Poor man's OCR: need performance improvement tips

[qvx]

I also have 0 OCR experience, but the case is simple enough.

I know about scipy but I have 0 experience with it. I was actually
hoping somebody who knows about it might have some recipe.

I also tried psyco, but unfortunetly, the speedup is only few percent.

I will check out ADaM's site.

I was hoping to replace matrix comparison with something more efficient
(minimal code change). I like the idea of dictionary lookup but it
requires more code to be changed. If nothing else comes up I will
probably try this method. Of course I will have to check the wider
characters first so there will be presumably several lookups for each
position. The only problem here is how to efficiently transform
portions of input picture into suitable format (some kind of
list/buffer).

Thanks.