DSLR SCANNER No.7

Frank Petronio's suggestion of a DSLR scanner reminded me of my own ideas on the subject, and on the basis that I had enough materials and equipment lying around already, and that it wasn't going to cost me anything, I took it on. Eventually, I did have to buy something, since my idea of a lo-fi pellicle was unworkable. So a tiny piece of two way mirror blew the entire budget for contingencies, at €15.


Attachment 70602



I made this at the end of January, and made some tests with it, enough to show that it had some problems that needed attention. Then other things got in the way, and it's been lying around idle ever since, until now.

I designed it as an optical tube (like an optical bench, but hollow) so that components could be positioned with some accuracy, for repeatable results. I tried different combinations of components, but the internal layout and concept remained the same- using a flash unit for exposure in combination with a beamsplitter to allow the use of a separate focusing light. The big idea I had was to use a reflective film as the primary mirror (Prism? Reflector?)- the type used for road signs, that uses glass microspheres to reflect your headlights directly back at you. By putting the light source in the reflected position of the camera lens using an angled mirror, illumination can be increased at the outer edge of the field of view, even compared to a collimated source. And to counteract the falloff in the lens itself, a centre filter was made by stippling a pattern on the reflective film, to match the pattern observed on a test exposure. Falloff isn't much- according to the histogram, centre luminance varies from the edge luminance by a maximum of 15/255. I've yet to make a conclusive test of this setup, but that's the theory... It's also possible to get quite close to even illumination in post processing, without using a centre filter.


Attachment 70601


Attachment 70600


This reflective film was the source of the problem in January- it has a very glossy surface, and it acted like a plane mirror, giving a huge hotspot. This can be fixed in a number of ways, by covering the surface with magic tape, which also provided a good surface to add the stippling, or by introducing a diffusing layer between flash and mirror.

The base uses a large porcelain tile as a positioning stage; the film is taped to a piece of glass that glides over a card layer attached to the tile- the card layer allows for different masks to be cut, depending on coverage required.


Attachment 70599


The positioning system is very basic. A T square that is positioned using a pair of interlocking racks controls the y axis; a similar arrangement attached to the glass plate controls the x axis. When I started, I might have thought that some kind of automated positioning system would be necessary, but now that I've used it, I find that making the exposures is by far, the least time consuming aspect of the exercise.

Attachment 70604


The camera is supported by a column, to which is attached a Manfrotto 410 head, and a cheap macro rail. If anyone would like to find out how much their tripod head actually moves, a setup like this will provide a good illustration. However, the macro rail was never going to be precise enough for focusing at this level, so an additional fine focus screw was incorporated- bearing on the end of the rail, and pushing against gravity. Another screw controlled the pitch of the head, and the final assembly is very solid, and quite immobile when locked down.


Attachment 70605


I built this to accept a wide range of negative sizes, I think up to 16x20 should be possible. That was as a direct consequence of having a 2' square tile lying around, but a smaller version dedicated to 4x5 has been started already- or rather, I've cut a tile down. The 4x5 version would occupy ¼ the space of this one.

I've made a scan, I need to process it now-
I should have it sometime within the next six weeks or so...

The results I'm getting are not perfect, but the machine could easily be perfected if I could dedicate a little more time to it-
I'm just not sure I've got that time right now-

More to follow-

Joseph,

Thanks very much for posting this. You've got some great ideas!

Thank you Peter- hope you're enjoying your break...



This brings me on to what will be, for many of you, the most important component of the system. The lens. Look away now.
I don't have a macro lens, and have neither the desire nor the budget to get one. I do have a reversing ring, and some nice old Olympus primes, so I used them. There were many design problems associated with this project, however it's nice to know that there is much room for improvement by simply specifying a better lens.

Reversing a lens is a low rent way of getting into macro photography- the reversing ring is one of the more satisfyingly cheap pieces of equipment you can buy. Focus is fixed, for any given lens, though you can make extension tubes out of old filters for extra magnification. Usually, a drawback of using a reversed lens is the fact that you must get very close to your subject; for this application, that's not a problem at all.

