(1) That is the paradigm of classification, don't knock it, and

(2) I did some work on a pre-product startup that wanted to build something like this about 10 years ago, when visual recognition was much less developed.