Reading drinks menus as geometry, not text
Why WhoozGot reads each drinks menu by its layout as well as its words, keeping whichever reading suits that document best.
What I tried
WhoozGot reads real drinks menus so the price you see is the price on the menu. The words themselves were never the hard part.
Why I chose differently
Which price belongs to which drink lives in where things sit on the page. A menu read straight across loses that on anything with columns.
What I chose
I built a reader that follows the layout and kept the simple one too. Each menu is read both ways and the better result is kept, which beats either alone.
The principle
Work out what kind of problem you have before reaching for a better tool.
- PDF extraction
- Document layout
- OCR
- Data quality