papero-pdf-text-extractor

papero-pdf-text-extractor is a fast and open-source API designed for efficient PDF text extraction. It provides detailed document structure extraction features without requiring heavy machine learning models. This tool operates entirely on CPU, making it accessible for use in various environments including browser applications, Python scripts, and REST APIs. Notably, it ensures that files are not stored, so users maintain complete control over their documents.

Key Features

• Flexible Output Formats: Convert PDFs to multiple formats such as Markdown, JSON, Word, and Excel while preserving reading order, tables, formulas, figures, and precise block positions.
• High Performance: Designed to perform efficiently on standard laptop CPUs, it handles the extraction and layout reconstruction in a matter of milliseconds for multi-column dense documents.
• No Machine Learning Required: By relying on traditional geometry, papero avoids the complexity and overhead of machine learning models, providing a straightforward solution for text extraction needs.

Detailed Functionality

Document Structure Extraction

The tool reconstructs various elements from PDF documents:

• 📖 Reading Order: Support for two- and three-column documents, maintaining proper reading flow. Non-essential elements like headers, footers, and logos are ignored during extraction.
• ▦ Real Tables: Captured tables, whether ruled or borderless, are accurately represented as rows and columns, including multi-line cells, and can be exported to CSV or Excel formats.
• ∑ Formulas: Mathematical expressions, including fractions and symbols, are captured as LaTeX format along with their rendered images. For example, 5/12 is output as \frac{5}{12}.
• 📍 Positioning Information: Every element's bounding box is tracked, allowing for precise referencing in results or further processing.
• 🖼 Captured Figures and Charts: Images and charts extracted are cropped with their respective captions intact, ensuring context is preserved.
• 📝 Word Output: When exported to Word or HTML formats, the layout and formatting of the original document are preserved, maintaining the integrity of two-column formats.

Getting Started

To utilize the extraction functionalities:

pip install papero-extract

Then, in Python:

from pdf_text_api import extract

doc = extract("paper.pdf")
print(doc.to_markdown())

Alternatively, for a hands-on experience, users can directly access the application in the browser: Try it now.

Output Examples

Upon extraction, users receive well-structured JSON data with detailed formatting, ensuring that every block of text maintains its identity and location. For instance:

{
  "type": "table",
  "bbox": [56.7, 294.8, 481.9, 374.2],
  "rows": [["Model", "Accuracy"], ["Base", "0.81"]],
  "caption": "Table 1: Comparison between models."
}

Benchmark Performance

In benchmark tests with dense academic papers, papero showcased impressive performance metrics, particularly in maintaining reading order and structural accuracy while operating 0 failures on extraction attempts for 54 papers.

How It Works

The extraction process employs two main engines that operate simultaneously:

1. Layout Engine (PDFium): Reads glyph positions and reconstructs document elements, factoring in their column structures.
2. Metadata Engine (Apache Tika): Enhances the extraction with metadata, OCR functionalities, and support for various document formats.

Contributing

Users are encouraged to report any issues regarding PDF extraction by submitting relevant files, which greatly aids in improving the tool's functionality.

Keywords

Some related topics include: PDF to Markdown, PDF parsing, layout analysis, formula extraction, bounding boxes, and document preprocessing for LLM applications.