If perfected, face recognition technology could aid law enforcement with spotting known terrorists or other criminals walking through a mall or an airport, for example.
So far, however, the technology has not been perfected. Even the best systems available today only work well using photographs taken in good light without shadows.
How Face Recognition Tech Will Change Everything
But now researchers have developed a face recognition technology that works in utter darkness.
Computer scientists Saquib Sarfraz and Rainer Stiefelhagen from the Karlsruhe Institute of Technology, Germany, created a system that analyzes dozens of infrared images of a person's face and then compares them to dozens of images taken in daylight.
The comparisons are made with a computer program that works using a so-called deep neural network system designed to imitate the function of a human brain.
In a study published this week, Sarfraz and Stiefelhagen explain how the deep neural network analyzed 4,585 images taken in both infrared and visible light, and was able to establish a match in just 35 milliseconds.
"The presented approach improves the state-of-the-art by more than 10 percent," Sarfraz and Stiefelhagen told MIT Technology Review.
Pay For Stuff With Your Face
But like other face recognition systems, this one is not perfect either.
Like daylight images of a person's face, infrared images can also change with the environment. If it's hot outside or if the person just finished a session at the gym, the heat map of the face could look very different from when it does when the person is cool and relaxed.
In Sarfraz's and Stiefelhagen's study, the 4,585 images represented 82 people and although the speed of the computer was fast, it was only about 80 percent accurate and worked best when it had many visible light images to compare with the infrared. In cases where it had only one visible light image, the system accuracy dropped to 55 percent.
Scientists are never discouraged by these kinds of results, though, because a bigger database and a more powerful neural network are not impossible.
And, as Technology Review points out, "interested customers are likely to be the military, law enforcement agencies and governments who generally have deeper pockets when it comes to security-related technology.
via MIT Technology Review