## OCR-D: An open ecosystem for improving OCR on historical documents <ins>[Clemens Neudecker](https://twitter.com/cneudecker)</ins>, Konstantin Baierer, Matthias Boenig, Elisabeth Engl [OCR-D](https://ocr-d.de/) @ [ELAG2020](https://elag.org/category/elag-2020/) | 20-10-2020 --- <!-- .slide: data-background="https://i.imgur.com/oVWGouk.png" --> ### Wait, is OCR still not solved? ---- * Not for the variety of historical documents * layout analysis/segmentation * lack of suitable models and/or sufficient quality and volume open training data * pecularities * _typo_**graph**_**ical**_ features * ſpecial charæcters --- <!-- .slide: data-background="https://i.imgur.com/UOnRIyV.png" --> ### What is OCR-D? ---- * OCR-D is a diverse ecosystem aimed at the improvement of OCR for historical documents * [coordination](https://ocr-d.de/en/about) project * [module](https://ocr-d.de/en/module-projects) & [implementation](https://ocr-d.de/de/phase3) projects * active user & dev community * Funded by the [DFG](https://www.dfg.de/en/index.jsp), we are mostly a German community - but we do speak English too ;-) * Visit [ocr-d.de](https://ocr-d.de) for further details <!-- ## What is OCR-D? * 3 funding phases: * Phase I (2015-2017): gap analysis of the state-of-the-art * Phase II (2017-2020): development of technical prototypes * Phase III (2021-2024): implementation scenarios in GLAMs ---- --> --- <!-- .slide: data-background="https://i.imgur.com/7ZgBINs.png" --> ### Main outcomes of OCR-D ---- * :scroll: [Specifications](https://ocr-d.de/en/spec/) and conventions for CLI, metadata, formats and standards * :snake: [Implementation](https://github.com/OCR-D/core) for Python * :notebook: [Documentation](https://ocr-d.de/en/use) for users and developers * :pushpin: [Publications](https://ocr-d.de/en/publications#publications) and [presentations](https://ocr-d.de/en/publications#presentations) ---- <!-- grep '"executable"' ocrd_all/*/ocrd-tool.json |wc -l --> * :hammer: ~60 [*Processors*](http://kba.cloud/ocrd-kwalitee/) providing Python modules for various subtasks in the OCR workflow * :wrench: [Setup](https://ocr-d.de/en/setup), [User](https://ocr-d.de/en/user_guide) and [Workflow](https://ocr-d.de/en/workflows) guide * :floppy_disk: [Installation](https://github.com/OCR-D/ocrd_all) of all supported modules ```bash git clone https://github.com/OCR-D/ocrd_all && cd ocrd_all sudo make deps-ubuntu make all ``` * :briefcase: Or use Docker ```bash docker pull ocrd/all:maximum ``` ---- * :black_nib: [Guidelines](https://ocr-d.de/en/gt-guidelines/trans/) for the transcription of document images for use as Ground Truth * :bank: [Repository](https://ocr-d-repo.scc.kit.edu/api/v1/metastore/bagit/search) with Ground Truth for text & layout recognition * :factory: [Models](https://ocr-d.de/en/models) for OCR engines & [tool support](https://github.com/OCR-D/okralact) for training new models --- <!-- .slide: data-background="https://i.imgur.com/2zHruKc.png" --> ### Open & Community-based development in OCR-D ---- * Main principles: * we :heart: open source & open data * we aim to involve and include the community from the start * we follow established standards and common [best practices](https://ocr-d.de/en/dev-best-practice) &#8594; transparency, replicability, collaboration --- <!-- .slide: data-background="https://i.imgur.com/Y7LH8M9.jpg" --> ### Ways to get involved in OCR-D ---- * try out [``ocrd_all``](https://github.com/OCR-D/ocrd_all) * create an [issue](https://github.com/OCR-D) * send us a [pull request](https://github.com/OCR-D) * add info to the [wiki](https://github.com/OCR-D/ocrd-website/wiki) * participate in our [open calls](https://github.com/OCR-D/ocrd-website/wiki#discussions) * say hello in our [chat](https://gitter.im/OCR-D/Lobby) --- <!-- ## Challenges & TODO's * Evaluation/quality assessment **without Ground Truth** * Ease of installation/deployment * Improve usability for non-experts * Scalability and performance * Maintenance and sustainability --- --> <!-- .slide: data-background="" --> ## Thank you for your attention!
{"title":"OCR-D An open ecosystem for improving OCR on historical documents","tags":"ocr-d, elag"}
    1262 views