---
license: mit
language:
- ar
task_categories:
- image-to-text
pretty_name: KHATT_v1.0
dataset_info:
  features:
  - name: image
    dtype: image
  - name: text
    dtype: string
  splits:
  - name: train
    num_examples: 4672
  - name: validation
    num_examples: 963
  - name: test
    num_examples: 1038
  dataset_size: 220M
tags:
- atr
- htr
- ocr
- historical
- handwritten
- arabic
---

# KHATT_v1.0 - line level

## Table of Contents
- [KHATT_v1.0 - line level](#KHATT_v1.0_dataset)
  - [Table of Contents](#table-of-contents)
  - [Dataset Description](#dataset-description)
    - [Languages](#languages)
  - [Dataset Structure](#dataset-structure)
    - [Data Instances](#data-instances)
    - [Data Fields](#data-fields)

## Dataset Description

- **Homepage:** [johnlockejrr's personal project](https://huggingface.co/datasets/johnlockejrr/KHATT_v1.0_dataset)

## 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.