Local OCR Image Text Recognition

Problem

Text in screenshots, scanned documents, and form images is often hard to copy, search, or enter into structured data. A local OCR path is useful when images contain sensitive content and should not be sent to a cloud service, and when an API key should not be required. This skill provides a local OCR entry point: it uses Tesseract.js to recognize Chinese and English text in images and returns plain text or JSON.

How it works

• Local execution: recognition relies on local Tesseract.js and does not require an API key.
• Language selection: --lang selects language codes. The default is chi_sim+eng; supported codes include chi_sim, chi_tra, and eng, which can be combined with +.
• Output format: plain text is returned by default; --json produces JSON for downstream parsing.
• Caching: the first run downloads roughly 20MB of language data per language, then caches it locally to reduce repeated downloads.

Limits

It works best with clear, high-contrast printed text such as web screenshots, digital document scans, and table images. Accuracy may vary for blurry images, complex backgrounds, low-resolution captures, or handwritten text. Use chi_tra explicitly for Traditional Chinese, or eng when English-only recognition is enough.