AI could soon be advanced enough to recognise faces concealed by clothing, according to a new research paper from the 2017 IEEE International Conference on Computer Vision Workshops. As CB Insights reports, the researchers created an AI that achieved a 67 percent accuracy rate while identifying faces concealed with scarves.
To privacy groups, the development of this technology is a serious threat. It could put an end to the limitation that has accompanied all previous facial recognition systems: if you cover your face, you can’t be seen. AI will soon be able to guess who you are in the same way you can recognise a friend or family member, even if you can’t see their face.
The technology developed by the researchers works by identifying 14 facial “keypoints” that are always visible. These are features which remain somewhat noticeable, even when the face is covered. The nose creates an outward protrusion, for instance, above the shallow depression of the mouth.
Using a neural network, the AI managed to identify obscured faces by looking for the 14 features it had been trained to spot. It compares their locations on the subject face to those on non-disguised faces previously used for training. It’s then able to recognise the patterns in the way keypoints move when obscured.
READ NEXT: Apple’s Face ID might allow cops to unlock your phone
Disguised face identifications systems could have a big impact on society. By removing the ability to effectively disguise yourself, they would make it much harder to evade detection. Researchers have already warned that facial recognition tech will soon be able to identify your political orientation. Soon, it could also spot you attending political rallies and protests, even if you cover your face and blend in with the crowd.
The system devised by the researchers is currently limited. It’s been trained on a small dataset and reduces in accuracy when the subject is against a complex background. However, the 67 percent overall success rate suggests the technology could be much more effective when running at a larger scale. If it was trained using all the images in a national database of photos, the AI’s potency could be significantly increased.
Although the researchers say the system is meant to be used for good, such as identifying suspects after terrorist attacks, there’s evident potential for malicious groups to use the tech nefariously. The AI, created by students from the UK and India, currently remains in development.
