Making a scanner with a DSLR

[Old-N-Feeble]

All these possibilities keep swimming around in my mind... what's left of it. We could spends many thousands of dollars on either a good drum scanner or this proposed DSLR method. For me, and I suspect for most of us, its viability balances on cost and complexity. But the OP and most others already alluded to that.

Being the spoiled brat that I am... I want the best... but I'm also unwealthy and stoopid. So, I'll be settling on something middle-of-the-road. I'm watching these related threads with great interest and anticipation. I know that many very good options will be developed by several people. I'm hoping to avoid spending $2K on a DSLR with enough resolution to make this method really shine. If I had to though... at least I could use it as a DSLR too.

[Peter De Smidt]

This is a 4x section of Acros developed in Pyrocat-MC taken with a D200 and a 4x microscope objective.


Is that large enough grain to have the effect we're discussing?
What if any implications does film's self-dithering quality have for our project?

[peter ramm]

Interesting observation Peter and Rick. Indeed, film has no set bit depth. It is dependent upon the RMS noise present in the sampling window. Over a small distance, grain becomes dominant and a few more exposed grains in one area will yield a very different value than in an adjacent area. SNR is lousy. Pull back a bit and film becomes very precise.

We used to make calibrated measurements from grainy radiographic film. Over small areas (e.g. 1 mm2) you might get only 6-8 bits precision, but one can get a true 10 bits out of a well processed film if you measure areas larger than about 1 cm2. Of course, film only responds well within its linear range. As soon as you get into the tails of the distribution bit depth compresses pretty quickly. Our eyes are tolerant of that. Quantitative systems are not.

I realize that measurement has nothing to do with photography. I guess the question there is whether it is necessary to resolve grain to archive a neg, given that you just going to be viewing it. I don't think anyone really cares about the grains. They are just an indication that the MTF of the system is clean enough to render non-granular detail well. You do see that in the print and it is the area that the DSLR will really excel in - relative to the Epson.

[Struan Gray]

So, what about long, fast medium-format lenses? I have a Pentax 200/4 that covers 6x7 and a CZJ Sonnar 180/2.8 that covers at least that. Is the idea that these could be arranged nose to nose with a stop in between them? That's sort of an inside-out Super Angulon, which is really a pair of reversed telephotos arranged tail to tail. Something about the concept is not clicking in my brain.

Using MF optics on smaller formats will help considerably with evenness of illumination - assuming that you can control flare from the non image forming light. The purpose here is more subtle - to find lenses with the right focal length but large front elements. Before anyone queries that, I hold up as evidence my 50 mm f2.8 for the Pentax 110 camera and my 50 mm f2.8 for Hasselblad 2000 series. The change in lens design widens the field and enlarges the elements.

For the deep side canyon into which I seem to have dragged this thread, the point is that a telecentric lens needs to have a front element which is a little bit larger than the thing it is trying to photograph. MF lenses generally have physically larger front elements for the same focal length compared to 35 mm lenses, so they are useful candidates for the front of a homebrew telecentric. An Aero Ektar is another candidate if you want large field sizes.

You can't just reverse one lens and put it on the other. You'll need a tube of some sort. Telephotos for SLRs have an additional problem that the nodal points may be badly placed - for example, most of them won't work as a simple macro with a reversing ring to mount them on the camera because the front focal plane is way out in front of the lens, so when you turn it round it ends up behind the film/sensor and you can't focus on anything at all.

I should probably move this lens-only discussion to the DIY lens thread, but I'd really like to tinker a bit with my lenses first so I have some experience to relate as well as speculation. In brief though, you could use two of your MF SLR lenses with a male-male thread coupler. Macro users know that if you mount one lens on another the repro ratio is just the ratio of the focal lengths. 50 mm reversed on a 150 gives you 3:1. To make the system telecentric you move the 50 mm outwards on a tube, bellows, or similar. The first telecentric setup is when the front lens produces an image at infinity of the aperture of the rear lens. With thin lenses that means it's 50 mm away. With thick telephotos the distance will be greater. The second telecentric setup is when the rear focal plane of the front lens (what was the front focal plane before you reversed it) coincides with the front focal plane of the second. Place a stop on this plane and you become bilaterally telecentric. The magnification remains at 3:1 the whole time.

[Struan Gray]

Grain aliasing exists, and can make grain look larger than it really is. The signature is that it goes away if you digitise on a finer pitch. Kodachrome 200 is a good film for testing because it has large, sharp-edged grain. The apparent grain gets worse and then better again as you change the pixel spacing - in this case, the repro-ratio of your images.


