Extract Images from PDF

Two modes, and how the tool picks one

A PDF does not keep a folder of pictures. It keeps drawing instructions, and an image appears in the list of instructions as a reference to a stored object. So the first pass here walks every page's instruction list looking for the operation that paints an image object, then pulls each referenced object out and writes it to a PNG at whatever pixel dimensions it was stored with. The label under each thumbnail shows those dimensions, so you can see immediately whether you have a usable original or a low-resolution placeholder.

Plenty of documents contain no such objects at all. A chart drawn as vectors, a text-only report, a form produced by a typesetting program: none of these store a single bitmap. Rather than showing you an empty grid, the tool switches to a second mode and renders each page as a PNG at double scale, roughly 144 DPI, so an A4 page arrives as about 1190 by 1684 pixels. The banner above the grid always states which mode ran, so you are never guessing whether the output is an original asset or a picture of a page.

The switch is all or nothing. If even a single embedded image is found anywhere in the document, the first mode wins and no page renders happen, even on pages that contained nothing.

Scanned documents: why you get whole pages

If you scan a document, each sheet is stored as one large image that covers the full page. That image is genuinely embedded, so the extractor finds it and reports embedded mode, and what lands in your grid is a stack of full-page scans rather than the individual photographs or signatures you may have hoped for.

That is correct behaviour, not a failure. There is nothing smaller inside the file to find. The picture of a signature at the bottom of page 3 is not a separate object, it is just some dark pixels in the middle of a page-sized one.

The useful part is the resolution. A sheet scanned at 300 DPI arrives as roughly 2480 by 3508 pixels for A4, which is the original scan data with no resampling. Save it, then crop out the part you want with Crop Image. If what you actually need is the words on the scan, run it through Image to Text instead.

Why the count is higher than you expected

Duplicate detection works within a page, not across the document. A logo placed in the header of a 40 page report is one stored object referenced 40 times, and you will get 40 identical PNGs. Delete the extras after downloading.

Some producing software also slices a single photograph into horizontal strips, often to keep individual objects small. When that has happened you get a set of tall thin images that reassemble into the picture you were after. There is no way to detect and rejoin those reliably, so the tool hands you what the file contains.

Anything smaller than 8 pixels in either direction is discarded, which removes the single-pixel spacers and hairline rules that some layouts scatter through a document.

Known rough edges

Transparency is the main one. When an image is stored with its cut-out mask held as a separate object, the mask is not merged back in, so a logo that looked like it floated on the page can arrive with a solid white or black rectangle behind it. You will spot this straight away in the thumbnails.

Output is always PNG, which is lossless but not always smaller. A photograph stored in the document as a 400 KB JPEG is decoded and re-encoded as PNG, and can easily come out at several megabytes. The pixels are identical to what was in the file, but if you want the size back, convert the result with Image Converter.

Colour spaces other than the common ones are skipped rather than guessed at, so a print-oriented document using separations may yield fewer images than it visibly contains. Each image object also has a four second window to decode, and a very large one on a slow device can be passed over.

Everything runs in your browser, so your files never leave your device. Password-protected files are refused on opening, and a warning appears for documents over 100 MB, which on a phone is a realistic point for the tab to run out of memory.

A quick comparison

If you want the assets that a designer placed into a document, this page is the right one. If you want a faithful picture of each page as it looks when printed, including text, borders and backgrounds, PDF to JPG does that directly and lets you choose 72, 150 or 300 DPI rather than being fixed at double scale. The fallback mode here is a convenience, not a substitute for that control.

Questions

Why does the banner say pages were rendered?

Because the scan found no bitmap objects anywhere in the document. That normally means the visuals are vector artwork or text, which have no pixels to pull out, so each page was drawn to a PNG instead.

Are the extracted images at full quality?

In embedded mode, yes. They come out at the pixel dimensions stored in the file, which is the most detail that exists. If those numbers look small, the document itself only ever held a small version.

Can I pick which images to pull out?

The scan always covers the whole document, but nothing is saved until you choose. Download individual PNGs from the thumbnails you want, or take the lot as a ZIP and sort them afterwards.

Why do some images look stretched compared with the page?

A document can scale an image to any size on the page independently of its stored pixels. A 200 by 200 image squeezed into a wide banner slot looks correct in the document but square when extracted, because the square version is what is actually stored.

Does this work on a PDF full of vector charts?

It will fall back to rendering pages, which gives you usable pictures of the charts but not editable vectors. Recovering true vector artwork needs a desktop illustration application.