Computers

New face-aging technique could help locate missing persons

View 2 Images
A series of test images reversing the algorithm to examine how accurate its estimates are when placed next to real photographs
University of Bradford
A series of test images reversing the algorithm to examine how accurate its estimates are when placed next to real photographs
University of Bradford
These are the images of missing boy Ben Needham that the system created from a single photograph of the child at 21 months
University of Bradford

Earlier this year, a smartphone app called FaceApp that can take your photo and apply filters to make you older, younger or even a different sex went viral. For weeks we saw amusing modified images of celebrities flying around the web. The software joined numerous apps already on the market doing similar things, but no one could really make a case for any of them being much more than a time-passing fad. Now, a more sophisticated method for aging faces in photographic images has been revealed, with a much more serious application in mind.

In the UK alone, nearly 300,000 people are recorded missing every year, and as time passes those missing persons inevitably age in ways that make the photographs used to try and identify them less and less relevant. A research team at University of Bradford was motivated to develop technology that would improve on current techniques of aging images of missing people.

The team developed a neural network that was fed facial feature data from a database of individuals at various ages. The system mapped numerous facial characteristics, such as mouth, forehead and even the shape of one's cheeks, and then learned how a variety of human features change with age. As it consumes more data, the system is able to improve its estimations.

To test the system's accuracy, the team employed a series of de-aging experiments, running the algorithm backwards to create images that could be compared to actual photographs of subjects at a young age.

These are the images of missing boy Ben Needham that the system created from a single photograph of the child at 21 months
University of Bradford

The researchers are also using a real missing person case from 1991 as a case study, highlighting how the technology could practically be employed in real-life scenarios. Ben Needham was a toddler who went missing from the Greek island of Kos when he was 21 months old. Ben has never been found, but over the years investigators have generated several images hypothesizing what he would look at as he grew into his teens and twenties.

This isn't the first time we have seen a serious attempt at making an algorithm that can age faces. In 2014 a team from the University of Washington developed a similar technique, able to estimate changes in physical appearance up to the age of 80.

The new method from the University of Bradford looks to be a huge generational step forward from that model three years ago. It is able to model more factors, such as datasets from relatives and siblings, into greatly individualized aging results that are presented in photographic quality images.

"Current methods that exist use linear or one-dimensional methods whereas ours is non-linear, which means it is better suited for the individual in question," explains Professor Hassan Ugail, the study's lead author.

The team's results are published in the Journal of Forensic Sciences.

Source: University of Bradford

  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
2 comments
BeinThayer
"Current methods that exist use linear or one-dimensional methods whereas ours is non-linear, which means it is better suited for the individual in question," . "....ours is non-linear, which means it is better suited..." . ...ah, no. That is not what 'non-linear' means. While the paper appear to have utilized this as gospel, the results do not support the assertion. . Look at those simulations versus the image of the real person. Not only are the simulations incredibly indistinct and vaguely similar to many people's faces, the simulations are remarkably unlike the image of the real person.
Terence Hawkes
Well that doesn't work!