I'll Be Watching You
Government Computer News reports that improved computer face recognition algorithms have reduced the false recognition rate 200 fold between 2006 and 2002. "In the 2006 test [sponsored by the FBI and Homeland Security] ... the accuracy of face recognition software was documented to exceed that of humans."
The new algorithms exploit the geometric signature of the human face and the ability to read micro-patterns -- swatches of skin display a structure of pores and texture -- rather like a fingerprint. Now you can look forward to a future where cameras can scan your face to the entrance of every stadium, theater, venue and public building that a network can reach. Yipee? Well as the Police (the band not the cops) once put it:
Every breath you take
Every move you make
Every bond you break
Every step you take
I'll be watching you.
Every single day
Every word you say
Every game you play
Every night you stay
I'll be watching you.
posted by wretchardthecat at 6/05/2007 06:02:00 PM
7 Comments:
This will be great for people who need alibis.
"See, honey, the surveillance camera will show you I was at the baseball game at the time. That fling with Samantha is over..."
Seriously, it should be possible for someone who understands biometric technology to create a fake facial profile for the purpose of fooling biometric scans. It's a means for a criminal to create false positives to make it appear he is one place when he is really somewhere else.
I'm reminded of an episode from Ballykissangel where artificial sheep were put out to pasture to fool the EU's sheep counting satellite.
Looks like the art of "Make-UP" as practiced by Hollywood and the gals in the neighborhod is about to take on a whole new dimension.
Can't you just see artificial skin pore patterns sold in see through microthin polymer sheets being all the rage on the shopping channels late at night and laws being passed against their unlawful distribution and a whole underground economy abuzz with brown paper bags full of the stuff.
The "New World Order" is on its way.
Oh Dear!
Technology can be used by good people to catch bad people, and it can be used by bad people to catch good people.
As I look at the world today, I wonder if maybe we should embrace technology and just have it out between us, anything else is just capitulation to bad people.
From the current issue of Government Computer News:
“The FBI’s biometric technologists have consulted with their counterparts abroad to develop regional biometric information repositories…”
“For example, some Middle Eastern countries seek to build a regional biometric database of criminals and other social enemies, and the Bureau has advised them…”
Shucks,… by the time this is all done, Sara Conner won’t even be able to run to Mexico…
Wretchard:
A 200-fold reduction in anything is a mathematical absurdity, as is any factor greater than 1x (when speaking of reductions). One worships your intelligence and clarity but is relieved to see that you can make a mistake, too, now and then, in terms of what you post/edit.
Aslam,
Maybe my arithemetic is off or the source is wrong. The actual claim is:
The results of the test, released in March, showed improvements in recognition accuracy of an order of magnitude, or 10 times better than in the previous test in 2002.
The FRVT 2006 results showed a false recognition rate (FRR) of only 0.01 while maintaining a false acceptance rate (FAR) of 0.001. That compares to an FRR of 0.2 in 2002.
My calculation was FRR 2006/FRR 2002 or .01/.2 =.05 or 20 times. I pushed one decimal point too far in doing the arithmetic. But maybe I'm getting it wrong in that the text claims only a tenfold improvement.
Wretchard:
First, yours is without doubt my favorite blog.
So, for those who are listening in, my comments are offered in the spirit of continuous learning rather than a cheap salvo in a puerile debate.
A 20-fold improvement in the FRR means a 95% reduction in that error rate just as a 2-fold improvement in the FRR would have meant a 50% reduction in the error rate.
The operative word/concept is REDUCTION. When a figure (a ratio, a stock price, whatever) falls from 0.2 to 0.01 then, mathematically, the REDUCTION in the thing being quantified is 95%.
Presuming the data are correct, one could legitimately say that things in 2002 were 20x times worse than they were in 2006.
That's fair.
But, it is mathematically incorrect to simply invert that and say that the FRR in 2006 was 20x lower than it was in 2002.
It was not. It could not be.
It was 95% lower. (1 - (1/20)).
Post a Comment
<< Home