Difference between revisions of "Tibetan OCR"

From Digital Tibetan
Jump to: navigation, search
(Tesseract 4 (by Google, with neural networks))
m (Google docs supports Tibetan OCR)
 
Line 7: Line 7:
 
It just seems the 'Translate' button doesn't work yet with Tibetan.
 
It just seems the 'Translate' button doesn't work yet with Tibetan.
  
==Google docs supports Tibetan OCR==
+
==Google Docs supports Tibetan OCR==
 
The easiest method for Tibetan OCR is simply to open a PDF with Tibetan in Google Docs. Wood block PDFs are supported!
 
The easiest method for Tibetan OCR is simply to open a PDF with Tibetan in Google Docs. Wood block PDFs are supported!
  

Latest revision as of 15:35, 29 November 2019

GoogleLens.png

Tibetan OCR with mobile phones

Latest Google Photos support translating images of Tibetan text into editable text. Simply take a photo of some Tibetan text, open Google Photos (both Android and iOS versions work!), and select the 'Lens' function. You can now copy editable text.

It just seems the 'Translate' button doesn't work yet with Tibetan.

Google Docs supports Tibetan OCR

The easiest method for Tibetan OCR is simply to open a PDF with Tibetan in Google Docs. Wood block PDFs are supported!

Example page:

Woodblock.png

After "Open with Google docs":

zབས་དང་བརྒྱུད་འདེབས་སབསྡུས་སྐབས་བསྟན་སྔོན་དུ་འགྲོ་བས། དངོས་གཞི་ལ་བདག་བུམ་གཉིས་ལས། དང་པོ་ལའང་སྐྱབས་སུ་འགྲོ་བ་ནི། སྤྱོད་རྣམ་དག་དྭངས་མའི་ །
eg སྐུ།ཉོན་མོངས་རྣམ་དག་ཡེ་ཤེས་ལྔ།, །སྐྱགསུམ་མཛད་པ་མཐར་ཕྱིན་པའི། །དཀོན་མཆོག་གསུམ་ལ་རྟོགས་པས་འདུད། །ཅེས་ལན་གསུམ། སེམས་
བསྐྱོད་ཐུན་མོང་བ་ནི། བདག་དང་མཐའ་ཡས་སེམས་ཅན་རྣམས། །ཡེ་ནསསངས་རྒྱས་ཡིན་པ་ལ། །རྟོག་པའི་དབདགིས་འབོརབ་ལས། །བྱང་ཆུབ།ཚོག l
ཏུསེམས་བསྐྱེདདོ། །ལན་གསུམ། ཁྱད་པར་གྱིསེམས་བསྐྱེད། བསམདུ་མེད་དེ་སེམས་ཉིད་ནམ་མཁའཚེ། །སེམས་ཉིདནམ་མཁའ་བསམ་པ་ཀུན་དང་་་་་་
བྲལ༑ །བསམ་དུ་མེད་པའི་སེམས་ཉིད་ནམ་མཁའ་ནི། །ནམ་མཁའ་མི་དམིགས་པ་ཡང་དམིགས་སུམེད། །ཨོཾ་དྷརྨ་དྷ་ཏུ་ཛཱན་བཛྲས་བུ་ཝ་ཨཏྨ་རྟེན་
དྲིལ་བཅས་སྙིང་ཁར་དམགྱི་བཅའ། མཆན། ཀོཉཧཾ། ཞེས་མི་དམིགས་པའི་དང་དུདལགཅིག་བཞག་ལ། སླར་རིག་གདངས་ཆུ་ལས་ཉ་འཕར་བ་བཞིན་དུ། ྋསོད་པའི་དད་ལས་ཧཱུཾ་ཧཱུཾ་

While there are a number of errors (some artefacts generated by the left caption, some Sanskrit weaknesses, and some problems with difficult-to-read parts), this is some remarkable progress.

Namsel Ocr

Available for Windows, Mac and Linux and now on Docker

Work great on Docker: Docker Namsel Ocr For the main version: Namsel Ocr

Sources are available on the Github page https://github.com/thubtenrigzin/namsel-ocr for the main project and https://github.com/thubtenrigzin/docker-namsel-ocr

OCR at dharmabook.ru

dharmabook.ru offers a free OCR service, uploaded texts are converted within a few days:

This service seems to work well with wood-block pechas.

See Buddhist Library Project for more information about the Project

Tesseract 4 (by Google, with neural networks)

Note: Tesseract mostly supports linux. Technical computer skills are required for usage.

Tesseract 4.0 supports OCR (optical character recognition) for Tibetan.

The new version adds a new OCR engine based on LSTM neural networks. It initially works (well) on x86/Linux. Model data for 101 languages (including Tibetan and Dzongkha) is available in the tessdata repository.

Installation

Refer to the Tesseract repository for detailed installation instructions.

In addition to tesseract, you will need trained language sets.

Language training sets

Results

Recognition works quite well for printed Tibetan texts, however, the recognition rate for wood-block pechas is still poor.

History

A good overview of different endeavors in Tibetan OCR is given at the Namsel project: