NSW Police abandons Cognitec’s facial recognition

New South Wales police have deactivated facial recognition capabilities in their suspect identification system following concerns about its accuracy.

The facial recognition software developed by German facial recognition company Cognitec was disabled within the PhotoTrac Suspect Identification in February, the police said in a response to a Budget Estimates inquiry by the New South Wales (NSW) Parliament. The identification tool compares images of potential suspects caught on CCTV against mugshots and is part of the Suspect Identification System (SIS) used by the NSW Police Force.

The police did not list a reason for deactivating Cognitec’s algorithm, which has been in use since 2011. The move, however, comes after local media uncovered that Cognitec’s technology misidentified West Africans almost seven times more often than it did Europeans in NIST testing performed in August last year. The Face Recognition Technology Evaluation (FRTE) Demographic Differentials report shows the highest false match rate (FMR) for Cognitec’s algorithm among all demographic groups tested is 0.538 percent for West African females age 65 to 99.

The NSW Police Force uses a variety of facial recognition systems. In the Budget Estimates response, Police and Counter-terrorism Minister Yasmin Catley clarifies that Cognitec has released numerous updates to the product since its deployment, but the police did not purchase them. As with other developers, Cognitec’s legacy algorithms have higher error rates for various demographic groups.

In 2018, the NSW Police Force adopted an existing open-source facial recognition system called FaceNet for use by the Facial Recognition Unit, internally branding it as NYX, she adds.

The response also revealed that the police force has used facial recognition to generate leads for investigation more than 560 times during the financial year 2023-2024 and more than 370 times from the start of the current financial year.

Last year, NSW Police Deputy Commissioner David Hudson told Budget Estimates that the police do not rely solely on the technology for their work.

“We don’t use facial recognition or facial matching services as the only evidence that we will use to charge someone,” says David Hudson. “We use it as an investigative tool to give us an indication of matching, say, a CCTV photograph with our offender photographs.”

Questions raised over police use of video analytics

Aside from its use of Cognitec’s algorithms, media reports have also raised other questions related to the NSW police’s use of video analysis technology.

The police force, for instance, has been using Microsoft’s AI-powered video analytics platform Insights, which includes features such as vehicle recognition, text-to-speech, object recognition, video stitching and biometric recognition – but does not use facial recognition.

NSW Police Force has been experimenting with the AI tool for crime prevention in predictive policing and video surveillance. In a blog post published by Microsoft in 2021, Gordon Dunsford, CITO and Executive Director of Digital Technology and Innovation for NSW Police, explains that the system has helped police speed up investigations.

“Using computer vision, it can search to recognize objects, vehicles, locations, even a backpack someone has on their back or a tie a gentleman is wearing,” says Dunsford.

In a November 2021 meeting with the police, the AI Review Committee (AIRC) raised concerns that Insights’ surveillance feed could over-represent people who frequent areas with higher crime rates, making them more likely to become suspects.

The state government-appointed body also recommends that guidelines be set up for the secondary use of Insights data in crime prevention, The Saturday Paper reports.

Article Topics

accuracy  |  Australia  |  biometric identification  |  biometrics  |  Cognitec  |  criminal ID  |  facial recognition  |  New South Wales (NSW)  |  police