---
title: "HICMA — paper by Ismail et al. (2023) — summary and ideas"
description: "Arabic manuscripts and calligraphy are difficult to transcribe automatically. 17 references, cited by 1 paper here."
authors: ["Anis Ismail", "Zena Kamel", "Reem Mahmoud"]
year: 2023
venue: "Proceedings of ArabicNLP 2023"
source: "https://aclanthology.org/2023.arabicnlp-1.3/"
doi: "10.18653/v1/2023.arabicnlp-1.3"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/ismail-2023-hicma-handwriting-identification-calligraphy-manuscripts/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ismail-2023-hicma-handwriting-identification-calligraphy-manuscripts/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/) / HICMA: The Handwriting Identification for Calligraphy and Manuscripts in Arabic Dataset

# HICMA: The Handwriting Identification for Calligraphy and Manuscripts in Arabic Dataset

**Authors:** Anis Ismail, Zena Kamel and [Reem Mahmoud](https://md.research.elamur.ai/free/en/authors/reem-mahmoud/) · **Year:** 2023 · **Venue:** [Proceedings of ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/2023/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2023.arabicnlp-1.3/) · [PDF](https://aclanthology.org/2023.arabicnlp-1.3.pdf) · [DOI 10.18653/v1/2023.arabicnlp-1.3](https://doi.org/10.18653/v1/2023.arabicnlp-1.3) · [OpenAlex](https://openalex.org/W4389518363)

**Citations:** 6 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read

## Summary
Arabic manuscripts and calligraphy are difficult to transcribe automatically. The script is cursive, letter shapes change with position, and real artworks surround the text with ornaments, diacritics and textured backgrounds. Earlier resources either cleaned calligraphy into digital-pen drawings on white backgrounds, as in Calliar, or kept authentic images but labelled only the style. The authors built HICMA, a public dataset of 5,031 real images from three permission-cleared sources. Each image is paired with a sentence-level transcription and a style label covering Thuluth, Diwani, Muhaqaq, Naskh and Kufic. Two teams of trained volunteers transcribed the images, and each team checked the other's work. The data was then deduplicated, normalized and split 80/10/10. Three current Arabic OCR systems were evaluated with an open benchmarking tool using error-rate metrics. Accuracy was reasonable on Naskh but fell sharply on ornamental styles, which the authors link to training on synthetic font-rendered text. HICMA gives a first realistic baseline for full-sentence recognition of Arabic manuscripts and calligraphy.

## Key points
- HICMA has 5,031 images in five styles. Every image carries an image–text pair and a style label, and the data is released under CC BY-NC 4.0 at hicma.net with a GitHub benchmark (HICMA-benchmark).
- The images come from three sources:
  - 106 Qur'an pages of about 15 lines each;
  - 202 photographed pages from a private 17th-century Naskh collection;
  - a further source.
- Eleven volunteers worked in two teams on all three sources in turn, with each team validating the other's labels.
- Post-processing removed duplicates, and pyArabic stripped diacritics and punctuation before the 80%/10%/10% train/validation/test split.
- Evaluation used CER and WER, both built on Levenshtein distance via python-levenshtein; either can exceed 1 when there are many insertions. EasyOCR, trained on synthetic Amiri and Noto Sans Arabic text, does best on Naskh and degrades on the decorative styles.

