Research has been revealed that masks are effective in preventing droplet infection, such as new colon virus infections (COVID-19), but has made it difficult to identify people due to face recognition algorithm.
NISTは顔写真にデジタル処理でマスクを追加し、89の顔認識アルゴリズムをテストした提供：NISTAccording to a report published by the US Standard Technology Research Institute (NIST) on July 27, the mask will even prevent the identification of the very advanced face recognition algorithm.According to the performance of the algorithm, the error rate is 5-50 %.
As the number of people wearing masks in response to the global epidemic of COVID-19 is increasing, it is a great result for the industry of facial recognition technology that is promoting the development of algorithms that can identify people only with eyes and nose.。
Musks are essential to suppress the spread of COVID-19 infections, and many state governments are obliged to wear masks in the United States.However, the mask has caused a problem with face recognition software, and technology companies are forced to respond.For example, Apple provided an update so that even if a user is wearing a mask, it can be unlocked quickly on a "Face ID" device.
The face recognition algorithm needs as many data points as possible from the image of the person, but the mask makes many important information in identifying the person.Even the algorithms have the aspects that are not well aware of the algorithms and the angle of light, and the research has revealed that the mask will further reduce the success rate.
In the masked person image, the error rate is 0 even in the most accurate algorithm..It jumped from 3 % to 5 %.In this study, we tested the effect of 89 face recognition algorithms using masked images.
This test was performed by comparing an image of a person with another image that the person masked to examine the algorithm's "one -to -one".NIST uses 6 million images for the survey, and has added various variations masks to the images in digital processing.
The survey also revealed that the more the mask hides the nose, the lower the algorithm identification rate.The result is that black masks are more likely to cause algorithm error than light blue masks.
According to NIST, a plan to conduct a series of tests on face recognition and masks starting this time, and will test an algorithm developed in consideration of a mask face this summer.
"Now that the pandemic has arrived, it is necessary to understand how the face recognition technology processes the face wearing the mask," said Mei Ngan, who wrote this report by NIST."So I started by investigating how the algorithm developed before the pandemic could be affected by the subjects covered by a mask."
According to NGAN, NIST predicts that the ability of algorithms to identify people with masks on their faces will improve in the future.
This article edited an article from overseas CBS Interactive by Asahi Interactive for Japan.