A more obvious problem is the lack of resolution compared to a lens that has been designed to work at this close range. However, since we're tiling multiple images, you might get the same overall resolution by making more tiles at higher magnification with a lower resolution lens, compared with a higher resolution lens, and less tiles. Make sense? If the goal is the absolute largest number of fully resolved pixels, then using the best lens at the highest magnification would be the way to achieve this. However, if the goal is to achieve a higher resolution than that available from any given scanner, a V750, for example, than that target is substantially lower.

More important than resolution, for this application, is field flatness, and lack of distortion. True, the stitching software can correct for some distortion, but starting out with the least amount of geometric distortion is far more preferable. If you can't cover the format of your DSLR without losing your edges to a curved field, then the resolution ceases to exist at all.

I've tested all my old Olympus lenses, the 24mm f/2.8 is good, but very high magnification, it would need about 250 frames on a full frame sensor to cover a 4x5. Similarly the 28 3.5. It's possible, and you can get right into the grain, but you'd want to be a little crazy. The 50mm f/1.8 covers a nice area, about 46mm wide, iirc, but suffers from terrible distortion and bad curvature. Similarly with the Nikkor 50mm f/1.2, though it does improve by stopping down to f/16. Its magnification is smaller than the Olympus 50mm too.

The best lens turned out to be an Olympus 35mm shift. It shows very little distortion, and curvature is controlled by stopping down to an indicated f/11. It projects an image approximately 26mm wide onto a Nikon Full Frame. Including the overlap, it requires 56 frames to cover a 4x5- sounds like a lot, but it only takes a couple of minutes to shoot on this setup.

Stitching- although I was a proponent of DoubleTake in the beginning, I've since purchased a copy of PTGui, so I've used that for the latest stitch. I might revisit DoubleTake in the future, but all these things take time… I used the stitching method linked to in Peter's scanning and Stitching thread, the mosaic method. The software identified most of the frames, but there were still some that had to have control points manually assigned.

The stitching allows for export at full size- in this case, around 21.5k pixels on the long side. About 415 Megapixels. It's obvious that these are only pixel dimensions, and that it's not possible to have a fully resolved image at this scale, so I thought that a more reasonable test would be to output a file size similar to that which is commonly accepted to match the real world maximum resolution of the Epson V750, and rounded it up to 2500dpi. Yes, I know, should be spi, but considering it's a consumer flatbed, you're bound to piss someone off, whatever term you use…


More to follow...

Quote:

---

Originally Posted by jb7

Thank you Peter- hope you're enjoying your break...



This brings me on to what will be, for many of you, the most important component of the system. The lens. Look away now.
I don't have a macro lens, and have neither the desire nor the budget to get one.

---

Would an enlarger lens work for this purpose? If your camera could accept a screw mount lens, an enlarger lens might give you the close-focus flat field you want.

Good stuff, Joseph. It looks like a number of affordable lenses should work well for this project. When I started, I thought it'd be easy to out-perform my 40 year old 55mm F3.5 Nikkor-P, but that's not the case! It's been at least the equal of the lenses that I've tried so far. Reversing adapters have become super cheap. I bought some on ebay for about $5 shipped. I'm not sure how that can be profitable. The quality is pretty good. By chance, I checked Amazon, and they had similar prices to Ebay, and the quality is better than my Ebay version. The Amazon one is marked Fotodiox Macro Reverse Ring for Nikon 52mm. It push fits into 2" black abs tube. Add some flocking and you can make a cheap extension tube of whatever length you'd like.

Enlarger lenses do work for this purpose. They are best used reversed. When you do so, they give their best performance at about 4-5x magnification, according to the macro folks. That's a little much for our purposes, but the performance at 1:1 is probably still quite good. I found that my Screen holder limited max detail acquisition, and so I won't be doing any tests until I get a new negative carrier built. But I'm on sabbatical....I've gotta go start work on a new fence.....

Thank you - yes, an enlarger lens might work, if you found a good sample-
but it's riskier than using something you already have, that works.

I have a few old 80mm enlarger lenses, but mounting them in perfect axial alignment, so there's no tilt or offset, might be beyond me.
or, I might need quite a stack of filters, with the glass removed...
A 50mm would be better for my purposes, but the point of this exercise was to build it with what was available, and not to start another shopping list...

This is a picture from the Zuiko- the squares are 5mm.
Distortion is minimal, at least compared with the other lenses I tested, and there's adequate resolution over most of the image. Overlap in the stitching software removes the edges anyway. This picture also illustrates the evenness of the illumination-


Attachment 70619


Next section- Illumination and exposure.

