A $1200 digital scanning 120Mp camera

[QT Luong]

Here is the abstract of:

UBC ScanCam: an inexpensive 122-million pixel scan camera (http://www.cs.ubc.ca/~szwang/Research/ScanCam/scancam.html) by
Shuzhen Wang and Wolfgang Heidrich, UBC:

"In this work we present the design of a low-cost, very high resolution scan camera which can be used in image-based modeling and rendering, cultural heritage applications, and professional digital photography. The camera can take color/near-infrared images with the resolution of 122 million pixels while the camera itself can be built from off-the-shelf components for only $1200. We discuss the issues in constructing the hardware setup and the calibration process of the camera. Finally, We compare the image quatlity of this camera with a commercial SLR digital camera, and obtain a quantative measurement of the actual light sensitivity and the optical resolution."



If you already have the 8x10 (they use a $400 bender) and lenses (they use a $600 Nikkor 300M),
your cost in parts would be $200 ?

[Legolas]

Great imagequality for an 80 €uro "Scanback". The only bad thing is the color in the highlights; would it reduce, if they remove the glassscreen from the scanner?

[David A. Goldfarb]

I've thought of trying such a thing. I presume you need to disconnect or black out the scanner bulb. I could imagine that getting the focus right might be tricky.

I'm not sure why they are using a three-pass method rather than a single pass. A three-pass scan should have more subtle tonal gradation, but I suspect the registration problems may not be worth the trouble.

[Ken Lee]

If that is "large format", then perhaps this image is too. Perhaps soon we will need to re-define what is meant by the term "large format".



See Breaking the Gigapixel Barrier (http://www.tawbaware.com/maxlyons/gigapixel.htm" target="_blank). This person stiched together 196 6-megapixel images of Bryce Canyon, to create something that would print quite sharply at 90 x 136 inches.

[QT Luong]

Scanning backs were always considered to be in the real of LF photography, as they use normal cameras and lenses. The last issue of VC had an update about the "Better Light" backs which was interesting to me in that it showed that prices have already dropped to the point that they are now competitive with MF digital backs (and even some 35mm digital cameras). However, after seeing in the article I linked what you can do with a $100 back, I think there is still room for prices to go down a lot.

[fred arnold]

So the inevitable question is: If I put a good enlarging lens or similar optic on this instead (Polaroid Ysaron, El-Nikkor), and aim it at a negative on a light table, how will it compare to a pre-built Epson 3200 or similar scanner?

Would it be better to simply project an enlarged image onto the scanning area, or is the quality difference caused by the underlying scanner such that it won't result in any significan improvment?

[David A. Goldfarb]

On that principle, if you could turn off the scanner bulb without preventing the scanner from scanning, I suppose you could put a scanner under an enlarger and work that way.

[Alan Agardi]

I'm sorry to burst the bubble of the good folks at UBC, but that wheel's already been invented.

Go to http://www.geocities.com/capecanaveral/cockpit/8296/tech/scanner.htm

The concept is clever, and the UBC folks - presumably with a much larger budget - have done a very creditable job.

Ciao.

Alan

[QT Luong]

Thanks for the interesting link, Alan. Should be:

http://www.geocities.com/capecanaveral/cockpit/8296/tech/scanner.html

What the UBC folks have done is making a cheap scanning LF back, while in this
link, what he built was a scanning camera with mobile (35mm) lens. There is nothing
revolutionary about the UBC project, after all scanning backs have been available
commercially for many years, but what's interesting is that you can make one yourself
for a budget which is actually very small (if you already own the photo gear).

[Kerry L. Thalmann]

Did anyone else take the time to actually download the high res versions of the images on the UBC site? I did (I have a T1 connection), and they were VERY underwhelming for a supposed 122 Megapixel image. The image of the Chinese wall ornament was very soft. The details in the red ropes and the green medallion (for lack of a better word) were very unsharp. To my eye, images taken with a $1000 Canon DSLR look sharper (but still nowhere near as sharp as a 4x5, or ecen medium format transparency).

Now the Gigapixel image of Bryce Canyon (and the 800 Megapixel image of the Library of Congress on teh same site) were much more impressive. I think this technique holds more promise, but will need more CPU MIPS to become practical (check out how many DAYS it took to stitch and proces the final image).

Kerry

[Kerry L. Thalmann]

Oops, I forgot to mention, that in addition to being VERY soft, the high res image of the Chinese wall ornament had a large number of vertical streaks that looked almost like scratches running from the top to bottom of the image. They are easily visible in the red ropes and the green medallion. I'm not sure what caused these. Obviously, they aren't scratches (as there is no film to scratch), but that's what they look like.

