[dimatura]

It's true that these functions are in some sense very different to traditional hash functions, because they are designed to create "collisions" (or at least be similar) for elements that are similar. But I still think it is intuitive to use the term, maybe qualified as a "similarity hash" (I've also seen "perceptual hash" or "semantic hash"). In both cases we are describing a (potentially) large object with a short code. In normal hashes, the hash is supposed to be the same for objects that are exactly the same - in similarity hashes, that is softened to similar hashes for similar objects.

[pcthrowaway]

Functions that create collisions can still hashes can't they?

Things like CRCs, Checksums are loosely considered hash functions: https://en.wikipedia.org/wiki/List_of_hash_functions

MD5SUM is a classic example of a hash function with a low (but still reasonable) chance of a collision