PDF to Markdown

PdfToMarkdown turns any PDF into GitHub Flavored Markdown: headings, paragraphs, bulleted and numbered lists, bold and italic text, links and pipe tables. It works for PDFs from any tool, not only those PdfPecker makes, and is meant for search indexes, language models and anything else that reads text.

// The whole document as one Markdown string.
string markdown = PdfToMarkdown.Convert(File.ReadAllBytes("statement.pdf"));

// Mark where each page starts, so a passage can be traced back to its page.
string withPages = PdfToMarkdown.Convert(pdf, new PdfMarkdownOptions { IncludePageComments = true });

// File to file, or into a stream.
PdfToMarkdown.ConvertFile("report.pdf", "report.md");

What it finds

The structure is the same one the structured PDF to HTML finds. A tagged PDF, such as one written for PDF/UA, is read from its structure tree, and other PDFs from their layout: headings from larger or bolder text, lists from their bullets and numbers, and tables from the borders and rules the PDF draws and from text that lines up in columns.

Several header rows of a table are joined column by column, and a spanned cell's text is in its first row and column.
Line breaks the PDF made on purpose, such as in an address, stay as line breaks.
Text that would read as Markdown, such as a leading # or *, is escaped, so the output reads back as the same text.

Options

`PdfMarkdownOptions`	Default	What it does
`Pages`	every page	The pages to convert, such as `PdfPageRange.Parse("1-3")`
`Password`	none	The user or owner password of an encrypted PDF
`IncludeLinks`	true	Text that links to a web or e-mail address becomes a Markdown link
`RenderAnnotations`	true	The text of annotations, such as form field values, is included
`IncludeImages`	false	Pictures, such as a logo, as `data:` images where they sit in the text
`IncludePageComments`	false	Each page starts with `<!-- page 3 -->`
`MaxPages`	10,000	More pages throw `PdfLimitException`
`MaxDegreeOfParallelism`	processor count	Pages converted at the same time

Pictures in table cells, pictures that fill most of a page and pictures too small to see are left out even with IncludeImages. Many Markdown viewers do not show data: images, and they make the text much longer, so they are off by default.

What it is not

It reads the text a PDF holds; a scanned page without a text layer gives no text, since PdfPecker does no OCR. The order and structure are a reading of the page, so an unusual layout can come out in another order than a person would read it.