Trueface, a biometrics firm, has a 98.5% accuracy rate on MegaFace’s 1 Million Face Challenge and a 99.85% accuracy rate on Labelled Faces In The Wild, according to the company’s CEO and Co-Founder Shaun Moore. The company works directly with law enforcement, government, healthcare agencies, schools, and other for-profit and non-profit organizations. Moore couldn’t go into detail about the company’s work with government agencies, although there will be an announcement within the next couple of weeks.
In schools, the technology is used to scan for unauthorized visitors, while regular students and teachers’ faces will be blurred, Moore said in an interview with Security Baron. The software can also detect knives, guns, or any other weapons in hopes of making schools safer.
The two biggest misconceptions about facial recognition technology regard its gender and racial biases as well as surveillance, Moore told Security Baron. When this technology was originally released in 2015, it was based on data skewed towards cis white men, resulting in lower accuracy rates for women, people with darker skin, and/or trans people. However, with diversification of photo databases, facial recognition technology has become more accurate for non-whites and women.
As far as surveillance goes, Moore describes facial recognition as “a tool in a tool box” for police, not something to make final decisions with. For example, a police officer could use facial recognition technology to verify the identity of a suspect, greatly narrowing down their suspect pool. Moore said to Security Baron,
“If you take away facial recognition from law enforcement, you’re reducing their ability to do their jobs. The conversation needs to be around regulation, not a ban of facial recognition technology”.
Not all facial recognition technology has the same accuracy rates, however. Amazon’s Rekognition, for example, could only classify 81.27% of females and 84.89% of darker subjects correctly, with the accuracy rate for darker females dipping at low as 68.63%. Like Trueface, Rekognition is being used by local police departments, specifically in Washington County, Oregon. However, without a more diverse data set, the software could make women, trans people and/ or darker subjects more likely to be misclassified.