It really looks like US7921296B2, which has a priority back to 2001 is prior art for 7,627,477.
They both create a database of id strings which are located at specific locations within a file. They then sample the same locations in the file to be compared. The only real differences is that the US7921296B2 patent uses hash strings, while Landmarks patent uses "fingerprint object". The US7627477 does not does NOT describe how those "fingerprint object" are created. In their description they state(with a prior citation to Knuth) "One way of preparing the data to allow for fast searching is to encode the fingerprint objects into numeric tokens, such as 32-bit unsigned integers, and using the numeric tokens as a key for sorting and searching". I wonder what method are going to use to create that 32-bit uint, without collisions LOL.<p>And whoever wrote 7,627,477 was a moron, unless the examiner required them to restrict their comparisons to strictly audio files. But since their earliest grant also is dedicated to audio, I'm guessing they were not thinking ahead. US7921296B2 on the other hand applies to multimedia files in general.<p>What get's me is that it was the examiner that cited US7921296B2.<p>Have I misinterpreted this?