Multiple scans: benefits?

[Marco Annaratone]

I am used to do multiple scans of the same picture because "this reduces noise." Tired of waiting to get my scans done I decided today to run an experiment. Setup: Epson 4870 with Silverfast Ai. Scanning positive film with 1, 2, 4, 8, 16 scans. No ICE, 3200dpi.

Result: the noise is indeed progressively reduced from 1 to 16 passes and one can see that on the screen at magnifications above 100%. However, the reduction in noise goes together with a reduction in sharpness. I was able to take the picture scanned once and reduce its noise at the same level of that scanned 16 times by applying in Photoshop a gaussian blur with a radius of 0.7 to it. (Of course I also lost some sharpness in doing so.)

Bottom line: it seems that scanning a film once and applying a tiny bit of blur produces the same result of scanning the film 16 times. Or, to put it another way, that scanning the film more than once buys you nothing: you might be better off in fact simply scanning it once and applying some selective blurring to high noise areas.

What am I missing here? Any word of wisdom?

[tim atherton]

"What am I missing here? Any word of wisdom?"

At least two things in the case of flatbed scanners (and which don't seem to be as much as an issue with dedicated film scanners in smaller formats)

1. film movement - usually do to heat - during the longer time it takes for each pass (could be reduced by some form of wet mounting - which in itself usually seems to give siginificant improvemetns to flatbed scans)

2. poor registration of the scan head between passes, which seems to be the case with many cheaper prosumer scanners - they were never usually designed to do multiple passes with accurate registration. This can vary a lot from scanner to scanner, even of the same model. They just aren't built to those tolerances. One 4870 can seem to do pretty well, another is afwul, for example

[julian_4860]

One advantage of my Minolta film scanner is that it does 'multiple samples' rather than passes, so the registration is spot on. I recently scanned some 4x5 on an old polaroid 45 film scanner, which was very accurate. These are a good buy refurbed - parts are still available from microtek - I've seen them going for just over 1k euros

[tor kviljo]

Julian - I have a Polaroid Sprintscan 45 (bought it to complement my old arcus 1200 for better Dmax), and is slowly learning to use it - but I have not found how to make multiple scan with it. Could You tell me two words about it to guide me in the right direction?

(by the way - do You know if these scanners were made by Microtek, or did they just take over parts & service after Polaroid left scanner-marked?)

Please excuse my interrupting of the original discussion!

regards

Tor Kviljo

[julian_4860]

Hi Tor,
I'm not an expert in this scanner, as I was doing some scanning for a friend in return for some pritns on his epson 9600. I was using vuescan and there is a box to tick for multiple passes on the first page. I didn't see much advantage to it, so I didn't use it.
These scanners were made by microtek, I have a boxed up unused sprintscan4000 which were the same generation. The 45 had two incarnations I believe the later 'ultra' and the original 45

[tor kviljo]

Thanks alot for info, Julian!. My Sprintscan 45 have some density-issue/banding in dark areas, which I would like to have resolved, but multiple scan is probaby not a way to go on that problem. Nice to hear it's made by Microtech, now that Polaroid seemes to go back to its roots and leave the electronic part of imaging to others.

Regards

Tor K.

[Donald Qualls]

One potential benefit from multiple pass scanning, if your software supports it or you're willing to do the work in Photoshop or whatever to make it happen -- it's very true to say that subsequent scan passes won't perfectly register with the first, and this can be a *good* thing. If you have the capability to align these later passes at sub-pixel resolution (as, for instance, by resampling to 400% and aligning the results with single pixel accuracy), then average the images after aligning, you can not only reduce noise without loss of sharpness, but actually *increase* effective resolution up to the limit of the resampled resolution you used to align the multiple scans -- though the resolution gain is limited by the number of frames available to combine and, somewhat, by the original resolution of those frames. Some software seems designed to do this automatically -- VueScan, for instance, may do this (I have the evaluation version, and it certainly does *something* with multi-pass scans that the TWAIN scanning software that came with my Arcus 1200 doesn't), but any editing software that supports layers can do it if you expand the canvas a little relative to the actual image size, resample to 200% or more, and use the single pixel nudge function (whatever it might be in your software) to match up the images in multiple layers.

Techniques for combining images this way were developed to a high level a few years ago in connection with video capture astrophotography -- the idea was to hook up a video camera to a telescope, capture several minutes of video tape with little or no effort to keep the image still in the field, select frames with good "seeing" (minimal atmospheric distortions), and combine those to produce a final image with both less noise and higher resolution than the original video; motion between frames was *essential* to making this work, in order to synthesize sub-pixel resolution. I've seen images created this way with a small telescope and 525 line video camera that equal those made with a single long exposure on twenty thousand dollars worth of large aperture, professional grade optics and chilled sensor multi-megapixel astrocamera.