Persian Pixel: A large-scale synthetic OCR dataset for Persian language

2026-07-22 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, quick

Summary

Persian Pixel is a new, large-scale synthetic OCR dataset designed to overcome challenges in Persian language optical character recognition, a field less developed than Latin-script OCR. Persian script's complexity, including cursive connectivity, context-dependent glyphs, and extensive ligatures, combined with a scarcity of annotated data, has hindered progress. This dataset comprises over 343,000 high-fidelity image-text pairs, covering sentence, paragraph, and full-page layouts. It was generated from a seven-million-word Persian corpus using the SynthOCR-Gen framework, accurately modeling typographic features like contextual character joining and diacritic placement. To enhance realism, over twenty-five stochastic degradation models simulate real-world document artifacts such as ink bleed, blur, and scanner imperfections. Persian Pixel provides an openly available resource for training and fine-tuning modern OCR models like TrOCR and Donut, supporting research in Persian document analysis and digitization.

Key takeaway

For Machine Learning Engineers developing OCR solutions for low-resource or complex scripts, Persian Pixel demonstrates a viable path forward. You should consider synthetic data generation as a primary strategy to overcome data scarcity and annotation costs. This approach allows you to train robust models like TrOCR and Donut without extensive manual labeling, accelerating progress in areas like document digitization for languages with intricate writing systems.

Key insights

Synthetic data generation, exemplified by Persian Pixel, offers a scalable solution for OCR in typographically complex, low-resource languages.

Principles

• Complex scripts require specialized typographic modeling.
• Synthetic data can bridge real-world domain gaps.
• Costly manual annotation can be replaced.

Method

Generate image-text pairs from a large corpus using a rendering framework, modeling script typography. Apply stochastic degradation models to simulate real-world document acquisition artifacts.

In practice

• Train/fine-tune TrOCR and Donut models.
• Advance Persian document analysis.
• Digitize historical Persian manuscripts.

Topics

• Optical Character Recognition
• Persian Language
• Synthetic Data Generation
• Low-Resource Languages
• Document Digitization
• OCR Datasets

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Persian Pixel: A large-scale synthetic OCR dataset for Persian language — Synthetic data generation is a scalable, cost-effective solution for OCR in low-resource, complex scripts like Persian.
• KazakhOCR: A Synthetic Benchmark for Evaluating Multimodal Models in Low-Resource Kazakh Script OCR — Current MLLMs struggle with OCR and language identification for low-resource Kazakh scripts, especially Arabic, performing worse than traditional OCR.
• Why Nepali OCR Is Brutally Hard — And How We Finally Solved It — A hybrid deep learning pipeline overcomes complex script challenges to automate document data extraction and form pre-filling.
• LV-ROVER-MLT: Low-Resource Maltese OCR by Synthetic Fine-Tuning and Multi-Stream Arbitration — Synthetic data and multi-stream ensembles significantly improve low-resource OCR, especially for diacritic-rich languages.
• Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods — Recognizing highly customized WordArt demands specialized datasets and models beyond general Scene Text Recognition.
• Archivists Turn to LLMs to Decipher Handwriting at Scale — General-purpose LLMs now reliably transcribe historical handwriting, outperforming specialized software in speed, cost, and accuracy.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.