Better Facial Recognition, Says NIST

Interesting stats.   Our own experience says it depends strongly on the contextual aspects of the image acquisition.  And often out of complete control.    Introduce the metadata of context?  But the improvement continues.

Facial Recognition Algorithms Are Getting a Lot Better, NIST Study Finds   in FedScoop   By Tajha Chappellet-Lanier

The U.S. National Institute of Standards and Technology (NIST) determined facial recognition software has made huge gains in accuracy over the past five years. NIST said the technology has undergone an "industrial revolution," making certain algorithms about 20 times better at searching databases and finding matches. NIST researchers tested 127 algorithms developed by 45 vendors, using a primary database of 26.6 million reasonably well-controlled portrait photos of 12.3 million individuals; when provided with good quality photos, the most accurate algorithm could identify matches with only a 0.2% error rate. The same test found at least a 4% failure rate in 2014, and a 5% failure rate in 2010. NIST said this improvement can be attributed to the widespread adoption of convolutional neural networks, which were not being used in 2014.  ... "