Step-by-step procedure for scanning film

[Lorenzo J. Lucchini]

I'd like to know if the following checklist for scanning film makes
sense and is complete enough.

In particular, I'd like comments on the order of the steps, and on
[sometimes] [possibly] useful steps I haven't included -- keeping in
mind that I only want things that can be done without chiefly subjective
judgments or an "artistic eye".


1) Take a preview of the whole scan area
2) Find each picture in the preview and take note of positions and sizes
3) Find each channel's whitepoint for each picture in the preview, and
take note
4) Scan each picture with an exposure time for each channel that is
inversely proportional to the whitepoint found (possibly with some
safety margin to avoid clipping); if scanning at less than the scanner's
native bit depth, then also tell the scanner to apply curves and levels
corresponding to the film's characteristics, as well as to the scanner
profile if available
5) Repeat step 4 if multi-pass scanning
6) Align each scan if multi-pass scanning
7) Invert the images if scanning negative film, unless this was done in
step 4 using the scanner's curves settings
8) Save the current images for archival if desired
9) Correct dust and scratches, using infrared channel if available
10) Apply noise reduction if needed
11) Adjust curves and level by film's characteristic curves and scanner
profile, if not done in step 4
12) Apply sharpening (*)
13) Crop, resize if desired, quantize to 8 bpc if desired, save


(*) If sharpening amound and kind was computed from measured scanner
characteristics (if you don't know what I mean, nevermind ;-) ), then
step 12 should be done *before* curves adjustment; however, this is only
possible if curves adjustments were done in step 11, and not if they
were done in step 4.
So, perhaps sharpening parameters should be adjusted depending on the
curves used.


by LjL
(e-mail address removed)

 

[Don]

3) Find each channel's whitepoint for each picture in the preview, and
take note
4) Scan each picture with an exposure time for each channel that is
inversely proportional to the whitepoint found (possibly with some
safety margin to avoid clipping)

That's essentially what Auto Exposure does, post-scan (i.e. in
software).

The only thing to keep in mind is that if image highlights are not
really white this will introduce a cast! You probably know this, but
it's an important thing to note.

Of course, by scanning each channel using its optimal exposure (even
if that does result in a cast) you get the most data giving you the
most elbow room for any subsequent adjustments. And that's a Good
Thing! However, it may require more skill - and certainly more work -
in post processing later.

; if scanning at less than the scanner's
native bit depth, then also tell the scanner to apply curves and levels
corresponding to the film's characteristics, as well as to the scanner
profile if available

That's a matter of choice. Personally, I would not apply any editing
at the scanning stage and - somewhat controversially - this includes
the scanner profile!

In general, a scanner profile may also sometimes go against the image
and would, essentially, be undone in post processing. In your specific
case, a scanner profile may undo all the careful work you have done to
get the exposure right!

But more importantly, a scanner profile and individual channel
adjustment don't get along - at all!!! The exposure adjustments are
linear while the scanner profile is (virtually always) non-linear!
Editing such an image afterwards is going to be a nightmare!

Don.

 

[Lorenzo J. Lucchini]

That's essentially what Auto Exposure does, post-scan (i.e. in
software).

Uh, but doesn't "auto exposure" actually adjust *exposure* (i.e. at scan
time) in most scanners?
It almost sounds like fraud to call it "auto exposure" if exposure is
not in the slightest involved!

The only thing to keep in mind is that if image highlights are not
really white this will introduce a cast! You probably know this, but
it's an important thing to note.

Of course, by scanning each channel using its optimal exposure (even
if that does result in a cast) you get the most data giving you the
most elbow room for any subsequent adjustments. And that's a Good
Thing! However, it may require more skill - and certainly more work -
in post processing later.

I do realize that a cast can be introduced. However, if you had taken
note of the whitepoint values, you can always restore the original
whitepoint, although this is certainly best done with high bit depth
data (using a 8-bit image scanned with the correct exposure time and
then twiddling too much with levels doesn't sound like a fine idea).

That's a matter of choice. Personally, I would not apply any editing
at the scanning stage and - somewhat controversially - this includes
the scanner profile!

But I was thinking about scanners that apply these changes to a high bit
depth version of the image, before sending the 8 bpc data to the
computer -- as you know, my own scanner does this.

With such a scanner, I think it's definitely better to apply curves and
levels at scantime when scanning at 8-bit, as you have less data to work
with afterwards.

By the way, I'm not too sure that my scanner is a rarity in this
respect, at least among flatbed scanners: do you remember all those
flatbets that were (are?) sold as "36-bit external, 24-bit internal"?
Well, they did exactly what my scanner does, i.e. apply curves and
levels to a 36-bit image but send a 24-bit image to the computer --
except that my scanner also allows sending the 36-bit (48-bit, actually)
image directly.

