Re: face recognition flop why do I not believe this

[David Farber <dave () farber net>]

Begin forwarded message:

From: Tom Goltz <tgoltz () quietsoftware com>
Date: July 26, 2007 11:13:39 AM EDT
To: dave () farber net
Cc: Brian Randell <Brian.Randell () ncl ac uk>
Subject: Re: [IP] face recognition flop why do I not believe this

At 08:20 PM 7/25/2007, you wrote:
> While the media got it wrong including the BKA's own president Jörg
> Ziercke - from a scientific perspective - these tests were a success
> With minimal human interference - false positive ratio was kept
> quite small - 0.1 percent

I don't believe it either.

Let's do the numbers.  If you have a stadium that holds 50,000  
people, and you screen all of them with this technology, at a false- 
positive rate of 0.1%, you'll get 50 false positives.  Assuming that  
you detain all 50 for further investigation, and some significant  
percentage of those people are members of parties who choose to  
remain with the detained person, you'll be detaining and handling  
between 100 and 200 people, most of whom will NOT be happy about the  
situation they find themselves in.

The article goes on to suggest that the solution to this high false- 
positive rate is to have a human double-check the match before  
sounding the alarm.  The catch with this is that unless you're going  
to design a fairly long gauntlet to run everyone through single-file,  
you won't have TIME for a person to confirm the match before the  
subject is lost in the crowd, unless you detain the suspect person  
("sound the alarm").

What also is not mentioned is the false-negative rate associated with  
this test.  In all of the realistic field tests I have seen to date  
of face-recognition technology, the only way they've managed to get  
the false-positive rate down to one-tenth of one percent was to set  
the match requirements so tight that the false-negative rate  
skyrocketed to such high levels that the system became essentially  
useless.  The article claimed that both the false-negative and false- 
positive rates were at 0.1%, but neglected to mention if both were  
held to that low rate at the same time with the same settings.

Sporting events and mass-transit stations are particularly  
problematic for computerized facial recognition technology, as  
depending on the season you'll need to get your subjects to  
cooperatively remove their hats, scarves, and sunglasses and all look  
into the camera under consistent lighting conditions.

The only place I'm aware of facial recognition technology has met a  
measure of success has been in casinos, where the indoor controlled  
environment, the large number of cameras, basically stationary people  
and large security forces has resulted in usable results.




-------------------------------------------
Archives: http://v2.listbox.com/member/archive/247/=now
RSS Feed: http://v2.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com