The content-recognition bakeoff


This article was originally on a blog post platform and may be missing photos, graphics or links. See About archive blog posts.

There was one obvious winner in the MPAA’s test of content-recognition technologies, and that would be Hollywood. The association disclosed a brief summary of the results at a conference on online entertainment held Thursday in cooperation with the University of California. Paradoxically, though, the results might prove to be a boon for user-generated content sites and the people who love them.

Here’s the background. The MPAA and MovieLabs, its R&D arm, spent months testing a dozen software programs designed to identify copyrighted videos from brief samples of their data. Each ‘fingerprinting’ program was fed about 1,000 test files of wide-ranging format and quality, including a number of camcorded movies and other bootlegs downloading from the Net. Of the technologies submitted (11 by firms and one by a Scottish university), the MPAA said, three successfully identified more than 90 percent of the files with no false positives. In other words, when they erred, it was in not spotting a copyrighted movie within a file, rather than identifying a movie that wasn’t there.


Steve Weinstein, ceo of MovieLabs, summarized the results as follows: ‘This stuff works.’ Of course, that’s just in a laboratory setting. The next step for MovieLabs and the MPAA, in addition to testing more iterations of the technology, is to try out the technologies in the field, where commercially important issues such as scaling (how well they handle a large volume of ID requests) can be measured.

Weinstein and other officials declined to say which firms had scored highest in the test. That honor was claimed by Santa Clara-based startup Vobile, although executives from two more established competitors, Gracenote and Audible Magic, also said their technology performed well. For the studios, though, the important thing is that several firms actually seem to be able to do what they claim. That could raise the pressure on networks with a large amount of unauthorized copying, such as user-generated video sites and college campuses, to deploy fingerprinting technologies. In fact, Joshua Metzger, a senior vice president at the UGC site, said at the conference that his company would add fingerprinting technology within 30 days.

A lot of details remain unsettled, and the courts have yet to weigh in on whether copyright law actually obliges anyone to install fingerprinting technologies on their system. But a fingerprinting technology that works well enough to satisfy the studios could be the key to advertiser-supported sites such as Veoh and YouTube striking major content deals and ramping up their ad sales. After all, one of the things holding back revenues at UGC sites is uncertainty over liability. If fingerprinting settles that issue, that’s a powerful motivator for companies to deploy it. On the other hand, if it chokes off so much content that users run away, then it’s a lose-lose situation.