Note that we're talking about fuzzy image-matching hashs that attempt to match an image to its source even after it was cropped at the edges, maybe got a new watermark somewhere and was recompressed with JPEG a few times. I don't know the false positive rate of these algorithms but I imagine they're much bigger than that of a cryptographic hash function (that's what analogies like yours usually refer to).