Extracting images from a PDF copies out the pictures stored inside the file, at the size they were stored, without redrawing the page. You get the original photo, not a screenshot of it. You also get a few things you did not expect, like a logo nine times and a black-and-white mask file, and this guide explains why.
Extraction or page conversion?
Pick the job by what you want to end up with:
| What you want | Use | What you get |
|---|---|---|
| The photos, logos or scans inside the PDF | Extract Images | One file per stored image, at its stored pixel size |
| A picture of each whole page, text and all | PDF to Image | One PNG or JPG per page |
| The words, to reuse or search | Text extraction | Plain text, no pictures |
The quick test: if you want a chart exactly as it looks on the page, labels included, convert the page. Most charts and diagrams are vector shapes, lines and fills described by coordinates rather than pixels, and image extraction skips them entirely.
Saved, not screenshotted
The extractor runs pdfimages -all from the Poppler PDF library. It does not draw the page. It finds every image object the PDF stores and writes out the pixel data as it is held in the file. A viewer scales, crops and layers each image when it renders a page; none of that is applied here. That one fact explains almost every surprise in the results.
For JPEG photos the result is an exact copy. When I compared an extracted photo with the JPEG that had been placed in the test PDF, the two files were byte-for-byte identical. No quality is lost, and none can be added.
What a three-page test returned
I built a 1.45 MB PDF with the same JPEG photo and the same transparent PNG logo on all three pages, then ran it through the extractor. The PDF stores each image once. The output was 9 files totalling 4.32 MB: for every page, the photo as a 1.44 MB .jpg, the logo as a 1.1 KB .png, and the logo's transparency as a separate 890-byte grayscale .png.
Three things are going on there. pdfimages writes an image again for every page that uses it, so a reused logo multiplies. A transparent image is stored as a color image plus a separate mask, and you receive both; in my test the transparent area of the logo came out black in the color file, because black is what is stored under the mask. And the photo came out at its full pixel size regardless of how small it was drawn. In another test file, a 2480 x 1754 pixel photo was placed only 60 mm wide, which pdfimages reported as 1050 pixels per inch on the page, and it was extracted at all 2480 x 1754 pixels.
The reverse also happens: a small logo stretched across a page comes out at its small stored size and looks blurry on its own. Cropping is not applied either. If the document shows only part of a photo, you get the whole stored picture, hidden parts included.
File types you may find in the ZIP
Because -all keeps each image's native encoding wherever it can, one ZIP can hold several file types. The pdfimages manual lists them; this is how they show up in practice:
| Stored in the PDF as | Saved as | Opens in |
|---|---|---|
| JPEG (DCT) | .jpg, an exact copy | Any image viewer |
| JPEG 2000 | .jp2 | Many viewers; some need a plug-in |
| JBIG2, common in black-and-white scans | .jb2e plus .jb2g for shared symbol data | Specialist tools only |
| CCITT fax | .ccitt with a .params file | Specialist tools only |
| CMYK color, not JPEG | .tif | Most image editors |
| Anything else, such as Flate-compressed RGB or gray | .png | Any image viewer |
If a file will not open, it is almost certainly JBIG2 or CCITT data, saved in its original form so nothing is lost. Converting that page with PDF to Image gives you a plain PNG of the same content.
Scans and the "one image per page" case
A scanned PDF usually stores each page as a single image. My 3-page grayscale scan gave three 2480 x 3508 pixel single-channel JPEG files, one per page, with no text in them. If what you actually want from a scan is a readable, searchable document rather than page images, that is a job for OCR, not extraction.
Expect more files than the pictures you noticed, too. Icons, rules drawn as images and tiny tiles are all image objects, and they are all extracted.
How to run it
- Start with the best copy you have. A PDF that was compressed or re-scanned already holds lower-quality images, and extraction cannot recover what was thrown away.
- Open Extract Images and upload the file. Free accounts can add up to 3 PDFs in one batch, Pro and AI up to 50. Each file counts as one daily check; on Free, with 100 checks a day, a batch that needs more checks than you have left is refused before any work starts.
- Wait. Extraction of a single file stops after 2 minutes, which only heavy files approach.
- Preview the results. Download images one at a time, or everything as
extracted_images.zip. - Compare the output with the pages and delete the duplicates you do not need.
If you would rather stay on your own machine, the same command works in a terminal: pdfimages -all input.pdf img writes the files next to you with an img- prefix. Acrobat has an Export All Images option that does a similar job with a different set of format choices.
When nothing, or the wrong thing, comes out
- No Images Found. The PDF holds text and vector drawings only. Charts, diagrams and logos drawn as shapes are not images; convert the page to PNG instead.
- "Failed to process PDF. The file may be corrupted or password-protected." pdfimages could not open the file. With my password-protected test file it stopped with an incorrect password error. Remove the password with the PDF password tool first, if you are allowed to.
- The same picture several times. It is used on several pages. Delete the extras after downloading.
- Far more files than you wanted. Split out the pages you care about and extract from those.
Before you reuse what you extracted
Being able to extract an image gives you no right to publish it. Photos in a brochure or report usually belong to the author, the publisher or a stock library, so check the license or ask. Look at the content, too: scanned pages can carry signatures, account numbers and addresses, and JPEGs copied byte for byte keep whatever camera metadata they had, including the date the photo was taken.
If an image ends up on a web page, give it alt text that says what it shows. The W3C's PDF1 technique sets out the same rule for images inside accessible PDFs.
What happens to your files
The uploaded PDF is deleted as soon as extraction finishes. The images and ZIP sit under a random link until an hourly cleanup removes anything older than 24 hours. The results page and the ZIP download open only in the browser session that uploaded the file; individual image files are served from public storage, so a direct image link works for anyone who has it until cleanup. The free plan accepts files up to 50 MB; Pro and AI allow 100 MB. If the real problem is that the PDF itself is too big because of oversized images, the PDF optimizer is the tool, and our guide to converting PDF pages to images covers the settings that keep rendered text sharp.