Extract images from a PDF

The pictures, not pictures of the pages

There are two different things people mean by getting images out of a PDF, and they are worth telling apart because the wrong one wastes your afternoon.

This one pulls out the image objects that are embedded in the file — the photographs, logos, charts and scans that were placed into it. They come out at the resolution they were stored at, which is often considerably higher than the page appears on screen, because a 300 dpi photo scaled into a small frame keeps all its pixels.

If what you want is a picture of each page as laid out — text, vectors and all — that is a different operation and PDF to JPG does it. Running this on a text document with no embedded pictures correctly finds nothing, and says so rather than handing you a zip of blank pages.

Every image once, not once per appearance

A logo in a page header is stored in the file once and referenced from every page. A naive extractor walks the pages and writes it out forty times.

Here each image object is identified and written once, so a forty-page report with a header logo gives you one logo rather than forty copies of it. The names carry the page an image was first found on, so p003-001.jpg is the first image on page three, which makes a large set easy to work through in order.

The minimum width, and why to raise it

Real PDFs are full of images that are not pictures: rule lines drawn as one-pixel-tall bitmaps, gradient strips, bullet glyphs, spacer images, the dot in a logo. A production document can contain hundreds, and they arrive mixed in with the four photographs you actually wanted.

So anything narrower than the minimum is skipped. The default of 64 pixels clears out most of the furniture; raise it to 300 or so when you want only the substantial photographs from a design-heavy file.

Formats are chosen to avoid a second loss

Images with transparency or a colour palette come out as PNG, because writing them as JPEG would flatten the transparency onto a guess at a background. Everything else comes out as JPEG at quality 92.

Being straight about that: a JPEG that was already inside the PDF is decoded and re-encoded rather than lifted out byte for byte, so it goes through one more compression cycle. At quality 92 that is not visible, but it is not literally the original file, and a tool that claims otherwise is usually only claiming it.

Locked files

A PDF with an open password needs it supplied. A PDF with only an owner password — the kind that permits reading but claims to forbid extraction — is read without one, because that restriction is a request to the viewer rather than encryption, and every library ignores it.