• What makes infrared facial recognition hard is that there's no linear relationship between the way someone's face looks in visible light and the way they look in infrared light. Often, military or surveillance cameras take infrared images, offering a logical application for the technology.
  • One challenge is that variables like the temperature of the air or skin can greatly affect infrared images. Plus, infrared images are often much lower resolution than standard photos or video. The researchers used a data set composed of 4,585 images of 82 different people taken in both visible light and infrared light.
  • The method isn't perfect yet. Its accuracy is over 80 percent when it has a wide range of previous data, but the approach only "improves the state-of-the-art" by 10 percent.

Share This Article