In less than a minute, a child’s face can be aged to show what he or she will look like when older. Researchers at the University of Washington have developed a software that automatically generates images of the face of a young child to age through life. The technique is the first fully automated approach for the aging of the image of a person since it is baby to an adult that works with bad lighting, expressions and poses.

Ira Kemelmacher-Shlizerman, Professor of computer science and engineering, explains that aging very young children from a single photo is considered very difficult. “We took photos of children in conditions without restrictions and we find that our method works very well.”

The shape and the appearance of a baby’s face – and the variety of expressions – often change drastically in adulthood, so change is difficult to model and predict. This technique uses the average of thousands of faces of the same age and sex, to then calculate the changes in vision between the groups as they age and apply those changes to the face of a new person.

More specifically, the software determines the available average of pixels between thousands of photos of faces at random on the internet at different stages of age and gender. Then an algorithm find correspondences between the averages of each section and calculates the average facial shape variation and aged appearance. Then, these changes are applied to the image of a new child to predict an aspect for any age until age 80.

“Our studies show that the progression of age results are so convincing that people can not distinguish them from reality”, says co-author Steven Seitz, Professor of the computer science and engineering. “When shown the image of a child who has been the progression of age and another of the same person in adulthood, volunteers are not able to identify reliably between the real photo.

Photos of real-life children are difficult to use for the progression of age, partly due to bad lighting, shadows, funny expressions and even milk mustaches. To offset these effects, the algorithm first automatically corrects inclined faces, head turns and inconsistent lights, and then applies calculated changes for shape and appearance to the child’s face.

Researchers believe that their technique can be useful to search for children who have been missing for a long time. Progression can be run on a computer standard, and it takes 30 seconds to generate results of a face. It is more complicated in children less than 5 years. In the future, the researchers hope to incorporate other identifiers to perfect the model, such as ethnicity and aesthetic factors such as bleaching hair and wrinkles.

age-progress