“The atmosphere moves and shifts around and your image gets sheared and blurry. Applying my heat dispersal model gets rid of that turbulence and brings it back closer to a focused, sharp image.”
A pair of researchers have managed to drastically improve the performance of a facial recognition algorithm for the Australian Department of Defense. The study was specifically concerned with facial recognition in adverse conditions, including long distances and extreme darkness.
The research was conducted by Sau Yee Yiu and Dmitri Kamenetsky, who both work for the DoD’s Department of Science and Technology. After reading the existing literature on the subject, Yiu and Kamenetsky came up with a heat dispersal theory about the distortion of images over long distances. They then applied a series of long lenses and algorithmic filters to improve the quality of those images.
“The atmosphere moves and shifts around and your image gets sheared and blurry,” explained Yiu. “Applying my heat dispersal model gets rid of that turbulence and brings it back closer to a focused, sharp image.”
By doing so, Yiu and Kamenetsky were able to identify people’s faces at distances up to 250 meters, and in dark tunnels where researchers were barely able to see their own hands. The algorithm can be tailored to different environments through the use of a set of sliders.
The news obviously has major implications for anyone deploying facial recognition for surveillance purposes. The Australian government has previously pushed for a more sweeping facial recognition database, although those plans have been derailed in recent years. NEC Australia recently took the government to court after a major biometric program was unceremoniously cancelled in 2018.