NSW Police face questions on use of facial recognition algorithm from 2011

Australia’s New South Wales Police used a legacy facial recognition algorithm for over a decade and declined repeated offers of updates from the vendor, instead replacing it with a system adapted in-house from open-source code.

New information indicates that the decision to discontinue the use of the algorithm may be based largely on cost considerations, contrary to prior reports.

The PhotoTrac Suspect Identification System (SIS), which is used by State Intelligence Command’s Facial Recognition Unit and NSW Police’s Real Time Intelligence Centre, used a biometric algorithm from Cognitec. NSW Police used the Cognitec software from 2011 through the beginning of this year, despite having never contracted technical support or updates for the software, the company tells Biometric Update in an email.

The facial recognition system began using an algorithm known as NYX, adapted from open-source facial recognition model FaceNet, in 2018. FaceNet was originally developed by Google researchers back in 2015.

Media reports and a Green MP have linked the decision to stop using the legacy algorithm to the discovery that the accuracy of facial recognition systems differs based on demographics. Cognitec’s representative told Biometric Update in the email that NSW Police never complained directly to the company about any problems related to bias. While independent figures for the demographic differentials of Cognitec’s 2011 algorithm are not available, from a 2018 update to 2024, the company’s false match rate for the demographic group with the most errors, West African females 65 years and over, fell from 1.9 percent to 0.5 percent in NIST testing.

Cognitec is supplying facial recognition algorithms bundled with other software and cameras to Australia’s major international airports.

Article Topics

algorithms  |  Australia  |  biometrics  |  Cognitec  |  FaceNet  |  facial recognition  |  New South Wales (NSW)  |  open source  |  police