Skip to content
Preprint

When Do VLMs Help Arabic Manuscript OCR? A Cross-Dataset Study

Aug 2026 · 0 citations · 38 references
Computer Science

TL;DR

Traditional OCR, general-purpose VLMs, Arabic-specialized VLMs, and OCR-conditioned VLM correction are evaluated across eight Arabic text datasets spanning historical manuscripts, aged printed books, clean print, multi-domain documents, and handwriting.

Abstract

Vision-language models (VLMs) are increasingly being used for document understanding, yet their role in Arabic and Islamic manuscript recognition remains underexplored. To address such a gap in this paper, we evaluate traditional OCR, general-purpose VLMs, Arabic-specialized VLMs, and OCR-conditioned VLM correction across eight Arabic text datasets spanning historical manuscripts, aged printed books, clean print, multi-domain documents, and handwriting. The results show that no single approach dominates across setups. On line-level historical manuscripts, VLMs are close to Tesseract; on page-level manuscript images, they perform better; and in several settings, an OCR-conditioned corrector improves over both standalone OCR and standalone VLMs. The central finding is an OCR-prior recoverability principle: OCR conditioning helps when the OCR output remains visually and textually recoverable, providing anchors that the VLM can refine against the image. It improves recognition on aged print, clean print, mixed-domain Arabic, and some Naskh manuscripts, but degrades performance when the prior is script-mismatched or systematically misleading, as in Maghribi manuscripts and realistic student handwriting. Additional diagnostics show that Arabic VLM-OCR is sensitive to diacritics, preprocessing, generation budget, and repetition loops. These findings support an adaptive OCR-VLM workflow that routes pages according to script, OCR-prior recoverability, length diagnostics, and failure-mode indicators.

View source

Similar papers

Preprint Oct 2026

MuSTEL: A Large-Scale Multilingual Synthetic Text-Line Image Dataset for OCR

Optical character recognition (OCR) systems depend strongly on the scale and diversity of their training data. One approach to document OCR first analyzes the page layout and extracts text-line images, which are then transcribed by a recognition model. Multilingual line-level recognition involving Persian, Arabic, and...

Majid Adibian, Hossein Saberi, Hossein Zeinali · 0 citations
Open access Oct 2026

AI‑based multilingual OCR and OCR-enabled collection diversity analytics in libraries

The digitization of multilingual library collections requires text recognition that copes with diverse scripts, writing styles and document structures. Optical Character Recognition (OCR) remains central to this process, but recognition of cursive scripts such as Arabic and Urdu is still difficult. This paper presents...

Yi-Jun Chen, Ling-Hong Wu, Chun-Ping Zhang et al. · 0 citations
Review Open access Sep 2026

The Optimal Model for Arabic Optical Character Recognition(OCR) Text A Comprehensive Study

Optical Character Recognition (OCR) technology has evolved significantly over the past few decades, enabling the conversion of printed or handwritten text into machine-encoded text. While OCR systems for languages like English have reached a high level of accuracy, Arabic OCR remains a challenging task due to the langu...

Sabri Mansour, Nasser Telesi · 0 citations
Open access 2026

A Dataset of Wolof Ajami Manuscripts for HTR and OCR

The first ever dataset of manually segmented and transcribed Ajami manuscripts written in Wolof is presented, and several Arabic-script recognition models intended for historical manuscripts are evaluated and it is found they produce character error rates (CER) of 61–81%.

O. Yousuf, Elhadji Djibril Diagne, Christian Høgel et al. · 0 citations
Conference Sep 2026

SwinAraBERT-Fusion: A Multimodal Transformer Architecture for Robust Arabic OCR and Handwriting Recognition with Diacritic-Aware Contextual Modeling

Arabic Optical Character Recognition (OCR) and Handwriting Recognition (HWR) remain persistently challenging due to the script's inherently cursive nature, context-dependent positional letter forms, dense diacritic system (tashkeel), and high inter-class visual similarity among characters. While transformer-based archi...

Sarra Eltayeb, O. Khalifa, A. H. Hashim · 0 citations
#artificial intelligence Preprint Sep 2026

How Far Can Synthetic Data Take Thai OCR?

It is found that synthetic-only training can be competitive, and the 0.9B-parameter PaddleOCR-VL-1.6 is adapted into Wayu-Paxa-OCR-Zero, a Thai OCR model adapted without OCR labels from real Thai document pages, showing that synthetic-only training can be competitive.

Kunat Pipatanakul · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.