I outlined the idea behind the lighting in an earlier post- using separate light sources for focusing and exposing. The focusing lamp is a LED, runs very cool, and it's a long way away from the film anyway. No heat gets to the film.


Attachment 70618


To make the exposures, I'm using a Nikon SB900 flash unit. Because of losses from the diffuser, and despite it being zoomed out to cover the FOV of a 200mm lens, it needs to operate at around half power, giving enough flashes to scan around four frames from a set of batteries. It's possible that it could be made to be more efficient by designing an internally mirrored mixing box to match the size of the diffusion film necessary to illuminate the frame, but I think that might be a finesse that's a little bit beyond diminishing returns. At the moment I'm happy enough to have a V8 that only does 80 mph…

A custom white balance can be set, to compensate for any colour cast that might be given by the glass in the system- though it's likely that you're going to want to fine tune the colour in post anyway. The sensor in a digital camera is optimized for daylight colour balance, and its range is diminished by deviating from that, so it makes sense to start from as close to 5000k as possible.

The flash output seems quite consistent, although it's early days yet, and I haven't tested sufficiently to prove this. However, I'm quite confident that this piece of equipment will not cause a problem.

Exposures are made at f/11 at 1/200 sec- too fast to allow the light from the focusing lamp to register.

I found a piece of free software called sofortbild http://www.sofortbildapp.com/ that allows tethered shooting, including live view. Files are imported directly into Apple Aperture, though there are any number of software combinations that will allow similar functions. Being able to operate the exposures directly from the computer is a far more comfortable setup than having to operate the camera directly. As I said earlier, once planarity and focus is set up, making the 56 exposures only takes a couple of minutes.

I'm taping the 4x5 onto the glass, I find that transparency film is seldom flat, especially if it's been clipped in hangers for processing. The clips tend to deform the film, so the taping becomes necessary. Film I process myself, without clips, is flatter, but usually there's still some curl, and anything that isn't flat is outside your depth of field- which is, technically speaking, tiny...

Led focusing lights are the way to go. I have a little pocket one I got free from the hardware store on my birthday. It is plenty bright.

Regarding flash consistency, Charles Krebs tested a bunch of speedlights for consistency recently, and all but one was within 1/1oth of a stop.

See: http://www.photomacrography.net/foru...sh+consistency

Thanks Peter-

Yes, that does seem consistent. The stitching software can make automatic corrections too, so I think that's covered-

I made a scan, which has pointed up a few things-

Attachment 70666

Firstly, the vignetting hasn't been completely sorted out yet, but it shouldn't be too hard to fix. You can see the anomalies in the sky area. Part of the problem might just be down to the size of the overlap, but there might also be an issue with the mirror producing an oval vignette. Needs some more investigation.


Attachment 70667


This is the stitching pattern- as you can see, I missed my positions a few times- the vignette is more apparent when the overlap is reduced. Now that I know what I'm doing better, I can calibrate the racks- there are lots of positioning marks on it, and it's all a bit of a jumble.

On the other hand, it illustrates that super accurate positioning is hardly necessary- some of these pictures are out by 3mm, and apart from the vignetting, which should be possible to fix, it hardly seems to make any difference, and the software can handle it. Using Flash, and allowing a reasonable time to recharge to half power, hand positioning with this system is quick and undemanding.


The workflow is completely different. I'm using an older computer to scan, since I'm not going to give Silverfast any more money to update software that should have been updated years ago. I made a comparison scan on the V750 at 4000dpi, and it took 40 minutes- is this normal for a 4x5? I could have had about six DSLR scans made in that time, including taping the film down, and removing it.

Once the images are captured, adjustments can be made to the individual frames, then lifted and stamped onto the others. Working on a large file is a sluggish experience, unless you're on a fast computer- the larger the file, the faster the computer you need. With this method, a lift and stamp for 56 files takes only seconds, and it's possible to make many small adjustments quickly and consistently. With a little bit of experience, it should be possible to make most adjustments before the files go anywhere the stitcher, leaving only the tedious spotting for later. And there's a lot of that, way too much. Perhaps my light source should be more diffuse...

