Facial Recognition for verification *only*

In 1997, I was the lead programmer for an OTC DL/ID (over-the-counter driver license and ID) 
system delivered to the State of West Virginia.  Each "capture station" was equipped with a
video 
camera, fingerprint camera, and signature pad.  WV residents could refuse to give a
fingerprint 
when getting a license, so we couldn't depend on fingerprint minutiae to confirm indentity.
This 
was the first time that a state-level OTC DL/ID system used biometric information as a way to 
make an *issuance decision* rather than just confirm identity.

We generated an identification "template" from each facial image captured by the video camera.

When the same person came to get another document, we compared the new template against 
the stored template and used the result to either grant or deny issuance of the license.  Both

images were displayed on the screen so the operator could also see the current and "new" 
images.  Operators couldn't override the system in cases where the template comparison 
indicated "no match".   Local managers *could*, and a record of every override was uploaded to

the central storage location in the capital.

There was only one goal: to try to cut down on fraudulent license issuance.  The original
system 
issued to WV did not have the capability to search through images and return demographic data.

Nor did it have the ability to display multiple historical images: only the "current image on
file" 
could be displayed.  Hence, it was intentionally limited in capabilities.

Our image template was generated on an image taken at a specific distance from the camera, 
and the image was captured against a specific colored background.  The same camera (or 
model), the same background, and the same distance from the camera (size of face within the 
image boundaries) had to be *identical* to the previous image.  If any of the specifications 
differed, the resulting template wouldn't produce a suitable match against the template on
file.

And, even with all of those limitations, we still had a significant false-positive and
false-negative 
error rate.

Knowing what I knew then, and having closely followed the industry, I can say that we are 
probably still decades away from practical biometric "face in a crowd" identification systems.

And, knowing what I do about how easily some of these mechanisms can be either faked or 
spoofed, I would also require *multiple* biometric identifications for legal identification.
Some 
combination of multiple fingerprints, palm print, retinal check, signature, and facial
template - 
at least 3 - should be required for legal (non-passport) purposes.