OmarMDiab/Egyptian-Handwriting-Dataset

Egyptian Handwriting Dataset

A dataset of 11k+ handwritten Arabic words from Egyptian writers, extracted and tightly cropped from scanned paper forms. This dataset offers diverse handwriting samples ranging from children to elderly contributors, making it ideal for training robust Arabic handwriting recognition models.

• Each form contains 6 unique words, resulting in 24 handwritten word images per form.
• Each word is written four times by the same writer to capture writer-variability.
• The dataset includes diverse handwriting styles, primarily featuring Naskh and Ruqaʿa (Riqʿa).

📚 Source of Words

To ensure high-quality and culturally appropriate vocabulary:

• Holy Quran (Non-Vocalized Frequency List)

  • Diacritics (shadda, fatha, damma, kasra, sukun) were removed to simplify the model input and reduce variability.
  • Only unique words were retained for vocabulary clarity and focus.

• Arabeyes List

  • A manually curated clean dataset of Modern Standard Arabic words.
  • Introduces non-religious vocabulary for better generalization to everyday language.

👥 Writer Demographics

• EGY SET 1: 500 handwritten forms

  • Total Writers: 89 Egyptian contributors
  • Total Word Images: 11,200+
  • Age Distribution:
    • Children (<13): 7 writers (youngest: 6 years old)
    • Adults (≥13): 82 writers (oldest: 73 years old)

• EGY SET 2: "To be uploaded"

This age diversity introduces valuable natural variation in stroke width, alignment, slant, and writing pressure — all essential for improving generalization in generative handwriting models.

• Dataset extraction and contribution are ongoing — more words and writers will be added in future releases.

🤝 Acknowledgments

Special thanks to Fotopia Technologies
for sponsoring the scanning and digitization process of the dataset.

With deep gratitude to Dr. Ahmed Ibrahim,
Director of AI R&D at Fotopia,
for his invaluable technical and logistical support during the data creation and collection phases.

📝 Citation

If you use this dataset in your research or applications, please cite it as:

@dataset{egydataset2025,
  author       = {Omar Diab},
  title        = {Egyptian Handwriting Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/OmarMDiab/egyptian-handwriting},
  note         = {Dataset published on Hugging Face Hub}
}