PS: resolving or detecting grain is one thing. Stitching so that the grain lines up is another - and probably too ambitious for all but the most megabuck projects.

[jb7]

I'd have to agree with this-
The stitching on my first test is really imperceptible-
However, if had resolved into the grain, I'm quite sure that the grain would have broken down at the stitch junctions, and would have been noticeable for at least some of the scan, and possibly on every join.

Lighting has a big influence on how the grains are resolved too-
I think I might try to diffuse my source for the next scan-

[Ben Syverson]

Is that large enough grain to have the effect we're discussing?

Oh definitely. You don't even need to go that far.

The upshot is that we can do 100% of our processing at 8 bit versus 16 bit. 16 bit is overkill if you can see any kind of grain texture.

[Peter De Smidt]

Getting back to the telecentric idea. Wouldn't this only be helpful if one employs focus stacking, since the whole point seems to be keeping the subject the same size when changing focus? If we don't change focus during the scan, then my very cursory understanding of the idea is that there wouldn't be any benefit. Is that correct?

[rdenney]

Getting back to the telecentric idea. Wouldn't this only be helpful if one employs focus stacking, since the whole point seems to be keeping the subject the same size when changing focus? If we don't change focus during the scan, then my very cursory understanding of the idea is that there wouldn't be any benefit. Is that correct?

I think it would also be helpful to correct any subtle misalignments in the negative. If one side of the negative is slightly further away than the other side, it will be subject to less magnification than when it's on the other end for the next tile. Those misalignments might result from negative curvature.

Let's do some math to see if it's an issue:

I have measure a maximum curvature in my holder of 0.3mm in the center of the frame. That is actually not bad, according to what others have said. Assuming (conservatively, since the scan frame doesn't at all cover the whole range of curvature) that this is the maximum error, one end might be 0.3mm further from the camera than the other end. Let's assume 1:1 target magnification with a 100mm focal length. At 1:1, the film should be 100mm in front of the lens's rear node (I think I have that right). So, a 0.3mm error means that one end of the negative being scanned will be 0.3% farther from the node than the other end. And the project size of the negative will therefore be 0.3% smaller, using similar triangles. That's 8.7 pixels across the width of my Canon's sensor, and it has big pixels. A telecentric lens, because the magnification does not change, would eliminate that problem altogether.

Or, let's say the negative stage is out of square with the camera by 2 thousandths across the width of the 4x5 frame. 0.002" = 0.05mm, and one 36mm-wide frame might encompass half of that. That error is smaller than film curvature. If one side of the frame is 100.03mm from the node (along the parallel line, of course) while the axial distance is 100.00mm, then that's a 0.03% error. That means the negative on that end will be 0.03% smaller than when it's on the other end. 0.03% of my 2900 pixels is about a pixel. In that case, it's okay. Getting the negative square within 0.002" of the camera is no minor task, however.

Since I don't have a telecentric lens, I'm hoping my calculation above is conservative enough to render it not important, or that it can be corrected easily in practice. I haven't gotten there yet.

Rick "correcting focus but moving the camera for each frame might solve the problem, but create others" Denney

[Struan Gray]

I've been tinkering. I'll post in the lenses thread when I have some real numbers, but FWIW and IMHO, here are some qualitative experiences.

1) Dan Fromm knows what he's talking about when it comes to photomacrography. I'm surprised he hadn't yet told us all to stop messing about and read the Lefkowitz book.

2) There are no free lunches. And you can't make a silk purse out of a sow's ear. If you want real quality, suck it up and buy a real macro lens.

3) Many of the manual focus 70s-80s era primes I have at my disposal have their entrance pupils quite deeply recessed behind the front of the lens barrel. Quite often a simple stacking arrangement is already close to telecentric on the object side.

4) In particular, two stacked Pentax 50 mm f1.7 lenses work very well at 1:1. (but my DSLR is only 6 mpix, so I'm aiming for 2:1 :-).

5) Stacked MF lenses give a lovely even illumination. Distortion can be surprisingly high though, even with Zeiss/Hasselblad primes. [a side discovery: my orphaned Kowa lenses might make good candidates for telecentric Kohler illumination, even if they are not worth the hassle as taking lenses].

6) An arrangement which is telecentric on both the object and the image sides is going to have a long tube length.

7) Telecentric arrangements need careful alignment, particularly for angles. Any working setup will need to be locked down and left alone.

That seems like a bunch of negatives, but it's really a narrowing of parameter space to the point where I can ask some real questions. I've been cobbling together ways of mounting lenses, cameras and apertures on my Sinar's rail, so if I'm not hiding eggs round the garden the entire time, I'll see if I can take some test shots over the Easter weekend.