However, I've had this transparency out a lot, so it's kind of sacrificial. Perhaps a virgin transparency might exhibit less dust, but this was a Fuji Quickload, and there does seem to be a lot of dust embedded in it. That shouldn't happen. It doesn't show up as much in the Epson, but it's a big problem with the DSLR. Could be a deal breaker...

Attachment 70668

Attachment 70669

Attachment 70670

Attachment 70671

Some details. I decided it might be instructive to see them at close to the output resolution, so I made comparison files on the Epson at 4000dpi. It turns out that the DSLR files needed to be reduced to 87% to match. These are screenshots from a 24" monitor, so they're reduced here by more than 50%. I might put up some full res sections if there's any interest.

See if you can tell which one is the Epson...

Good stuff! I'm going to guess that with the first one the dslr scan in on the left. With the second and third one, the dslr is the right frame, and with the last image the dslr is the left frame, at least I hope so. :)

Dust is going to be a problem with color materials, since ICE works pretty well with them, and it's very unlikely that we'll be able to get something like ICE working with a dslr scanner, although I'd love to be wrong about that. With BW, the problems should be about equal, given an equally diffused light source.

Well, if there wasn't going to be at least some improvement, I might just have given up, and not posted anything-

The Epson is much more diffuse, even without ICE, which I never use.
One possibility is to introduce another diffuser, below the negative stage, perhaps about an inch below. That might work...

Anyway, that'll be for another time. This has been an interesting exercise, but the scan I made equates to a 5' high print at 360dpi, and I don't get to make many of those from large format. It's good to know that it's there, and can be made to work, but I doubt it'll get very much use- but you never know...

Joseph,

I really appreciate that work you've gone too. Thanks!

Thanks Peter-

So, apart from yourself, there doesn't appear to be any interest in this thing- but I'll post these here anyway, for the sake of completeness.
I'll post them in your comparison thread too-

These are screenshots of a 4000dpi V750 scan against a background of the DSLR stitched output, as outlined above.
They're about 600kb each, and obviously, not cleaned up at all.

For the sake of those who might not have been following, the originals were taken with a reversed 35mm lens, not a dedicated macro optic, so there's a lot of room for improvement.



http://img209.imageshack.us/img209/1...omparison1.jpg

http://img220.imageshack.us/img220/8...omparison2.jpg

http://img717.imageshack.us/img717/5...omparison3.jpg

http://img714.imageshack.us/img714/1...omparison4.jpg

Joseph, there are lurkers interested too. Well, at least one.

Thanks for your efforts.

Make it two, really interesting thread.

Ian

Thank you, two is good, three including Peter...

What I like about these prototypes is that as new models of DSLR are brought out, the sensor portion of the scanner cdan be up dated or improved. For example: Nikon is rumored to have a 36 MP pro-sumer body waiting for introduction "soon".

I'm not sure that more pixels will improve things-
In fact, I think the 12Mp full frame is a kind of a sweet spot. Not the only one, however...

Already, the scan here represents around 350Mp, and it's obvious that those pixels aren't the best quality.
Trebling the number of pixels isn't going to improve the image, but will treble the file size.
You could choose to image at a higher magnification, leading to more film detail, again at the cost of a bigger file size, but there comes a point where you're imaging the grain, or dye clouds, and not the image-

The image can be improved- by the combination of-
1. Using a better lens- flatter field, less distortion, higher resolution at larger aperture.
The lens used here was at an indicated f/11, probably somewhere north of f/16, taking account of the magnification- so probably at the diffraction limitation of the sensor.

2. Scanning a flatter original- especially at larger apertures, and higher magnification, depth of field will decrease. If you're only scanning flat film, then this is not much of a problem, but not all film is flat, especially if it's clipped into hangars for processing, as is the case with the image above- Drum scanning involves taping the film to the drum, then encapsulating it within another stretched sheet, and something like this might be necessary to flatten the film, to place it within the depth of field of a faster lens.

If those two conditions are met, then a number of further options are available.

You could choose to scan at a lower magnification, leading to increased depth of field, a larger available aperture, and smaller image and file sizes for the same level of detail-
this is probably the preferred option for prints that aren't going to be enlarged more than about 4' on a side.

