What we probably need is a client-side neural net within the browser to notice that certain high information-content IMG URLs (i.e. those with random-looking garbage in the URL) from certain domains tend to result in very low information-content imagery. And that's the signature of a likely tracker.