Dedicated AI solutions for rare documents and languages

Unlock the richness of your collections with models specialized for your needs, from text recognition to data structuring.

Our expertises

...

OCR and HTR - Text recognition

Calfa developed an Artificial Intelligence for Arabic, Armenian, Georgian, Syriac etc. This technology is able to analyse the page layout and to read the text from the document, giving it a new, digital potential.

Discover

Support and dedicated projects to showcase collections

From the study of your collection to the delivery of the corpus, we run tailor-made projects for libraries, archives and research teams: models trained on your documents, reference transcriptions, catalogues structured into searchable databases and integration into your digital library.

...

Open-source datasets, models and tools

We provide open acces tools and models for research. We also make available for downloading multiple datasets and research materials.

Discover

Funding heritage preservation projects

We lead or co-finance projects to preserve written or printed cultural heritage. Discover our digitization and conservation actions.

Discover

They entrusted us with their projects

BnF DataLab BULAC Ghent University UNIL Lausanne Peeters Publishers Nubar Library

Our scientific partners

GREgORI CNRS-IRHT GIS Moyen-Orient et Mondes Musulmans EANC DIGILIB FSL

Specialized models for every technical need

From image to data: line and layout detection, text recognition, object detection and linguistic analysis, on your own documents.

Armenian manuscript in cursive script on yellowed paper, with the writing on the back showing through.
Cursive Armenian manuscript with ink bleed-through

Segmentation · OCR/HTR

Text recognition (OCR/HTR)

Manuscripts and prints, generic models or models trained on your documents. Curved or vertical text, damaged pages, mixed alphabets on the same page.

Page of the Armenian daily Haratch from 1931: text columns, headlines and advertisements highlighted in different colors by type.
Haratch (Paris, 1931): detected regions and content identification

Layout

Layout analysis and data extraction

Headings, notes, columns, reading order; catalogues turned into databases.

Sheet of round and octagonal Armenian stamps and seals, black and red, automatically extracted from manuscripts.
Stamps detected in Armenian manuscripts

Object detection

Illuminations, initials and stamps

Detection of illuminations, decorated initials and stamps, including from a IIIF manifest. Try the detector (free)

და ვითარცა ნისლმანნისლი · noun რაჲმანმე მოიცვამოცვა · verb გულიგული · noun ჩემი და დაჰფარადაფარვა · verb

Old Georgian: content words linked to their lemma and part of speech

NLP · linguistic analysis

Linguistic analysis

Lemmatization, morphological analysis and named entities for Classical, Eastern and Western Armenian, Old Georgian, Syriac and Ancient Greek; annotation of corpora of several million word forms.

Discover our projects

Aerial view of the island of San Lazzaro in Venice: the monastery and its gardens surrounded by the lagoon.

99.2% recognition accuracy

The San Lazzaro collection catalogue turned into an online catalogue

Printed catalogue in 8 volumes (6,250 pages, over 2,000 manuscripts) recognized by OCR and full-text searchable, with the Mekhitarist Fathers.

Browse the catalogue Case study

Lithographed page in Arabic on orange paper: a table in columns (detail).

97.2% average recognition accuracy

TariMa: 10,000 pages of Maghreb history made searchable

Manuscripts, lithographs and prints from the 17th–20th centuries held at the BULAC, in Maghrebi scripts, now searchable in BINA, the BULAC digital library.

The project on the BULAC website Dataset

Page of one of Antoine d'Abbadie's notebooks: list of names in Ethiopic script with their Latin transcription, and handwritten notes in French.

3,000 handwritten pages transcribed by AI

Antoine d'Abbadie's notebooks: several scripts on the same page

Notebooks from his journey in Ethiopia (1837–1848) mixing Ethiopic, Latin, Arabic, Hebrew and Greek scripts, transcribed by a triple HTR model, with the BnF and CNRS–IMAF.

Case study

Stay tuned with our projects

Receive the latest news and updates on our projects and conferences by email