What Extract images looks for inside a PDF

The rules behind JPG or PNG

Each image object in a PDF carries a note about how its pixels are compressed and what color system they use. Extract images reads that note before deciding what to save.

A picture is copied out byte for byte as a .jpg only when all of these are true:

• It's compressed as a JPEG, possibly wrapped in an extra text encoding such as ASCII85. That outer layer is undone first, and the JPEG inside is left alone.
• Its colors are gray or RGB, including the calibrated and ICC-profile versions of those.
• It has no soft mask, no color-key mask and no Decode array (a setting that can invert an image's tones when it's drawn).

Anything else is drawn by pdf.js onto a canvas at its own pixel size, with a transparent background, and saved as a PNG. That covers palette images, Lab colors, JPEG 2000 photos and black-and-white fax-style scans, as well as the masked and CMYK cases.

When the PNG is bigger than the PDF

A PNG made this way always has 8 bits per channel plus an alpha channel. That's fine for a logo, but it can balloon other things:

• A 1-bit scan stored with CCITT or JBIG2 compression takes very little space inside the PDF. The same page as a full-color PNG can be many times larger.
• A JPEG 2000 photo is lossy and compact. Saved as a lossless PNG, it has to keep every pixel exactly, so it grows.
• 16-bit images come out as 8-bit, because a browser canvas can't hold more.

None of this loses detail you could see, but a ZIP of extracted scans can easily outweigh the PDF it came from. If you only need to read those pages, rendering them with PDF to JPG and the "Black and white" color setting usually gives smaller files.

Images it skips on purpose

Some image objects aren't really pictures. A stencil mask is a 1-bit shape with no color of its own: the page fills it with whatever color is current, a bit like spraying paint through a cut-out. Extract images leaves these out, since the file has only the outline.

This matters for scans saved with heavy "mixed" compression, which some scanners and PDF makers use to shrink color documents. The text is often kept as a sharp black stencil over a softer, low-resolution background picture. You'll get the background, sometimes with the text missing. For those files, render the page instead.

Images hiding in comments and stamps

The tool walks each page's own content and any reusable graphics that content calls on, such as a letterhead or a page imported from another PDF. It doesn't open annotations. A photo added as a stamp comment, or a signature image placed by an app that stores it as an annotation, sits outside the page content and won't be listed.

The same applies to images in form field appearances, such as a photo box on an ID application form.