Facial recognition algorithms are good at identifying people in a crowd, but they struggle with age.
According to New Scientist, facial recognition algorithms don’t handle wrinkles and natural feature changes well. If no new face photographs are recorded, it only takes five years before they have trouble identifying people.
This is the finding of Norwegian University of Science and Technology researchers lead by PhD candidate Marcel Grimmer. AI-generated human faces were aged to test facial recognition systems.
More From Us:Oracle Sued For Tracking 5 Billion People
Open-source facial recognition techniques were employed because commercial systems don’t reveal their algorithms. The team found that algorithm accuracy decreased with age, with five years being the threshold at which identification failed. They had little hope of success after 20 years.
Age affects facieal recognition accuracy. Because ageing progresses so quickly, it’s harder to identify those under 20 or beyond 60. Grimmer says, “Babies change in two months, therefore a new snapshot every month would fail. Faces alter even till age 20.” Over 60s “the head shape changes again, and you have more pronounced wrinkles,” thus the failure rate is increased.
The easy answer is to capture new photos of each face regularly. How realistic that is depends on how many people a facial recognition system monitors and how privacy regulations change. Getting older isn’t often considered as a plus, but it could preserve your privacy.