If you do feel the need to go larger than that, then scanning at a higher magnification might be an option-

A 36Mp sensor could have advantages, in that you might hit the same quality threshold using a smaller number of less magnified higher resolution tiles, but since there is no physical limitation on the number of tiles used to make up an image, pixel count on a sensor is not going to be the final arbiter-

In my opinion...

Joseph, your last comment is what interests me. DSLR imaging at 0.5X allows a four fold increase in the field of view. A 24 X 36 mm field at 1:1 becomes a 48 X 72 mm. field at 0.5X. The detail referred to the sensor of course is reduced by 0.5X but a higher full frame pixel count can offset this - although a factor of 2 is a lot to compensate for. But full frame at 0.5X will nearly cover a 4X5 with four frames. The catch is probably in the quality of the lens. To make use of the intrinsic resolution of a 7 um pixel pitch sensor at 0.5X one needs a lens capability of close to 100 lp/mm at some reasonable contrast.

The only item I have to get started with is a 90mm, f/2.8, Vivitar Series 1 macro with 1:1 adaptor; although still highly regarded for its unusual resolution capability approaching the 100 lp/mm level. It is of notably low contrast however. I should try to find some resolution data on it. These are still available in the $400 to $500 range I believe.

On the other hand 36 MP X 4 will yield a 140 MP count on a 4X5, and GB+ files. I think most 4X5 images don't contain anywhere near that kind of detail. BTW that would be about 4 times the resolution I would normally get from my Epson 750 where I see a spot size equivalent of between 20 and 25 um. at reasonably high contrast. The full frame 20+ MP Canons seem ideal for this application as well as a new Nikon D800 - although very pricey - for me anyway.

But this would be my direction if financially feasible.

Nate Potter, Austin TX.

There's some info regarding all of this at: http://www.photomacrography.net/foru...=101164#101164

Interesting testing over there Peter, looks like there's a lot more resolution in the aerial image-
but there are perhaps a few flaws in his reasoning, and methodology-
Finer sensor pixel pitches will still be limited by diffraction themselves, and a more accurate test might have seen him use sine test patterns, rather than the slightly slanted linear ones.

Also, all his testing seems to be done pretty much on-axis, and there's no corresponding off-axis results- and that's an area we're going to need to make use of. The smallest amount of field curvature will eat all the resolution away from the centre, in our case, making the centre correspondingly useless, on its own.

Nathan, you're correct- a high resolution lens, used at an aperture that isn't going to limit it, coupled with a high resolution sensor, would be a good start. Assuming of course, that lens also has a flat field, and that the film is flat. I measured a piece of transparency film in a holder last night, and the deviation from edge to centre was 1mm (actually 0.99mm, but nobody is going to believe that) A piece of b+w negative, drum processed without clips, was much flatter, but still had some curl.

You're correct also, in that it's main use might be in being able to make less frames to perform a complete scan.

As I mentioned earlier in the thread, because these scans are made up of any number of frames you choose, using any magnification ratio, they are potentially limitless- you don't need the best lens, or the the highest resolution sensor- you just need to put in the effort with the exposures and stitching. True, the pixels are not going to be very efficient, but downsampling will help that- perhaps even before you get to the stitching stage, so using the D800 will definitely be more economical in terms of time spent. Though I think, if you're going to be using a program like PTGui to stitch, the magnification have to be more like 0.4x, to allow for overlap.

Regarding the sensor, I think a long range of sensitivity is going to be important- the D700 did get into the shadows on this rather dense transparency, but bringing detail out of them might need local attention- just as bringing detail out of the low range on a drum scan might need local attention- all the processing on the scan in this test involved global adjustments.

And regarding the lens- if you're going to scan a 4x5 in 4 shots, then flare might become more of an issue- depending on how resistant to flare your lens is-
photographing small parts of the transparency means that flare is controlled locally, in 56 locations, in this case- It might become an issue for some scans of larger areas, and some lenses...

I'd be in complete agreement regarding the level of detail contained in a 4x5- which, of course, is going to be limited by the taking lens and the film. It's difficult to see how it would be necessary to present more than 140 million pixels (actually, more like 100-110 million, taking account of overlap)- unless you wanted to scan at a microscopic level, where the image is made up of the grain structure of the film, which is a kind of a quantum leap beyond the resolution of the taking lens...

