VueScan or NikonScan?

[Kennedy McEwen]

I tried some scans, and I can't find a negative that really show something
wrong. There is a small difference at the extremes (shadows and high
lights) but without a clear example, it is just theoretical.

I did notice that NS does weird things with the frame boundaries. I scanned
a negative as negative, and selected a crop. Switch to positive, and the
next preview after switching, the frame was completely off. Very weird.

Vuescan makes some strange errors with framing if you tell it you have
positive film and feed it negative or vice versa too, its almost
inevitable given what the software is looking for as a frame boundary.

Anyhow, in past I had some scans where NS made a mess of things. But it
could have been NS 2.x or the LS-2000. I'll try to find a frame that shows
the difference.

This is starting to sound like a far cry from your original position
that this was some sort of intentional bug... ;-)

The thread has, from the beginning been about current, or at least very
recent, versions of the software in the subject line. However, I go
back to at least NS1.6, and have never seen the 'bug' you describe
although, of course, the software was very different, and much less user
friendly, then.

In the past I have had lots of scans that Vuescan made a mess of -
mainly because early versions of Vuescan did not include any form of
preview, indeed Ed argued vociferously against its need at all for a
while, but it was eventually included for various reasons including user
demand. That failure of early versions of Vuescan hasn't caused me, or
anyone else, to decry current implementations of the program as
intentionally bug ridden. It did, however, cause me to persevere with
early versions of NikonScan, which has paid dividends in rapidly coming
to terms with later versions.

I still look at the latest releases of Vuescan now and again to see if
there is anything I am missing, but have never been sufficiently tempted
to adopt it as my primary scanning interface. I accept that it is
easier to get a good first scan from even for a novice, and I see that
this could be a major attraction, but I have never found it to be able
to scan anything that I cannot also scan with the Nikon software.

 

[Dave]

I think you're right, Don. Good catch. I didn't do any cropping, but
VueScan has that autocropping feature that I didn't turn off. I
should have caught that myself.

The jury is still out on which program to use. I'm getting better at
scanning with more knowledge and practice (Funny how that works!) and
now I'm getting decent scans with both programs. Both have advantages
and disadvantages. No winner yet.

 

[WD]

Interesting point. I do not know whether or not luminance or
composite RGB is used in Nikon Scan auto-exposure. But in any case
there is still no easy way to resolve this issue with Nikon Scan.
It is a shame, because Nikon Scan has a number of good things going
for it (quick 6 frame preview, access to ICE4 algorithms).

WD

 

[Raphael Bustin]

Interesting point. I do not know whether or not luminance or
composite RGB is used in Nikon Scan auto-exposure. But in any case
there is still no easy way to resolve this issue with Nikon Scan.
It is a shame, because Nikon Scan has a number of good things going
for it (quick 6 frame preview, access to ICE4 algorithms).

Even if you do use Auto Exposure in NikonScan,
there's nothing to keep you from pushing out the
white point or black point of any channel, or the
whole RGB composite.

Ie., you can elect to have the whole image
mildly compressed, and make sure that no
pixels anywhere are clipped at either end.

That's assuming of course that there's no clipping
within the film or due to the scanner's dynamic
range -- ie., the "raw" histogram isn't clipped to
begin with.


rafe b.
http://www.terrapinphoto.com

 

[Philip Homburg]

This is starting to sound like a far cry from your original position
that this was some sort of intentional bug... ;-)

There is some additional 'auto-contrast step' that does destroy some
(invisible) information. Setting the black point above 0 is also
intentionally, and it is annoying. (And it is not documented, so it could
be considered a documentation bug).

However, it is possible that it is the more natural operation of the analog
gain in positive mode that results in better scans, and not so much the
automatic processing of NS.

As I said, I'll look for good examples. But you are right, I should
not claim anything about NS unless I put the sample images on-line first.

 

[WD]

The white and black points cannot be pushed out in Nikon Scan that is the issue.

 

[Dave]

And the winner is...drum roll, please...VueScan! Both programs can
make good scans from newer negatives and slides, and NikonScan's User
Interface is better, but I was working with a 13-year-old strip of
negative film and trying to save one particular photo of me standing
in the desert in front of some tents. With NikonScan, I had the
choice between turning all color correction off and accepting a
desert-brown cast on everything or turning minimal color correction on
and seeing myself with a purple face. Minimal color correction with
VueScan fixed the other colors in the photo and gave me a slightly
sunburned look. I chose sunburned pink over diseased purple.

My next project will be comparing scans of me with that salmon in
Alaska. It's a much newer piece of film, so I'm sure it will be much
easier.

 

[Kennedy McEwen]

The white and black points cannot be pushed out in Nikon Scan that is
the issue.

