SuperNet‑108: OCR support for new languages, more reliable results for old scanned documents, and more

OCR improvements

We are now supporting text recognition for two new languages:
  • Greek
  • Georgian (modern alphabets, Mkhedruli and Mtavruli)
We have also improved recognition for challenging old scanned documents. To illustrate, the following paragraph of text:
Sample of old document scan
will now be recognized as:

Posto tal Letterario contante in vista di chi legge non saprei dir, di qual peso, è da notarsi, che questo nome sustantivo proprio Livorno , il quale in latina lingua dicesi Liburnus Labro , specificatamente , anzi individualmente indica la Città , e. Porto di Livorno come si legge in Cicerone, nel Volterrano , e nell' Alberti (1) , ma che sotto quel nome di Liburnus esser può soggetto a più significazioni, che lo rendono equivoco.
Finally, we are changing our output to be more Unicode aware, and to output Unicode symbols instead of Latex commands where appropriate (in text). For example, the following image:
Sample of text containing ★
will now return the following response:

Text sentence that contains ★ will be recognized without Latex command for a star.
instead of the response that we used to return:

Text sentence that contains $\star$ will be recognized without Latex command for a star.
Note, that while many other Unicode symbols are supported in-text context, when present in math context, we will still return proper Latex commands for equations.

Output formatting changes

Output formatting changes are related to our efforts to have more Unicode friendly results. We are changing the way form fields are represented in the output:
  • We are now using □ in cases where \square was used (e.g. checkboxes, and some other form boxes)
  • We are now using ◯ in cases where \bigcirc was used (e.g. radiobuttons)