Machines That Read Faces to Guess Real Age

Artificial intelligence can study a face.

The study looked at clues in a face. It tried to guess how old a person looks. That is the age other people see. It is not the same as a person's real age. It is not the same as their biological age. The study wanted to use a set of photos. It wanted to test a machine process. It wanted to check marks on a face. It wanted to learn from those marks.

People who check faces have a hard job. Checking face clues can take a long time. It can also vary from person to person. One expert may not see the same thing as another. The study wanted a steady method. The study wanted a machine method. The machine method should give similar answers each time. It should not depend on one person. It should use a clear scale. It should aim for a fair result.

The study used full-face photos. It had 468 women. Their ages ranged from 18 to 80. They were in Dallas and Phoenix. Fifteen trained experts took part. Some were dermatologists. They looked at 13 aging signs. They used Skin Aging Atlas scales. A separate group of 253 people took part. Their ages ranged from 20 to 60. They were in New York City. They gave age estimates. The system used facial landmarks. It used models to learn. The models copied the average expert score. The models graded each sign.

The models were tested on 90 percent of the image set. The other 10 percent was for checking. The team used error scores. They used close estimate counts. They used links with age. The study showed one way to build such a tool. The team used the last photos for its own check. The work focused on a machine method. The method aimed to copy expert judgment. The study looked at visible age. The study looked at facial aging.