July 28, 2015 -
Researchers at the Institute of Anthropomatics & Robotics and Karlsruhe Institute of Technology have developed a new facial recognition method that uses thermal signature to recognize faces in the dark, according to a report by Gizmodo.
In the new study, the researchers use a new method that matches a regular visible light image of a face with an infrared image counterpart of the face.
The researchers used a “deep neural network” — computer programs that are designed to emulate the thinking patterns of a human brain — to discover that there is little one-to-one correlation between the two types of images.
When a deep neutral network is trained with a significantly large data set it is able to create connections based on a complex series of factors, just as a human would.
In this case, there are only few existing datasets of visible light images with corresponding infrared images.
The researchers used a University of Notre Dame set that included groups of images which include individuals with different facial expressions, in various lighting conditions, along with many images of the same person over a period of time.
By using the neural network to find correlations, researchers were able to significantly improve the accuracy of face matching.
However, the researchers admit that they are still a long ways away from developing a truly reliable system.
In cases where the system had many visible light images to compare to the thermal image, the network made the correct match 80% of the time. On the other hand, when the system only used one visible image, the matching percentage dropped to 55 percent.
The researchers said the facial recognition method is mostly useful for covert surveillance.
When paired with existing visible image databases containing mugshots and driver’s license photos, the new technology should allow government agencies to identify individuals even when their face is not visible, such as in low light settings or even at night.