Persian Arabic Textline OCR: 100K-1M Multilingual Image Samples

Persian_Arabic_TextLine_Image_Ocr_Small contains between 100,000 and 1,000,000 text-line images for Optical Character Recognition tasks, published by mohajesmaeili in June 2025. The collection covers multiple languages using the Arabic script, including Persian, Arabic, Urdu, and Pashto. It is delivered in Parquet format for compatibility with high-performance data processing libraries.

Use Cases

• Training OCR engines to transcribe Persian text-line images
• Benchmarking multilingual script recognition for Urdu and Pashto
• Developing text-to-image alignment models for Arabic-based scripts

Strengths

• Scale of 100,000 to 1,000,000 records
• Multilingual support for four Arabic-script languages (fa, ar, ur, ps)
• Optimized for Polars and Dask via Parquet format

Limitations

• Undocumented annotation methodology and source material
• Unknown license status may restrict commercial application
• Small designation suggests this is a subset of a larger, unspecified corpus