic issues surrounding exposure
and "proper" RAW scanning, but I'll post those separately.

Also, one of your comments left me with the impression that NikonScan
might be capable of "scanning" files. Is this the case?

It can import some properly tagged TIFF, BMP and JPG files. Most
Photoshop output is OK, but I have found a few images that it balks on.
Some PaintShop Pro produced jpgs create some strange artefacts when
opened in NikonScan, but that might be just a bug in very old version of
PSP. However, NikonScan is not an image editor - it is fine if that is
the application that you are controlling scanner and output with, but I
certainly would not recommend it as a general purpose image editor.

 

[Kennedy McEwen]

"Howard" posted:
"...
In a nutshell (not a 55-gallon drum), it seems that your position is:
if Photoshop CS were truly, fully 16 bit (instead of the bug-induced
15-bit maximum), and if the scanner software were operated in full
(ignore potential future technology) 16/48-bit mode,
..."

Picture Window Pro is (or at least claims to be) a *full* 16-bit program.

http://www.dl-c.com/Temp/products/pwspecs.html

= = = = = Begin Quote = = = = =
"Full support for 16-bit Black and White and 48-bit Color Images

Picture Window Pro fully supports scanning, read, writing, and editing
images with up to
16 bits per component. Increasingly, scanners are being introduced with
the ability to
deliver 30-bit, 36-bit, or even 42-bit color data to your computer.
This extended dynamic
range gives you more leeway to adjust the brightness and contrast of
the image without
losing the smooth tonal range in the original image. If your scanner's
TWAIN driver
supports this feature, Picture Window Pro not only lets you read 48-bit
color images into
your computer, but all of its tools, transformations, and commands can
be used to save,
load, and edit these images without first reducing them to 24 bits.

Picture Window Pro can read and write 16-bit black and white images in
either TIFF or FITS
format, and it can read and write 48-bit color images in TIFF format."
= = = = = End Quote = = = = =


How does this figure into the discussion?

As a candidate that needs testing to validate its claims. ;-)

In fairness, Photoshop could make exactly the same claims - it also
works with images with 16-bits per channel, it just truncates them to
one level more than 15-bits first. Since Adobe have been peddling
16-bit compatibility myth for years and getting away with it, I would be
reluctant to take any software house at its word.

 

[Kennedy McEwen]

A few of things to keep in mind...

Even if you have a true 16-bit scanner IMHO it's still better to scan
raw and do all image editing in Photoshop.

Namely, when you view things in context, the loss of 1 bit in
Photoshop is more than offset by vastly larger toolbox and general
usability of Photoshop vs. scanner software.

<Snip> none of which is a valid argument for *not* processing with full
16-bit precision first using those tools which are available (and which
Howard defined earlier in the thread) *before* transferring the result
to Photoshop to take advantage of the wider range of tools. It is the
same argument as was used in the 8-bit PS days - for the same reason,
get your major adjustments out of the way first using the highest
precision tools available. When you have an image that is as close to
the end result as those tools enable, tweak the result with the wider
toolset since their limited precision is less significant at that stage.

The only thing I would harp on is to scan as raw as possible i.e. not
use any of the scanner editing tools. The only thing I personally use
in scanner software is exposure i.e. Analog Gain and focus.

Got Photoshop 15-bit multiscan working yet?

I look at it this way: Using scanner software editing tools is
comparable to taking a Polaroid picture, while scanning raw is
comparable to shooting negatives. In the former you'll do all your
"editing" and decision making at the time of taking the picture and
won't be able to meaningfully improve afterwards.

I believed we had established that this was untrue - your example
indicated misinterpretation of the indicators available to you, not a
limitation of the workflow.

Using the latter
will give you the opportunity to improve afterwards and repeat this
until you get satisfactory results.

Assuming, of course, that you don't need full bit depth for the result
you intend to produce - then you will be repeating it until Adobe
redesign PS! Anyone for multiscanning? ;-)

 

[Howard]

Thanks, Kennedy.

But it seems that, yet again, a follow-up is in order. Fortunately,
there are only two, and they are brief.

1) You wrote:
“ . . . make sure that you use the scanner software to optimise the
exposure or analogue gain as part of the production of the "wide open"
scan.â€

What does it mean “to optimise the exposure or analogue gain?†If it
helps, I’m now using an Epson 4870 scanner.


