AI Agent Hub
Back to skills
Smart Image-to-Text OCR icon

Smart Image-to-Text OCR

Office Efficiency Updated 2026.08.30

Paste the following prompt into your AI chat to install this skill:

Please follow https://skillhub.cn/install/skillhub.md and install @user_8a153b62/smart-ocr.

About this skill

Problem

When .doc, .docx, or .pdf files are mostly scanned pages, generic OCR often produces look-alike characters, stray spaces in mixed Chinese and English text, garbled fragments, and page-number noise. This skill is built around a practical cleaning chain for image-to-text extraction rather than a single basic OCR call.

Workflow and Limits

It first converts documents to images, then uses ocr_pipeline.py to run Tesseract multi-strategy recognition, apply common corrections from ocr_errors.json, and remove inter-CJK spaces. ocr_deepseek.py then calls DeepSeek for a 0-10 quality score and constrained correction; if DEEPSEEK_API_KEY is not set, the pipeline falls back to Tesseract-only output. The skill also skips obviously mosaiced, blurred, tiny, or textless images and generates a .docx file with ▎Image N labels. It fits scanned bilingual documents, image text extraction, and second-pass OCR proofreading. It is not a replacement for precise table reconstruction, complex layout parsing, or fully offline OCR without AI correction.

Use Cases

  • Convert scanned PDF pages into editable bilingual text and export a .docx.
  • Process a .docx with embedded contract images, extract clause text, and fix look-alike characters.
  • Score noisy Tesseract output and apply conservative DeepSeek corrections to garbled scans.
  • Skip mosaicked, tiny, or unreadable images and mark them in the final document.

Best For

  • Document specialists processing scanned contracts or reports need editable text with fewer character errors.
  • Technical writers maintaining knowledge bases need to extract bilingual paragraphs from multi-page scans.
  • Platform engineers comparing OCR quality need a pipeline that falls back to Tesseract without API access.
  • Archivists handling archive images need to skip mosaicked or invalid images while preserving filenames.