Both are determined by exposure and can be adjusted in either direction
- as demonstrated in the scan data I published yesterday.

 

[Kennedy McEwen]

There is some additional 'auto-contrast step' that does destroy some
(invisible) information.

I have already mentioned the control for the auto-contrast function,
which is the same control for the black and white point target level and
population percentage exclusion. Auto-contrast is performed, and indeed
with a negative *can only* be performed, after the data has been
captured.

Setting the black point above 0 is also
intentionally, and it is annoying. (And it is not documented, so it could
be considered a documentation bug).

Why do you consider it to be a bug at all? It is offset from zero
because zero it the equivalent of a full scale signal on the negative
and, since the colour negative has an orange mask, the base emulsion
level is lower than full scale. Level manipulation, using the curves
control, after the raw data has been captured is required to pull this
level back to black. Since each type of colour negative film (as well
as each type of colour negative process) produces a slightly different
orange mask, so each combination of film type and process requires a
different black level adjustment. NikonScan is not provided with any
predefined film settings, so it is up to the end user to find those
levels themselves either by using the autocontrast button, the
black/mid/white eye droppers or manual adjustment of the data. When
defined, it is a simple matter of saving this as a named profile for
that specific film and process combination. Vuescan does provide some
predefined negative film types, but relies on auto white balancing to
adjust for variations that occur in film processing chemistry.

However, it is possible that it is the more natural operation of the analog
gain in positive mode that results in better scans, and not so much the
automatic processing of NS.

I am not sure I understand this comment.

 

[Don]

I think you're right, Don. Good catch. I didn't do any cropping, but
VueScan has that autocropping feature that I didn't turn off. I
should have caught that myself.

It's very easy to overlook in the confusing VueScan's so-called "user
interface". This comment tells you which side of the fence I'm on...
;-)

But seriously, that's why I'm always very wary of all "auto" features
and have a tendency to turn them off. They're fine but you have to
constantly keep them in mind or they'll bite when you least expect.

The jury is still out on which program to use. I'm getting better at
scanning with more knowledge and practice (Funny how that works!) and
now I'm getting decent scans with both programs. Both have advantages
and disadvantages. No winner yet.

Although the crop thing above was not that problematic by itself, it
illustrates two of the many VueScan problems:
1. Very bad (non-existent?) user interface so individual options get
lost, especially as they are continuously flashed on and then hidden
virtually at random.
2. Options spread widely over different tabs affecting each other in
virtually unpredictable ways.

That's why the author's cure-all advice is always:
1. Delete VueScan.ini
2. Click Preview
3. Right click on a neutral gray area
4. Click Scan

So, apparently, even he can't keep track of all of the options and
their obscure interaction.

Don.

 

[Don]

In the past I have had lots of scans that Vuescan made a mess of -
mainly because early versions of Vuescan did not include any form of
preview, indeed Ed argued vociferously against its need at all for a
while, but it was eventually included for various reasons including user
demand.

Arguing *against* preview!?

This puts things into perspective regarding his kicking and screaming
against individual Analog Gain adjustments - also implemented, in the
end, after all the "argumentoids" ran out.

I thought the irrational Analog Gain intransigence was a one-off but
apparently it's a pattern.

Don.

 

[Dave]

I understand the "delete VueScan.ini" part, but why right click on a
neutral gray area? Does that re-set something?

I may have to rescind the "Best Software" award for the time being.
The standard settings gave VueScan the edge over NikonScan, but now
I'm working with the NikonScan image in Elements and it's much better
than the VueScan image. The next step is to use Elements on a VueScan
image and compare. Elements will be the constant and the starting
points will be the variable. This is fun, but very time consuming.

 

[WD]

Kennedy,

Where is the scan data you 'publlished yesterday'? I would like to
take a look.

We are going around in circles here, so let me state what I believe
Nikonscan
is doing.

1. It determines an exposure (probably does this adequately)
2. It scans the negative and collects all the data
3. If you can imagine a full dynamic range histogram, as you
stated,
the dynamic range of the scanner is far beyond a typical
negative's
dynamic range. Nikon Scan analyzes all the data. It internally
determines that much of the 'data' at either end of the
histogram
contains no useful information and 'throws it out'
4. Nikon Scan 'presents' what it has determined to be the part of
the data
(i.e. some portion of the full scanner dynamic range)
which contains the full dynamic range of the negative scanned.
5. What the user gets to operate on in terms of curves, white point
black oint targets etc. is this data
(with what NS has deemed to be the extremes of scanner dynamic
range
beyond the negative dynamic range not included).

It is my assertion that in steps '3/4' NS in fact throws out
some data
which actually does contain useful information from the
negative.

