How many wrinkles will you have within 20 years? There are already techniques to simulate the appearance that someone will have in the following decades of their life, but they usually take time and have a high cost. The computer expert Grigory Antipov, from Orange Laboratories, in France, has developed together with his team an algorithm that allows to do it with ease. The technique not only ages young faces but creates younger versions of the older faces.
Antipov has developed software of deep learning (ie, a set of machine learning algorithms) to solve a problem in the above systems, which allow changing the faces of different ways, but keep the personal characteristics of the individual. That is, the traits of a 12-year-old child were not identifiable in “aged” photography.
The programmer’s approach involves two machines that work together: a face generator and a face discriminator. Both learn how faces look as they get older through the analysis of photographs of people from six age groups (0-18, 19-29, 30-39, 40-49, 50-59 and over 60).
The system was trained with 5,000 faces in each group, taken from the Internet Movie Database and Wikipedia and then labeled with the age of the person. In this way, the machine learns the characteristic features in each age group and applies them to other faces so that they appear to be of determined age. As this can cause the loss of identity, the second machine – the facial discriminator – analyzes the aged image to know if the individual can be recognized in it. If you can not, the image is automatically rejected.
The team applied the technique to 10,000 faces in the database that they had not used in the training phase of the system and then tested the before and after images with software called OpenFace, which can tell if two images show the same person or not. The software detected a precision in 80% of the images, while other facial aging techniques have 50% fidelity to the features of the people photographed.
Antipov says his technique could be used by security and intelligence teams to identify criminals wanted by the police or people who have been missing for many years. “The aged version of the photo provided by the family, for example, can provide an idea of that person’s appearance today,” he explains. The expert points out that the algorithm can also have a possible application in the world of entertainment.
The programmer and his colleagues do not intend to make the public algorithm, for now, but point out that the description of the system in his article is “sufficiently detailed” how to facilitate its implementation by an interested person. “And we are always willing to provide additional details of our models,” says Antipov.