A Scanner (not too) Darkly
March 4, 2011 8:50 AM   Subscribe

I've got 50+ rolls of slide film that I want scanned. I have a film scanner. What software should I be using to do the scanning?

I've done some scanning in the past but never had the time to tweak the settings to make the scanned image look like the image on film. I've used Vuescan so that is the software I know. It seems good enough, but apart from Xsane it is the only software I know.

My plan is to set the software up and then have my wife do the actual scanning when she has time.

So here are my requirements:

1) I want the best images my scanner can provide because there will definitely not be time to re-scan these in the future.

2) The process should be easy for the person doing the scanning. All the tweaking will either be done before-hand or by batch processing afterwards.

3) No Macs in the house yet so it's gotta be Windows or Linux.
posted by any portmanteau in a storm to Computers & Internet (6 answers total) 7 users marked this as a favorite
My Nikon Coolscan V came with perfectly acceptable software -- in fact, the Digital ICE, ROC & DEE, Etc. that it ships with are pretty essential to me. I've tried a demo of Silverfast but I didn't see much of anything it did that was better than the Nikon software.

I use mine as a plug-in through photoshop, so that once the scanner is done scanning, the tif opens within Photoshop for editing. Most scanners should ship with Photoshop plug-ins and instructions on how to install them.
posted by Devils Rancher at 9:22 AM on March 4, 2011

Hey, I run an archive that used to be a slide library; we scan 35mm slides constantly. We use a Nikon coolscan 5000 scanner with a batch feeder, and the Nikonscan software. I've tried Silverfast and Vuescan but the Nikonscan does the job for free and it has been trouble-free for years (until they stopped making some of the scanners, but that's another story). We also use an Epson 10000xl flatbed with slide templates using the epsonscan software launched through photoshop. It gives good results, no complaints. We crop and color-correct in Lightroom, and touch up in photoshop.

I have to say, 1800+ slides is a big project. The Nikon scanning at full resolution using ICE dust-removal runs about a minute and a half per scan, and that's without any cropping or color work yet. With a big flatbed scanner the 35mm slide templates might hold 15 or 30 slides (our Epson 10000xl does 30 in two templates at once) but at high resolution you're talking 15 minutes per batch scan, plus further processing time.

You might consider farming this out to a vendor - there are plenty of scanning labs that specialize in mass digitization of slides - just pick one that scans in their own lab and has good quality control (not an overseas send-us-a-box and we-do-the-rest type place).

The Nikon scanners are getting hard to find and any dedicated slide scanner worth the money is pretty expensive. (most are not worth the money)

Feel free to memail me with specific questions.
posted by gyusan at 9:35 AM on March 4, 2011

I've used most of the big ones (though I'm a Mac user, so there are surely some Windows-only options I haven't tried), and IMHO VueScan is the way to go. Good quality scans, all the flexibility you should need, relatively inexpensive (compared to Silverfast, for example). It also works with every scanner I've tried it with, which is nice. The UI isn't super intuitive, but really none of them are, and once you have it figured out it's easy to use. Most of them have free demos, so it's not that hard to try them and see what works best for you.

There *are* differences in the quality of scans you'll get from various software (or, more correctly, how difficult it is to get a good scan). I've especially found that with marginal originals (e.g. massively under- or over-exposed, or with weird color balance problems), some software makes it almost impossible to get an acceptable result while others don't. In my experience, the manufacturer's software is rarely the best in this regard, but it's been years since I've played with NikonScan.

All that said, I do agree that you should consider outsourcing this project. It's a lot of work, and not particularly fun.
posted by sharding at 9:48 AM on March 4, 2011

1) I want the best images ...
2) The process should be easy...

These are seldom compatible expectations. For best quality, turn off all auto exposure, histogram balancing, sharpening, turn it all off.

Try to find a single exposure control, and test scanning a normal picture. Check the histogram to see you have the full range, and aren't clipping at the top or bottom.

Scan each image four separate times at the highest bit rate, probably nominally 256 shades of red green and blue.

The rest can be batched, once you figure the right values.

When processing, open a picture and using curves in photoshop, or some similar tool, set the picture to black, and start switching the lowest colors (color 0, color 1, color 2, etc) to white. See if those white info that comes in conforms to image data, or to a noisy or fixed pattern that is only the image sensor itself.

Almost certainly, colors 0 and 1 will be complete noise. Often colors up to 8, or even 16 and higher will be complete noise. By noise, I mean not the slightest amount of picture is there at all, just, usually, stripes along the entire image, or random snow with no hint of image. Most eight bit scanners do a maximum of six bits, with the lowest 2 bits, or 4 values, being pure.

Return to normal color tones, and raise the black level up to, or just slightly below, the lowest pure noise setting. If the noise in the lowest values has a pattern to it, just wipe out the lowest values by raising the black, if it is pure snow, then do a gaussian .3 blur, and wipe out the bottom colors.

Don't expand the range or anything like that, but moving the bottom point in curves will slide the whole color-mapping line with it, so maybe look closely at the results to make a decision. You may want to consider how you adjust that 45 degree line plot you see in curves, by pinning a control point at 32 or 64, so that nothing moves above that.

The process of clipping the noise from the black can be done separately for each of red, green, and blue. The blue will usually have a lot more noise down there, so you would clip that more. All three, r,g,b,should be pinned at the same place, as described above, so the relative mix of the colors in unchanged above 32 or 64.

Look at the histogram afterward, if it looks like a comb, you did it wrong. Do a gaussian blur of .3 to .5 pixel width on each scan, lesser value if the results already look sharp, higher if it's noisy or blurry. Using 16 bit per pixel Photoshop, add the four separate scans of each picture together. Do the lightest amount of sharpening that you can. The picture will look dark at sixteen bits, so again in curves, pull the top corner of the graph down to 2048, or whatever doesn't clip the bright whites. Another .3 gaussian blur. This would be a good time to reduce the picture to half its size with the best interpolant option. (Your scanner cannot capture good quality at its highest res, unless it was very expensive.)

From here on you'd be eyeballing it, looking at the histogram or other tools.
posted by StickyCarpet at 1:43 PM on March 4, 2011 [2 favorites]

"Most eight bit scanners do a maximum of six bits, with the lowest 2 bits, or 4 values, being pure. CRAP"
posted by StickyCarpet at 1:45 PM on March 4, 2011

Should mention that once you have found the settings in curves to clip the black noise, save that curve and re-invoke it in the batch script. That curve is not to enhance any particular image, it is to adjust for the properties of the scanner. You can adjust each picture all you want later.

You could also pre-sort you slides into normal, underexposed, and over exposed, and set a single scanner exposure and adjustment curve for each batch of exposures.
posted by StickyCarpet at 1:58 PM on March 4, 2011

« Older I'm not sure I can fit under my car, though.   |   Lithuanian names, spellings and family names Newer »
This thread is closed to new comments.