Arabic handwriting OCR data
Arabic as people actually write it: quick notes, filled-in forms, letters and exam answers, in naskh and ruqʿa hands, joined and broken, with and without dots and diacritics. Every line is boxed and transcribed exactly as written.
Specification
- Writing
- Notes, forms, letters, exam answers and dictations, by many writers of all ages
- Hands
- Naskh and ruqʿa styles, neat and hurried, on lined, plain and squared paper
- Images
- Scans and phone photos at original resolution, JPEG or PNG
- Annotation
- Line and word boxes, exact transcription with the writer’s own spelling, optional diacritics
- Privacy
- Names and personal details replaced before delivery
- Quality check
- Several native reviewers read every line
Example of the data
- Line 1
- الاجتماع يوم الأحد الساعة ٩
- Line 2
- إحضار التقرير المالي
- Style
- ruqʿa, hurried
What it is used for
Handwriting recognition
Train models that read real Arabic handwriting, not only print.
Form digitisation
Read handwritten forms in banks, clinics and government offices.
Education
Read and mark handwritten answers in Arabic.
Archives
Make handwritten records searchable.
Questions
Is the transcription corrected?
No. Lines are transcribed exactly as written, including the writer’s spelling, with a corrected version added on request.
Are diacritics included?
When the writer used them, yes; a fully diacritised transcription can be added on request.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Related datasets
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples