Making a scanner with a DSLR

[jb7]

He suggests that evenness of illumination will be critical, because people can see macro repeating patterns very easily in an image--far more easily than seeing those patterns when they do not repeat. He believes we'll have a hard time weeding out those patterns. He agrees that our use of an enlarger-head light source might help, but he suggests that every component of the system (lens, sensor) will add falloff even with the most even possible illumination. Sure, we can correct that with software systematically, but that may take some effort to do, and it will need to be done before stitching.

He also thinks stitching will be difficult in large, flat expanses of tone. He does not think the grain patterns will provide the necessary detail to provide accurate stitches. But he thinks we are on the right track using software-driven stitch algorithms rather than attempting the mechanical precision necessary to avoid it--that imprecision will cause more artifacts that the stitching will, especially if we overlap more rather than less.

I've built the platform, and have done some initial testing. I've also got a much improved design for a Mk II, but until it can be proven that the combination of uneven illumination (light source and lens) can be corrected, I doubt if I will build it.

Regarding the bed; I'm using a 10mm porcelain tile, very flat. I have a piece of 6mm glass that is positioned on top of it, with the transparency taped to that. The glass is guided using some Mod 1 racks, these allow incremental repositioning in multiples of 3mm. Wet mounting might be better, particularly with film that has been clipped in frames; the clips and the heat of the process tends to distort colour films.

I'm going to do at least one more run of tests, but I don't hold out too much hope of making an acceptable scan, for an acceptable level of commitment...

The magnification of the image that's possible is quite scary, right into the dye clouds-but not as scary as the magnification of the dust...

[Peter De Smidt]

Joseph,

Thank you very much for building a prototype.

Do you have any pictures?

[rdenney]

I still don't see the need for a new scanner. There are many scanners that out resolve large format film. High end drum and flatbed scanners are able to produce scans with zero or close to zero added noise.

It just depends on how much money you have. When those high-end scanner mechanical parts wear out and can't be replaced, can you afford to fabricate replacements? Do you have the space for the sort of scanner you describe, plus the computer museum you will need, say, 10 years from now? What will you do if the circuit board fails? What if you can't afford over a hundred bucks per scan to those willing to maintain the obsolete technology?

Better to consider alternatives now, before they become an absolute necessity, and for those for whom the scanners you describe are impractical or infeasible, especially considering they might have a life's work in negatives that they hope to mine in future decades.

Rick "glad early photographers were a bit more experimental" Denney

[Kirk Gittings]

In addition to what Rick said above, if this works out, as time passes and scanners are going away we will have ever better dslrs to make this solution continually improve.

[jb7]

Thanks Peter- no, I haven't photographed it yet, I will when I have it set up for testing again. Not sure when that will be, I'm busy on other things at the moment.

Stitching is another thorny issue, quite apart from the repeating vignette problem.

I made the scanner bed oversize- since I may have need to scan larger images in the future, larger than 8x10. A smaller size, the MkII, would have better ergonomics.

First things first though, need to improve the light source...

[Nathan Potter]

Rick said it nicely. Many of us have fine DSLRs already paid for along with a fine macro lens. The image capture and signal processing within the DSLR is phenominal, so worth making use of. Setting up to copy film by backlighting is not such a big deal, depending on how fancy you want to be. The stitching has to be worked out and is manually time consuming but the high potential quality of scan would be limited to only those images that deserve it, so it would be used sparingly.

Any 6um pixel DSLR used at 1:1 with a macro lens compatible with that resolution will yield a digital frame at 4200 spi and at an astoundingly high quality, especially with a reasonably collimated light source. That would easily match what you can get from a Nikon 5000 ED type device. We're talking home made lashups here not engineering a commercial product.

Nate Potter, Austin TX.

[Peter De Smidt]

Here's my Light Source Prototype 1 (LSP-1).







It's built like the inside of an enlarger light box, i.e. it's lined with white styrofoam, and the side opposite the light source is at a 45* angle.

[Peter De Smidt]

I did a quick test by photographing the light source with a 55mm Nikkor at 1:1. The middle readings were all L (luminance channel LAB) 71. One far edge dropped to L 69, and the other edge dropped to 67. I can tray and position the flash a little better, and if that doesn't work, I can add a fresnel to or more diffusion in front of the flash, but I'm certainly open to suggestions.

[Asher Kelman]

Well it seems older drum scanner used photomultiplier tubes. I'm betting you can't get those anymore. Newer ones use CCD arrays which are not easy for amateurs to deal with. It seems more practical to put a whole bunch of APS-C sensors on a device that can position them. You take one picture with the whole bunch of sensors, shift the sensors (because there will be borders between the sensors), take the picture again and stitch the result. If this could be done with an 8x10 cameras I'm sure the resulting digital image would be in the Gigapixel range easily.

Two lines of sensors with the second one covering the gaps in the first. That would then scan over a flat pattern. Nothing complicated.

Asher

[argos33]

I did a quick test by photographing the light source with a 55mm Nikkor at 1:1. The middle readings were all L (luminance channel LAB) 71. One far edge dropped to L 69, and the other edge dropped to 67. I can tray and position the flash a little better, and if that doesn't work, I can add a fresnel to or more diffusion in front of the flash, but I'm certainly open to suggestions.

Peter,
A few things I thought of looking at the prototype:

1. The diffusion plane/light source doesn't necessarily have to be where the negative sits. You can use any materials to get a good diffused light source, and use a flat piece of glass, etc for the negative to sit on above that. Perhaps you can use a thicker and/or more opaque piece of plexiglass or acrylic sheet for the top panel?

2. If there is just a little falloff on the edges perhaps you can just mask the surface so that the light source area is smaller?

3. Could you scoot the flash back a little bit so it's cone of light is wider? I assume you have the flash set on the "wide" setting?

Looks good though, that's a good idea adding the 45 degree bounce, it could save a lot of vertical room that way.

Evan