2) You wrote:
“. . . what was a huge file 3 or 4 years ago that very few systems
could handle now commonly fits within the RAM complement of most
machines, so the optimum compromise has moved towards making best use
of the information in the larger file.â€

Is this your way of saying that, today, the best compromise is (NOT to
scan at the next highest ppi level which is an integer divisor into
the scanner's maximum optical that will provide enough pixels for the
desired print size, and let the printer resolution fall where it may,
but instead) to (i) scan at the scanner’s highest optical resolution,
(ii) resize the image in Photoshop as desired without resampling, then
(iii) resample to the preferred resolution (such as 360 for an Epson
printer)?

Thanks again,

Howard

 

[Howard]

Don, Thanks for your response.

I do have just one question, which I just posed to Kennedy a few
minutes ago.

While it seems that Kennedy has a more flexible view of using the
scanner software's editing adjustments than scanning raw, wide open,
similar to you he specifically noted that, even if one prefers to scan
"wide open" and do all editing in Photoshop, it it still important to
use the scanner software's "analog gain" function (and you added
"focus" as well).

What are these tools? I note that I use only a flatbed scanner for
reflective material (I do photo restorations). From what I've read
elsewhere, it seems that adjusting "focus" is important in a film
scanner, but I'm not at all familiar with "analog gain". Is it, too,
a tool for film scanners?

Thank you.

Howard

 

[Kennedy McEwen]

Thanks, Kennedy.

But it seems that, yet again, a follow-up is in order. Fortunately,
there are only two, and they are brief.

1) You wrote:
“ . . . make sure that you use the scanner software to optimise the
exposure or analogue gain as part of the production of the "wide open"
scan.â€

What does it mean “to optimise the exposure or analogue gain?†If it
helps, I’m now using an Epson 4870 scanner.

In that case, ignore my comment since I don't think that scanner offers
exposure or analogue gain control - at least not with the standard Epson
software. Vuescan might though.

2) You wrote:
“. . . what was a huge file 3 or 4 years ago that very few systems
could handle now commonly fits within the RAM complement of most
machines, so the optimum compromise has moved towards making best use
of the information in the larger file.â€

Is this your way of saying that, today, the best compromise is (NOT to
scan at the next highest ppi level which is an integer divisor into
the scanner's maximum optical that will provide enough pixels for the
desired print size, and let the printer resolution fall where it may,
but instead) to (i) scan at the scanner’s highest optical resolution,
(ii) resize the image in Photoshop as desired without resampling, then
(iii) resample to the preferred resolution (such as 360 for an Epson
printer)?

Not quite. The reasons for downsampling by an integer factor are still
the same - spatial discontinuities. Although this is much less
significant than the way most scanners implement downsampling (ie. just
dropping CCD samples and overstepping) it still occurs and can
occasionally be visible, especially on slightly inclined edges.
Sometimes there just is no alternative to non-integer scaling to get
the image size you want though.

Apart from that though, if you have the memory in the system to handle
it, and you have implemented all of your levels and whatever adjustments
so that you don't run foul of PS bit depth limitations, then resampling
in PS with bicubic is the better compromise over doing so in the scanner
these days.

For an Epson desktop printer the final output should be sent in an
integer division of 720ppi (eg. 360ppi as you suggest) because the
printer driver resamples everything to 720ppi first using linear
interpolation. But that should not be achieved by using odd resampling
values unless you really have to produce a very specific size and cannot
do so by cropping the image slightly.

Integer resampling ratios always give better results and you should only
need to resample once.

 

[Don]

<Snip> none of which is a valid argument for *not* processing with full
16-bit precision first using those tools which are available

But when those tools are hampered (by the keyhole view, lack of
analysis utilities, etc) the loss of 1 bit is more than offset by the
myriad of Photoshop benefits which scanner software doesn't have.

Again, this is only the case when the scanner produces true 16-bit
output. When the scanner produces less than 16-bits, it's no contest.
Scanning raw and post-processing in Photoshop is always superior
barring a "lucky shot" adjustment in scanner software when - at best -
it can only hope to match raw + Photoshop.

The key, as always, is context: A ruler with cm in bright daylight
will always be more accurate than a ruler with mm in pitch darkness...

Got Photoshop 15-bit multiscan working yet?

