Face recognition techniques are not bad when identifying the right people in the crowd, but ageing makes it a very difficult task for them, in some cases it becomes impossible. In the absence of new images of a person, it takes only five years to seriously impede the operation of the facial identification system.
A group of scientists from the University of Natural and Technical Sciences in Norway came to a disappointing conclusion with the help of the artificial intelligence algorithm, which generated 50,000 human beings and artificially aged them to see how facial identification systems could cope with this task.
The designers of commercial identity recognition systems do not disclose the operating principles of their algorithms, so open source solutions were selected as subjects. During the test, researchers found that as the sample age increased, the accuracy of the systems was reduced, and when the five-year difference between the sample and the image was reached, there were significant failures. When the 20-year threshold was reached, the chances of a positive result were extremely low.
An important aspect was the age of the samples: the systems were more difficult to deal with people under 20 or over 60 years of age; the age changes are faster in these periods; Mr. Grimmer explained that the faces of infants change very much in even two months -- you can take a picture of a child every month, and there will be no guarantee of results. Small changes in the face occur until the age of 20. Similar processes begin after the age of 60: "anew changes in the shape of the head, and there are more wrinkles", which means that the proportion of positive effects is decreasing.
The simplest solution to support the regular operation of facial identification systems is to regularly update sample images, but it depends on the specific task, and when it comes to confidentiality, age changes can be a natural form of privacy protection.