Kerry

[QT Luong]

The problem with the stitching technique is that you need (a) to have light which is stable enough to take the 200 or so photos without too much variation, and to have (b) subject matter such that the lens can be focussed at a fixed distance. Those two requirements will eliminate most of the dynamic light and compositions often associated with great landscape images. At least the current scanning backs take just a few minutes to capture the images, and let you use view camera movements.

[Kerry L. Thalmann]

Tuan,
You make a valid point...

At least the current scanning backs take just a few minutes to capture the images, and let you use view camera movements.

However, many of the distortions corrected using view camera movements (i.e. converging verticals) can also be corrected in image editing SW. Also, the example of the gigapixel image takes things to the extreme. To get something equivalent to the 122 megapixel image from the scanning back would require far few stitched images (nearly an order of magnititude less). This would still require a static subject, but as long as the light wasn't changing too fast, it wouldn't be an issue. If the light does change a little, it can also be corrected in SW. For the 1.09 gigapixel image, the 196 photos that were stitched togther took the photographer 13 minutes to complete. The 25 - 30 images required for a 122 megapixel stitched image would probably take 2 - 3 minutes to capture with a DSLR and the proper tripod head.

I'm not a digital advocate (yet) and still prefer large format film for my photography. However, I have seen several very impressive stitched images (check out some of the other images on Max Lyons' web site) and think the technology is promising. It still has practical limitations, but that may be fixed in time as the technology evolves.

Kerry - still shooting film (although not as often as I'd like)

[QT Luong]

The most critical missing camera movements would be tilts for focussing. The focal length used in the Bryce 1Gp photo is 135mm. Even with the lens stopped down to f22 (no f64 setting in the small format world !) on hyperfocal distance, I think the sharpest foreground you'd be able to get would be still fairly far. In his forum, Max Lyon says that if the lens is not focussed at a fixed distance, you get strange artefacts thare are very difficult to fix. Of course, this is not a problem for vast vistas focussed at infinity, which seem to make up most of the images on his site.

[Legolas]

I've download both HighRes Images, and like Kerry said, they are not perfect. The UBC Cam use a cheap 1.200 dpi Scanner. With a better 3.200 dpi Scanner, one could make large (32.000 x 25.600) 819,2Mp Images for only ~400 €/$. One of the highest resolution Scanbacks on the market "Picture Gate 8000 Da Vinci" from Anagramm, have only (14.400 x 17.400) 250,6Mp, but cost 40.000 Euro/Dollar.

[David A. Goldfarb]

I suspect the sharpness issue, if the back is in proper registration with respect to the groundglass plane, has to do with the registration of the three images together. I have an old Nikon LS-4500AF scanner that I picked up for cheap that has a slight registration problem of this type (it's a three-pass scanner). It produces great B&W scans, and the color scans should be outstanding, but at 1000 dpi, my Agfa Duoscan is often better for color, I suspect because it's a one-pass scanner. With this scanning back, it may be that the system isn't rigid enough and the stepping motor in the scanner isn't precise enough to produce three images that can be aligned in perfect registration.

[Kerry L. Thalmann]

Tuan,

Actually, I believe Max Lyons used a 280mm lens for the 1.09 gigapixel Bryce Canyon image. However this is purely a function of the number of images he planned to stitch togther to get the final view. If you were to attempt a more reasonable stitched image in the 100 - 150 megapixel range, you would use a much shorter lens. Something in the 20mm - 70mm range depending on if you wanted a wide angle, or telephoto view. Assuming you wanted a wide angle shot with a dramatic, leading foreground, you'd be in the same 20mm - 35mm range typically used with a 35mm film camera.

Kerry

[QT Luong]

Kerry, you're of course right about that particular lens. I don't follow you for the rest of the argument. If you want a final view which is, let say equivalent to 28mm on 35mm with a 150 Mp resolution, you need to stitch together 150/6 = 25 = 5x5 images. Now each of these images have to be taken with a lens whose field of view is 5 times less than your original lens (maybe not 28 x 5, but I'm a bit too lazy to work that out :-)) which is well into the telephoto range and definitively not in the 20-35mm range.

[Vivek Iyer]

A friend pointed to this thread. You may want to take a look at Davidhazy's article (year 1999)here: http://www.rit.edu/~andpph/text-demo-scanner-cam.html

He does address many of the issues related to converting a scanner to a scanning camera.

I agree with the many well founded observations above that point out the problems with the (publically funded)UBC "Re-Search" project.