Comment by 3wolf

They're using perceptual hashing, not cryptographic hashing of raw pixels. So it's invariant to variable bitrate, compression, etc.