Yup! A while back. However, I see no real benefit of multiscanning.
Although noise is masked there is no extra detail in the shadows. I
can achieve almost identical results - in a fraction of the time
needed for multiscanning - by simply selecting the shadows and
applying 0.3 Gaussian Blur.

I have now settled on my own version of contrast masking and - so far
- I'm getting excellent results. By adjusting histograms before
merging I have eliminated the major "problem" of contrast masking:
gradients i.e. color imbalance. And the detail in shadows is amazing
when compared to multiscanning.

I believed we had established that this was untrue - your example
indicated misinterpretation of the indicators available to you, not a
limitation of the workflow.

No, we have not established it was untrue. We have established it's
not your preferred workflow.

Judging by the long and detailed message Howard obviously likes to
think about these things and would like to have all the facts. I have
just provided the rest of the facts.

Now it's up to Howard (or anyone else reading this) to decide - based
on his requirements - whether he chooses to discard these additional
facts or take them into consideration.

Don.

 

[Don]

Don, Thanks for your response.

I do have just one question, which I just posed to Kennedy a few
minutes ago.

While it seems that Kennedy has a more flexible view of using the
scanner software's editing adjustments than scanning raw, wide open,
similar to you he specifically noted that, even if one prefers to scan
"wide open" and do all editing in Photoshop, it it still important to
use the scanner software's "analog gain" function (and you added
"focus" as well).

What are these tools? I note that I use only a flatbed scanner for
reflective material (I do photo restorations). From what I've read
elsewhere, it seems that adjusting "focus" is important in a film
scanner, but I'm not at all familiar with "analog gain". Is it, too,
a tool for film scanners?

Ah sorry... In my never-ending fight with Kodachrome slides I have a
film scanner on the brain... ;-)

However, the gist of raw vs. "cooked" scanning applies to flatbeds as
well.

Analog Gain is just another word for exposure. When you scan you're
really "taking a picture of a picture". So, just like when you exposed
the film in the first place, you have to expose the "digital film",
i.e. the CCD elements in your scanner, when scanning.

Therefore, exposure is just as important in scanning as it was when
shooting the original image. Perhaps even more so, because when you
shoot reality with your camera you have a much more elbow room
because, by comparison, reality has a very good dynamic range as does
your film, etc. When you "shoot" a film or a photo with the scanner
you are already working with "second hand reality" which makes correct
exposure even more important.

As I think Kennedy mentioned I'm not aware of focus being available on
flatbeds but don't take my word for it. (Anyone who knows otherwise
please chime in!) I think this may be because on flatbeds the image is
always in the same place (pressed by the cover against the glass) so
the focus is pretty much known and fixed. On film scanners, however,
the film "floats in space" as the light is shone through it so the
focus is very important. Indeed, curved film has been known to make
grown men cry... ;o)

Another reason for lack of focus on conventional flatbeds may be
because they have a pretty good depth of field which is why you can
even scan real life objects which, in part, may hover above the glass
and still come out sharp. The only exception, to my knowledge, is
Canon Lide series of scanners which use a different technology and
have no optics like conventional flatbeds but the image is acquired
directly like on a fax machine. I'll now tag Kennedy and pass it on to
him to explain the details... ;-)

Don.

 

[Don]

1) The goal is to archive as much of the original image data as
faithfully as possible for my own and perhaps others' unforeseen
future use. In other words, I worry that making even basic
adjustments to the original might somehow reduce a future user's
flexibility. Particularly relevant for faded older photos, where even
the slightest differences in HSL could be useful to someone. I'm not
sure how the initial difference between the original and the scanner's
perception of it factors into this.

Indeed, that is the key but there are a couple of thoughts which
crossed my mind since I started wrestling with "digitizing my life".

Even if today's Photoshop is "only" 15-bit, this is bound to go up, if
for no other reasons than for "one-upmanship". So saving an
unadulterated, raw image, uncorrupted by scanner software adjustments
is essential in that respect so that the image can be re-edited when
the tools catch up.

On the other hand, one also has to consider the context and order of
magnitude. The former being how the images are viewed. If they are
printed (something I don't do at all, so take this with a grain of
salt) the narrower CMYK gamut can't even cope with what current
scanning technology is capable of. But even if/when printing
technology catches up we come to the latter (order of magnitude) which
also applies to display technology. Can one really tell the
difference? Even if the relevant technologies keeps improving! Sort of
a reality check question which I for one keep forgetting in my quest
for perfection. For example...

