How can the scanner even see gaps between grain clumps?

Hi,

The highest resolution film is probably something like Kodak Techpan.

I understand it can resolve 220 lp/mm under ideal conditions. This

means it can see a gap under very high contrast circumstances of

about 45 microns. I also understand that when they measure rms

values, they use a photo cell that is also 45 microns. Maybe this is

no coincidence, maybe the sensor is sized to look at the "gap" of the

highest resolution film.

Anyway, the grain is always significantly smaller than the

resolution. On another photo.net thread it was suggested that grain

was 3 microns and grain clumps (3-6 grains) were 6-9 microns,

roughtly 1/0th the size of the gap between line pairs.

If this is true, however, then it should be very difficult to ever

see grain clumps because of the resolution of a 4000 ppi scanner is

at best 2X the resolution of the commonly used film, but the grain

clumps are 10X smaller still and yet we see something. I read

elswhere on photo.net that we are seeing the gaps between grain

clumps or dye clouds. (This is starting to sound very much like a

lp/mm discussion, except now we have clusters (like lines) and gaps.)

So, the photocell that is used to measure rms, is a pretty good way

to measure granularity because it is only when the density of grain

clumps or dye clouds becomes high enough that it becomes visible. If

rms is an absolute measure vs a relative one, then shouldn't we be

able to correlate it to scan resolution and build curves. For example

at 2700 ppi the rms must be 4 or less to be grainless, and at 4000

ppi the rms must be 8 or less. Or, if the rms is 10 then you must

scan at 5000 ppi to get rid of it. Etc...

So, I am just trying to find a good way to define the "film space"

for my scanner without doing it empirically. Supposedly Kodak and

Fuji are producing film for scannability. Certainly they must have

some way to measure it.

Scott Eaton, Norman Koren and Struan Gray can you guys comment?

Thanks.

Mike

Try learning about the phenomenon of grain aliasing. See

www.photoscientia.co.uk/Grain.htm

The site Leanard points to is another classic example of billiard ball math trying to describe problems that are caused almost entirely by density range discrepencies and NOT sampling. All it concludes is I don't want that guy doing my film scans.

I have no idea how you can make a direct scan from a neg and then a print and conclude that extra grain in the neg scan is caused by 'grain aliasing'. Talk about BAD science. This is like making B/W prints in a dark-room, using a diffuser under the enlarger lens on some of the images, and concluding the sharper prints with better Dmax were the funtion of those sheets of paper originally having more silver in them.

A problem here is a constant misconception that film is a perfect 2-dimensional strata and film scanners take perfect 'slices' (samples) in the time domain. In fact, neither are ideal or quite perfect. When a film scanner makes a recording in the form of a pixel it's merely an approximation based of the energy being relayed to the sensor at that time. The actual granularity of the film being sampled can be smaller than the physical step the scanner makes, but this does not makes those grain clumps invisible to the sensor device. This might be a brutally simplistic analogy, but if you drive on a gravel road with big fat tires you'll still be able to tell you are driving on a gravel road vs fresh pavement.

One way to conceptually play with this in real time is to use Photoshop's noise generator, which does a pretty good job of emulating the grain structure of film, at least in a quasi real 2-D matrix. If you then apply various filters such as gaussian blur etc., even at a degree larger that the actual radii of the pixels, you can still see the original grain clumps, although at decreasing resolution.

The gaussian blue acts kind of the same way a film scanner does by sampling the incoming data and assigning it a bit value in an organized, progressive fashion. As long as the data is within the physical read range of the device and not way outside it's dynamic range you'll see evidence of grain. If your original film emulsion is even several factors finer than the 1:1 sampling resolution of the scanner you'll still see evidence of grain because the granularity threshold is still high enough to trip the CCD at certain points, unless you are dealing with perfect monochrome regions of no tonal change.

I used to take high rez film scans and run them through a 3-D topography mapper which would show density detail in a really neat 3-D 'hills and valleys' representation. If you imagine the scanner sampling only the biggest hills and valleys at low rez, and the smaller hills and valleys with increasing resolution you can still get a fairly good sampling of the original grain in the image regardless of scanner sampling frequency. This is why crisp, 1200 dpi film scans can usually show evidence of the original media being film based, even if it's a very fine grained film.

Another thing to take into account is that all films and emulsions are based on a similiar if not identical molecules of grain in the form of dye or silver crystals/clumps. Doesn't matter if the original film is techpan or 2475 recording. Virtually the same molecules, and the wavelengths of light and laws of optics are the same as well. A coarse grain film can resolve a 400lpm chart along with a fine grain film. The fine grain film just does a more efficient job at it by concentrating either dye or clumps of silver in tighter, more discrete packets that are better at showing the regions of abrubt density change.

This is roughly how you make a film more scanner friendly. You engineer it to be far more resistent to halation and any other effects that cause the dye cluds or silver clumps to spread their density to adjacent regions with no good cause. The drawback can be increased contrast sensitivity which is a big reason that color print films become increasingly more contrasty as they evolve to become more scanner friendly within a film line. Or, try to show me a low contrast slide film with the resolution of Velvia or Provia. I'm not even sure if it would be technically possible to make such a thing.

