Scanned PDF back to JPG: get the scans out as they were saved

Why extracting beats converting for a scan

When a scanner or a scan app makes a PDF, it takes a photo of each sheet, compresses it (almost always as a JPEG) and wraps a PDF page around it. The text you see is part of the photo. So the picture you want back is already inside the file, whole.

There are 2 ways to get it out, and they give different files:

Extracting copies the stored picture out of the file. A JPEG stays the exact same JPEG: same pixels, same size, no second round of compression. On our test scan, a lease page scanned at 200 dpi in greyscale, extraction returned a 1165 x 826 pixel JPEG of 33.9 KB, the scanner’s own file.

Rendering (what PDF to JPG does) draws the page again as a new picture at the resolution you choose. At 150 dpi the same page came out 874 x 619: smaller than the scan, so detail is thrown away. At 300 dpi it came out 1747 x 1239: bigger, but every extra pixel is a guess between 2 real ones, and the file grew to 96.5 KB. Neither is the scan; both are copies of it, compressed once more.

Steps: get the scans out

1. Drop the scanned PDF on the tool above. It opens in your browser; nothing is sent anywhere.
2. Leave Format on As stored. JPEGs come out exactly as stored; any other kind of scan (a black and white fax-style scan, for example) comes out as a lossless PNG. All PNG turns everything into PNG, which you rarely need.
3. Leave Smallest at 50 px. It skips specks, lines and logos smaller than 50 pixels on both sides. A full-page scan is far bigger, so it always comes through.
4. Pick the pages if you do not want all of them: 1-3, 7 in Pages.
5. Press Extract pictures and download. 1 picture downloads on its own; several come as 1 zip.
6. Check the result line. It says how many pictures came out and how many as the original JPEG, for example “Took out 12 pictures at their own resolution, all as the original JPEG files”.

What the numbers tell you about the scan

The extracted image’s size tells you how the page was scanned. Divide its width in pixels by the page’s width in inches:

• our test page is 5.8 inches wide (an A5 page on its side) and the scan is 1165 pixels wide: 1165 / 5.8 = 200 dpi.
• an A4 page is 8.27 inches wide; a scan 2480 pixels wide is 300 dpi, 1654 pixels is 200 dpi, 1240 pixels is 150 dpi.

Your computer shows the pixel size in the file’s properties (Windows) or in Get Info (Mac). 200 and 300 dpi are the usual settings of document scanners and scan apps. That number is the most detail you will ever get out of this file.

Many receipts in 1 PDF

Karol Michał scanned 12 receipts into 1 PDF in Łódź and his expense tool wants 1 image per receipt. Extraction gives exactly that: 1 picture per page, named after the file and the page, receipts-page01-1.jpg, receipts-page02-1.jpg and so on (the page number gets a leading zero when the PDF has 10 pages or more, so the files sort in page order), in a zip called receipts-images.zip. The last number counts pictures on the page, so a page with 2 pictures gives -1 and -2.

One thing to know: the same picture comes out once. When we built a test file where page 3 was the same scan as page 1, we got 2 images, and the result said “Skipped 1 copy of pictures used more than once”. A real duplicate scan is caught too; 2 scans of 2 different receipts never are.

When to render instead of extracting

Extraction gives the stored picture, not what the viewer shows. Most scans are the same either way, but scan apps sometimes apply changes on top of the picture as page settings:

• Rotation. We turned a sideways scan upright with the page’s rotation setting, the way scan apps do. Extraction returned the picture sideways, 826 x 1165, as it was stored; PDF to JPG returned it upright, 1165 x 826.
• Cropping. If the app cut the page down to the sheet’s edges by setting a crop, extraction gives you the whole uncropped picture, table edge and all.
• Things added on top. Notes, highlights and form entries added after scanning are not part of the scan, so extraction leaves them out.

If you want the page as you see it, use PDF to JPG at a resolution close to the scan’s own: for our 200 dpi scan, 300 dpi keeps every original pixel, 150 loses some.

The honest part: the scan is the ceiling

Extraction is lossless, but it cannot be better than the scan. A page scanned at 150 dpi has 150 dots per inch of detail, and no tool recovers small print the scanner never captured; any “upscale” invents pixels rather than finding them. If the extracted image is too rough, the only real fix is to scan the paper again at 300 dpi.

Some PDFs that look scanned are not: a document printed to PDF with a photo on it has text that is real text, plus a picture. Extraction then gives you the picture without the text, and a page with no picture gives nothing. If the tool finds nothing at all, the file has no stored pictures, and PDF to JPG is the way to get the pages.

And if what you really want is the words, not the picture, extraction is the wrong job. OCR PDF reads the text in the scan and adds it to the file so you can search and copy it; the guide to making scans searchable walks through it. For everything else about getting images out of PDFs, see PDF to images.