SAFR Details Face Mask Performance of Facial Recognition Platform

Amid the COVID-19 pandemic, SAFR is working to make its facial recognition algorithm more effective at identifying people with masks – and seeking to be transparent with clients about what masks mean for its technology’s accuracy.

It’s all detailed in a new blog post on the RealNetworks business’s website, written by Senior Director of Product Management for Face Recognition & Security Solutions Eric Hess. To cope with the growing preponderance of face masks, Hess asserts, “we’re enhancing our occlusion logic even further to ensure we can maintain the highest accuracy and best performance under challenging conditions.”

In the meantime, he says, the SAFR facial recognition platform already has a feature called “Grouping” that lets administrators include multiple profile images for a single registrant. This means that if a company uses SAFR for employee recognition, each employee can be registered with one normal face image, and one in which they wear a mask, helping the system to maintain accuracy in the latter case.

Hess also notes that the SAFR platform is already quite accurate when it comes to face masks as it stands. Noting that SAFR attained a True Identification Rate of 99.87 percent in the University of Massachusetts Labeled Faces in the Wild dataset, with a false positive rate of one in a million, Hess says that internal testing with fully cooperative subjects wearing face masks yielded a positive identification rate of 93.5 percent, and a false positive rate of 1:3,760.

That should prove reassuring to the growing number of organizations looking to take advantage of the benefits of contactless biometrics as social distancing practices are embraced around the world.

–

April 9, 2020 – by Alex Perala