Extract or render — which do you want?
Extract when you want the pictures: a logo to reuse, a chart from a report, the photographs from a brochure, the scans inside a bundle. You get the original files at their original pixel dimensions.
Render when you want the pages: a page as a thumbnail, a poster as a JPG, a certificate to post as an image. That is PDF to JPG, where you choose the DPI.
What quality do the extracted images have?
Exactly the quality inside the PDF, which is not always what the photographer took. A designer placing a 24-megapixel photo into a layout usually has it downsampled to 300 DPI at print size, so extraction returns that, not the raw original.
JPEG images come out as JPEG with no re-encode at all. Everything else — PNG-like transparent graphics, JPEG 2000, JBIG2 scans, masks — is decoded and saved as PNG, which keeps every pixel.
Why do I get more images than I expected?
A single visible picture can be stored as several objects: a photo plus a transparency mask, a background tile repeated across the page, or a scan split into horizontal strips by the scanner’s software.
The minimum size filter deals with most of this: raising it drops the rules, bullets and tiny decorations and leaves the real pictures.
Can you extract images from a scanned PDF?
Yes — and in a scan each page is one big image, so you get one file per page at the scanner’s original resolution, which is usually better than rendering the page again.
That also makes extraction the right way to recover a scan someone has "PDF-ified" without losing a generation of quality.