NIST rolls up its sleeves to test facial recognition used for biometric entry and exit program
The federal government’s chief technology standards body is working with the U.S. Customs and Border Protection agency to test facial recognition software used in its biometric entry and exit program.
The National Institute of Standards and Technology, or NIST, will perform full operational testing of the border agency’s facial recognition process, including an upgraded version of the NEC algorithm now used at airports and an unidentified number of land crossings.
NIST also has agreed to perform ongoing tests of facial recognition algorithms for the CBP.
The developments are happening downstream from a U.S. House of Representatives committee meeting last week called to examine the use of the technology at the nation’s boundaries.
Three government experts testified before the homeland security committee about perceived threats the biometric technology poses to the civil rights of U.S. citizens through possible racial or gender bias.
One of the witnesses, John Wagner, deputy executive assistant commissioner of the border agency’s field operations, said at the hearing that although an older iteration of NEC’s algorithm has been used to date, results have met expectations.
Of the 44 million people scanned by the system, only two percent to three percent have not been matched to photographs that people had previously submitted to the government for identification documents. (It has spotted 252 people fraudulently using documents, include 75 stolen, authentic U.S. papers.)
Not matching means that the software could not find a match, not that anyone was erroneously matched with another person’s photograph. Wagner would not rule out that a “handful” of false-positives occurred, but he said he was unaware of them.
The Congressional panel could not say if missed matches were the result of bias, technical problems or other possible factors.
When that has happened, Wagner said, border-patrol agents have asked the individuals for their passport to be sure the person was who they said they were.
The software being used will be upgraded to the so-called NEC-3 algorithm in March, a product that NIST rated among the best in terms of bias after examining 189 facial-recognition algorithms written by 99 software developers.
Until now, NIST personnel have not looked at how the border agency’s program is working. They will examine how face scanning functions, including the how well the NEC-3 algorithm operates, in the field.
There are many factors that can cause a face to not be recognized, including lighting, inferior camera equipment, the size of the gallery against which a scan is compared and others. NIST will be looking at these and possible demographic — race, gender, etc. — possibilities on an ongoing basis.