SAFR reports biometric recognition algorithm adapts to work with masks

Company training tech for higher accuracy with occluded faces
SAFR reports biometric recognition algorithm adapts to work with masks

Performance accuracy of SAFR biometric facial recognition technology from Real Networks is not affected by people wearing masks to protect themselves against the spread of COVID-19, explains Eric Hess, senior director of product management for Face Recognition & Security Solutions in a company blog post.

Hess says the algorithm can easily adapt to the situation thanks to the occlusion detection feature which is included in the SAFR platform. The company is now focusing on improving occlusion logic to enable high performance and accuracy even in the toughest scenarios. The technology could then be easily deployed for healthcare workers moving from one area to another, and essential service providers and potential security threats can still be identified.

When an exact match is not possible, there are procedures and security protocols that can be followed such as multifactor authentication. Hess warns that accuracy depends on sample size and the range of facial traits. The more information a sample contains, the more chances for similar features to be detected.

Some people are easier to recognize based on the unique features in the upper half of their face, while others are recognized based on the unique features around their mouths or jawline. This would affect accuracy rates when that part of the face is covered. SAFR is working on training the algorithms to maintain high accuracy despite occlusions.

“Yes, we too can match faces when people are wearing masks, but we feel a responsibility to be honest about the limitations, and even better performance can be achieved when SAFR understands how a person may appear when wearing a mask,” Hess writes.

Accuracy depends on image quality, user cooperation in terms of image capturing, lighting and environment. “SAFR’s current capability on the University of Massachusetts Labeled Faces in the Wild (LFW) dataset is a 99.87 percent True Identification Rate with just 1:1,000,000 False Identifications (false positives),” he adds.

Based on internal benchmarking, if the subject is wearing a mask, the true positive identification rate is 93.5 percent, with less than 1:3,760 false identifications, but the company wants to keep improving it.

Eric Hess is a biometrics industry expert who was appointed SAFR Senior Director of Product Management in February. He is an expert in the use of video analytics and facial recognition technologies for criminal investigations, surveillance, loss prevention, and identity solutions.

Related Posts

Article Topics

 |   |   |   |   |   |   |   | 

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Most Read This Week

Featured Company

Biometrics Research

Biometrics White Papers

Biometrics Events

Explaining Biometrics