Doing a little more online research, it seems like I am looking for a free/open-source solution for OMR (optical mark recognition) using a generic image scanner or PDF (instead of a specialized device that just scans OMR forms).

I wonder if there are computer/machine vision hackers who might know where to find (or put together) a good solution.