Clearly, life would be easier for the owners of video-sharing sites if copyrighted content could be blocked before it’s posted. And that’s what Guba claims Johnny can do. The company was founded in 1998 to build a tool that could search for still images and videos posted to Usenet discussion groups. But as soon as the company began to aggregate such content, it started receiving complaints and takedown requests from copyright holders. “We wanted to make it easier to find things, but we ended up demonstrating how much copyrighted content is really out there,” says McInerney.
That inspired them to switch gears and start developing Johnny (named after the Keanu Reeves character in “Johnny Mnemonic”). “We needed a system that could identify and classify [copyrighted] video without human assistance,” McInerney says.
The centerpiece of the system is a huge database of digital fingerprints for copyrighted video. Each fingerprint is created using wavelet compression technology that distills the video signal into a few compact mathematical representations. It does the same for the audio track, and it uses computer vision technology to measure the frequency of scene changes, providing a kind of time signature. The compressed video and audio signals and time signature together make up the file’s fingerprint.
Johnny extracts a fingerprint from every video uploaded to Guba, and if it matches a fingerprint already in the database, the file gets quarantined and flagged for review by a human. The system is so effective, McInerney says, that only one percent of the flagged video files turn out not to be copyrighted.
Guba may eventually license Johnny to other video-sharing sites; but for now the technology – and the company’s commitment to copyright protection – are giving it an advantage in negotiating with networks and movie studios for the rights to sell downloads. Already, Guba is hosting downloads of full-length films from Warner and Sony.
Nevertheless, digital rights management experts aren’t convinced that fingerprinting technology will be a silver bullet against video piracy, even though it has been used successfully by the music business. One problem is simply the burgeoning amount of copyrighted content needing fingerprinting. “The universe of music tracks to be fingerprinted is relatively tractable, compared to copyrighted video clips such as every day’s newscasts on television networks and stations throughout the world,” writes Bill Rosenblatt, editor of Jupitermedia’s DRM Watch. “Updating such a large and fast-growing fingerprint database, and making it efficient enough to be used in the filtering of copyrighted material from a site like Guba, seems utterly impractical.”
But to survive, says Guba’s McInerney, video download services “are going to need to make efforts to scrub the copyrighted stuff off their sites – and we think the best solution is a technological solution.”