Multilingual table detection improves using masked images to ignore script details
EMBLEM: Enhancing Multi-script Table Detection through Masking
Machine Learning
Summary
Detecting tables inside documents is important for many tasks like searching and understanding content. Existing technology works well for English and Chinese but struggles with documents in many languages because scripts and fonts vary so much. To solve this, the authors created a new dataset with documents in 18 languages and 15 scripts, then developed a method called EMBLEM that hides script-specific details in images. This helps models trained on English to focus on the layout rather than the script, improving detection on multilingual documents by a big margin. Their approach also performs well on standard English benchmarks without needing extra multilingual training data.
table detectiondocument analysismulti-script documentsmasked imagesmachine learningdatasetlayout analysisF1 scorepretrainingmultilingual OCR
Authors
Dhruv Kudale, Udhay Brahmi, Ganesh Ramakrishnan
Abstract
Table detection is a core task in document analysis, supporting downstream applications such as information retrieval, document reconstruction, and visual question answering. While existing deep learning models perform well on English and Chinese documents, they struggle with multilingual, multi-script documents due to script diversity and the limited availability of labeled data. To address this challenge, we introduce MANDALA (Multi-script Annotated Documents for Table Detection), a manually curated dataset of 2,323 table-containing pages spanning 18 languages and 15 scripts across diverse domains. We also propose EMBLEM, a masking-based paradigm for Multi-script Table Detection (MTD). EMBLEM generates masked images that conceal script- and font-specific details, enabling models pre-trained on abundant English documents to focus on script-agnostic page layout. Experiments across three table detection architectures show that EMBLEM consistently outperforms strong baselines on MANDALA while remaining competitive on five standard English-dominant benchmarks. Using only English masked images for fine-tuning, with no multi-script training data, EMBLEM achieves an absolute F1-score gain of 20.8% on MANDALA. We release MANDALA along with the accompanying code and models at https://github.com/IITB-LEAP-OCR/EMBLEM.git.