Making a scanner with a DSLR

[el french]

Nearly every image I've shot in the past 4 or 5 years has been a macro panorama. Most have also been focus stacked with thousands of individual images. The biggest difference between what I've been doing and what is required for this project is the difference between orthographic stitching and spherical stitching. Spherical stitching nearly always requires some overlap (10% to 20% or higher for macros) in order to warp the individual frames to fit the projection. The stitching software will usually require landmarks to do accurate stitching. Orthographic stitching on the other hand will not require any overlap if the XY positioning is accurate to 1 pixel. However, for the absolute best results, lens abberations will need to be corrected before stitching, so some overlap will be desireable to enable blending the seams. Good orthographic stitchers allow you to set the XY coordinates for individual frames and/or allow the individual frames to be warped to fit. AutopanoPro and Microsoft ICE do this (I'm not sure about PTAssembler or PTGui and PhotoShop just hasn't worked well for me as a stitcher).

Featureless areas may not have much contrasting detail, but they may have subtle color shifts that are quite visible from frame to frame. Blue sky is one of the hardest things to stitch, not because it's featureless, but because of the color variations. Stitching software can be set to blend the colors between frames, but then you have to ask, is that the same image I would get by using an enlarger? The stitching and focus stacking software can also do a good job of evening out the exposures, but again it's a question of how accurately do you need to reproduce the image. I see quite a bit of variance in the light from my flashes especially when the batteries start to get low, but you won't see it in the final stitched image.

The images of film shown earlier in this thread have proven beyond a reasonably doubt the resolution of a DSLR with a good macro lens is more than adequate for the task. I think the only remaining questions are how can you reliably move the film or camera and how can you ensure even lighting from the first frame to the last and how much tolerance is there is in each of these to get an acceptable final image? Yet another question is how much post-processing do you want to do? I usually inspect each seam at 100% to 200% and manually blend the incongruities. The current crop of stitchers has reduced the manual blending required significantly. The more accurately you can move the film or camera and the more stable your light source is will determine how much post-processing you need to do. Manual movements will probably require more post-processing, but the end result should be very nearly the same.

Another gratuitous sample: http://photosynth.net/view.aspx?cid=...a-3f46feefa9c5

[SURF]

Very well said el french.

1."However, for the absolute best results, lens abberations will need to be corrected before stitching."

2."Stitching software can be set to blend the colors between frames, but then you have to ask, is that the same image I would get by using an enlarger?"

3."I think the only remaining questions are how can you reliably move the film or camera and how can you ensure even lighting from the first frame to the last and how much tolerance is there is in each of these to get an acceptable final image?"

I can add that scanner use several calibrations before good scan be possible. Before each scan (!) scanner calibrate the light and certain calibration is needed in scanning with DSLR: actual light and lens geometry profiles are needed to be created somehow.

As for moving parts: it can be easier to move film in X direction and camera in Y direction.

[jb7]

It has to be vertical because we bought up all of the remaindered 7,369 Bowens Illumintrans to repurpose [...]

Oh dear, looks like Frank might have put a deposit on that Gulfstream V already-

There goes the pension plan...

[Frank Petronio]

Let's just hope the sandblasting for the new cosmetic treatment doesn't mess up the optics.

[rdenney]

Let's just hope the sandblasting for the new cosmetic treatment doesn't mess up the optics.

The only word you stupid engineers know is "can't". Sheesh.

Rick "get with the program!" Denney

[peter ramm]

You know, Frank, if you check out my description of the stitched spider above you can probably save a bundle on models and the type of scanner you use wouldn't matter much.

[sethlatimer]

I appreciate everybodies comments, some great ideas. I am new to LF so I don't have a ton of experience, I just ordered a light table over the weekend!

In my head I have been imagining one of the xy stepper tables moving the slide over the light table, with the camera and the table stationary. With that setup each frame shot will have the same piece of light table under it. If our stitching algorithm first took a shot of the light table with no photo on it, it could use that to normalize each photo's exposure for any unevenness. The lens doesn't change so it can correct abberations before the stitch too.

Finally the hardest part is the blue sky. If there is an array of 20 photos of the slide, then there are many different slide orders we can stitch the thing together. The fastest would be to stitch them together in the order they were shot. Now when you use automatic stitching the computer identifies control points, and generates a parameter about the quality of each of the automatically generated control points. This way the computer can distort and bend the photo in smart ways, relying on the best information. If we can alter these algorithms slightly to get all the control points from all the borders and then stitch them in order (highest quality to lowest) then by the time we get to the lowest quality control point borders the algorithm will have more sides to align with.

I am just beginning to digest the panotools source code. I still think this is a manageable project. But I still have no experience with the stepper motor/arduino setup. My plan is to take some 20 shots when my new light table arrives then Ill have data to work with to mock up the merging.
Seth

[Jay DeFehr]

I haven't kept up with this thread very well, so please excuse me if this has already been linked:

http://theagnosticprint.org/future-of-scanning/

Seems relevant.

[rdenney]

I haven't kept up with this thread very well, so please excuse me if this has already been linked:

http://theagnosticprint.org/future-of-scanning/

Seems relevant.

Yes, it was mentioned, and I've read it recently so somebody linked it in this or another thread.

It makes the same assumption Ctein's article makes, which is using a single digital copy photo of an entire negative. To get a good enough copy image for larger formats would require a better DSLR than we own. This article was primarily interested in copying 35mm, where something like a 5D2 is probably fine. Ctein wasn't so specific, but he didn't speculate on the larger formats.

The step we are taking is to use a lesser camera than the 'Blad H4 (one of which I held in my hands today, but it was still the Pentax 645D that made me wet my pants, because I already have a most useful collection of lenses for it), like, say, a Canon 5D, and then move the film under the camera to make a scan in tiles. Each tile would be made 1:1 (or greater). Even at about 15% less magnification, I've already proved to myself that even with modest equipment and a makeshift setup this approach has the potential to significantly better an Epson V750. I think there is plenty of reasonable potential to match a dedicated film scanner.

The article seems correct that the magic will be in the light source. There is concern that flare will put a ceiling on what is possible, given that the light source can't be pushed through an orifice or slit the way it is on a drum or film scanner. But my test was with a highly diffuse light source and the results were usable; much more collimated light sources are easy-peasy, using a condenser head from an enlarger. We'll see.

Tiling also imposes quite an image assembly challenge, especially given the machines we make at home will not likely provide sufficient precision to align based on pixel-accurate registration. Thus, a stitching algorithm that has to match the pattern is likely, and it's still an open question whether existing products will suffice. Again, we'll see.

Where I think the article is correct is that we simply have no choice given what we, as individual photographers, have available to us. We don't represent a sufficient market for a fully integrated machine, and few of use could afford such even if it was made. We will never be able to achieve custom sensing hardware or other specialized products. We might be able to write the software for ourselves, or more likely find ways to work efficiently with what's already out there. But most of us could cobble together a decent DSLR and a quality copy lens--I got my results with a 13MP 5D and an older EL-Nikkor 105 on an adapted bellows.

Rick "hopeful" Denney

[Dave Jeffery]

It seems as though creating a template with numbers along the edges would help the stitching software along the sides.

I wonder if very, very small letters or markers had to be placed on the film to help the stitching software if Content Aware Fill would work well enough to repair the scan?