copyable-pdf

A lightweight, dependency-minimal bash script to convert scanned PDFs into searchable PDFs using Tesseract OCR.

copyable-pdf takes a PDF input, converts each page to an image, performs OCR (Optical Character Recognition) using Tesseract, and merges them back into a single, searchable PDF document.

Features

• OCR: Make scanned documents searchable and copyable.
• Preserve mode: Keep the original pages untouched and add an invisible text layer on top, instead of rebuilding the PDF from images.
• Parallel Processing: Uses multiple cores for faster OCR.
• Text sidecars: Also write the recognised text as .txt or .md.
• Interactive mode: Run it with no arguments and it asks for each setting.
• Dependency Check: Automatically checks for missing tools, and offers to install them and any missing language pack.
• Customizable: Set language, DPI and the number of parallel jobs.

Installation

Via Homebrew

brew install maxgfr/tap/copyable-pdf

Manual Installation

1. Clone the repository:

   git clone https://github.com/maxgfr/copyable-pdf.git
   cd copyable-pdf

2. Make the script executable:

   chmod +x script.sh

3. (Optional) Move to your bin directory:

   mv script.sh /usr/local/bin/copyable-pdf

Dependencies

Ensure you have the following installed:

• tesseract: For OCR.
• poppler: For pdftoppm, pdfunite, pdfinfo and pdftotext.
• qpdf (optional): Only needed for --preserve.

On macOS (Homebrew):

brew install tesseract poppler
brew install qpdf            # optional, for --preserve

On Ubuntu/Debian:

sudo apt-get install tesseract-ocr poppler-utils
sudo apt-get install qpdf    # optional, for --preserve

Usage

copyable-pdf [options] input.pdf

Run it with no arguments for interactive mode, which asks for the input path, the language, the DPI, the sidecars and the output name in turn.

How it works

1. Every page is rendered to an image at the chosen DPI, and Tesseract recognises the text on it, several pages at a time.
2. By default those recognised pages are merged into a new PDF. The result is searchable everywhere, but it is a re-encoded copy of the original.
3. With --preserve, only the recognised text is kept and stamped onto the original file as an invisible layer, so nothing about the original changes.

If a page fails to be recognised, the run stops and writes nothing, rather than handing back a document that is quietly missing pages.

Options

Option	Description	Default
-l, --lang <code>	Language code(s), + separated (e.g. eng, fra, eng+fra)	eng
-o, --output <path>	Custom output file path	input_ocr.pdf
-d, --dpi <num>	DPI resolution for OCR	300
-j, --jobs <num>	Number of parallel jobs	Auto-detect
-t, --text	Generate an additional .txt file	false
-m, --markdown	Generate an additional .md file (layout-preserved plain text)	false
-p, --preserve	Keep the original pages, add an invisible text layer on top (needs qpdf)	false
-k, --keep	Keep temporary files (debug)	false
-v, --verbose	Verbose output	false
-h, --help	Show help message	-
-V, --version	Print the version and exit	-

Examples

Basic usage:

copyable-pdf document.pdf

Specify language (French) and higher DPI:

copyable-pdf -l fra -d 600 document.pdf

Explicitly set output filename:

copyable-pdf -o searchable_doc.pdf scan.pdf

Keep the original pages untouched:

copyable-pdf --preserve scan.pdf

Preserve mode

Rebuilding a PDF from 300 DPI renderings costs you the original. A scan stored as JBIG2 or CCITT can grow several times over, and vector text, bookmarks and metadata do not come back.

--preserve avoids that: images, vectors, bookmarks, metadata and file size stay exactly as they were, with the recognised text selectable on top. It needs qpdf, and only when the flag is used, so the default path still depends on nothing but Tesseract and poppler.

Use it for anything worth keeping. Use the default when qpdf is unavailable and a plain searchable copy is enough.

License

MIT