In general, a scanner profile may also sometimes go against the image
and would, essentially, be undone in post processing. In your specific
case, a scanner profile may undo all the careful work you have done to
get the exposure right!

About the profile adjustments being undone in post processing, I see
this can happen. However, now I was mostly thinking about a scanning
procedure that could be done automatically by the computer, with no or
as little as possible human intervention.

In this case, certainly the output image should be as color correct as
possible, even though this might do more ill than good if the image is
to be edited afterwards.

But, there's always 16-bit scanning that can greatly help with this kind
of things.

But more importantly, a scanner profile and individual channel
adjustment don't get along - at all!!! The exposure adjustments are
linear while the scanner profile is (virtually always) non-linear!
Editing such an image afterwards is going to be a nightmare!

That's a problem, yeah.
But I think there are formulas around -- even though I don't quite know
them -- for, well, correcting the color correction matrix depending on
the exposure time.

Perhaps [some] scanners even do this automatically, who knows, I should
test mine for this -- doubt it, though, as my SANE driver has hardcoded
color correction values that are different for negative (high exposure
time) and positive (low exposure time) film.


by LjL
(e-mail address removed)

 

[Don]

Uh, but doesn't "auto exposure" actually adjust *exposure* (i.e. at scan
time) in most scanners?
It almost sounds like fraud to call it "auto exposure" if exposure is
not in the slightest involved!

I was referring to "Auto" mode in Photoshop Levels or Curves, for
example. It uses the same principle/concept to set the white point.
It's that principle I was referring to (outlined in the subsequent
paragraph).

The point I was making is that trying to set the White Point by simply
adjusting all channels individually until they line up (whether with
exposure or with editing) does not necessarily set the white point
correctly (if the highlights in the image are not pure white).

I know you're aware of this, but it's just something which is
important to mention.

About the profile adjustments being undone in post processing, I see
this can happen. However, now I was mostly thinking about a scanning
procedure that could be done automatically by the computer, with no or
as little as possible human intervention.

I know. That's exactly what I'm after, too! An optimal procedure I can
automate and leave all the "thinking" and difficult esthetic decisions
for later.

In this case, certainly the output image should be as color correct as
possible, even though this might do more ill than good if the image is
to be edited afterwards.

I constantly flip-flop between the two approaches. One, have the scan
reproduce the film as faithfully as possible, even its shortcomings.
Two, ignore all that and just get the most (dynamic range) out of the
film.

The problem with the first approach is that using linear exposure
(i.e. Analog Gain) to correct non-linear casts (be it Kodachrome, or
scanner introduced) is physically impossible.

The problem with the second approach is that it goes against the
"prime directive" above (automated scanning) because each image must
be examined separately. It's also much more work to edit such an image
afterwards.

That's a problem, yeah.
But I think there are formulas around -- even though I don't quite know
them -- for, well, correcting the color correction matrix depending on
the exposure time.

Adjusting (i.e. removing or adding) exposure in software is easy
because it's linear (although it may result in loss of data, of
course, if there is clipping). But the question then becomes why apply
exposure in the first place in hardware if it's going to be removed
later in software?

The only solution is to evaluate each image separately before scanning
to determine if the exposure adjustment of individual channels should
be made. But that goes against the "prime directive" of automated
scanning. Back to square one... :-/

Perhaps [some] scanners even do this automatically, who knows, I should
test mine for this -- doubt it, though, as my SANE driver has hardcoded
color correction values that are different for negative (high exposure
time) and positive (low exposure time) film.

A scanner profile combined with a film profile are *supposed* to do
just that.

However, in actual usage - at least as far as my Nikon goes - it just
doesn't go far enough. It's (sort of... :-/)) OK at the nominal
exposure (absolute exposure of 0) but once that is changed/boosted the
channels go right out of sync. Back to square one... :-/

Don.

 

[false_dmitrii]

On Mon, 10 Oct 2005 22:10:33 +0200, "Lorenzo J. Lucchini"

That's a problem, yeah.
But I think there are formulas around -- even though I don't quite know
them -- for, well, correcting the color correction matrix depending on
the exposure time.

Perhaps [some] scanners even do this automatically, who knows, I should
test mine for this -- doubt it, though, as my SANE driver has hardcoded
color correction values that are different for negative (high exposure
time) and positive (low exposure time) film.

A scanner profile combined with a film profile are *supposed* to do
just that.

