Making a scanner with a DSLR

[peter ramm]

Straun, thanks for posting the link. I love weird and wonderful systems. Wonder if Harvard actually ended up using it for the specified project. What often happens with things like this is that a grad student or post-doc puts something together as a feasibility study, publishes it (and uses it for an academic requirement) and then the results of that experience are used to work with a commercial supplier who can provide something that will do the job with minimum fuss. They have a lot of plates to truck through. Do you know the outcome here?

Couple of comments. Telecentricity at the field lens is not required for film. However, it is useful at the detector plane because tele lenses can have some strange color aberrations. We built a singly telecentric lens that worked great in monochrome (what it was designed for) but multiwavelength imaging took some careful calibration and was never really aberration-free.

Yes, the optics tend to be huge. One reason is that the greater the reduction factor the more aberration and deviation from true telecentricity - so the lens becomes more like a pipe. Sill has done well with its lenses and they are used in a lot of deep imaging applications (e.g. microwell plates, machine parts).

Note that telecentric lenses are not immune to vignetting. They collect rays in parallel at the field lens, but the efficiency of collection varies from point to point. Also, any lens design is a compromise. Telecentric lenses are designed for measurement (low distortion), not optimal modulation transfer. You can get reasonable MTF out of one of these but the tolerances become difficult (= even more expensive).

Wow, that is quite a sample movement system. Fun to build if you have the budget.

[Nathan Potter]

I believe Struan is referring to the Depth of Field advantage of a telecentric setup. The entrance pupil is located essentially at infinity by placing an auxillary stop at (or near) the exit pupil of the macro imaging lens. The auxiliary stop, in effect, selects mostly on axis rays exiting the main lens rendering the selected rays from the subject to near infinity corrected and thus providing hugh depth of field. Cute stuff but of course with sacrifices. The biggest disadvantage for scanning is much reduced FOV (Field Of View) due to restricting the diameter of the entrance pupil.

The reduced FOV is less significant for the stepped macro folks because their subjects are tiny to begin with, so this is a neat application for that endeavor and provides near zero perspective distortion when scanning through the desired depth. Increasing the field of view requires increasing the physical diameter of the imaging lens. Restricting the effective aperture also will reduce the resolution of the imaging lens due to diffraction. As always there are optical tradeoffs.

Peter, I hadn't thought about the advantages in reducing chromatic distortions but seems that could be quite significant.

All interesting stuff indeed. Thanks Struan.

Nate Potter, Austin TX.

[jb7]

Ingenious stuff- however, if I had to employ an Ivy League University research department, develop a custom telecentric lens, ccd, and lighting system, crane in a 2,4000lb linear stage, and then have to consider the project a failure if I couldn't hit the target of scanning one 8x12" plate per minute, well, I might just have to stick with the V750. At least I wouldn't have to upgrade my current air conditioning system- which reminds me, those windows could do with a clean...

I think there's a danger with over-complicating things. It's fine if you're NASA, or have access to a large budget and/or unlimited time, but problems don't have to be that daunting. If something isn't working, it's often more productive to scrap the idea and move onto the next one. Maybe with less moving parts...

Even the Telecentric lens, it's a solution to a problem that may not exist. True, for the project you mention, which uses a 46mm ccd, it was probably vital, but using a DSLR instead, makes it less so.

Regarding the robotic stage- again, I don't think it's essential, though I suppose that depends on the amount of scanning you might have to get through. If you were the Astronomy Department at Yale, and you had to scan your entire back catalogue of plates, then yes, you have a point- but speaking personally, I might spend more time making the robotic stage than I would spend scanning pictures I need to reproduce at five foot high.

I'm torn between reading your post as an actual indication of what you think is necessary to make a scan with a DSLR, or if perhaps you have your tongue firmly in your cheek-

I'm slightly reminded of the alleged $10 million that was spent developing the Space Pen, while the Russians went up with a pencil...

[jb7]

I hope someone can prove me wrong regarding the Telecentric lens, maybe someone already has the equipment lying around, and can test to find an ideal setup- but it sounds like quite an involved project in itself...

[Peter De Smidt]

The gigamacro system works at up to 5x magnification. They don't use a telecentric lens, and they seem to get pretty good results.

I agree with Jacob that automation is unnecessary. If you aren't doing lots of scans, or working at magnifications above 1x, it's a simple enough matter to slide the negative carrier from place to place using a guide. That's what I did for the lighthouse picture, and it works fine.

That said, I'd like to automate the stage, not only because it might help with stitching difficult negatives but, well, it's fun.

I appreciate new suggestions, even esoteric ones. Some of it might be beneficial down the road. Right now I'm still working on the basics.

[Struan Gray]

Joseph, I hope neither you nor anybody else feels that I'm slighting your efforts. Linking the Harvard plate scanner was a way of showing that the engineering can be taken to a level that rivals or surpasses drum scanners.

In fact, once you see the trouble and expense you need to go to to get true micron-level positioning and repeatability it becomes abundantly clear that a homebrew setup needs to use software compensation for inaccuracies in positioning or angle setting. I have used stages like the one shown here in various forms of scientific instrumentation, and they simplify all sorts of calibration and positioning issues, but they are in no way simple or cheap.

Peter, the Harvard system has been churning out scans for a while now. The project status page says they are up over 21 000 scans, so it's not just tinkering. I still think it's amusing that one of the most stringent design goals was the ability to get the thing into the building :-)

Telecentric lenses are not a universal cure, but they do simplify some of the issues that crop up in stitching. The benefits are greater if you are measuring 3D objects, but I can see enough reasons to see if I can roll my own. A film scanner is going to be working at photomacrography repro ratios, so there's enough space to put auxiliary apertures into the light path.

