> That’s good info about #1, because it seems to be an O(n * m) problem where n is number of existing copyrighted works and m is the number of uploads.
This is likely very naive of me to ask but is it actually O(n*m)?
I don't know how the fingerprinting works but if it's anything like a hash I don't see why it couldn't be closer to O(n+m) since you only have to hash each copyrighted video once and then compare O(m) uploaded videos against the hashes.
I can imagine you can't actually compare against hashes of all copyrighted videos in O(1) since they can't just be kept in memory but I don't see why it can't be done in O(log n).
Of course this is very far out of my expertise so please someone explain to me why this is completely wrong =]
Hm, good point. I was thinking of the case of manual checking, but good point nonetheless.
Of course, the checksum would have to be invariant to pitch shift, rotation, horizontal flipping, framing, clipping, and length changes — all of which I know have been used to evade copyright filters on YouTube. (Also must resist mild noise audio or visual noise, mild convolution filters, and probably hue shifting, but I don’t know if those have been used by pirates yet)
This is likely very naive of me to ask but is it actually O(n*m)?
I don't know how the fingerprinting works but if it's anything like a hash I don't see why it couldn't be closer to O(n+m) since you only have to hash each copyrighted video once and then compare O(m) uploaded videos against the hashes.
I can imagine you can't actually compare against hashes of all copyrighted videos in O(1) since they can't just be kept in memory but I don't see why it can't be done in O(log n).
Of course this is very far out of my expertise so please someone explain to me why this is completely wrong =]