NIST launches ongoing face recognition test to evaluate accuracy
20 January, 2017
category: Biometrics, Corporate, Government
In February, NIST will launch a new evaluation of face recognition systems. Called the Face Recognition Vendor Tests (FRVT), the goal is to provide independent evaluations of commercially available and prototype face recognition technologies. Using the results, government and law enforcement agencies should be better able to determine where and how to deploy facial recognition.
Unlike prior NIST face testing that occurred every few years – most recently in 2013 – this project will be ongoing, enabling vendors to submit products for evaluation at any point and have algorithms tested on a first-come-first-served basis.
Results will be posted to the NIST website as each is completed.
The FRVT will evaluate face recognition technologies applied to 2D still images. The evaluation will use very large sets of facial images to measure both commercial and academic algorithm performance. Each image will contain only one face, and systems will be graded on accuracy, speed, storage and memory consumption, and resilience.
According to NIST the aim is to assess, “performance of automated face recognition technologies applied to a wide range of civil, law enforcement and homeland security applications including verification of visa images, de-duplication of passports, recognition across photojournalism images, and identification of child exploitation victims.”
The initial assessment is geared solely around face verification, but the program will extend to other areas in the future, including: evaluations of open-set one-to-many identification accuracy, face detection accuracy, and also age and gender estimation.
Last year, NIST tested face recognition systems applied to video images rather than the still images that will be used in the FRVT. In the video test, called Face-in-Video-Evaluation (FIVE), the effort measured a system’s ability to correctly identify or ignore persons appearing in video sequences.
The goal was to determine viability in operational use-cases including:
- High volume screening of persons in the crowded spaces (e.g. an airport);
- Low volume forensic examination of footage from a crime scene (e.g. a convenience store);
- Persons in business meetings (e.g. for video-conferencing); and
- Persons appearing in television footage.
The final FIVE report is currently under review and expected to be published later in January 2017.