id3 facial recognition shows combination of accuracy and low resource use in NIST testing
Facial recognition algorithms from id3 Technologies have scored respectable accuracy with a small template size and without benefit of a GPU in the NIST Facial Recognition Vendor Test (FRVT).
A pair of proprietary id3 algorithms based on deep convolutional neural networks were found to have template sizes among the twenty smallest of 100 algorithms, and also placed in the top half in matching time. The id3-002 algorithm achieved the seventh-best false non-match rate in the webcam and selfie categories, while the id3-003 algorithm was found to be in the top 40 for FNMR in the visa 0.0001 and mugshot categories. The id3-003 algorithm successfully enrolled all images in several categories, while id3-002 placed anywhere from 25th to 110th in failure to enrol rate.
“More than just a technology provider, id3 Technologies’ goal is to build strong partnership with its customers and to support them in the development of innovative solutions,” comments id3 Technologies President Jean-Louis Revol.
The company has deployed its facial recognition technology in a large-scale public safety system to detect intrusions in critical areas, and says its technology enables unconstrained real-time face detection and recognition within a crowd regardless of subject gender, age, or race.
Earlier this year id3 was granted biometric operator status by Columbia’s National Civil Registry office, and opened a regional office in the country.
algorithms | biometrics | facial recognition | id3 Technologies | NIST