When I got my flatbed scanner I (like most people) wanted one that
"goes to 11" so I went for a scanner with the highest resolution
(4800). But as I learned more, I realized that due to the nature of
photographs the consensus among those in the know is that anything
above 600 is a waste of time simply due to the nature of photos. I'm
still busy with slides so all that is still ahead of me...

2) I'm still too inexperienced to trust my own judgment when
"correcting" color, contrast, etc. My fear is that at a later date I
would think a photo had been adjusted toward the wrong direction. For
instance, I boosted saturation in the scanner software for a number of
early scans. Now I'm not so sure, especially since it often caused
apparent clipping at the high end of the red channel. Scanning RAW
would allow me to approach an image differently without rescanning the
original.

I'm in the same boat. I agonized literally for months - more than a
year, actually - over Nikon's inability to scan Kodachromes correctly.
There's lots of pulled hair around my scanner and scratchmarks on the
walls which I climbed in exasperation... ;o)

That's why my own goal all along was a workflow which, preferably,
eliminates the most unreliable part (me!) or at the very least
minimizes my (negative) influences. So I always favored a workflow
which was based on objective rules rather than based on my subjective
opinion. This involved numerous tests (and I mean NUMEROUS!) as well
as agonizing over the results, always trying to keep reminding myself
to pull back and think of the context instead of getting lost in the
minutiae of whether a fraction of a pixel matters.

3) At present, I'm most interested in archiving my images. Current
output quality is a secondary concern. One can expect full 16-bit
editing to become available in consumer graphics software at some
point in the (distant?) future; from your snipped post I gather that
the editor-adjusted RAW file could then fully match the point-of-scan
adjusted output. (Can anyone reading this comment on the 16bpc
post-scan adjustment quality of Vuescan and Silverfast HDR, either of
which already has the potential to replace Photoshop CS for RAW image
adjustments?)

I'm also doing all this for archiving purposes and, just like you, I'm
not concerned with output as that is bound to change. But, by the same
token, so will the input technology as scanners improve.

For example, having gone up from LS-30 to LS-50 I expected all of my
problems with the LS-30 to go away. And during the honeymoon they did.
But as I continued testing some came back and new ones appeared. In
particular, the wider dynamic range of the LS-50 did not solve the
problems in shadows of my Kodachromes and I still have to contrast
mask.

So, considering the durability of photographic media (fading of photos
notwithstanding) I also ask myself if all the time I'm putting into
this is really productive considering that "new, improved" 24-bit
scanners are "just around the corner"...


Anyway, sorry about waxing lyrical, but I think many people go through
the same process. In the end, I think it's really a question of
balance and priorities. Although different people have many identical
and overlapping concerns I think in the final analysis it's a personal
decision based on each person's individual context and requirements.
Sounds wishy-washy and non-committal, but it's true...

Also, one of your comments left me with the impression that NikonScan
might be capable of "scanning" files. Is this the case?

I think you may be referring to NEF files. That's Nikon's version of
Raw scans which you can import back into Nikon scan to apply curves,
etc. I haven't really done much with them since my LS-30 days. That's
an 8-bit scanner which internally uses 10-bits. However, NEF files
were still 8-bit and - in my book - that's not raw! So I personally
ignore NEF files and just make my own raw scans.

BTW, even if NEF files were true raw files I still see no advantage to
loading them back into NikonScan because I'd rather use Photoshop
which has a much wider toolbox and a far better working environment.
And since Photoshop (at least my v. 6) doesn't support NEF (which is a
proprietary file format and therefore - for that fact alone - a very
good reason to avoid it at all costs) I simply make my own raw TIF
files (which, to boot, can also be loaded back into NikonScan, not
that I would want to do that).

Don.

 

[Kennedy McEwen]

But when those tools are hampered (by the keyhole view, lack of
analysis utilities, etc) the loss of 1 bit is more than offset by the
myriad of Photoshop benefits which scanner software doesn't have.

As you would have discovered by now had you undertaken the calculations
I suggested what you consider to be "hampering" is much less significant
that the 1-bit difference.

Again, this is only the case when the scanner produces true 16-bit
output. When the scanner produces less than 16-bits, it's no contest.

