Andrea Bozzi in Pisa had been working on a ms project which did image recognition to locate the exact word. I don't think it ever progressed beyond a research project, but was interesting work nevertheless. It was published, but I don't know where.
And have you looked at the IATH's inote? it may be able to do what you want. It's an image annotation program.
Scholarly Technology Group
Providence, RI 02912