Philips has developed a system that can scan video files, comparing them to a database, in order to identify copyright infringing material, among other uses. This is similar to how music fingerprint service Shazam works.
Clearly there is interest among the more conscientious video sharing sites to automate this process, if only to stem the tide of work being generated by complaints from copyright holders. Just yesterday for instance, Microsoft acknowledged that it had asked YouTube to remove the Ricky Gervais Microsoft training video from its system, a video featured on the WorldTV charts. (At the time of writing it is still available on Google Video).
The challenge with these systems is being resilient to situations where the content is significantly altered, for instance adding a logo, re-encoding with a much lower quality compression scheme, cropping, etc. It’s much easier to identify music, because music still has to sound the same to the end user, and there is less data to process, whereas video can be put into black and white, sytlized, have noise added, etc, without really affecting the user experience.
Philips claim the system can overcome these challenges and is pushing it as a solution to copyright infringement on video sharing sites and P2P networks. They also suggest that it can be used within broadcast environments for remote triggering of advertisement inserts etc. I can think of several other applications including media monitoring for businesses, video library organization, security applications and automated billing for stock footage usage.
No system like this is perfect, but it doesn’t need to be. If it can flag content for manual review with just an 80 or 90% success rate, that makes a huge difference to labour costs. If it gains sufficient traction however, and is used on a wide range of sites, it will undoubtedly come under substantial attack from those wishing to subvert it.