Here’s a python snippet showing the inconsistencies of the OCR with regards to underscores:
param list = (
param_name, param_desc, param species, param_sound, param_behavior, param color,
param align
All the parameters within the parentheses were to have underscores in their names. However, as we can see, only half came through OCR correctly.
Another example:
elif user_input in ('4', 'S', '6'):
This took a little while because I was viewing my code on a smartphone screen. Only when opening on my (admittedly not so large but still reasonable) 13 inch Chromebook did my eyes catch the capital S.
I think that OCR performance could be massively improved just by introducing context. That is, if it were to ‘know’ that the text was a coding language, for example, it would be directed to pay more attention to underscores and other programming-intense elements.