This repository has data and code for the paper: Muharaf: Manuscripts of handwritten Arabic Dataset for cursive text recognition.
Mehreen Saeed, Adrian Chan, Anupam Mijar, Joseph Moukarzel, Georges Habchi, Carlos Younes, Amin Elias, Chau-Wai Wong, Akram Khater
-
Check out our latest model trained on 3382 images and its demo on 🤗HuggingFace. More details to follow soon.
-
🤗 Now on HuggingFace! Muharaf-public line images
-
Published at NeurIPS 2024
The dataset has two parts.
- The public part of Muharaf has 1,216 images and can be downloaded from Zenodo. We distribute this dataset under the CC BY-NC-SA 4.0.
- The restricted part of Muharaf has 428 images distributed under a proprietary license. It can be downloaded by writing to Carlos Younes at Phoenix Center for Lebanese Studies, USEK. This part of the dataset is distributed under a proprietary license with the condition that it will not be re-distributed and only used for research purposes.
ScribeArabic is the annotation software used to create the Muharaf Dataset. Read its manual to get an idea of how it works.
The ScribeArabic software natively uses JSON files. You can conver them to PAGE-XML using the source code provdided in the page-xml folder.
All image files accompanied by their corresponding _annotate files containing annotations, transcriptions and tagging can be viewed using the XML viewer in the page-xml converter repository.
The HTR results reported in the paper can be reproduced using the Start, Follow, Read — Arabic code System. This code is adapted from Start, Follow, Read System and its Python3 version.
We thank Stephen Randall Filios from Family Search for initiating discussions and providing feedback on tagging page elements in document images. We thank Elham Abdallah who is the Assistant University Librarian at USEK for providing support and coordinating work between NC State and USEK.
We acknowledge the computing resources provided by North Carolina State University High- Performance Computing Services Core Facility (RRID:SCR_022168). We also thank Andrew Petersen for his assistance and technical guidance on running jobs on the HPC.
This work was supported in part by the National Endowment for the Humanities (FAIN: ZPA-283823- 22), Family Search, and the ECE Undergraduate Research Program at NC State.
@inproceedings{
saeed2024muharaf,
title={Muharaf: Manuscripts of Handwritten Arabic Dataset for Cursive Text Recognition},
author={Mehreen Saeed and Adrian Chan and Anupam Mijar and Joseph Moukarzel and Gerges Habchi and Carlos Younes and Amin Elias and Chau-Wai Wong and Akram Khater},
booktitle={The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
year={2024},
url={https://openreview.net/forum?id=1s8l1tnTXW}
}