Skip to main content

Bella Domify

An open-source document parsing Python library from Lianjia Tech. It can be imported as a Python library package or run as a service, supporting parsing and conversion of multiple document formats.

Features​

Supported File Formats​

  • PDF
  • Word Documents (DOCX/DOC)
  • Excel Spreadsheets (XLSX/XLS)
  • CSV Files
  • PowerPoint Presentations (PPTX)
  • Text Files
  • Image Files

Parsing Capabilities​

  • Layout Parse: Extracts basic layout structure of documents, including text blocks and image blocks
  • DomTree Parse: Builds detailed document object model for further processing and analysis
  • Markdown Conversion: Converts parsing results to Markdown format

Advanced Features​

  • Image Processing: Built-in OCR capabilities using large models to extract image information
  • Table Processing: Parses table structure and content
  • Header/Footer Recognition: Automatically identifies and filters headers and footers
  • Multi-process Parsing: Uses multi-process parallel processing to improve parsing efficiency
  • Evaluation and Annotation: Built-in evaluation module for annotating PDF parsing details pdf_marked

Advantages​

From the evaluation data shown below, Lianjia's self-developed parsing capabilities are very strong with higher accuracy rates (based on Lianjia's limited test set)

image2

Acknowledgments​

This project is based on secondary development of pdf2docx. We thank the original authors and their team for their outstanding contributions. pdf2docx extracts raw data such as text, images, and vectors based on PyMuPDF, and parses layout and styles of sections, paragraphs, tables, images, text, etc. based on rules. For specific features, please visit its GitHub address. It provides an important technical foundation for our document parsing functionality.

PDF Parsing: A Journey from Vision to Structure Reconstruction