scanning technology

[robc]

As scanning seems a hot topic currently(for me anyway) and just for interest, here's a post from another list(some while back) from Mark Sauerwald.

<< start of post

A few thoughts on this.... But first a bit of
background. I am an amateur photographer, and do the
bulk of my photography in B&W. For a profession, I am
an electronic engineer, and have spent much of my
career working on Analog To Digital (A/D) converters,
many of which are used for image digitization.

When scanning an image, there are two, completely
separate things going on. You are sampling spatially,
and you are measuring the opacity of the film. In
sampling spatially, the determining factors are the
size of the spot which is sampled - which you rarely
if ever see specified on scanners, and the number of
samples per unit width. There is a theory (Nyquist
Theorem) which dictates how small this spot must be,
and how many samples you need to be able to sample the
image without loss of information, and the basic
answer is that the spot needs to be about half the
size of the smallest grain particles on the film.
This is where there is a significant difference
between colour film and B&W - in the colour film
process, the grains of silver are converted to dye
clouds, and they are on different layers of the
emulsion, with the end result being that there is not
as much high spatial frequency information on a colour
negative or transparancy as there is on a B&W
negative. Most scanners are designed to scan colour
images, and as a result, have their sample sizes
optimized for the colour die clouds. If you were to
design a scanner purely for B&W you would want to have
a much smaller sample size, and a greater number of
samples per inch than what you find for colour
scanners.

The second dimension to the scanning process is the
dynamic range of the image. This is another place
where there is a real problem. On a microscopic
level, a B&W negative is just that - BLACK and WHITE -
it is only when we step back, and average the
transmission through the negative over an area that is
large compared to the grain size that greys appear.
The ideal scanner for B&W would only need a 1 bit A/D
converter, which would look over a very small area
(smaller than the grain size) - to generate a digital
black and white file, which could then be manipulated
digitally to give us what our eye sees when we look at
a good print. Alas, this is not what we have, what we
have is an A/D converter which averages the
transmission of light over an area large - but not
very large compared to the grain. The result is that
there is a relatively large amount of noise that ends
up in the image, which is created by exactly how the
grains within one of the samples from the scanner.

There are other things that can complicate matters -
there is something called ICE which is used to reduce
such things as scratches and bits of dust on colour
negatives or slides. It works on the theory that
negatives and transparancies are transparant to IR,
regardless of the density in the visable part of the
spectrum. This assumption does not hold true for B&W
or for Kodachrome - so if using a scanner with ICE -
turn that feature off for scanning B&W.

Where I have had the best results, is not to try
scanning the negative, but to generate a print in the
wet darkroom, then scan the print.

Mark Sauerwald

end of post >>

[Ed Richards]

This would make a lot of sense if you wanted to scan an image of the grain. What most of us what is a scan of the image formed by the grain, and for that the Nyquist scanning frequency is 2x the smallest image detail we need to resolve, which is a lot bigger than b&w grain in LF. That also means we are averaging lots of grain so it is analog. Scanning a print could work, if the print were about as big as a billboard.

[Bruce Watson]

Would that life were actually this simple. There is much left out of the above discussion, and the conclusion is not supported by the argument.

I'm not going to write a book on how scanning works and publish it here in a post. Y'all would run screaming from the room. I'd be right behind you.

Suffice it to say, you might possibly describe the shape of film grain using fractal mathematics. A digital pixel, on the other hand, is nothing more or less than a square. Compared to a pixel, film grain is orders of magnitude more complex. Further, physical grain and grain clumps are, from the point of view of the scanner, distributed across the film stochastically. When you scan the film, the scanner lays a virtual grid over the film and looks through the holes (pixels) in the grid. In other words, scanning is deterministic.

These two are never going to play all that well together. You can't make a sample size small enough to let you realistically capture the film grain's size and shape. Nor can you get the film grain in question to reliably sit in the middle of your pixel. That's just the reality of the situation - just the laws of physics. And... it's not important that either of these things is accomplished. Not at all.

What you do get when you scan is a scanner looking through the holes in it's scanning grid. It averages the amount of light coming through this pixel and assigns it a value. That's all it does, and that's all that's needed. The reason this is sufficient is, basically, because detail is captured on film by lots of grain clumps. You don't need to see individual grains to be able to capture the image detail on the film.

If you must digress into Nyquist theory, you must also include the vast range of sizes of the film grains in the negative's structure. Smallest to biggest must be several orders of magnitude. Since you are going to sample all this with a single, fixed, pixel size... where will you go with this argument anyway?

Where I have the best results is scanning the negative with a drum scanner.

[Neal Shields]

Kodak has some very good white papers on this that are very hard to find on their site.

A good start is here although there are more papers in the non-linier chip section which I can't find right now.

http://www.kodak.com/global/en/digital/ccd/products/linear/linearFamilyPublications.jhtml?id=0.1.6.10&lc=en

[Brian Ellis]

"Where I have had the best results, is not to try scanning the negative, but to generate a print in the wet darkroom, then scan the print"

IMHO if he's getting his best scans from a print rather than the negative there's something wrong somewhere. My guess is that he's tried to scan 35mm negatives with a flat bed or maybe a very low grade film scanner. Otherwise he should be getting much better results by scanning the negative rther than the print. The print has already lost a lot of the information that was in the negative. Then still more information is lost in the scanning and printing process.