metadata

language:
  - ar
configs:
  - config_name: default
    data_files:
      - split: Amiri
        path: Amiri/*.csv
      - split: Sakkal_Majalla
        path: Sakkal_Majalla/*.csv
      - split: Arial
        path: Arial/*.csv
      - split: Calibri
        path: Calibri/*.csv
      - split: Scheherazade_New
        path: Scheherazade_New/*.csv
features:
  text:
    dtype: string
csv_options:
  delimiter: ','
  quotechar: '"'
  encoding: utf-8
tags:
  - dataset
  - OCR
  - Arabic
  - Image_To_Text
license: apache-2.0
task_categories:
  - image-to-text
pretty_name: 'SAND: A Large-Scale Synthetic Arabic OCR Corpus for Vision-Language Models'
size_categories:
  - 100K<n<1M

SARD: Synthetic Arabic Recognition Dataset

Overview

SARD (Synthetic Arabic Recognition Dataset) is a large-scale, synthetically generated dataset designed for training and evaluating Optical Character Recognition (OCR) models for Arabic text. This dataset addresses the critical need for comprehensive Arabic text recognition resources by providing controlled, diverse, and scalable training data that simulates real-world book layouts.

Key Features

Massive Scale: 743,000 document images containing 662.15 million words
Typographic Diversity: Five distinct Arabic fonts (Amiri, Sakkal Majalla, Arial, Calibri, and Scheherazade New)
Structured Formatting: Designed to mimic real-world book layouts with consistent typography
Clean Data: Synthetically generated with no scanning artifacts, blur, or distortions
Content Diversity: Text spans multiple domains including culture, literature, Shariah, social topics, and more

Dataset Structure

The dataset is divided into five splits based on font name:

Amiri: ~148,541 document images
Sakkal Majalla: ~148,541 document images
Arial: ~148,541 document images
Calibri: ~148,541 document images
Scheherazade New: ~148,541 document images

📋 Sample Images

Each split contains data specific to a single font with the following attributes:

image_name: Unique identifier for each image
chunk: The text content associated with the image
font_name: The font used in text rendering
image_base64: Base64-encoded image representation

Content Distribution

Category	Number of Articles
Culture	13,253
Fatawa & Counsels	8,096
Literature & Language	11,581
Bibliography	26,393
Publications & Competitions	1,123
Shariah	46,665
Social	8,827
Translations	443
Muslim's News	16,725
Total Articles	133,105

Font Specifications

Font	Words Per Page	Font Size
Sakkal Majalla	50–300	14 pt
Arial	50–500	12 pt
Calibri	50–500	12 pt
Amiri	50–300	12 pt
Scheherazade	50–250	12 pt

Page Layout

Specification	Measurement
Left Margin	0.9 inches
Right Margin	0.9 inches
Top Margin	1.0 inch
Bottom Margin	1.0 inch
Gutter Margin	0.2 inches
Page Width	8.27 inches (A4)
Page Height	11.69 inches (A4)

Usage Example

from datasets import load_dataset
import base64
from io import BytesIO
from PIL import Image
import matplotlib.pyplot as plt

# Load dataset with streaming enabled
ds = load_dataset("riotu-lab/SARD", streaming=True)
print(ds)

# Iterate over a specific font dataset (e.g., Amiri)
for sample in ds["Amiri"]:
    image_name = sample["image_name"]
    chunk = sample["chunk"]  # Arabic text transcription
    font_name = sample["font_name"]
    
    # Decode Base64 image
    image_data = base64.b64decode(sample["image_base64"])
    image = Image.open(BytesIO(image_data))

    # Display the image
    plt.figure(figsize=(10, 10))
    plt.imshow(image)
    plt.axis('off')
    plt.title(f"Font: {font_name}")
    plt.show()

    # Print the details
    print(f"Image Name: {image_name}")
    print(f"Font Name: {font_name}")
    print(f"Text Chunk: {chunk}")
    
    # Break after one sample for testing
    break

Applications

SAND is designed to support various Arabic text recognition tasks:

Training and evaluating OCR models for Arabic text
Developing vision-language models for document understanding
Fine-tuning existing OCR models for better Arabic script recognition
Benchmarking OCR performance across different fonts and layouts
Research in Arabic natural language processing and computer vision

Acknowledgments

The authors thank Prince Sultan University for their support in developing this dataset.