German scientists have developed a deep neural network that enables facial recognition for infrared surveillance video, matching blurry thermal images with those captured in visible light.Saquib Sarfraz and Rainer Stiefelhagen at the Karlsruhe Institute of Technology say the innovation lies in training a network to recognize visible light faces by looking at infrared versions.In “Deep Perceptual Mapping for Thermal to Visible Face Recognition”, they say the technique could prove crucial for law enforcement using night-time or low light surveillance footage where the image is captured discretely or covertly through active or passive infra-red sensors.”Thermal-visible face recognition is a very difficult problem due to the inherent large modality difference. Our presented method is very effective and has benefits for many related applied computer vision and domain adaptation problems from image matching, detection to recognition,” write the researchers.To run the comparisions, the team used 4,585 images of 82 people taken either in visible light at a resolution of 1600 x 1200 pixels or in the far infrared at 312 x 239 pixels. The data set contains images of people smiling, laughing and with a neutral expression taken in different sessions to capture the way people's appearance changes from day to day, and in two different lighting conditions.Overall, the team improved the best published state-of-the-art performance on the “very challenging dataset” by more than 10%. “Our results show that this accounts for bridging the performance gap due to modality difference by more than 40%.”

Also in the news:

Default ThumbnailFingerprints reports increased earnings Default ThumbnailKenya’s biometric election hits henna snag Default ThumbnailValid USA wins biometric contract for Washington driver’s license