AIDC is a known field, Layout recognition too. But
Structured Text does not get the attention it should.
Funny, because it is needed in almost every field of
business.
So, following this post and the hundreds of buns (and twits,
shares and likes on facebook following the HB success),
there
will emerge a python framework for fast structured
text recognition originally built around OpenCV which will
then be ported to Java and CSharp. It will do the
following:
a. Discover different possible text areas and direction of
text in the image.
b. Mark those areas - and show them (if asked) to the user.
c. Discover the font and guess its "size" and attributes.
d. Work out objects that are NOT in the same font (so for
example :: New York :: won't be understood as ii New York
ii)
e. Optionally give font information for the recognized text
f. Give the recognized document structure and layout.
g. Cleanup and pre-process document areas separately. -
also giving correction information for the image as a
whole.
h. Each recognition area should attempt several pre-
processing directions and compare results for achieving
ground-truth results.
i. Optional: pre-input of expected text or expected format
of text for interpreting results
j. Optional: dictionaries with possible texts to expect -
assisting in interpreting results - (with a score for supposed
understanding)
k. Optional: Pre-input constant texts that are to be located
in certain areas.
l. Optional: Pre input constant layout.
I'm sure that I'll get at least one bone from someone who
will show me that it is widely known to exist, and I could
have found it with a simple google search if only I knew
the correct terms.