I ended up doing something like that... However it seems that the slices are not necessarily vertically contiguous (i.e. where there was a vertical gap on the page, the scanning software skipped it). Therefore, after assembling the slices, I obtained pages of varying height.
The fact that pdfimages doesn't report which images came from which page was a big nuisance (the original PDF had some pages with 4 slices and some with a single slice).
At the very least I wish pdfimages could report the location (page, x and y) of the various images it pulls out of a PDF, instead of dumping a heap of sequentially-numbered image files. It seems like such a silly omission. The other thing that I found annoying is imagemagick's propensity to silently resample images.
Oh well, I guess we're lucky to have even these tools...
Last edited by danutz; 12-21-2009 at 05:12 PM.