Online video piracy is a big deal. Google’s YouTube, for example, is being sued for more than $1 billion by Viacom, a media company. But it is extremely hard to tell if a video clip is copyrighted, particularly since 24 hours of video are uploaded to YouTube every minute. Now a new industry standard promises to be able to identify pirated material with phenomenal accuracy in a matter of seconds.
The technique, developed by NEC, a Japanese technology company, and later tweaked by Mitsubishi Electric, has been adopted by the International Organisation for Standardisation (ISO) for MPEG-7, the latest standard for describing audio-visual content. The two existing methods do not do a very good job. One is digital “watermarking,” in which a bit of computer code is embedded in a file to identify it. This works only if content owners take the trouble to affix the watermark—and then it only spots duplicates, not other forms of piracy such as recording a movie at a cinema. A second approach is to extract a numeric code or “digital fingerprint” from the content file itself by comparing, say, the colours or texture of regions in a frame. But this may not work if the file is altered, such as by cropping or overlaying text.
NEC’s technology extracts a digital signature that works even if the video is altered. It does this by comparing the brightness in 380 predefined “regions of interest” in a frame of the video. This could be done for all or only some of the frames in a film. The brightness is assigned a value: -1, 0, or +1. These values are encapsulated in a digital signature of 76 bytes per frame.
The beauty of the technique is that it encompasses both granularity and generality. The 380 regions of interest are numerous, so an image can be identified even if it is doctored. At the same time, the array of three values simplifies the complexity in the image, so even if a video is of poor quality or a different hue, the information about its relative luminance is retained. Moreover, the compact signature is computationally easy to extract and use.
Nec says the system could be used to automate what is currently a manual procedure of checking that video uploaded to the internet is not pirated. The technology is said to have an average detection rate of 96% and a low rate of false alarms: a mere five per million, according to tests by the ISO. It can detect if a video is pirated from clips as short as two seconds. And an ordinary PC can be used with the system to scour through 1,000 hours of video in a second. There are other potential uses too, because it provides a way to identify video content. A person could, say, use the signature in a clip to search for a full version of a movie. Piracy will still flourish—but the pirates may have to get smarter.