With a lens on an extension bellows (or your DSLR mounted on your LF camera) going telecentric can be as easy as adding an aperture at the focal plane of the lens you would be using anyway. Worth a try, at least if you like tinkering. More complex but also more versatile is to add an auxiliary lens in front of your regular one, which is a variant on the old macro trick of reversing a normal lens in front of a short telephoto but with a spacer between the two lenses (again, an LF camera works well here). John Hallmén is a Swedish photographer specialising in insect macros who has got good results out of the Computar lens I linked, which costs a couple of hundred dollars. That might seem pricey, but it's comparable to, say, custom machining if you need to make lens mounts and other gadgets.

http://www.flickr.com/people/johnhallmen/

FWIW, a telecentric doesn't have more depth of field than an equivalent normal lens. There is a subtle effect if you modify an existing lens with an external aperture because you will almost certainly change the pupillary magnification, but that's not what I meant in my original post. The advantages for this application come from the projection geometry - you get constant magnification of the negative even if it shifts a bit towards or away from the lens. Telecentrics can still have distortion (although lenses built as telecentrics tend to be designed for low distortion at macro repro ratios) and because distortion can vary if you move the aperture around in a lens design the homebrew methods can in principle make distortion worse.

The form of telecentrics I am interested in are the ones with two lenses placed back to back with an aperture in between. If the front lens is reversed you find yourself needing two lenses which are sharp at infinity rather than one which is sharp at macro ratios. For me, the former is easier and cheaper to arrange. The lens closest to the object/film needs to be physically large, but for an APS-C camera the Kowa and Hasselblad MF lenses I have lying about anyway will be more than adequate. Another possible advantage is that the telecentricity is fairly robust if the two lenses are not perfectly aligned (with the resolution requirements of this project at least) so it may prove possible to compensate for changes in the film height by letting the camera autofocus the rear lens.

Of course, as an inveterate lens tinkerer I have several APO-ronars lying about too. It will be interesting to see how much combinations of MF and 135 lenses compare with an existing lens designed for low distortion and 1:1 imaging. My hope is that the homebrew solutions will allow for larger apertures, but a face off is one of the things I intend to test.

[jb7]

Thanks Struan, and thanks for your comment elsewhere-

No, no slight taken- the scanner you linked to has very demanding accuracy and performance criteria, and I'm not sure how any of it would translate to a kitchen table version. It might have exactly the level of precision necessary in order to make meaningful astronomical measurements and calculations, but I don't think anyone is going to be measuring the distance between my dust particles. (except, perhaps, for PTGui) I think that if this thing is going to work for scanning photographic images, it has to be a simple and achievable design, and there's a chance that someone might be put off starting a project like this if they're told that it's impossible from the outset, without hugely expensive optics and control systems.

True, there might be some trickle down technology, your post is the first in this discussion that has mentioned telecentricity, for example- and I hope someone follows up on that and can cobble together something that works- but minimizing the depth of field necessary to photograph a piece of film might be achieved in other ways too. If it turns out that I can connect a couple of Zuiko primes to perform that function, then I'll be very happy to spend the €5 on the connecting ring. I'd be less happy to spend large amounts of cash on optics if they were to be useless for any other purpose, and if there was a good chance that the first few attempts wouldn't produce usable results at all. I think I might just be happier buying a Micro Nikkor, and taping the transparency down...

I think the difficulties with a normal lens are field curvature- which the telecentric lens might inherently cure? Distortion? Would positioning an external aperture in the right position remove distortion from a lens that already exhibited it? And evenness of illumination, would it cure that, or might some of the falloff be due to the inefficiencies of the sensor in dealing with off axis rays?

I'm not for a minute saying that we shouldn't be having this discussion, you never know where ideas will come from, and all input is valuable in that regard- but I do think simplicity and accessibility are key qualities that shouldn't be needlessly circumvented-

[Struan Gray]

If it turns out that I can connect a couple of Zuiko primes to perform that function, then I'll be very happy to spend the €5 on the connecting ring. I'd be less happy to spend large amounts of cash on optics if they were to be useless for any other purpose, and if there was a good chance that the first few attempts wouldn't produce usable results at all. I think I might just be happier buying a Micro Nikkor, and taping the transparency down...

I'm in complete agreement with this philosophy.


Mind you, there is a certain 'whoa!' quotient in having a lens that looks like this: Xenoplan on ebay

[jb7]

'Whoa' - is right-

I've come across a less spectacular version- http://www.ccddirect.com/store/custo...roductid=17435

However, from what I've read, these lenses don't project a very large image circle-
though presumably coverage would increase in the close up region-
I suppose if this lens covered its native format at infinity, then there wouldn't be a problem at all.

Perhaps we shouldn't be looking at using a DSLR in the first place, micro 4/3rds seems to be the native format for this- though 1:1 might produce even more frames than I've used on mine already.

A lesser issue is the Flange Focal Distance, which is much shorter than Nikon's- however, again, for close up, this shouldn't be a problem-

An issue which I'd have no idea about, without having access to the lens, and a little screwdriver, would be remounting this in an F mount.
I do have access to one of those lenses here, for not very much money, but I have to get out of the habit of buying lenses...


btw, if you'd like another Whoa, check out the thread about pictures of photographers with their cameras in a couple of minutes...

[jb7]

Ah- I've just looked up the specifications of that Computar- 11mm ic, presumably at infinity. So, 22mm at 1:1.
Might be just possible on a crop sensor, but I really wanted to use the full dynamic range of the D700.
I don't think this is going to be easy...