## Interesting ideas
- Instead of a single annotator pass, two teams label disjoint halves and then audit each other, giving cheap quality control without full double annotation.
- Breaking errors down by style showed the model works where test data resembles its synthetic training fonts and fails elsewhere, so one aggregate score would hide the gap.
- The authors deliberately kept noisy, realistic context rather than sanitizing samples, because cleaned data overstates how well systems work in the field.
- Normalization (removing diacritics and punctuation) is applied consistently before scoring, and the metrics' ability to exceed 1 is stated explicitly, so readers interpret scores correctly.
- The dataset ships together with its scoring tool, so future work is measured identically.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Silver UD trees or frame labels from one witness parser or LLM tier could be audited by a different tier, and the disagreements kept as a repair-rule signal.
- **I2** (learn): Evaluation could be reported per genre or register (fables, school tests, maths problems) to expose where training data stops resembling the input.
- **I3** (tokens): The tokenizer and tagger test sets could include unnormalized text (odd punctuation, line breaks, OCR-like noise) alongside clean text to measure robustness honestly.
- **I4** (tokens): Lemma and morphology scoring should document exactly which normalizations apply before comparison, so that results stay comparable across versions.
- **I5** (answer): Each answer benchmark (QA on tales, word problems) could come with a frozen scorer, so improvements are not confounded by changes in evaluation.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [KU Leuven](https://md.research.elamur.ai/free/en/universities/ku-leuven/) — University, Leuven, Belgium · 42 works · 5,839 citations
- [Lebanese American University](https://md.research.elamur.ai/free/en/universities/lebanese-american-university/) — University, Beirut, Lebanon · 5 works · 37 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works · [⚙️ Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/) · 133 works · [🖱️ Human-Computer Interaction](https://md.research.elamur.ai/free/en/shelves/computer-science/human-computer-interaction/) · 141 works · [⚙️ Media Technology](https://md.research.elamur.ai/free/en/shelves/engineering/media-technology/) · 34 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Anis Ismail, Zena Kamel and Reem Mahmoud. 2023. HICMA: The Handwriting Identification for Calligraphy and Manuscripts in Arabic Dataset. Proceedings of ArabicNLP 2023.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{ismail2023hicma,
  title = {{HICMA: The Handwriting Identification for Calligraphy and Manuscripts in Arabic Dataset}},
  author = {Anis Ismail, Zena Kamel and Reem Mahmoud},
  booktitle = {Proceedings of ArabicNLP 2023},
  year = {2023},
  doi = {10.18653/v1/2023.arabicnlp-1.3},
  url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.3}
}
```

- [← previous GYM at Qur'an QA 2023 Shared Task: Multi-Task Transfer Learning for Quranic Passage Retrieval and Question Answering with Large Language Models](https://md.research.elamur.ai/free/en/papers/mahmoudi-2023-gym-qur-qa-2023-shared/)
- [next → Hierarchical Fusion for Online Multimodal Dialog Act Classification](https://md.research.elamur.ai/free/en/papers/miah-2023-hierarchical-fusion-online-multimodal-dialog/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/)

## References

17 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Volker Märgner and Haikal El Abed. 2009. Icdar 2009 arabic handwriting recognition competition. In 2009 10th International Conference on Document Analysis and Recognition, pages 1383–1387. IEEE. [↗](https://doi.org/10.1007/s10032-010-0117-5)
2. S Abdelazeem and E El-Sherif. 2017. The arabic handwritten digits databases: Adbase & madbase. [find](https://scholar.google.com/scholar?q=S+Abdelazeem+and+E+El%2DSherif%2E+2017%2E+The+arabic+handwritten+digits+databases%3A+Adbase+%26+madbase%2E)
3. Kalthoum Adam, Somaya Al-Maadeed, and Ahmed Bouridane. 2017. based classification of arabic scripts style in ancient arabic manuscripts: Preliminary results. In 2017 1st International Workshop on Arabic Script Analysis and Recognition (ASAR), pages 95–98. IEEE. [find](https://scholar.google.com/scholar?q=Kalthoum+Adam%2C+Somaya+Al%2DMaadeed%2C+and+Ahmed+Bouridane%2E+2017%2E+based+classification+of+arabic+scripts+style+in+ancient+arabic+manuscripts%3A+Preliminary+results%2E+In+2017+1st+International+Workshop+on+Arab)
