190 MB
6,679 files
Updated about 1 month ago
Name
Size
data
.gitattributes2.31 kB
xet
README.md4.47 kB
xet
config_files.tar.bz2396 kB
xet
README.md

KHATT_v1.0 - line level

Table of Contents

Dataset Description

Dataset Summary

KHATT (KFUPM Handwritten Arabic TexT) database is a database of unconstrained handwritten Arabic Text written by 1000 different writers. This research database’s development was undertaken by a research group from KFUPM, Dhahran, S audi Arabia headed by Professor Sabri Mahmoud in collaboration with Professor Fink from TU-Dortmund, Germany and Dr. Märgner from TU-Braunschweig, Germany.

The database includes 2000 similar-text paragraph images and 2000 unique-text paragraph images and their extracted text line images. The images are accompanied with manually verified ground-truth and Latin representation of the ground-truth. The database can be used in various handwriting recognition related researches like, but not limited to, text recognition, and writer identification. Interested readers can refer to the paper [1], and [2] for more details on the database. The version 1.0 of the KHATT database is available free of charge (for academic and research purposes) to the researchers.

Database Overview:

  • Forms written by 1000 different writers.
  • Scanned at different resolutions (200, 300, and 600 DPIs).
  • Writers are from different countries, gender, age groups, handedness and education level.
  • Natural writings with unrestricted writing styles.
  • 2000 unique paragraph images and their segmented line images (source text from different topics like arts, education, health, nature, technology).
  • 2000 paragraph images containing similar text, each covering all Arabic characters and shapes and their segmented line images.
  • Free paragraphs written by writers on any topic of their choice.
  • Paragraph and line images are supplied with manually verified ground-truths.
  • The database divided into three disjoint sets viz. training (70%), validation (15%), and testing (15%).
  • Promote research in areas like writer identification, line segmentation, and binarization and noise removal techniques beside handwritten text recognition.

For futher information about the database go through:

[1] Sabri A. Mahmoud, Irfan Ahmad, Wasfi G. Al-Khatib, Mohammad Alshayeb, Mohammad Tanvir Parvez, Volker Märgner, Gernot A. Fink, KHATT: an open Arabic offline handwritten text database , Pattern Recognition.[http://www.sciencedirect.com/science/article/pii/S0031320313003300]

[2] Sabri A. Mahmoud, Irfan Ahmad, Mohammed Alshayeb, Wasfi G. Al-Khatib, Mohammad Tanvir Parvez, Gernot A. Fink, Volker Margner, Haikal El Abed, KHATT: Arabic offline handwritten text database, 13th International Conference on Frontiers in Handwriting Recognition (ICFHR), pp. 447–452, 2012. [Best Poster Award Winner] [http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6424434&tag=1]

Languages

All the documents in the dataset are written in Arabic.

Dataset Structure

Data Instances

{
  'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=4300x128 at 0x1A800E8E190,
  'text': 'رفاظ قيار يؤل نب فوؤر هبحصب ماغرض رفظم حون بهذ'
}

Data Fields

  • image: a PIL.Image.Image object containing the image. Note that when accessing the image column (using dataset[0]["image"]), the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0].
  • text: the label transcription of the image. The text was intentionally flipped from RTL to LTR because of PyLaia library limitation to LTR.
Total size
190 MB
Files
6,679
Last updated
Oct 22
Pre-warmed CDN
US EU US EU

Contributors