PDF扫描件 OCR 转换技能 📄

配置

• 百度 OCR API Key: vOBOM7tO0lL8cKMJdZy453Ai
• 百度 OCR Secret Key: bib8MvDPTfXXdPz4JyzIyDCvCeKxtpyu
• 免费额度:1000次/月（1次=1页），592页以内一次免费跑完
• 接口: 通用文字识别（高精度版）accurate_basic

依赖安装

pip install pymupdf python-docx pillow

使用方法

python3 {baseDir}/scripts/pdf_to_docx.py <PDF路径> [输出目录]

输出文件在[输出目录]/xxx_全文_ocr.docx，文件较大时用脚本压缩图片：

python3 {baseDir}/scripts/compress_docx.py <docx路径> <输出路径>

处理策略

Known limitations

• Pages with mixed text and images(Text within charts/tables): OCR will recognize text inside charts/tables as body text, requiring manual replacement
  • Solution: User identifies the problematic page, informs the PDF page number, takes a screenshot, and manually replaces it
• White-background table of contents pages: Will not be automatically recognized as special pages; will be OCRed (with average results)
  • Solution: Manually replace the table of contents page with an image after conversion

Practical case ("The Book of Predictions" 592 pages)

• Processing time: Approximately 20 minutes (including 0.6s/page interval)
• Original output size: 303MB (embedded 144 images)
• Compressed size: 3.4MB (images reduced to 600px width, 60% quality)
• Recognition effectiveness: High accuracy for body text, chart pages require manual processing
• Automatically save progress every 50 pages to prevent mid-process crashes

Precautions

• Free version QPS=2, script already includes a 0.6-second/page interval
• Cropping ratio (header 6%/footer 4%) can be adjusted at the top of the script
• After OCR completion, it is recommended to spot-check a few pages to verify accuracy
• Original high-definition version is retained on the server; compressed version is for distribution