همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

An Extended Table Detection Framework for Enhanced Recognition of Persian Tables in OCR Systems

نویسندگان :

Aras Valizadeh¹ Bita Baroutian² Fateme Taraghi³ Zahra Mohammadi⁴ Atefe Aghaei⁵ Mohsen Ebrahimi Moghaddam⁶

1- Shahid Beheshti University 2- Shahid Beheshti University 3- Shahid Beheshti University 4- Shahid Beheshti University 5- Shahid Beheshti University 6- Shahid Beheshti University

کلمات کلیدی :

Persian OCR،table detection،document layout analysis،vision–language models (VLMs)،structural recognition

چکیده :

Currently, with the widespread usage of automated document processing and retrieval systems, the importance of accurate OCR has increased significantly. However, robust OCR for Persian documents remains a challenging task, particularly in the presence of complex table layouts that exacerbate the intrinsic difficulties of the Persian script. Existing OCR and VLM-based methods often face limitations in reliably identifying and interpreting tables. These approaches are further hindered by structural hallucinations that compromise downstream reasoning and extraction. In response to these challenges, a modular, OCR-agnostic framework has been proposed. This has been specifically designed to improve table recognition in Persian documents. The framework is designed to integrate with existing OCR systems, enhancing the robustness of table detection across a wide range of real-world layouts. A dedicated dataset of Persian tables was gathered from different sources to facilitate systematic evaluation. The experimental results demonstrate that the proposed approach achieves superior table recognition performance in comparison to state-of-the-art OCR and VLM methods. This establishes a strong baseline for future research in Persian table analysis.