Computers Outperform Humans at Recognizing Faces

seven of five writes "According to the recent Face Recognition Grand Challenge, The match up of face-recognition algorithms showed that machine recognition of human individuals has improved tenfold since 2002 and a hundredfold since 1995. 'Among other advantages, 3-D facial recognition identifies individuals by exploiting distinctive features of a human face's surface--for instance, the curves of the eye sockets, nose, and chin, which are where tissue and bone are most apparent and which don't change over time. Furthermore, Phillips says, "changes in illumination have adversely affected face-recognition performance from still images. But the shape of a face isn't affected by changes in illumination." Hence, 3-D face recognition might even be used in near-dark conditions.'"

Comments

Face the Consequences

[Doc Ruby]

When a human makes a mistake recognizing a face, they suffer the results. If that's identifying a criminal, they can be cross-examined, or even sued or jailed, depending on what they said that face did.

When computers mis-ID a face, do we cross-examine and maybe punish its programmers?

Re:Great, now commercialize it..

[Original Replica]

It is being turned into product. Has you flown into the US from overseas recently? They have (and use) fingerprint scanners, cameras and facial recognition software running in US Customs. Sure right now only people with foreign passport have to scan in and back out when they go through customs, but the cameras are right in your face in the US passport lines as well. This may be very new, I first saw them this last Monday in JFK. But apparently this has been going on for awhile.

I'm lucky

[Tribbin]

It's a question of time 'till there 's a law that forbids to wear anything that partially covers your face in certain public areas.

I think I have about ten years 'till computers are able to interpret my front-head as a 'face' so I'm safe.

Quite impressive..

[denoir]

This is actually a great milestone as we humans are really excellent at face recognition. In fact, we are so good at it that we produce tons of false positives and recognize faces where there are none (clouds, toasts etc).

A few years back (well, nearly a decade actually), I did my master's thesis in a lab that among other things did work on face recognition. The experts there assured me that perhaps in 50 years or so computers might be able to approach human face recognition capabilities. Apparently the development was far quicker than they could have imagined.

An interesting technical point is that in fact the algorithms haven't changed a lot since then - it's still mainly various adaptive systems such as neural networks and support vector machines. The really big breakthrough is in the data collection - in the sensors and scanners. What they couldn't imagine a decade ago was the type of accurate automatic 3d face modeling and measurements that can be done today. It's also how certain computing methods that were deemed unsuitable a few years ago are coming back big time (neural nets for instance). I guess the time wasn't ready for them the last time due to computing power and memory limitations (and of course sensors as in this case).

Automating Go

[handy_vandal]

Go is not yet as well-automated as chess, but it appears that go-playing software is rapidly advancing:

"Two Hungarian scientists have now come up with an algorithm that helps computers pick the right move in Go, played by millions around the world, in which players must capture spaces by placing black and white marbles on a board in turn.

"On a nine by nine board we are not far from reaching the level of a professional Go player," said Levente Kocsis at the Hungarian Academy of Sciences' computing lab SZTAKI.

The 19 by 19 board which top players use is still hard for a machine, but the new method is promising because it makes better use of the growing power of computers than earlier Go software."

Link

See also:

http://zaphod.aml.sztaki.hu/papers/ecml06.pdf

http://citeseer.ist.psu.edu/vanderwerf03solving.ht ml

http://www.primidi.com/2007/02/26.html

-kgj

Caricatures

[Philotic]

I'll be impressed when they can recognize caricatures as well as humans.

Other Race Effect

[fishbowl]

One of the academic research areas I've been involved in, is study of the so-called "Other Race Effect". There is some evidence that people have quantifiable error when asked to identify faces of people of other races than their own.

Computers won't be subject to this.