Code Read Against Source, 2026-09-10
How to Create Searchable Sandwich PDFs with Invisible Text in Python
How do you create a searchable PDF sandwich with invisible text in Python?
B.L.A.S.T. generates searchable PDF sandwiches with SearchablePDFGenerator, pairing PyMuPDF (fitz) with a ReportLab fallback. It overlays recognized text as an invisible font layer exactly over the raster image coordinates, preserving 100% visual fidelity while enabling search, highlighting, and copy-paste. Verified in blast_ocr/core/searchable_pdf.py.
CLI Quickstart
git clone https://github.com/Ibrahim-Salman19/OCR.git && cd OCR
pip install -r requirements.txt
python -m blast_ocr.cli scanned_contract.pdf --formats pdf
Python implementation: exact sandwich geometry
The high-level path is pipeline.process_job(..., formats=["pdf"]) — see the high-throughput guide. To call the sandwich generator directly with your own OCR detections:
from blast_ocr.core.searchable_pdf import SearchablePDFGenerator
# create_from_page_images() takes matching lists of page images and
# per-page OCR detections, and picks PyMuPDF if available, else ReportLab
output_pdf_path = SearchablePDFGenerator.create_from_page_images(
page_images=["samples/scanned_contract_p1.png"],
page_ocr_results=[
{
"details": [
{"bbox": [100, 150, 420, 175], "text": "CONFIDENTIAL SETTLEMENT AGREEMENT", "confidence": 0.98}
]
}
],
output_pdf_path="output/searchable_contract.pdf",
title="Settlement Agreement",
)
print(f"Searchable PDF generated at: {output_pdf_path}")
page_ocr_results[i]["details"] is one of two payload shapes SearchablePDFGenerator accepts natively (_extract_text_boxes) — a flat list of {bbox, text, confidence} items per page.