Suppose your target post a picture from their iPhone to a social network. You just have to add this picture perceptual hash to the offensive database for bad things to happen to them.
To do this : you grab their picture, overlay some offensive content on top rendered very transparently so that the new image is like 95% original + 5% offensive content. You also generate a second image with ratio 98% original + 2% offensive content.
Then you post the pair of images on a public site that usually get scanned for offensive content. (You may even add the original image for good measure). Given that the offensive and original image can be exactly reconstructed, and that even at low transparency the human eye can still distinguish the offensive image it won't take long before all images get labelled as offensive. Their hashes added to the database, and a collision with the picture of your victim registered.
Launch a script and do it for every pic they post. Of course there are plenty of variations of the above technique.