I’m interested in creating a video fingerprinting system for viral videos. Basically, it would track the videos across all the major video uploading sites. Naturally, this entails creating a huge database of videos, and this would result in lots of processor time being used, which is expensive. So naturally, I’m interested in knowing how YouTube does their video fingerprinting for their ContentID system. I’ve been searching around, but it seems that there isn’t any information available that goes into detail on how the system actually works. Does anyone have information relating to this? Specifically, I’m interested in knowing how the fingerprint is actually matched in the content database, since that’s the major implementation problem I foresee.
