You render an image on the client browser, how this comes out will depend on screen-resolution, video driver etc, then you hash the result and store it on the server. When the mark visits a different site, you do the same, and if the hashes match, a good chance it's your target.
This is the HTML from Amazon's front-page after I logged in (and the Firefox fingerprint blocker stepped-in)
This is the HTML from Amazon's front-page after I logged in (and the Firefox fingerprint blocker stepped-in)
<img src="//us-u.openx.net/w/1.0/cm?id=e818ca1e-0c23-caa8-0dd3-096b0ada08b7&r=https%3A%2F%2Faax-eu.amazon-adsystem.com%2Fs%2Fecm3%3Fex-rc%3DWA1GheMVTG2hSi76AJLwBg%26ex%3Dopenx.com%26id%3D" width="1" height="1">
Search "Browser fingerprinting" for more info