---
license: apache-2.0
---
# TrOCR Base Fine-Tuned for Czech Historical Vital Records

This is a fine-tuned TrOCR-Base model (`microsoft/trocr-base-handwritten`) specializing in Handwritten Text Recognition of 19th-century Czech vital records (birth, marriage, death registers).

The model was trained as part of the Master's Thesis: **"Automated Transcription and Search in Historical Records Using Handwritten Text Recognition"**.

It was developed on an original, manually annotated dataset of historical Czech scripts and is designed to be used inside the full historical document processing pipeline (layout analysis → text detection → recognition → post-processing).

For detailed performance metrics, evaluation, and the full pipeline description, please refer to the thesis text.

## Citation

```
@misc{palkovic2025htr,
      AUTHOR = {Palkovič, Radoslav},
      TITLE = {Automated Transcription and Search in Historical Records Using Handwritten Text Recognition},
      YEAR = {2025},
      TYPE = {Master Thesis},
      INSTITUTION = {Masaryk University, Faculty of Informatics},
      LOCATION = {Brno},
      SUPERVISOR = {Michal Batko}
}
```