Skip to main content

Course 1: Text Recognition

Andreas Fischer · HES-SO / University of Fribourg (now University of Fribourg)

Lecturer

Andreas Fischer

Andreas Fischer is Professor of Computer Science (Software Engineering) in the Department of Informatics at the University of Fribourg, which he joined in September 2025, and heads the research group AI for the Benefit of Human Experts (AIBEX). Before his appointment he combined a lectureship at the University of Fribourg with a professorship at the School of Engineering and Architecture of Fribourg (HES-SO). He obtained his PhD at the University of Bern and did postdoctoral research at Concordia University and Polytechnique Montréal. His research covers deep learning, large vision and language models, graph-based pattern recognition, and document and handwriting analysis, including the automatic transcription of medieval manuscripts.

Recorded while he was a professor at the HES-SO in Fribourg and a lecturer at the University of Fribourg.

🤖 Bio generated by AI from public academic profiles and checked against them on 20 November 2026. Homepage · ORCID

Lecture Overview

Overview

Fischer presents text recognition as a key access technology for written cultural heritage. His lecture moves from the motivation for automatic reading of historical documents to the sequence of tasks involved: locating text on a page, recognizing it line by line, and making it searchable afterwards.

Main Points

  • Text recognition matters because digitized images alone are not enough; researchers also need machine-readable text for search, navigation, and large-scale access.
  • The workflow begins with layout analysis, continues with line-level recognition, and often ends with retrieval tasks such as keyword search.
  • Fischer distinguishes between full transcription, keyword spotting, and transcription alignment, and explains that they require different amounts of training data and different levels of difficulty.
  • Modern print OCR is close to solved, while historical handwriting remains much harder because scripts, layouts, and languages vary strongly across sources.
  • The lecture frames machine learning as the practical basis for current systems, but stresses that success depends on training material, document structure, and the target use case.

Examples Mentioned

  • Medieval manuscripts
  • Papyri and engraved inscriptions
  • Historical correspondence
  • Abbey Library material and other heritage collections

Source transcript: intro-lecture/_src/transcripts/Course 1_Fischer_TextRecognition.txt

Further Reading

See Zotero collection for 5 selected publications by this lecturer.