Sigurdur/timarit-images

Icelandic Historical Documents - OCR Dataset

Dataset Description

Dataset Summary

This dataset contains images from Tímarit.is, the National and University Library of Iceland's digital archive of historic newspapers and periodicals. The images are sourced from historical Icelandic publications and are intended for training Optical Character Recognition (OCR) models on Icelandic text.

The dataset includes high-quality scans of historical documents featuring Icelandic text in various fonts, layouts, and conditions, making it valuable for developing robust OCR systems for the Icelandic language.

Supported Tasks and Leaderboards

• Optical Character Recognition (OCR): The primary task is to extract Icelandic text from historical document images
• Document Layout Analysis: Understanding the structure of historical Icelandic publications
• Historical Text Recognition: Processing degraded or aged text from archival materials

Languages

The dataset contains text in Icelandic (is).

Dataset Structure

Data Instances

Each instance in the dataset contains:

{
  'image': <PIL.Image>,  # The image file
  'image_id': '74227141',  # Unique identifier from timarit.is
  'image_url': 'https://timarit.is/files/74227141'  # Source URL
}

Data Fields

• image: A PIL Image object containing the scanned document page
• image_id: String identifier (8-digit zero-padded) corresponding to the image's ID in the Tímarit.is archive
• image_url: Direct URL to access the image on timarit.is

Data Splits

The dataset is provided as a single collection without predefined splits. Users should create their own train/validation/test splits as needed for their specific use case.

Dataset Creation

Curation Rationale

This dataset was created to support the development of OCR models specifically trained on Icelandic text. Historical documents present unique challenges for OCR systems:

• Varied typography and fonts from different time periods
• Degraded image quality due to aging
• Historical spelling conventions and vocabulary
• Complex layouts with multiple columns, illustrations, and mixed content

By training on historical Icelandic documents, models can better handle real-world archival materials and contribute to the digitization efforts of Iceland's cultural heritage.

Source Data

Initial Data Collection and Normalization

Images are sourced from Tímarit.is, which is maintained by the National and University Library of Iceland (Landsbókasafn Íslands – Háskólabókasafn).

The collection process:

1. Images are fetched from the Tímarit.is archive using their public file URLs
2. Each image undergoes validation to ensure:
   • Minimum file size of 1KB
   • Minimum dimensions of 500x500 pixels
   • Valid image format and integrity
3. Only images meeting quality thresholds are included

Who are the source language producers?

The text in these documents was originally produced by various Icelandic publishers, journalists, and authors whose work appeared in historical newspapers and periodicals in Iceland.

Annotations

This dataset does not include ground truth text annotations or transcriptions. It provides raw images suitable for:

• Semi-supervised learning approaches
• Pre-training vision models
• Creating custom annotations for specific research needs
• Testing OCR systems on real-world historical documents

Users will need to add their own annotations/labels if training supervised OCR models.

Personal and Sensitive Information

The images contain historical published materials that were publicly distributed in Iceland. As archival materials, they may contain:

• Historical names and references to public figures
• Dated social and cultural perspectives
• Historical advertisements and notices

Users should review the content and apply appropriate ethical considerations when using the data.

Considerations for Using the Data

Social Impact of Dataset

This dataset supports the preservation and accessibility of Icelandic cultural heritage by enabling:

• Improved digital access to historical Icelandic documents
• Better search and discovery of archival materials
• Support for research in Icelandic linguistics, history, and culture
• Development of NLP tools for under-resourced languages

Discussion of Biases

As historical documents, the materials reflect the perspectives, language usage, and social norms of their time period. Users should be aware that:

• Content may reflect historical biases and outdated viewpoints
• Language and spelling conventions may differ from modern Icelandic
• Coverage may be limited to materials that were published and preserved
• Document quality varies based on original printing quality and preservation conditions

Other Known Limitations

• No text transcriptions or ground truth labels are provided
• Image quality varies depending on the condition of original materials
• Coverage is limited to materials available in the Tímarit.is archive
• Dataset size is relatively small and may need augmentation for robust model training

Additional Information

Dataset Curators

This dataset was curated from materials provided by the National and University Library of Iceland through their Tímarit.is platform.

Licensing Information

The images are sourced from Tímarit.is and are used in accordance with their terms of service. Most materials in Tímarit.is are in the public domain or available under open licenses.

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

When using this dataset, please:

• Provide appropriate attribution to Tímarit.is and the National and University Library of Iceland
• Include a link to the original source materials when possible
• Respect any additional usage terms specified by Tímarit.is

Citation Information

If you use this dataset in your research, please cite:

@dataset{icelandic_ocr_timarit,
  title={Icelandic Historical Documents - OCR Dataset},
  author={Timarit.is and National and University Library of Iceland},
  year={2026},
  publisher={Hugging Face Datasets},
  howpublished={\url{https://huggingface.co/datasets/Sigurdur/timarit-images}}
}

And please also acknowledge the source:

Images sourced from Tímarit.is - The National and University Library of Iceland
https://timarit.is

Contributions

To contribute to this dataset or report issues, please contact the dataset maintainer or visit the dataset repository.

For questions about the original source materials or Tímarit.is, visit: https://timarit.is

---

Usage Example

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Sigurdur/timarit-images")

# Access an example
example = dataset[0]
image = example['image']
image_id = example['image_id']
source_url = example['image_url']

# Display the image
image.show()
print(f"Image ID: {image_id}")
print(f"Source URL: {source_url}")

Technical Details

Image Specifications

• Format: JPEG
• Minimum dimensions: 500x500 pixels
• Minimum file size: 1KB
• Color mode: Varies (color and grayscale)
• Resolution: Varies based on original scan quality

Data Collection Script

The dataset was collected using the prepare_data.py script included in this repository, which:

• Systematically scans for valid image IDs in the Tímarit.is archive
• Downloads and validates images based on quality criteria
• Packages images into a Hugging Face Dataset format
• Includes rate limiting to respect the source server