What proves this to me is the fact that both Vuescan and Silverfast
can provide
scans of the same negative on the same scanner which in fact includes
the parts
that are irretrievably thrown out by Nikonscan.

As further evidence, NS allows no way to retrieve the full dynamic
range
that the scanner sees when scanning a negative (barring tricks like
telling NS it is actually scanning a positive and inverting manually
later on
which in fact does capture the full dynamic range of negatives even
using
Nikonscan - further evidence).

 

[Kennedy McEwen]

I understand the "delete VueScan.ini" part, but why right click on a
neutral gray area? Does that re-set something?

I may have to rescind the "Best Software" award for the time being.
The standard settings gave VueScan the edge over NikonScan, but now
I'm working with the NikonScan image in Elements and it's much better
than the VueScan image. The next step is to use Elements on a VueScan
image and compare. Elements will be the constant and the starting
points will be the variable. This is fun, but very time consuming.

I don't think there has ever been any argument that Vuescan will give
the better scan straight out of the box without any user intervention.
However, both packages require practice to get the best scan possible,
and having done that there is very little to choose in terms of the
image content (though a lot of folk give up with NikonScan simply
because Vuescan gets closer to start with). It then comes down to which
has the better user feel, the better cleaning algorithm or the better
grain reduction.

I contend that I can get a better scan from NikonScan than I can from
Vuescan - whether that is down to more practice with the former,
slightly better capabilities or a better user interface is not something
I could quantify.

 

[Kennedy McEwen]

Kennedy,

Where is the scan data you 'publlished yesterday'? I would like to
take a look.

This same thread. Data included in post made on 21 June 2004 at
13:56:59 GMT.

We are going around in circles here, so let me state what I believe
Nikonscan
is doing.

1. It determines an exposure (probably does this adequately)
2. It scans the negative and collects all the data
3. If you can imagine a full dynamic range histogram, as you
stated,
the dynamic range of the scanner is far beyond a typical
negative's
dynamic range. Nikon Scan analyzes all the data. It internally
determines that much of the 'data' at either end of the
histogram
contains no useful information and 'throws it out'

Wrong - you assume that it does this - but there is no evidence at all
that it does and, as you will see when examining the scan statistics I
published yesterday, it does not throw information out in the first
place.

4. Nikon Scan 'presents' what it has determined to be the part of
the data
(i.e. some portion of the full scanner dynamic range)
which contains the full dynamic range of the negative scanned.
5. What the user gets to operate on in terms of curves, white point
black oint targets etc. is this data
(with what NS has deemed to be the extremes of scanner dynamic
range
beyond the negative dynamic range not included).

It is my assertion that in steps '3/4' NS in fact throws out
some data
which actually does contain useful information from the
negative.

What proves this to me is the fact that both Vuescan and Silverfast
can provide
scans of the same negative on the same scanner which in fact includes
the parts
that are irretrievably thrown out by Nikonscan.

They are not "irretrievably thrown out" - they are there and perfectly
recoverable. This is simply a matter of where the autoexposure
algorithms of NikonScan, Vuescan and Silverfast set the exposure level
in the first place.

As further evidence, NS allows no way to retrieve the full dynamic
range
that the scanner sees when scanning a negative (barring tricks like
telling NS it is actually scanning a positive and inverting manually
later on
which in fact does capture the full dynamic range of negatives even
using
Nikonscan - further evidence).

Rubbish. I have just repeated a similar exercise as yesterday with
another image which was flash lit and contained a white object in the
foreground, thus exceedingly overexposed on the film relative to the
main subject. Vuescan sets an autoexposure which reproduces the white
foreground object at just below peak white on the image. However, as a
consequence the main subject matter is, as expected, much darker in the
Vuescan image than on the actual print from the frame. NikonScan
however, using its default autoexposure setting, clips the white
foreground object but exposes the main subject matter perfectly! The
clipping occurs because the white foreground object is the densest part
of the negative and the autoexposure algorithm sets an exposure that is
too low for it to be detected by the CCD/ADC. No information is "thrown
away". Furthermore, by adjusting the master analogue gain by -0.8EV I
can produce a scan which is virtually indistinguishable from the rather
inferior, based on the main subject matter, Vuescan result. The main
subject matter can, from both scans, be lifted up to the correct level
as produced in the default Nikon scan using gamma adjustment. Viewing
the resulting histogram with the Optimax Wide Histogram plug-in shows
nearly identical combing in the black levels and the same range at white
- unsurprisingly proving that the images contain near as dammit the same
level of information.

