["stitching" multiple pieces of a scan together]
I had tried
that in the past, but the result weren't satisfying at all.
I've done many
scans of HP manuals that way with good (if time consuming)
results. See, for example, the schematic on page 8-9 of:
http://www.bitsavers.org/pdf/hp/64000/hardware/64050-90901_Feb-1983.pdf
I have only a US-letter-sized scanner (HP 5590), so that 30-inch by 11-inch
page was made from four or five scans spliced together.
E.g. you'll never get the lines crossing the
scan boundaries to match.
I obtained good matches if I used a program that
straightened each of the
partial scan images before assembly. No matter how carefully I aligned the
page on the scanner, there were always small angular displacements from
image to image. Joining them together without postprocessing would show
the problem that you mention. But after straightening, they would join
virtually seamlessly (i.e., with a displacement, at 600 dpi, of a pixel or
two across the image height).
I worked around the edge distortions by cropping the images before
assembly, although it often meant that one more scan was needed, as I was
only using the central 8 inches or so of the 8.5-inch scan.
My experience is that it is certainly possible, but it took me maybe 5-10
minutes of labor per wide page in cleanup and assembly to get good results.
Does any software currently wexist that is specifically for that purpose?
For analog images (photographs), there is some specific software available
specifically for it.
Look at gigapan
However, successful results require overlap, and the amount of overlap
needed might be an issue.
In the case of Gigapan, where they are photographs, there can be some
facinating artifacts wherever there was motion between one frame and
another, often resulting in missing heads, extra arms (ala Zaphod
Beeblebrox), etc.
On a panoramic phot sequence, I tried the Microsoft Image Composite
Editor, with mixed results. When there were a lot of images (high-res),
it got confused and made some false matches, eliminating the Richmond
refineries! Also, although it was JPG in and JPG out, the JPGs that it
output were not accepted as input (I tried to combine the images in
several groups, but it claimed that there were NO matches between the
multiple groups).
For schematics, or even text, there are different problems, and ideally,
the software for stitching them should NOT be exactly the same algorithms
as the search for overlaps in photo stitching.
A mismatch is potentially serious, so I would suggest including in the
image an overlay of a set of dotted? lines marking the edges of all
overlap areas. After manually confirming that the traces line up
properly, that overlay could be dropped.
Perhaps specialized software could be written that would look for and flag
any occurence of a line segment that ends without connection, particularly
in the overlap regions.
--
Grumpy Ol' Fred cisin at
xenosoft.com