Parents wondering what their sons or daughters would look like as they grow up may soon get a quick answer courtesy of automated age-progression software developed at the University of Washington.
The technique, which researchers call the first automated method of aging babies into adulthood, creates images of a young child's face from an original image as it would look as it ages over a lifetime.
The software can work from images containing various poses, expression and lighting, researchers say.
"Aging photos of very young children from a single photo is considered the most difficult of all scenarios, so we wanted to focus specifically on this very challenging case," says UW engineering and computer science assistant Professor Ira Kemelmacher-Shlizerman.
The appearance and shape of a young child's face, not to mention its expressions, can alter drastically as they mature, making it difficult to predict and model the changes.
The UW researchers have taken an average of thousands of different faces, focusing on the same gender and age to calculate visual changes normally found in groups as they age, then applied the changes to the face of the subject image.
A software algorithm scans pixel configuration of thousands of randomly selected Internet photographs in different gender and age brackets, then calculates a common change in facial appearance with any age up to as old as 80, the researchers say.
"Our extensive user studies demonstrated age progression results that are so convincing that people can't distinguish them from reality," study contributor Steven Seitz says. "When shown images of an age-progressed child photo and a photo of the same person as an adult, people are unable to reliably identify which one is the real photo."
Photos of young children often include unusual poses, shadows, variable lighting, odd expressions and, occasionally, the odd milk moustache. So, the computer algorithm corrects for turned or tilted heads and varied lighting before applying the computed appearance and shape changes to create the age progression.
The system could be useful in rendering aged images of missing children, a time-consuming and not always accurate process usually created by artists using photos of a missing child and of other family members.
The automated age-progression system is especially good at creating images for children below the age of five, an age when facial features are closer to those of a baby, the researchers say.
Running on a desktop computer, the system can generate aging results for a face in around 30 seconds.
Future work may include incorporating other identifying factors including ethnicity and also cosmetic features such as wrinkles and hair-color change to create a method for representing any human face at any age, Kemelmacher-Shlizerman says.
"I'm really interested in trying to find some representation of everyone in the world by leveraging the massive amounts of captured face photos," she says. "The aging process is one of many dimensions to consider."
