DSLR SCANNER No.7

[jb7]

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.






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.








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.





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.

[jb7]

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.





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-

[Peter De Smidt]

Joseph,

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

[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. 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...

[Moopheus]

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.

[Peter De Smidt]

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.

[Peter De Smidt]

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.....

[jb7]

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-





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.





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...

[Peter De Smidt]

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

[jb7]

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-



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.





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...