I've been using VueScan successfully for about a year and, although
I've had problems, it has been fine since some time this summer, so
please explain your comment above about "buggy and unreliable", this is
news to me. FHK
There is a constant stream of messages about VueScan bugs and
exasperated users scrambling for older versions of the program where
one or the other feature still worked. Some bugs, e.g. the notorious
Minolta streaks, seemed to be a permanent "feature" for at least a
year while other bugs reappear at almost regular intervals.
All in all, VueScan may be OK for casual use, but not if you care
about quality or accuracy. As someone once said, VueScan can at best
be charitably described as a "rolling beta".
Why would pixel or sub-pixel alignments affect my test as I'm working
with the histogram for almost the whole 6x7 cm film area? Is it that
the blurring effect will suppress the range of pixel values and give a
false low number for the "Standard Deviation" spread? And, disregarding
the scanner software, is Standard Deviation in the Histogram a
reasonable method of estimating or comparing film grain and dark noise,
at least among my film choices on my machine? FHK
Blurring is actually supposed to help but the point I was making,
really, is just to be aware of the whole picture.
This is because, in my experience, scanning is the trickiest thing I
ever came across. It's like those Russian dolls with an almost
unlimited number of gotchas hiding within other gotchas. However,
that's really relevant only if you are into "details" and by measuring
noise you - just like me - seem to have an "enquiring mind"... ;o)
The ScanMultiPro does single pass mutli-scanning. How does this affect
your comments above? FHK
If you have single-pass multi-scanning then you will achieve some
noise reduction. However, even though that's a "good thing" the
effects are still limited because you are also scanning noise multiple
times. I don't know what algorithms are used in your case (or my
NikonScan, for that matter) but noise reduction is a science in its
own right. For example, throwing away extreme values before averaging
(generally) produces better results than blindly averaging all values,
etc.
I personally find that scanning twice - once for shadows and once for
highlights - produces far more dramatic improvement in reducing noise.
I do realize this is beyond the scope of your question (accurately
determining noise levels) but, presumably, reducing noise is the end
goal.
What software do you use instead? And how is sub-pixel alignment done?
I had assumed that PS6, for example, only moved the image data in pixel
increments in any direction. FHK
I use NikonScan, which is my scanner's native software. However, I got
the Software Developer Kit recently so - time permitting - I'm hoping
to "roll my own" in not too distant future.
There is commercial software out there to do sub-pixel alignment. I
don't know the exact details of how they achieve it.
So, after some thinking, I devised my own method using PS6:
As the name suggests, subpixel alignment is when two images are off by
a fraction of a pixel. I determine this by blowing up the images to
100% or more (300% seems to work well) and then flipping between the
two. In Photoshop, if you check the "Resize windows to fit" the images
should be superimposed automatically once you blow them up. After that
I use Control/Tab to flip between the two. The "pepper spots" seem
particularly well suited to spotting the subpixel shift.
The idea, then, is to move one of the images this fraction of a pixel
in order to align them. In essence (in my method) I first blow up the
image, move it, and then reduce. Easiest explained with an example:
Say, you have an image of 100x100 pixels and want to move it by a 1/3
of a pixel. I would blow up the image by 3 to become a 300x300 pixel
image, move it by 1 pixel in the desired direction, and then shrink
the image back to 100x100. Bingo, the image has moved by a 1/3 of a
pixel!
Two caveats. When changing image size use the highest quality
possible. In my version of Photoshop (6) that's "Bicubic". The second
caveat is that any resizing of the image will blur it somewhat.
Because of that I move the image which has less content, in my
particular case (slides) that would be the shadows' scan. Also, if the
image is shifted in only one direction, say, horizontally - blow up
the image only on that axis (in above example to 300x100). That will
reduce blurring.
I can see that transparency film would have a much more severe problem
than color negative film and would require that approach. The tone
range of my current film, Portra 160, the one I did the test on, seems
to be well within the range of the scanner and I no longer make two
scan passes, as I did some time ago and when scanning my older film
(Kodachrome, TMax, Ektachrome, Ilford FP4+, Agfa 25, and some others).
FHK
In theory, my 14-bit scanner should also have more than enough dynamic
range for my Kodachromes - but it doesn't. My first scanner was a
Nikon LS-30 which is only 10-bits internally. So one of the main
reasons for buying the LS-50 was the dynamic range. However, my tests
showed that I really need about 17-18 bits of dynamic range to
penetrate those dark Kodachrome shadows without burning out the
highlights.
But I can see histograms in 16 bit without any plugin. FHK
The problem is you only see 256 "bins". Since each 16-bit image
channel data can have 256*256 distinct values, each of those bins you
see actually contains the sum of 256 distinct 16-bit value counts.
The above mentioned "Wide histogram" doesn't actually go that far, but
it increases the number of bins to 1024. However, the most useful
feature (to me, at least) is the ability to export data so you can do
your own calculations - like the standard deviation - instead of
guessing what PS does internally.
Don.