This is a very clear indication of *exactly* where the difference arises
- and I seem to recall Ed mentioning something of this ilk in the past.
NikonScan judges the *exposure* based on median level in the frame and
does not specifically correct for the orange mask variation - hence the
black data level requires user adjustment for each film and process
combination (which changes orange mask tone and hue) - an adjustment
which can be saved by the user and called up in the future for that
specific film and process type. Vuescan, however, uses a black and
white point colour balance algorithm (as well as other processes, I
would guess, which are proprietary to Ed) to clamp the orange mask level
to black and the highlight to white. The median level in the image,
then ends up where it happens to lie.

You are completely wrong in your claim that NikonScan does not capture
the full dynamic range of negatives and does not provide a mechanism to
adjust the exposure to bring either highlights or shadows that may be
significantly off the median level into the captured dynamic range. I
demonstrated exactly that yesterday, and published the scan stats to
prove it, and have just repeated the process in the exercise documented
above.

 

[Dave]

Makes sense. I've noticed a real improvement in the scans as I've
played with each software package. The hard part is deciding which
scan is best from a group of good scans that all vary slightly.

 

[WD]

K,

I found your stats. In that case there did not seem to be any
clipping.
In fact, maybe I wasn't clear, but many times Nikon Scan does not
clip.
But a significant amount of times I have had this problem. When I made
adjustments to exposure, I just found I clipped one end of the
histogram
vs. the other (i.e. lost detail in either highlights or shadows not
being
able to get detail in both.....unless I used Vuescan or Silverfast).

I have also raised this issue with Nikon Tech support (there denials
being
as vehement as yours) going as far as sending them a couple of
negatives to
demonstrate my point. Guess what, they couldn't get the full dynamic
range
contained in the Nikon Scan output either! They ended up making
statements
like; "...that's what's on the negative", "....you could probably use
curves to
retrive the detail..." etc. But the bottom line is they could not get
the full dynamic range of the very same negatives that I was able to
get using either
Vuescan or Silverfast.

I will not post anymore on this issue, my experience has been that
when it comes to capturing the full dynamic range of many negatives
without
having "brick wall" fall-offs at either end of the histogram with
corresponding
loss of either highlight or shadow detail in the scan, Nikon Scan
cannot match
the capabilities of either Silverfast or Vuescan.

Thanks for the spirited dialog

W

 

[Don]

I understand the "delete VueScan.ini" part, but why right click on a
neutral gray area? Does that re-set something?

Yes, this basically neutralizes any casts and (in theory) gives you a
color-balanced image. It's the same as clicking on a gray color
sampler in NikonScan Curves and then clicking on a neutral gray area.

The problem is VueScan has many picky settings and their interaction
is virtually impossible to predict. That's assuming you even know
which options are on - let alone, all the convoluted conditions and
dependencies between different options.

Because of all this, you're also supposed to set Media to "Image" in
the procedure I posted last time, so none of that interaction takes
place (or, at the very least, it's minimized).


Now, my personal approach to scanning is quite different. I don't want
to do any color correction or, indeed, *any* unnecessary processing
during scanning (with the exception of gamma conversion).

I want as "uncontaminated" a scan as possible with maximum dynamic
range i.e. capture as much data as possible. That's my "digital
negative".

I do all the corrections afterwards on a copy keeping my digital
negative unchanged.

That's why I find all the additional tools in scanning software (such
as curves and friends) pretty useless. All I want the scanner program
to do is... well... scan! And keep out of my way instead of offering
me all sort of bells as whistles such as DEE, GEM, ROC etc.

The only exception, of course, is ICE because that's hardware based.
In theory it doesn't have to be but, unfortunately, that's how Nikon
implemented it.

Don.

 

[Don]

This is simply a matter of where the autoexposure
algorithms of NikonScan, Vuescan and Silverfast set the exposure level
in the first place.

One thing which may play a part here is that VueScan has a setting (as
far as I remember) to exclude - or minimize - the effect of the border
around (or inside?) the crop when calculating Auto Exposure.

As far as I can tell (although, I haven't done any substantial tests
yet) but Nikon scan seems to come up with the same exposure regardless
of crop (I only did one cursory test, no crop, and crop around the
image - but no tests of small segments of the image).

Anyway, that may account for the difference in the Auto Exposure
between the two.

Don.

 

[Don]

The hard part is deciding which
scan is best from a group of good scans that all vary slightly.

Welcome to our (or, at least my) nightmare! ;o)

I think there are at least two "best" images depending on your goals:
One, reproduce what's on the film faithfully with minimum, preferably,
no "pollution" from the scanner (if the image has a cast - e.g. , as
in a sunset - keep it), and two, produce the image closest to what it
"should" look like (in the above example, remove the cast so white
looks white, and not pink).

The problem is there are no right or wrong answers. Take the above
sunset example. If the image is that of a face, in most cases you
would want to remove the "sunburned" look, but if the image is that of
a landscape with a setting sun, you most likely would want to keep
rich reds and oranges...

Don.