• Not for the variety of historical documents
  • layout analysis/segmentation
  • lack of suitable models and/or sufficient quality and volume open training data
  • pecularities
    • typographical features
    • ſpecial charæcters

• OCR-D is a diverse ecosystem aimed at the improvement of OCR for historical documents
  • coordination project
  • module & implementation projects
  • active user & dev community
• Funded by the DFG, we are mostly a German community - but we do speak English too ;-)
• Visit ocr-d.de for further details

• Specifications and conventions for CLI, metadata, formats and standards
• Implementation for Python
• Documentation for users and developers
• Publications and presentations

• ~60 Processors providing Python modules for various subtasks in the OCR workflow
• Setup, User and Workflow guide
• Installation of all supported modules

  ​​git clone https://github.com/OCR-D/ocrd_all && cd ocrd_all 
  ​​sudo make deps-ubuntu
  ​​make all
  

• Or use Docker

  ​​docker pull ocrd/all:maximum
  

• Guidelines for the transcription of document images for use as Ground Truth
• Repository with Ground Truth for text & layout recognition
• Models for OCR engines & tool support for training new models

• Main principles:
  • we open source & open data
  • we aim to involve and include the community from the start
  • we follow established standards and common best practices

→ transparency, replicability, collaboration

• try out ocrd_all
• create an issue
• send us a pull request
• add info to the wiki
• participate in our open calls
• say hello in our chat