Grain also introduces it's own noise factor because it is not a perfectly efficient recording medium - hey, nothing is. If I shoot a grey card with techpan vs 2475 recording, and make 4000dpi scans, I don't need lpm charts to show which film is what. Even when exposed to a theoretically even spread of photons the grain in the emulsion won't develop perfectly evenly. If the maxium resolution of a scanner is above this 'noise' threshold of film, and most are, this helps give away the grain signature of film.

I tire of Scott's constant 'grain-aliasing is a myth' posts, as he probably tires of those of us who say exactly the opposite. As I have better things to do on Xmas day, I won't bother repeating what I have said before, but I STRONGLY advise you to do your own research on this issue (experiment with black and white points, invert the film and see if the pattern repeats, view the film under a good quality microscope, look at what the dye clouds do in 'thin' areas, etc...) and then <b>make up your own mind</b>. Long and detailed answers are not always right.<P>

Over to you, Pete.. :o)

It is a myth, and I'm getting tired of people like Mark flaming me while blaming their lack of scanning skills on electronic manifestations that should be below the noise threshold in the first place. They want you to spend endless hours trolling through meaningless web-sites rather than simply learn to use your d@mn scanner. Any type of scanner artifact that shows up in a film scan immediatley gets blamed on 'grain aliasing' by the peanut gallery here on photo.net.

Some other advice is never trust ANY web site or statement about scanning from ANYBODY who doesn't have a single example of a high quality scan from a professionally looking image. I don't care WHAT degree or certification they claim to have. Note those who are more vocal about grain aliasing always seem to have the lowest quality scans containing castrated histograms and never a mention of the sRGB color space and other shortcuts they took in the process.

Gee, when are we going to start seeing scanners with 'comb filters' to correct for 'grain aliasing'?????? There's a sales pitch for you.

Example: grain aliasing was originally blamed as the problem for excessive black noise with the Nikon 8000.

Example: grain aliasing was originally blamed as the problem for streaking with the HP Photosmart and it's problem with electronic sheilding.

In the above examples, the problem wasn't grain aliasing, but another problem that the manufacturer identified (after we had zillions of posts identifying it as grain aliasing). So thanks to Mark, we've created a nice excuse for scanner manufacturers to possibly not fix their products.

A few more things to ponder. There are more scanners on the market than D/A converters and CCD types. If grain aliasing were such a common problem, it would affect ALL scanners using that CCD/DAC array and not specific models within in a product line.

Next, if grain aliasing was such a problem, there would be nothing you could do about. Since 'grain aliasing' by it's definition would be a hardware function of sampling frequency of the scanner related to grain size there would be nothing you could do about it other that switch films. Kind of strange that an amatuer A pops on photo.net with a problem they've identified as 'grain aliasing', and a more experienced photo.netter with the exact same scanner isn't having the problem.

You think that was a flame, Scott? Read your own post. In fact I invite readers to go and browse through *your* posts, and again MAKE YOUR OWN DECISION about who does the flaming. And now you have accused me of it, I might as well proceed..

You object to folk suggesting people do their own research??? Funny, but I thought forums were meant to be open for discussion, and different views were allowed. Now that I know otherwise, I will shut up, as it is clear that you know more than Nyquist and other scanning experts who have made it very clear that aliasing does NOT require two perfect patterns. And I'm sure that you have examined films under a lab microscope and compared specific areas to scans at both 2700 and 4000 dpi, and that you have checked every scanner's behaviour in detail, (not just that 'magic' Epson flatbed of yours - didn't it use to resolve Provia grain?). And of course, you know full well that I am just a worthless amatuer (to use your spelling) who knows sod-all about them 'castrated' histograms, black point issues, scanner noise, et al. Well spotted.

Scott, some of your stuff is really useful, but you really can be an..

Hmm. Must be hungover and grumpy. Like I said, I'll now shutup, and let others on the forum decide who is right here.

Hi everyone,

Merry Xmas!

Thanks for the responses again.

How about a few more answers though.

I asked specific questions in this last round.

Scott, I was thinking some more about what you said.

I wrongly thought that you can't see what you can't optically "resolve," that 4000 dpi is the optical resolution, the limit. From experience I know it isn't true. My thoughts are that the things between the lines in a test chart are visible if the contrast between those things is higher than the contrast between the gap the lines on the test chart. They might not resolve into nice neat lines very well but would still be visible. In my experience there is a always a lot of local contrast where there is grain, i.e. dark and sigficantly lighter grains mixed together. Also, I know that the lines from the test chart are in fact grainy when magnified not smooth like I might think. Does this make sense?

So, I would like to shift this discussion toward the scanner.How much can the Nikon 8000 ED see between those lines? What is the "resolution" of the CCD? Is it more than 10 pixels in that 0.045 mm gap? This is equivalent to saying that I can't see what the CCD array can't see. In the brochure it seems to imply that the CCD has 30,000 pixels but it doesn't say how big the CCD array is, so I can't calculate the CCD resolution. Obviously there are probably some optical issues here also to consider and LED things too.

I understand that this physical model is not very precise, that in reality it is a statistical thing and much more complex but maybe it is still helpful for a basic understanding of what's going on.

Thanks.

Mike

Scott, your "fat tyre on gravel" phrase is inspired, I couldn't give a flying f*** about grain aliasing but there's few things in this life more satisfying than a good analogy!

Anyhow, I've always tended to develop black and white negatives with a small bias towards low contrast to have an easier time in the darkroom. Are you saying that for optimum scanning I should increase the contrast?