Damn... I didn't realize there was so much porn in here. Amazing work as always Joseph. Great design. I shot some negatives against the window before I got my V700 using a D300 with both a 50mm 1.4 and a 105mmVR 2.5. Very badly arranged but image quality was better than expected.

What are your thoughts on the compared images?

Thank you Ramiro-

Yes, I copied some film with a point and shoot some years ago, and made a nice Blurb Album for the family for Christmas- so I was well primed for this exercise-

The compared images-

Well, they're much too large, but that stems from the DSLR Capture, and being able to compare at the native resolution.
The magnification of the taking lens was fixed, being a reversed prime- I could have gone for more magnification, but not less, at the time.

The V750 doesn't resolve at 4000dpi, and the reversed lens doesn't resolve at the the pixel pitch of the camera sensor, and the taking lens (72XL with centre filter) doesn't resolve to that level of magnification on film either. ( I think...)

But no matter, it might be instructive to see the captured information, knowing that improvements can be made-

I think it's quite clearly shown that the DSLR scan and stitch out-resolves the V750, but not by a whole lot.
It shows that the V750 employs a lot of smoothing, even when Digital ICE is turned completely off. I suppose this could also be a function of its uprezzing algorithms-

The DSLR stitch shows better native colour, I think- because I was able to edit individual frames in Aperture, then stamp the remaining frames in the set, allowing far greater control, at a greater level of detail. The Silverfast Software, (or the Epson) by comparison, is a very broad brush, and adjustments made post scanning are a lot more processor intensive. Scanning at a ridiculous 4000 dpi took around 35 minutes too, much longer than making the DSLR images, though using an Epson at 4000 dpi is not something you'd do often. Ever.

The DSLR Scan shows all the dust, the Epson removes most of it- though I'm going to have to try a more diffuse light source on the next run of tests.

Next time, I plan on making a dedicated test negative, with sine charts at centre and edge, using a valuable piece of T55, which is apparently capable of 180 lp/mm, according to some sources- I've got a new scanner lens to test, which looks promising, though I can't mount it properly yet, without extra components. It's designed for a much smaller sensor, but at the increased magnification I'm using it at, it should be fine. It's specified to have 100 lp/mm in the centre, 60 lp/mm at the edge, on an image circle of 22mm at 1:1, so I think that will work out just fine using the usual stitch overlap. I might have to go just beyond 1:1, or use a crop sensor camera, but initial tests, just holding the lens up to the camera, show a very highly resolved image, and signs of a very flat field.

I plan on making the test image using a Nikkor-M 300mm- it's probably the sharpest lens I have, though other options are 150 Symmar-S, 450 Nikkor-M, or some Super Angulon XL's - thoughts, anyone?

I have no doubt that a DSLR scanner is a viable option, and would hope that the next run of tests gets it into Drum Scanner territory-
based on this prototype, it might not take long to put the 4x5 version together, which will be whole a lot smaller-

I think the stitching doesn't have to be very onerous, and a slightly more generous overlap would allow the software to make a comprehensive assignment of control points, leading to a far less labour intensive output.

All in all, not a bad outcome.

If my V750 failed tomorrow, (which it won't) I don't think I'd replace it, but for most normal scans, including all proofing, it's just too convenient. If I had a need for higher res scans, I think the next generation of this thing will be able to provide them-

Quote:

---

Originally Posted by jb7

Thank you, two is good, three including Peter...

---

Four, but until the semester is over I can only passively contribute. I have done similar scans before, and using a Beseler Dual Mode Duplicator to scan slides for a local studio. What I envision for my scanner is an x-y table using linear guides and 17 frame steppers and ballscrews or leadscrews, an 8x10 glass holder, light pipe or light box type light source with strobe or MH for a light source and a smallish PIC controller to sequence the steppers and fire the release.

Eminently "doable" and I probably have most of the stuff sitting in the garage, sans the PIC stuff, but for the time being I can use a PC with a paralell port and Mach 3 to drive the table and release the shutter as I already have a machine set up with it as a controller.

Quote:

---

Originally Posted by Nathan Potter

On the other hand 36 MP X 4 will yield a 140 MP count on a 4X5, and GB+ files. I think most 4X5 images don't contain anywhere near that kind of detail. BTW that would be about 4 times the resolution I would normally get from my Epson 750 where I see a spot size equivalent of between 20 and 25 um. at reasonably high contrast. The full frame 20+ MP Canons seem ideal for this application as well as a new Nikon D800 - although very pricey - for me anyway.