However, in actual usage - at least as far as my Nikon goes - it just
doesn't go far enough. It's (sort of... :-/)) OK at the nominal
exposure (absolute exposure of 0) but once that is changed/boosted the
channels go right out of sync. Back to square one... :-/

Isn't this why profiles are supposed to be created from a scan of the
target at the exact same exposure settings that will be used for the
profiled images? Or am I misunderstanding something?

false_dmitrii

 

[Bart van der Wolf]

SNIP

Isn't this why profiles are supposed to be created from a scan
of the target at the exact same exposure settings that will be
used for the profiled images? Or am I misunderstanding
something?

No, you're correct, anything different from the exposure level for the
profile will introduce color shifts.

Bart

 

[Don]

Isn't this why profiles are supposed to be created from a scan of the
target at the exact same exposure settings that will be used for the
profiled images? Or am I misunderstanding something?

No, and that's exactly why the built-in Kodachrome setting on Nikons
is virtually useless (even at absolute exposure of 0 it still leaves
some of the blue cast - actually, absence of red).

What's inexcusable is that the characteristic curves of Kodachromes
are well known and documented. And Nikon - of all people - should be
able to create a dynamic implementation (just like tone mapping in HDR
images copes with different exposures).

Sure, because of many different versions of Kodachromes it would be a
"guess" to some extent, but it would be orders of magnitude better
than the "blue horror" it now produces.

Don.

 

[Lorenzo J. Lucchini]

SNIP



No, you're correct, anything different from the exposure level for the
profile will introduce color shifts.

But, is my theory (call it rather my hope) that the color profile could
be mathematically modified depending on exposure completely wrong?

Note that I'm thinking about profiles working with a 9x9 matrix, not
"numeric" profiles or whatever they're called.


by LjL
(e-mail address removed)

 

[Bart van der Wolf]

Bart van der Wolf wrote: SNIP

But, is my theory (call it rather my hope) that the color profile
could be mathematically modified depending on exposure completely
wrong?

Unfortunately, the color space shape is not symmetrical. The only way
to approximate its shape, is by using a look-up table with many
sampled points and interpolate/extrapolate for other levels.

Note that I'm thinking about profiles working with a 9x9 matrix, not
"numeric" profiles or whatever they're called.

They are called "TRC"s, Tone Reproduction Curves. Even if you use a
9x9 matrix you'd have to make look-up tables for different luminance
levels and they'd still be less accurate than the TRC versions.

Bart

 

[notsosimple]

By "whitepoint", did you mean true gray, or something else?

If there is no true gray in the image, wouldn't your suggested
adjustment introduce (or remove) color cast?

In your original post, there was no mention of adjusting a scanner's
focus.

 

[notsosimple]

Uh, but doesn't "auto exposure" actually adjust *exposure* (i.e. at scan
time) in most scanners?

Not necessarily. Some scanners provide ways to adjust the hardware
exposure, but some don't. But all scanners' user interface will provide
some way to adjust exposure which may mean separate hw and sw controls,
or only sw control. It is up to the users to find out and distinguish
between the two.

It almost sounds like fraud to call it "auto exposure" if exposure is
not in the slightest involved!

Boy, are you naive.

Oops, just noticed I repeated this.

 

[ljlbox]

(e-mail address removed) ha scritto:

By "whitepoint", did you mean true gray, or something else?

What do you mean by tryue gray?

No, by "whitepoint" here I mean this: after you've taken the preview,
find the highest pixel values for red, green and blue.
You'll come up with something like, say, R250 G100 B80.

Then, if you call your scanner's standard exposure "1", you'll set
exposure to red = 255/250, green = 255/100, blue = 255/80.

If there is no true gray in the image, wouldn't your suggested
adjustment introduce (or remove) color cast?

Yes, it would. But the point is that it would also maximize the signal
to noise radio, by using the longest exposure that doesn't cause
clipping.

You can always remove any cast afterwards in software, if you just
don't trash the values R250 G100 B80.

In your original post, there was no mention of adjusting a scanner's
focus.

There wasn't because my scanner is fixed focus, but it's certainly an
important step to add.
Only, all I could say for that step would be "adjust focus", as I don't
really know what kind of procedure is involved.


by LjL
(e-mail address removed)

 

[ljlbox]

(e-mail address removed) ha scritto:

:

[snip]

It almost sounds like fraud to call it "auto exposure" if exposure is
not in the slightest involved!

Boy, are you naive.

Guess so... but if it isn't hardware, it's "auto levels" (like just
abotu every graphics program calls it), so just call it auto levels!
Perhaps "fraud" is a big word, but what the heck.

[snip]

by LjL
(e-mail address removed)