U.S. Army researchers develop method for matching infrared facial images to existing databases
U.S. Army researchers have developed a technique for converting a thermal image of a person’s face to a visual one that can be compared with an existing image by a human or automated system. The method leverages artificial intelligence and machine learning to enable facial recognition in covert low-light or nighttime conditions, R&D reports.
“This technology enables matching between thermal face images and existing biometric face databases/watch lists that only contain visible face imagery,” said Dr. Benjamin S. Riggan, a member of the team of scientists from the U.S. Army Research Laboratory that developed it. “The technology provides a way for humans to visually compare visible and thermal facial imagery through thermal-to-visible face synthesis.”
A technical paper titled “Thermal to Visible Synthesis of Face Images using Multiple Regions” describing the technique was presented at the IEEE Winter Conference on Applications of Computer Vision in March. As part of the presentation, the researchers demonstrated a proof-of-concept using a FLIR (Forward Looking Infrared) Boson 320 thermal camera and a laptop, with the algorithm converting images in close to real-time.
The researchers performed face verification with an open source deep neural network architecture designed for visible face recognition, and found that their approach performed better than a generative adversarial network-based approach which had shown photo-realistic properties, according to R&D.
The research is supported by the Defense Forensics and Biometrics Agency.