OCR
Turn an image of a page (print, woodblock or manuscript) into Unicode text.
Definition
Reading the text on a page image and writing it out as Unicode, line by line, without correcting the author. Layout (which block is main text, which is a note or page number) is part of the task when the output keeps it.
Input and output
- Input: One page image (scan or photo).
- Output: The text of the page in Unicode, in reading order, with line breaks.
Examples
None yet. Add a short real example with its correct answer, and say where it comes from.
Edge cases
- Woodblock prints with worn or broken letters.
- Cursive (dbu med) manuscripts and abbreviations (bskungs yig).
- Interlinear notes in smaller script, marginalia, page numbers in the margin.
- Pages that mix scripts, such as Tibetan with Sanskrit in Lantsa.
Languages and traditions
- Tibetan (bo): measured on the BDRC Tibetan OCR benchmark.
- Other languages: no benchmark listed yet.
How it is measured
Tibetan OCR is ranked by BDRC on its own open leaderboard, which BuddhistAI lists but does not score: Tibetan OCR leaderboard, metric mean character error rate.
Help wanted
Benchmarks for other scripts and languages (Pali in Burmese, Sinhala or Thai script; Classical Chinese; Sanskrit manuscripts), each with page images and checked transcriptions. A first benchmark can be small: twenty to fifty items. See how to submit a benchmark.
Related tasks
Found in workflows
- Translating a Buddhist text (Kurt Keutzer)

BuddhistAI Knowledge base