Page MenuHomePhabricator
Create Task
Maniphest T345055

Add kraken OCR engine to Wikimedia OCR
Open, Needs TriagePublicFeature
Actions

Description

Wikimedia OCR currently uses the free Tesseract OCR engine (which only supports printed text) and the commercial Google and Transkribus OCR engines.

The free kraken OCR engine supports printed and handwritten text. Like Tesseract, kraken is used in the OCR-D project for OCR of historic prints. It is much slower than Tesseract, but sometimes gets better results and would be the only available non-commercial OCR engine for handwritings.

I suggest to start with my free models for German print and German handwriting (they are not limited to German, but can be used with other languages which use Latin script as well), but there exist many more models, for example for Arabic or Hebrew script.

I already have implemented a prototype and sent a draft pull request for Wikimedia OCR.

Related Objects
Search...

Event Timeline

sweil created this task.Aug 28 2023, 8:27 AM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptAug 28 2023, 8:28 AM
sweil updated the task description. (Show Details)Aug 28 2023, 8:28 AM
sweil updated the task description. (Show Details)
MusikAnimal removed a project: Community-Tech.Aug 30 2023, 2:08 PM
Restricted Application added a project: Community-Tech. · View Herald TranscriptAug 30 2023, 2:08 PM
MusikAnimal removed a project: Community-Tech.Aug 30 2023, 2:08 PM
Samwilson moved this task from Backlog to Kraken on the Wikimedia OCR board.Sep 15 2023, 5:38 AM
sweil added a comment.Sep 18 2023, 12:41 PM

The current implementation offers 3 different models for the text recognition.
Is there a need for non Latin scripts as well? Which ones? Arabic? Hebrew? Others?

Kraken also supports different models for the segmentation (region and line detection).
The segmentation model should be selectable from the web interface and the API, too.

sweil added a comment.Sep 19 2023, 3:07 PM
In T345055#9174507, @sweil wrote:

Kraken also supports different models for the segmentation (region and line detection).
The segmentation model should be selectable from the web interface and the API, too.

The implementation now supports different segmentation models for kraken, too. Currently either default or ubma_segmentation can be selected.

sweil closed subtask T346413: Install Kraken OCR (and web service) on a new Wikisource VPS as Resolved.Sep 28 2023, 11:37 AM
Content licensed under Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 unless otherwise noted; code licensed under GNU General Public License (GPL) 2.0 or later and other open source licenses. By using this site, you agree to the Terms of Use, Privacy Policy, and Code of Conduct. · Wikimedia Foundation · Privacy Policy · Code of Conduct · Terms of Use · Disclaimer · CC-BY-SA · GPL