Of course, that is why you have such success with multiscanning several
less than 16-bit images - such success that you can't even see the point
of it!

Scanning raw and post-processing in Photoshop is always superior
barring a "lucky shot" adjustment in scanner software when - at best -
it can only hope to match raw + Photoshop.

The key, as always, is context: A ruler with cm in bright daylight
will always be more accurate than a ruler with mm in pitch darkness...


Yup! A while back. However, I see no real benefit of multiscanning.

So you haven't got it working at all then! Being hampered by
Photoshop's limited bit depth is no reason to write off multiscanning,
which does work and does offer real improvements.

 

[Kennedy McEwen]

As I think Kennedy mentioned I'm not aware of focus being available on
flatbeds

I don't recall mentioning that.

but don't take my word for it. (Anyone who knows otherwise
please chime in!)

I am fairly certain that the Epson Expression (not the Perfection) range
of scanners have user adjustable focus.

 

[Don]

As you would have discovered by now had you undertaken the calculations
I suggested what you consider to be "hampering" is much less significant
that the 1-bit difference.

Theory is very nice - in theory... And as you know I always thirst to
learn more of it, but - as can be seen below - practice (i.e. context)
plays a significant part in real life situations. Which all reminds me
of...

In theory, a bumblebee was once declared insufficiently aerodynamic to
be able to fly. In practice, numerous bumblebees strongly disagreed...

And this bumblebee also strongly disagrees... ;o)
Do note that the stress should be on the "bee" part (as in "worker
bee"), not on the "bumble" part... ;o)

So you haven't got it working at all then! Being hampered by
Photoshop's limited bit depth is no reason to write off multiscanning,
which does work and does offer real improvements.

In theory...

According to you, due to Photoshop's 15-bit and integer math
"limitations", multi-pass multi-scanning only makes sense with up to 2
scans. And yet...

I have scanned 18 times and then I threw away 2 most extreme scans.
Next I used Photoshop to combine the remaining 16 scans in increments
of 4 to end up with: 4x, 8x, 12x and 16x multiscans. Comparing those
there is a clear and incremental reduction of noise at each junction
with 16x nearly eliminating all, perhaps with only about 1% of shadows
- if that - still having some very minor noise.

However, there was no increase in shadow detail. As I already
mentioned, a very similar effect can be achieved by simply selecting
the shadows (threshold = 32) and applying 0.3 Gaussian Blur. The
multiscanned images are still slightly superior in that they appear a
tad sharper.

However (and that's a big "however") I could only observe this at
300-400% magnification. Even at 100% this minor loss of sharpness of
Gausian Blur image was difficult to observe. So, for all practical
purposes, Gaussian blur of shadows (anything below 32) achieved
virtually the same effect as multiscanning.

Finally, for the purposes of full disclosure, I did this on an LS-50
ergo, 14-bit scans and, of course, Kodachromes. I pushed the exposure
until highlights touched the right edge of the histogram, but did not
clip. In spite of this conservative exposure, the dynamic range of the
LS-50 was sufficient so that anything below 16 contained no image
data. The shadows I selected were anything below 32 since anything
above 32 did not have any noise (inspected, again, at 300-400%
magnification).

Therefore, we are only dealing with about 16 clicks on the histogram
(~16-32) where noise was a problem. I find that, as a rule of thumb,
an Analog Gain boost of 2 notches shifts this part of the histogram
sufficiently (above 32) so that these 16 clicks in the histogram lose
all shadow noise. And that's what I use for my "contrast merging", as
I call it...

Comparing shadow areas (< 32) in the two scans:
A: 16x multi-pass, multi-scan
B: single scan but with 2 Analog Gain clicks boost, followed by
histogram adjustment to bring it down to same level as A.
The difference between the two was considerable as B revealed a
significant amount of detail not visible in A - and that's even after
B was "darkened" to bring it down to A levels.

Please repeat this (it's a genuine question because I really want to
know) and report if you see any difference between the two. I
understand if you can't, because sub-pixel alignment of multi-pass
scans is excrutiating and very time consuming. It took me about a 1
day per image!

Anyway, I'm curious for two reasons: the effect of single-pass
multi-scan, and NikonScan's floating point math.

BTW, do you have any references that NikonScan uses floating point
(when calculating multiscanned images) and not integer math like
Photoshop? I can't find anything in the manual.

Don.