Smart Image-to-Text OCR
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.
Related Skills
Convert teaching schedules from talent development plan PDFs into structured Excel files, with cross-page tables, merged cells, multi-semester course splitting, major metadata extraction, and batch output.
Turn post-meeting text into bilingual minutes, preserving structure and consistent terminology.
Interprets tender files, generates bid docs, and checks compliance via the Bailian API.
Create, repair, and recalculate xlsx, xlsm, csv, or tsv workbooks with formula-safe edits, validation, and recalculation.