pdf2md

PDF → Markdown extractor for Rust. Wraps pdfium-render with a layout-analysis pipeline that recovers tables, columns, headings, and figures from arbitrary PDF documents and emits clean Markdown.

What it does

Given a PDF on disk, pdf2md returns:

• Markdown with ATX headings, paragraphs, and inline image references
• Figures — embedded raster images and rasterized vector regions, each deduplicated by SHA-256 hash
• Heading depth — the maximum ATX level produced (0..=6)

The pipeline includes:

• Recursive XY-cut zone segmentation with table promotion
• Banner detection — strips repeating page headers, footers, and stray page-number digits
• Border/line detection — recovers table grids and clusters path segments into figure regions
• Heading classification from font size, weight, and italic flags
• Noise stripping for junk headings and stray glyphs
• Vector-figure rasterization at ~180 DPI for diagrams without embedded bitmaps

Requirements

libpdfium must be available at runtime. The crate uses dynamic binding via pdfium-render's bind_to_system_library().

OS	Install
Arch	yay -S pdfium-binaries
Debian / Ubuntu	sudo apt install libpdfium-dev
macOS	brew install pdfium
Windows	grab pdfium.dll from bblanchon/pdfium-binaries and place it next to your binary

Minimum supported Rust version: 1.85 (edition 2024).

Library use

The crate ships with default features (cli) on, which pulls in clap for the binary. Library-only consumers should disable defaults:

[dependencies]
pdf2md = { version = "0.1", default-features = false }

Basic extraction:

use std::path::Path;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let doc = pdf2md::extract(Path::new("input.pdf")).await?;
    println!("{}", doc.markdown);
    eprintln!("extracted {} figures", doc.figures.len());
    Ok(())
}

Custom image-directive emission (e.g. a custom Markdown extension):

use std::sync::Arc;
use pdf2md::ExtractConfig;

let cfg = ExtractConfig {
    image_emitter: Arc::new(|hash, alt| {
        if alt.is_empty() {
            format!("<image hash=\"{hash}\">")
        } else {
            format!("<image hash=\"{hash}\" alt=\"{alt}\">")
        }
    }),
};
let doc = pdf2md::extract_with_config(Path::new("input.pdf"), cfg).await?;

CLI

cargo install pdf2md
pdf2md input.pdf > out.md
pdf2md input.pdf --figures-dir ./figs > out.md

--figures-dir writes each extracted figure as <sha256>.png into the target directory; without it, the CLI just reports the figure count to stderr.

License

MIT — see LICENSE.