Greg's home page
Greg's diary
Greg's photos

Problems with Canon 9900F scanner

Last modified: $Date: 2008/02/10 23:50:18 $

This page illustrates the problems I have had with my Canon “CanoScan” 9900F scanner, which I bought for the express purpose of scanning in old slides and negatives.

The images on this page are reduced in size; to see the originals, click on the image in question with the mouse.

Summary

This scanner seems to work perfectly adequately from a hardware perspective. Unfortunately, the hardware is undocumented, so software like SANE does not support it. The only way to interface to the scanner is via the Canon-supplied ScanGear CS driver. Unfortunately, this driver has serious bugs. Canon support does not appear to understand the issues. I am asking Canon to:

I would prefer for them to do both. I am not interested in returning the unit, in which I have invested a large amount of time and effort.

The good news

The good news is simple: it's possible to scan with the software. If you're lucky, you'll get a thumbnail image like this:

Click on the picture for an enlarged version (128 kB)

This shows, roughly, thumbnail images of the 8 slides in the machine. It's still not perfect: it's difficult to see on the tiny images that the software provides (click on the image to get the 1024x768 original), but slides number 5, 6, 7 and 8 have a black band at the bottom. This is a relatively benign symptom of the problem: the software can't recognize the boundaries of the slides, even when given well-defined slides like this. You can compare the thumbnails with what the scanner scans by unticking the box “Display thumbnails” on the right. This then gives:

Click on the picture for an enlarged version (136 kB)

In this view, slide 1 is at top left, slide 4 bottom left, slide 5 top right, and slide 8 bottom right. There's nothing obviously wrong with the slides on the right-hand side; in particular, slides 7 and 8 are very well defined.

This framing isn't just a cosmetic issue: the scanner uses it to decide what to do during the final scan. For example, compare the raw scan and the finished image here:

Click on the picture to see a medium-size version in the index Click on the picture to see a medium-size version in the index

There's no excuse for this kind of breakage, but fortunately it's usually possible to fix it with software like xv, as long as the exposure hasn't suffered too much as a result.

The bad news

The previous section shows the best you can expect from the software. I have never seen a case where it recognizes all 8 slides correctly. In the following example, it's completely baffling that the software thinks that the slide is the other way around:

Click on the picture for an enlarged version (128 kB)

This problem isn't because the slide is inserted in landscape format: the software does cater for that, and it frequently makes the opposite error and assumes that portrait slides are in fact landscape.

Things get a lot worse if the slides have an even marginally dark background. In the following example, the scanner has managed to decide that frame 5 is the wrong way round:

Click on the picture for an enlarged version (0 kB)

If you move to the raw display, everything looks fine. It sometimes helps to turn the slide in question through 90°, as has been done in this example. The images look good, and the scanner appears to have been able to locate the first slide correctly, to judge by the frame around it. Note that it has been told that the scanner contains 35 mm slides (field on the right).

Click on the picture for an enlarged version (0 kB)

Sometimes this works, in other cases things go to hell, as in this example. Returning to the thumbnail screen we find:

Click on the picture for an enlarged version (0 kB)

Here the software appears to have decided that these are negative strips, although it has been told that they're slides. This is confirmed if you go back to the full view: the field now shows 35mm strip, though nobody has changed it. In other words, instead of taking the advice as to the kind of film in the scanner, it not only ignores it but changes it. What use is that?

To add insult to injury, when you return to the raw view and try to change it back again, it insists on clearing the previews:

Click on the picture for an enlarged version (0 kB)

It didn't do that when converting (without permission) in the other direction. Note also where it has moved the frame in this example. It's hardly even worth mentioning the bad grammar of this particular message.

I don't know how the algorithms work (or in this case, are supposed to work), but it seems probable that they recognize edges. In this case, this software has failed catastrophically: it has ignored the information at its disposal and spread the real edges of the slides around the “thumbnails”.

One theoretical workaround is to scan in the entire set of images as one image and use real software to dissect them later. That doesn't really work: the driver includes the black frames in the exposure calculations, resulting in a really washed-out image. You can also mark the frames individually and scan them like that. The latter takes for ever, and due to the low resolution of the thumbnails, about 50x70, it's also prone to incorrect exposure. The fact that the software chooses a line cursor for this function also makes things more difficult.

What this means

At the very least, this bug is a nuisance. But this kind of nuisance seems to be normal in the Microsoft world. It's more than a nuisance, though: at the very least, it means:

What Canon is doing about it

After a week of messing around with this software, I decided to contact Canon. They don't publish an Email address, so I called the local service phone line. After 5 minutes of waiting, was connected with a service representative who identified himself as Ejaz. He took some time to understand my problem report, but then suggested, in succession:
Greg's home page Greg's diary Greg's photos

Valid XHTML 1.0!

$Id: Canon-breakage.html,v 1.5 2008/02/10 23:50:18 grog Exp $