wondarchive

wondarchive › OCR vs HTR

OCR, HTR, automatic transcription: what’s the difference?

OCR (optical character recognition) reads printed or typewritten text; HTR (handwritten text recognition) reads handwriting, long relying on models trained on a given hand. Recent AI models read the whole page, printed or handwritten, without training: that is what wondarchive does.

What is OCR?

OCR, Optical Character Recognition, recognises the characters of printed or typewritten text. It excels on clean scans in regular type; it struggles with skewed photos, shadows, blackletter without a dedicated model, and handwriting. Tesseract, which is open source, and ABBYY FineReader are two examples.

What is HTR?

HTR, Handwritten Text Recognition, reads handwriting line by line. HTR tools such as eScriptorium with its Kraken engine, or Transkribus, segment the page into lines and then apply a model, often trained on pages in the same hand transcribed manually.

And AI reading models?

They read the whole image the way a person does: print, typescript, blackletter, handwriting, tables, double pages, without training on your handwriting. They understand layout and language, and can also translate and pick out names and dates. wondarchive uses them while always showing the text next to the photo, and recognises blank pages rather than inventing a text for them.

How do they compare?

OCRHTRAI reading
Print, typescriptyes, on clean scansyesyes, photos included
Handwritingnoyes, with a suitable modelyes, without training
Trainingnooften, on the handno
Skewed photos, shadowspoorlydepends on segmentationwell
Translation, names, datesnonoyes
ExamplesTesseract, ABBYY FineReadereScriptorium, Transkribuswondarchive

Which one does wondarchive use?

AI reading for every page, with an extra-strength reread for handwriting. And, as a free option, an open-source OCR that reads pages in your browser, with no AI and no credits: handy for sharp print (free transcription).

The account takes a minute to open, a hundred pages free for a year.

What does HTR stand for?

Handwritten Text Recognition: the automatic reading of handwriting.

Can OCR read handwriting?

Poorly: classic OCR is built for regular printed characters. For handwriting, you need HTR or an AI reading model.

What is the difference between transcription and OCR?

OCR is a technique; transcription is the result, the text of the page as written, whether produced by a machine or by a person.

Do you still need to train an HTR model?

To read and search a collection, no: recent AI models read without training. Training is still worthwhile for a line-by-line edition of a single hand.

See it on your own pages

A hundred pages transcribed and translated, for a year, no card required.