4. Andrew Morris, Viktoria Maier, and Phil Green. 2004. From wer and ril to mer and wil: improved evaluation measures for connected speech recognition. [find](https://scholar.google.com/scholar?q=Andrew+Morris%2C+Viktoria+Maier%2C+and+Phil+Green%2E+2004%2E+From+wer+and+ril+to+mer+and+wil%3A+improved+evaluation+measures+for+connected+speech+recognition%2E)
5. Sourour Njah, Badreddine Ben Nouma, Hala Bezine, and Adel M Alimi. 2012. Mayastroun: A multilanguage handwriting database. In 2012 International Conference on Frontiers in Handwriting Recognition, pages 308–312. IEEE. [find](https://scholar.google.com/scholar?q=Sourour+Njah%2C+Badreddine+Ben+Nouma%2C+Hala+Bezine%2C+and+Adel+M+Alimi%2E+2012%2E+Mayastroun%3A+A+multilanguage+handwriting+database%2E+In+2012+International+Conference+on+Frontiers+in+Handwriting+Recognition%2C+pag)
6. Kalthoum Adam, Asim Baig, Somaya Al-Maadeed, Ahmed Bouridane, and Sherine El-Menshawy. 2018. Kertas: dataset for automatic dating of ancient arabic manuscripts. International Journal on Document Analysis and Recognition (IJDAR), 21(4):283–290. [↗](https://doi.org/10.1007/s10032-018-0312-3)
7. Seetah AL Salamah and Ross King. 2018. Towards the machine reading of arabic calligraphy: a letters dataset and corresponding corpus of text. In 2018 IEEE 2nd International Workshop on Arabic and Derived Script Analysis and Recognition (ASAR), pages 19–23. IEEE. [find](https://scholar.google.com/scholar?q=Seetah+AL+Salamah+and+Ross+King%2E+2018%2E+Towards+the+machine+reading+of+arabic+calligraphy%3A+a+letters+dataset+and+corresponding+corpus+of+text%2E+In+2018+IEEE+2nd+International+Workshop+on+Arabic+and+Deri)
8. SR Allaf and R Al-Hmouz. 2016. Automatic recognition of artistic arabic calligraphy types. Journal of King Abdulaziz University, 27(1):3–17. [↗](https://doi.org/10.4197/eng.27-1.1)
9. Bodour Alrehali, Najla Alsaedi, Hanan Alahmadi, and Nahla Abid. 2020. Historical arabic manuscripts text recognition using convolutional neural network. In 2020 6th Conference on Data Science and Machine Learning Applications (CDMA), pages 37–42. IEEE. [find](https://scholar.google.com/scholar?q=Bodour+Alrehali%2C+Najla+Alsaedi%2C+Hanan+Alahmadi%2C+and+Nahla+Abid%2E+2020%2E+Historical+arabic+manuscripts+text+recognition+using+convolutional+neural+network%2E+In+2020+6th+Conference+on+Data+Science+and+Mach)
10. Sandip Sarkar, Dipankar Das, Partha Pakray, and Alexander Gelbukh. 2016. JUNITMZ at SemEval-2016 Task 1: Identifying Semantic Similarity Using Levenshtein Ratio. In Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016), pages 702–705, San Diego, California. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Sandip+Sarkar%2C+Dipankar+Das%2C+Partha+Pakray%2C+and+Alexander+Gelbukh%2E+2016%2E+JUNITMZ+at+SemEval%2D2016+Task+1%3A+Identifying+Semantic+Similarity+Using+Levenshtein+Ratio%2E+In+Proceedings+of+the+10th+Internation)
11. Najwa Altwaijry and Isra Al-Turaiki. 2021. Arabic handwriting recognition system using convolutional neural network. Neural Computing and Applications, 33(7):2249–2261. [↗](https://doi.org/10.1007/s00521-020-05070-8)
12. Ciprian Tomoiaga, Paul Feng, Mathieu Salzmann, and Patrick Jayet. 2019. Field typing for improved recognition on heterogeneous handwritten forms. ArXiv:1909.10120 [cs]. [↗](https://arxiv.org/abs/1909.10120)
13. Zaid Alyafeai, Maged S Al-shaibani, Mustafa Ghaleb, and Yousif Ahmed Al-Wajih. 2021. [Calliar: An online handwritten dataset for arabic calligraphy](https://md.research.elamur.ai/free/en/papers/alyafeai-2021-calliar-online-handwritten-dataset-arabic/). arXiv preprint arXiv:2106.10745. [↗](https://arxiv.org/abs/2106.10745)
14. Zineb Kaoudja, Mohammed Lamine Kherfi, and Belal Khaldi. 2019. An efficient multiple-classifier system for arabic calligraphy style recognition. In 2019 International Conference on Networking and Advanced Systems (ICNAS), pages 1–5. IEEE. [find](https://scholar.google.com/scholar?q=Zineb+Kaoudja%2C+Mohammed+Lamine+Kherfi%2C+and+Belal+Khaldi%2E+2019%2E+An+efficient+multiple%2Dclassifier+system+for+arabic+calligraphy+style+recognition%2E+In+2019+International+Conference+on+Networking+and+Adva)
15. Manal M Khayyat and Lamiaa A Elrefaei. 2020. A deep learning based prediction of arabic manuscripts handwriting style. Int. Arab J. Inf. Technol., 17(5):702– 712. [↗](https://doi.org/10.34028/iajit/17/5/3)
16. Benjamin Kiessling. 2022. The Kraken OCR system. [find](https://scholar.google.com/scholar?q=Benjamin+Kiessling%2E+2022%2E+The+Kraken+OCR+system%2E)
17. V. I. Levenshtein. 1966. Binary Codes Capable of Correcting Deletions, Insertions and Reversals. Soviet Physics Doklady, 10:707. ADS Bibcode: 1966SPhD...10..707L. [find](https://scholar.google.com/scholar?q=V%2E+I%2E+Levenshtein%2E+1966%2E+Binary+Codes+Capable+of+Correcting+Deletions%2C+Insertions+and+Reversals%2E+Soviet+Physics+Doklady%2C+10%3A707%2E+ADS+Bibcode%3A+1966SPhD%2E%2E%2E10%2E%2E707L%2E)

*Updated 1 November 2026*