Face masks offer double protection: They help against the spread of the coronavirus – and against being recognized by face recognition. According to Study by the US National Institute of Standards and Technology (NIST) Even the best of the 89 facial recognition algorithms examined do not deal well with masked faces. The error rate compared to the original image without a mask is between 5 and 50 percent due to the mask.
For the Investigation (PDF) The research team placed nine different mask shapes digitally on the original photos and used them to test the performance of the algorithms. The digital masks were black or in the light blue of surgical masks. There were masks that also covered the nose and others without. The team then compared the results with how the algorithms performed on unmasked faces.
The most accurate of the facial recognition systems examined had an error rate of 0.3 percent for unmasked faces. In the case of makeup images, the rate increased to five percent for the best algorithms, but to 20 to 50 percent for most of the algorithms examined.
The result was that the algorithms coped better with round masks and better with the light blue masks than with black masks. The more the nose is covered by the mask, the more difficult it is for the algorithms to recognize the face.
Algorithms are fed with masked faces
The current advantage of masked faces over face recognition could, however, be short-lived. Researchers and monitoring companies are currently working on to feed their computers with records of masked and unmasked peopleso that the algorithms can also better recognize masked people. NIST announced another study for late summer in which the algorithms would take more account of masking.
Facial recognition researchers and surveillance companies are also currently using Selfies of people with masks on Instagram accountsbecause there you will find pictures of the same people both with and without a mask. Research by netzpolitik.org on the company PimEyes revealed such browsing of images on the Internet.
Wearing masks is still causing a headache for security apparatus, at least at the moment. In an internal bulletin, for example, the US Department of Homeland Security (DHS) warned at the end of May that that the widespread use of masks is causing problems for police face recognition. That comes from the BlueLeaks, a nearly 300 GB large data set that comes from US police computers.
The bulletin states: “We believe that violent extremists and other criminals who have always had an interest in avoiding facial recognition are likely to be opportunistic [..] The wearing of face masks may be recommended to increase the effectiveness of facial recognition systems in public places [..] to hinder “.
Anti-fundamental rights technology
Face recognition is a technology that is hostile to fundamental rights. Among other things, this involves errors in recognition that are based on a racist bias in the data. Because facial recognition systems have a higher error rate with people of color than with white people, they are particularly affected by the use of the technology. Every false alarm can lead to the fact that actually unsuspecting people are monitored, searched and captured, which can be traumatizing and stigmatizing for them.
But not only the racist bias is a problem of the technology: Face recognition increases with it “The license plate on the face” the ubiquitous surveillance threatens fundamental rights such as freedom of assembly and ultimately abolishes privacy.
In Germany, among other things, the Alliance “Stop Face Recognition!” against this high-risk technology. Numerous digital and civil rights organizations belong to the alliance.