Document OCR Tool

简体中文 | English

An offline Windows desktop application that converts PDFs to editable Word documents and table images to Excel workbooks.

Features

• PDF to Word
  • Native layout conversion with pdf2docx
  • Offline OCR fallback with RapidOCR and ONNX Runtime
  • Automatic, forced-OCR, and native-only modes
  • Batch conversion, Chinese paths, page breaks, and optional source-page images
• Image to Excel
  • PNG, JPEG, BMP, TIFF, and WebP input
  • OpenCV grid detection for bordered tables
  • OCR-coordinate fallback for borderless tables
  • Single-file and batch .xlsx export
• Desktop focused
  • Tkinter interface with a top-level function switch
  • Background conversion and progress reporting
  • No Microsoft Word, Excel, PaddlePaddle, or Tesseract installation required
  • Dedicated Windows 7 SP1 x64 build workflow

Quick Start

Download

Download the latest Windows executable from GitHub Releases.

Run from source

git clone https://github.com/ADDA-acx/document-ocr-converter.git
cd document-ocr-converter
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
python -m pip install -e . --no-deps
document-ocr-tool

Windows 7 Build

The compatibility build uses Python 3.8.10 and frozen native dependencies.

scripts\setup-win7-build.bat
scripts\build-win7-onefile.bat

Output:

release\DocumentOCRTool-Win7-x64.exe

For the folder-based build:

scripts\build-win7-folder.bat

See Building for details.

Project Structure

src/document_ocr_tool/  Application and conversion code
tests/                  Automated tests
scripts/                Build and compatibility scripts
docs/                   English and Chinese documentation
models/                 Optional external OCR model files
assets/                 Icons and application assets
tools/win7-vm/          Optional manual Windows 7 test tooling
release/                Local release output (not committed)

Accuracy Notes

OCR quality depends on image resolution, skew, compression, font size, and table complexity. Clear, front-facing images with visible grid lines produce the best Excel output. Merged cells, handwritten text, and highly irregular tables may require manual correction.

Development

python -m pip install -e . --no-deps
python -m unittest discover -s tests -v

Contributions are welcome. Read CONTRIBUTING.md before opening an issue or pull request.

License

MIT