Extract the images from a PDF

A PDF that shows you a photograph contains that photograph. Extracting it is not the same as taking a screenshot of the page, and it is not the same as converting the page to an image: those give you a picture of a page, at whatever resolution you asked for, with the text and margins baked in. This gives you the picture the document was built from, at the size it was stored.

That difference matters whenever the image is the point. A scanned figure you need for a report, a logo you need to reuse, a photograph somebody sent inside a document instead of attaching it — in every case the version inside the file is better than any version you can make by rendering the page around it.

JPEGs come out untouched

Most photographs inside a PDF are stored as JPEG already. The format keeps them as a stream of exactly the bytes a .jpg file would contain, so those bytes are written straight out. No decoding, no re-encoding, no second generation of compression artefacts on top of the first.

Images stored in other lossless forms are decoded and written as PNG, because re-encoding them as JPEG on the way out would throw away the thing that made them lossless in the first place. So a chart or a screenshot comes out as a PNG, and a photograph comes out as the JPEG it always was.

Each image is extracted once, however many pages it appears on. A logo repeated on forty pages is one entry in the ZIP, not forty copies of the same file.

What gets skipped, and why it says so

Some images cannot be read out without guessing. Indexed colour stores a palette and a table of references into it; CMYK means four channels rather than three; an image put through more than one encoding has to be unwound in the right order. Getting any of those wrong does not fail loudly — it produces a picture with the wrong colours, which is worse than producing nothing, because you might not notice until it is in front of somebody else.

So those are counted and skipped, and the result tells you how many. A file whose images are all in a form that cannot be read reports that plainly rather than handing you an empty archive and letting you work it out.

Soft masks are skipped too, and that one is easy to miss. A transparent image in a PDF is stored as two images: the picture, and a separate greyscale image describing which parts of it are see-through. Extracting the second one gives you a black and white silhouette that looks like an image you had and is not one.

Vector art is not an image

A chart drawn as lines and shapes, a logo stored as vector paths, a page of text — none of these contain an embedded picture, so there is nothing to extract from them. A document made entirely of vector artwork will report that it has no embedded images, which is true, even though it plainly has pictures on it in the ordinary sense of the word.

If that is what you want out, converting the pages to images is the tool that does it, and it will render the vector art at whatever resolution you choose.

Nothing leaves the device

The document is opened, walked, and unpacked by a worker inside your browser. There is no upload, no account, and no copy on a server — which is worth knowing for the case where the images are the sensitive part of the file.