It's not a technical article. I think it's safe to assume that 1) The people implementing this aren't idiots, and 2) The "hash" is probably referring to a "perceptual hash", where very-similar images would show up as the same.