They aren't using file hashes like SHA1, but perceptual hashes and then a fuzzy metric like hamming distance to make a fuzzy match. It's entirely possible for two images that have nothing to do with each other to have similar hashes.
Yes, that's what I was alluding to. It's a perceptual hash, so you could probably take a flagged image and superficially alter it (contrast/color/crop/whatever) until it wasn't anything visually objectionable on it's own. But it would still match the perceptual hash.
The poor souls that do the work of entering the images into the database select identification blocks, not showing the victim (a lamp, wall paper, etc), that reviewers on the “client” side check against
