Those who handle security matters are facing problems with the software they normally use to identify individuals at sensitive locations. The trouble is with the face masks that have become an integral part of the dress code. They help stop the spread of Coronavirus but, simultaneously, act as obstructions to the facial-recognition software.

The National Institute of Standards and Technology NIST is a branch of the US Commerce Department. It has taken cognizance of the problem and has initiated action. Federal researchers are trying to establish the accuracy of facial-recognition algorithms that are in use.

Obviously, there is a need to evolve revised algorithms to take care of this development.

CNN explains the situation. The existing systems fail to make a proper match with a person wearing a face mask. The error could be between five percent and 50 percent with the general observation of failure rates hovering between 20 percent and 50 percent. That is what an expert reveals. She is Mei Ngan, a computer scientist at NIST and an author of the report.

The logic used by the facial-recognition systems is to rely on comparison to deliver results. It compares measurements of different facial features between the image without a face mask and one with it. When the protective covering blocks portions of the subject, the software is unable to make a perfect match.

Issue of face masks poses a challenge

Various services use this recognition system extensively in applications like unlocking the smartphone or going through a security checkpoint. However, the entry of face masks has complicated matters. These have become mandatory because the pandemic is expected to remain for some more time.

President Donald Trump agreed to wear face masks during one of his recent visits. CNN mentions that technical teams are experimenting with different combinations. They have conducted tests on multiple algorithms using millions of photos. The source of the photos was official – one was from applications for US immigration benefits, and the other of photos pertaining to travelers who crossed a border to enter the country.

Face masks will test the ingenuity of artificial intelligence

CNN goes on to add that NIST did not conduct testing of the algorithms on images of people who actually wore face masks.

This was because of the constraints of time and resources. Real time exercise is necessary because masks fit differently on different people. Moreover, their texture and patterns could affect the accuracy of the software. One option to get better results could be to target the region of the face above the middle of the nose. The mask is to arrest the spread of the virus via the nose and the mouth. Hence, it keeps other portions of the face exposed. Moreover, the masks come in innumerable designs and the facial contours vary from person to person. Needless to say, the tech teams have a tough job on hand.

Facial recognition algorithms go haywire because of face masks

According to The Verge, coronavirus has brought face masks into focus and everyone must wear one to prevent the spread of the disease.

However, it has created another problem that has the tech teams worried. This pertains to the facial recognition algorithms in vogue that fail when coverings hide the face. A study by NIST finds errors with black colored masks were more than blue colored ones.

Another obstacle relates to the extent of nose covering. Anyway, there are two broad groups for matching. First is the one-to-one, the other is one-to-many. The former is usually applicable to border crossings and passport control scenarios. The latter is for mass surveillance where the purpose is to scan the crowd to identify possible matches in a database. Mei Ngan, a computer scientist at NIST, says, “With respect to accuracy with face masks, we expect the technology to continue to improve.”