Text recognition in Rutba Office that never leaves your computer
Since 1.47 Rutba Office reads the words in scans and pictures itself, with the network off, in English, Urdu, Arabic, Hindi, Chinese and Japanese. Scanned PDFs become searchable, a photo’s words can be copied, and a printed table lands in cells.
· 3 min read · updated 15 October 2026

A lot of text recognition is a service: the scan goes to somebody’s server, and the words come back. For a contract or a medical letter, that is a lot to hand over to read a page. Rutba Office 1.47, released on 13 October, reads the words itself, on your computer, with no account and the network off.
6
languages: English, Urdu, Arabic, Hindi, Chinese and Japanese
~1 s
a page on a four-core laptop
2.3%
of characters wrong on real English scans (2.7% for Hindi)
Where you can use it
- In PDF, Recognise Text writes the words of scanned pages in as an invisible layer under each page, so Find, selecting and copying work on a scan, while the pages look exactly as they did. Pages that already have text are left alone.
- In Pictures and Image, Copy Text from Picture puts every word in a picture on the clipboard in reading order, and lays the recognised lines over the picture so a drag copies just the words you want.
- In Worksheets, Data from Picture reads a printed table, from a file or the clipboard, into cells. A preview shows each cell beside the picture, editable, with the cells it was unsure of marked for you to check, and numbers arrive as numbers.
How it reads
The reader is Tesseract 5, an open-source engine, compiled to WebAssembly. Its six language models come in the download, so nothing is fetched the first time you use it. Around it the page work is our own: straightening a crooked scan, evening out unevenly lit paper, finding lines, columns and tables, and putting right-to-left lines in reading order. Reading runs on threads of its own, one page a thread and never more than four, so no window waits while a long scan is read, and any reading can be cancelled.
The figures above are measured, not estimated. The suite’s own documentation gives the error rate, speed and memory for each language on pages drawn with known text and on real scans, and the tests prove that a reading happens on the reading threads and makes no network request.
Reading a page as a document
1.50 taught the engine to read a page as a kind of document: it tells an invoice, a receipt, a letter and a form apart, finds a bill’s supplier, number, dates, subtotal, tax, total and line items, checks that the sums agree, and reads a form’s tick boxes. The windows did not offer this in 1.50.
Download Rutba Office
Text recognition is part of the free suite, in every copy. No account, no telemetry, and nothing uploaded.
Download

