In trying to archive some prints I have before they inevitably yellow, I invested in a scanner and Silverfast Ai. They’re printed materials, so require descreening. It’s important to scan materials that need descreening at whole multiples of the native optical resolution. The best trade-off for mine ended up being 1280ppi.
Some of these prints are far larger than the scanner bed, so I scan them in multiple passes, with about 20% overlap. Then I use photoshop to stitch them together.
I discovered an unusual phenomenon in that the auto stitching mechanism sometimes had great difficulty getting a match, and when I attempted it manually it became apparent that the same sections, scanned multiple times, were different dimensions. So the overlap(s) were not perfect copies of one another.
I found literally ONE reference to this on the internet, so thought I’d place this information here in case that site ever goes down:
After digging into the issue further a few days ago, it seems to be related to CCD scanners specifically, as opposed to CIS scanners, due to the particular use of a lens to transfer the image data to the CCD sensor.
Related Links:
CCD or CIS (web – by the WideTEK scanner folks)
CIS vs. CCD (pdf brochure from Paradigm Imaging Group)So it appears that the best color capturing happens with CCD, but you also get lens distortion with that. I suppose if someone is mainly scanning documents, a CIS scanner may be the way to go (in actuality, probably just fine for photos, too).
Josh T. on forums.anandtech.com
The lens distortion is not uniform, nor symmetrical. The best option I’ve found is to use Photoshop’s Photomerge “auto”, which allows scaling, rotation and distortion (meant for photographs). I was really hoping to be able to pull off something approaching archival quality (I calibrated and profiled the scanner), but having to distort dimensions to successfully stitch is really disappointing.
I’m almost crazy enough to switch to the method of using a DSLR and macro lens, undoing lens distortion before stitching.
Almost…