Datasets:

OttomanNLP
/

Akis-Ottoman-Dataset

license: apache-2.0
task_categories:
  - image-to-text
language:
  - ota
tags:
  - ottoman
  - ocr
  - htr
  - document-transcription
  - handwriting-recognition
  - ottoman-turkish
  - historical-documents
size_categories:
  - 1K<n<10K

Akis-Dataset

This repository provides the test dataset used in the paper "Automatic Transcription of Ottoman Documents Using Deep Learning". It contains line segment images of Ottoman documents along with their corresponding transcriptions.

Dataset Overview

The dataset contains 8,037 image–transcription pairs of Ottoman handwritten document line segments.

Format 1: HuggingFace Dataset (Parquet — Recommended)

The dataset is natively available as a Parquet-format HuggingFace Dataset with two columns:

Column	Type	Description
`image`	Image	Line segment image of an Ottoman handwritten document
`text`	String	Corresponding Ottoman Turkish transcription

You can load it directly with:

from datasets import load_dataset

dataset = load_dataset("OttomanNLP/Akis-Ottoman-Dataset")
print(dataset["test"][0])

Format 2: Raw Archive (ZIP)

The raw files are also available in akis_dataset_test.zip with the following structure:

  .
  ├── images/         # Directory containing the line segment images (.png).
  ├── annotations/    # Directory containing the corresponding transcriptions,
  │                   # with each image having a matching .txt file.
  └── files.txt       # File listing the image/transcription pairs. Each line contains
                      # the image file and its corresponding transcription file, separated by a space.

Citation

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

Full Reference:

Tasdemir, E.F.B. et al. (2024). Automatic Transcription of Ottoman Documents Using Deep Learning. In: Sfikas, G., Retsinas, G. (eds) Document Analysis Systems. DAS 2024. Lecture Notes in Computer Science, vol 14994. Springer, Cham. https://doi.org/10.1007/978-3-031-70442-0_26

BibTeX:

@InProceedings{tasdemir2024automatic,
  author    = {Tasdemir, Esma F. Bilgin and Tandogan, Zeynep and Akansu, S. Dogan
               and Kizilirmak, Firat and Sen, M. Umut and Akcan, Aysu
               and Kuru, Mehmet and Yanikoglu, Berrin},
  title     = {Automatic Transcription of Ottoman Documents Using Deep Learning},
  booktitle = {Document Analysis Systems},
  editor    = {Sfikas, G. and Retsinas, G.},
  series    = {Lecture Notes in Computer Science},
  volume    = {14994},
  publisher = {Springer, Cham},
  year      = {2024},
  doi       = {10.1007/978-3-031-70442-0_26},
  url       = {https://doi.org/10.1007/978-3-031-70442-0_26}
}

Feel free to reach out if you have any questions or require further information about the dataset.