Facial Recognition Database Used By FBI Is Out of Control, House Committee Hears (theguardian.com)

The House oversight committee claims the FBI's facial recognition database is out of control, noting that "no federal law controls this technology" and "no court decision limits it." At last week's House oversight committee hearing, politicians and privacy campaigners presented several "damning facts" about the databases. "About 80% of photos in the FBI's network are non-criminal entries, including pictures from driver's licenses and passports," reports The Guardian. "The algorithms used to identify matches are inaccurate about 15% of the time, and are most likely to misidentify black people than white people." From the report: "Facial recognition technology is a powerful tool law enforcement can use to protect people, their property, our borders, and our nation," said the committee chair, Jason Chaffetz, adding that in the private sector it can be used to protect financial transactions and prevent fraud or identity theft. "But it can also be used by bad actors to harass or stalk individuals. It can be used in a way that chills free speech and free association by targeting people attending certain political meetings, protests, churches, or other types of places in the public." Furthermore, the rise of real-time face recognition technology that allows surveillance and body cameras to scan the faces of people walking down the street was, according to Chaffetz, "most concerning." "For those reasons and others, we must conduct proper oversight of this emerging technology," he said.

Comments

Re:which states?

[Anonymous Coward]

Alabama
Arizona
Arkansas
Delaware
Illinois
Iowa
Kentucky
Maryland
Michigan
Nebraska
New Mexico
North Carolina
North Dakota
South Carolina
Tennessee
Texas
Utah
Vermont

there is a reason for that

[Gravis Zero]

... are most likely to misidentify black people than white people.

Before you jump to conclusions about "racist software," I can tell you that the reason for this is very well known and understood: lighting and contrast ratios. Specifically, you get a much higher contrast ratio of faces with light skinned people from image sensors than do you for dark skinned people unless you have their face very well lit up. Simply put, camera sensors are a poor substitute for the human eye and this is one of the side-effects of that.

The reason for the racial misidentification

[Zibodiz]

The algorithms generally tend to identify an individual based on relative (i.e. comparison-based) facial structure. In order to have a facial structure to measure, it must identify the eyebrows, eyes, nose, jawline, hairline, etc. The way it identifies these features is based in contrast from one element to another; when hair and skin color are similar (such as with a platinum blond like me, or with a dark-haired, dark-skinned individual), there's a lot less contrast there to measure.
Furthermore, with darker skin, shadows aren't as noticeable, so the shadows that would otherwise allow measurement of eyebrow prominence or jawline will also be much harder to identify.