PDF Structured Text Extractor

A small PDF text extraction tool built on pdfplumber.

It tries normal text extraction first. If that fails (bad encoding, weird fonts, some scanned or broken text layers), it falls back to grouping words by their position on the page to rebuild the text.

Features

• Direct text extraction first, spatial grouping as a fallback.
• Groups words by distance to rebuild line order.
• Runs locally, no API calls or cloud dependency.
• One import, one function call.

Limitations

• Multi-column layouts can get mixed up. The grouping just looks at nearest-neighbor distance, with no real column detection. Two-column resumes (sidebar + main content) can end up interleaved.
• No OCR. If a PDF is a scanned image with no text layer, this returns "No text found." Use something like pytesseract alongside it if you need OCR.
• Threshold isn't tuned rigorously. The threshold value (PDF points, not pixels) works on the documents it was tested on, but hasn't been validated on a bigger dataset. You may need to adjust it for other layouts.
• Not a full layout engine. For serious table/column reconstruction, look at pdfminer.six (layout mode), PyMuPDF, or unstructured — they're more mature. This is a simpler fallback, not a replacement for those.

How It Works

1. Try pdfplumber.page.extract_text() directly.
2. If that returns nothing:
   • Get word bounding boxes.
   • Measure horizontal/vertical distance between words.
   • Group nearby words into lines.
   • Join lines in reading order.

Installation

pip install pdf-structured-extractor

Or from source:

git clone https://github.com/2005MohitInamdar/pdf_Text_Extraction
cd pdf_Text_Extraction
pip install -r requirements.txt

Usage

from pdf_resume_extractor import extract_text_from_pdf

text = extract_text_from_pdf("path/to/document.pdf")
print(text)

Comparison

Ideas for Contribution

• Basic column detection
• Optional OCR fallback for scanned PDFs
• A test set of PDFs (text, scanned, multi-column, tables) to check accuracy against

License

MIT (change if you're using something else)

Contributing

Issues and PRs welcome, especially PDFs that break the grouping logic.