Optical Character Recognition (OCR) With Tesseract3

[jpbro]

Cool. I did this a few years back,

I'm often late to the party. I should probably be "embracing AI vision models" for this now instead.

The idea for me, was to read information from invoices (which were stored as PDF's).

I'm also grabbing stuff from PDFs (converted to bitmaps first). I have a smorgasbord of PDFs to work with - there's such a wide range in quality (including scans of printouts), and I found it impossible to read information at the page level (or any level) and give users a chance to figure it all out. So instead, I ask users to pick a column/field of data they are trying to input (e.g. "$ Owed" for Record X) and then they drag the mouse over the number they are interested in. The software then converts the image->String->Currency and drops it into the appropriate field. The String->Currency bit was a bit tricky actually, but it seems to be working reasonably well now (albeit kinda fuzzy).

How did your approach work?