8 billion individuals, so each face is one of them — this is an easier problem than “what did he say?”
Well, yes. Faces are mostly created by biology. Language is created by culture. Culture evolves faster and in more complex ways than biology. It is harder for computers to learn culture than to learn biology.
I have a useful test for speech recognition: Take the recordings of public domain books made by librivox.org and run them through your tool. Compare the output to the published public domain work.
Computers don’t care about any of those distinctions. Recording the inputs are a much more complex problem for faces. The pattern recognition part is much easier. Faces form a much more fixed pattern than speech does.
I wouldn’t know which recognition is ultimately easier, but the commenter is applying logic better suited to human thought than computerized operations.
They are not matching a billion faces to a billion faces. They are matching a billion to a few thousand on a watch list. So they know how the few thousand are different, making the match problem simpler.
Maybe 3d-printing prosthetic noses and the like to fool the facial scanners will become a thing.
“It took Chinese authorities just seven minutes to locate and apprehend BBC reporter John Sudworth using its powerful network of CCTV camera and facial recognition technology.
This wasn’t a case of a member of the media being forcibly removed from the country. The chase was a stunt set up to illustrate just how powerful and effective the Chinese government’s surveillance system can be. It’s a stark example of the type of monitoring that China has invested heavily in over recent years with the aim of helping police do their job more efficiently.
Such systems are also used in private organizations, for example to monitor workers and processes in factories, but government critics have warned of the potential for abuse in the hands of the state.
China has the largest monitoring system in the world. There are some 170 million CCTV cameras across the country, and that’s tipped to grow more than three-fold with 400 million more set to be installed by 2020.
Beyond the sheer numbers of lookout points, China is harvesting information with a new-found focus on intelligence. The government also works with facial recognition and AI companies, such as unicorn Face++, which can pour through data to extract meaningful information such as faces, ages, registration plates and more.
The full video of Sudworth’s ‘capture’ is on the BBC website, with a snippet is below — hat-tip The Next Web.”
https://techcrunch.com/2017/12/13/china-cctv-bbc-reporter/