[nSLUG] Image utility for Linux?
jdkenney at gmail.com
Tue Apr 24 02:10:15 ADT 2007
Just to follow up, most digital cameras support some type of
"panorama" mode for taking several pictures and doing exactly what you
are describing if I understand right. I think the same functionality
is in Photoshop, which means there is probably a similar function in
the gimp, so at the very least there is code there to be had for
harvesting, if you can't find a library that will do it.
Additionally if they are line images, there are pretty straightforward
algorithms for reducing a binary image to one that contains only lines
of single pixel widths, which doesn't seem like it would be too bad to
make a fuzzy matching system for to find the co-ordinates in the two
images which match up (well, a common co-ordinate, relative angle and
zoom). In any case, a much simpler problem than the general case of
two images which are not identical.
On 4/24/07, Jason Kenney <jdkenney at gmail.com> wrote:
> I believe what you want is called "image registration".
> On 4/24/07, Mike Spencer <mspencer at tallships.ca> wrote:
> > Way back, I had a hand scanner on Windows 3.1. Since it could only
> > scan a narrow strip, the Logitech software included a utility to
> > automagically align and paste strips together. It worked surprisingly
> > well for me when I did some 8x10 B&W photos.
> > Now I have a flatbed scanner that works fine with SANE. But I want to
> > scan a portfolio of architectural line drawings that are just a tad
> > too big for the scanner.
> > Is there a Linux command line utility that will do this trick with two
> > overlapping PBM images? Or, for that matter, *any* Linux software that
> > will do this for *any* image data format?
> > The images are too big for manual manipulation on the screen and I'd
> > like to keep the original large format for archival purposes.
> > Failing that, is there a technical key word for doing this trick that
> > would help me google?
> > Shingling off even further onto the fog:
> > If it were guaranteed that the two overlapping images were bit for bit
> > identical in their overlapping areas, it would be more or less
> > straightforward, albeit tedious, to write something with libpnm,
> > especially considering that these are line drawings. But it's
> > reasonable to assume that the overlapping area will only be very
> > similar, not identical, which would mean a fuzzy matching algorithm.
> > Which is much harder. Ho hum.
> > Is there now a C algorithm (possibly well-known to everyone but me :-)
> > for doing this sort of thing? I recall talking to one of the genome
> > guys at NRC back about 1992. They were having a hard time doing this
> > kind of matching between DNA fragments, where they wanted to find
> > matches between fragments that were *nearly* the same. So is there
> > now some canned code that I could use if I want to try to write
> > something?
> > Or any other suggestions on how to accomplish this?
> > Thanks,
> > - Mike
> > --
> > Michael Spencer Nova Scotia, Canada .~.
> > /V\
> > mspencer at tallships.ca /( )\
> > http://home.tallships.ca/mspencer/ ^^-^^
> > _______________________________________________
> > nSLUG mailing list
> > nSLUG at nslug.ns.ca
> > http://nslug.ns.ca/cgi-bin/mailman/listinfo/nslug
More information about the nSLUG