But this would be my direction if financially feasible.

---

4x5 negatives contain one level of detail that has proven itself to me as worth pursuing--grain. If we can resolve grain, then we can restore a high level of MTF in the DSLR through what would normally be excessive sharpening. That does not create artifacts, because at that scale, it's all noise on the negative being scanned. But it means we get full use of the sensor in the DSLR. It also means tonal separations at larger scales will be based on grain dithering, rather than on pixel color depth. This should greatly improve dynamic range, though I have not yet really proved that to myself. This seems to me a very useful outcome.

For my 12.5MP Canon 5D and the lenses I've tested so far, I needed greater than 1x magnification--more like 1.8x--to be able to resolve grain well enough to test the theory. And that was marginal, but it did work. Peter was able to resolve grain (I think) using his 55mm Micro-Nikkor at 1:1, so I've picked up one of those in hopes I can improve on what I have and not have to use the higher magnification. I think the sensel size should be small enough in the Canon to render grain at 1:1 if the lens can, though the AA filter is a wild card in that assumption. The most convenient answer seems to me the lowest magnification that allows us to resolve grain well enough to oversharpen at the grain level enough to sharply resolve grain. If the sensel density and lens quality is good enough to do that at 0.5X, then that will be easier to stitch than doing it at 1x or 1.8x. If we ever get to the point of being able to do the whole 4x5 film with a single frame in the DSLR, 4x5 film will probably vanish from this earth soon after. But it's likely I won't be able to afford it--either the camera or the glass.

(I do not know how this thinking will translate to color film, though I think it will do okay with negative materials and be more challenging with transparencies. A system that is good enough to resolve grain in black and white film, though, will surely do well enough with color film, though perhaps without the benefit of using grain dithering to help with the dynamic range issue.)

Rick "whose 5D came as a result of a bonus, and who no longer gets bonuses" Denney

One thing to note, my example was with Fuji Acros developed in Pyrocat MC to a slightly lower than normal density, the equivalent of n-1. Most other film will be grainer. And if you use Bergger 200, for example, grain shouldn't be too hard to find. :)

Rick, our Linos lenses are probably optimized for 1:2 normally positioned and 2:1 reversed. They should be pretty good.

Yes Peter, perhaps choosing your grain size will be an important factor? I have some Bergger 200 stored, haven't used any yet, but had always expected that it would be for contact printing. Polaroid 55, which I mentioned earlier, that might turn out to be a no grainer...

Anyway, not particularly relevant, I'm sure the right size grain has been imaged before with a drum scanner. Perhaps it deserves another one of your scanning thread sub-sections? I see one potential issue, the stitching of such images, so it deserves to be studied in greater depth-

Erie, it sounds like you know what you're doing- so good luck with that-
As I might have mentioned earlier, an xy stage would be desirable, but not essential- however, that's the position of someone for whom the electronics and the software control would be quite alien; I'm sure if I had access to a stage, I would use it.

Quite early in the main thread I suggested using the motion stage from the RepRap 3d printer system. It uses threaded rod, nuts, standard bearings, and 3d printed parts for components, and should be the least expensive way to achieve a moving bed. Positional accuracy and consistency is not particularly important if you're using a stitching application which involves the assignment and comparison of control points for individual images, but might be overwhelmingly important if you choose to combine tiles using another method, and I'll be using a stitching application- it's actually impossible to manually position images in PTGui without having control points assigned, although you can replace the images in an existing project, which might work.

The RepRap bed uses open source design and software, and there's a lot of information available, so if I chose to automate it, I think that's the route I'd take. It's accurate enough to provide the 3d modelers with resolution down to 0.1mm, which is 30 times more accurate than I managed to achieve with some of my positioning in the test scan in this thread. I see the benefit of such a system, I just don't see it as an essential part of this particular project, which has more fundamental issues to deal with- such as optical performance, and lighting.

Good luck with your stage, look forward to seeing it- Maybe I'll change my mind when you come up with a foolproof and inexpensive solution, and I'll wonder why I was so ambivalent to it in the first place- I can change my mind easily, if I'm presented with something shinier...