Face mask popularity prompts new biometric datasets for sale, FaceID work around in China
Datasets consisting of images of people wearing facemasks are being sold online for miniscule prices in China, according to Abacus, with one seller reported to be offering a database of 20,000 images which could be used to train an AI facial recognition model for 1,000 yuan, which is roughly US$140, or $0.007 each.
Companies, particularly those based in China, have upgraded their facial biometric technologies to identify people with faces partially blocked by respiratory masks.
Some of the images are gathered from social media, and some collected by crawlers from public websites.
Users can hold a folded facemask up over half of the lower portion of their face, while moving their face back and forth, to register a masked version of their face. Following this, the device should be unlocked with the user’s unmasked face, and then will be able to log-in without removing facial protection.
Tutorial videos has already been circulating online, purporting to show iPhone users how to do the same thing, but Abacus points out the Tencent tutorial is the first method to be validated by a security research lab. Tencent does warn that the hack can decrease FaceID’s security.
An Abacus reporter was unable to make the hack work in an initial attempt, but discovered the next day that her iPhone recognized her with a mask on.
The trick works with surgical masks, N95 masks and those with breathing valves, due to the “mirror match” nature of FaceID, Xuanwu Lab says, without elaborating on how that works.