Researchers claim to have overcome one of facial recognition’s biggest hurdles: identifying people who have disguised their faces.
Donning a pair of sunglasses, cap or scarf is enough to fool conventional systems. But a new paper suggests a neural network can see past such obstacles.
Amarjot Singh, a researcher at Cambridge University, says the software is now 78 per cent accurate: “The system is not perfect, but it’s a proof of concept.”
Singh and his team at Cambridge and the National Institute of Technology in Warangal, India fed the network 4,000 photographs of disguised faces.
The images, which were captured for the project, were annotated to identify 14 markers of facial features, including parts of the eyebrow, eye, nose and lips.
The database consists of photographs of Indian and caucasian people, but Singh’s team is hoping to expand this to all ethnicities by the end of the year.
Singh is hopeful that the system will reach 90 per cent accuracy within the next three or four months. Once it has, the team plans to tout it law enforcement.
Facial recognition has become a popular tool for forces policing large public events. It was deployed at Notting Hill Carnival for the second time this year.
It’s also popular with social media networks and retailers. In 2015, a survey of 150 retail executives found that a quarter of shops in the UK used the technology.