Every two years, are released of Madeleine McCann, . Ten years later, photos of the little girl聽are unlikely to help find her today, a common problem for all long-term searches for missing children.
But a new AI could identify images of missing children even as they age.
Police currently have plenty of software options available聽for aging photos of people who have been missing for a long time, but these tend to work best when the missing person is 20 or older. That鈥檚 because age-related changes to adult faces are more predictable, and easier for an algorithm to simulate.
Advertisement
Even when聽artificially aged images are a good likeness,聽it can still be difficult for a person to match two images against each other. 鈥淕iven a face image of a child, it is extremely hard for a human to identify, visually, who the child is from a large dataset of child face images,鈥� says Debayan Deb, at Michigan State University in East Lansing.
Now Deb and his colleagues have created an algorithm to do this聽for them. They trained a face recognition algorithm on the Children Longitudinal Face (CLF) dataset 鈥� a collection of images of over 1,000 children, each of聽whom were photographed at least 4 times over a period of 6 years. The database comprises children aged between two and eighteen years old.
Best yet
The AI learned from the dataset to accurately identify photographs of children as young as 2 years old with 80 per cent reliability up to 2.5 years, automatically flagging them against a new image put into the system.
With one year between the original photograph, the method was 90 per cent accurate at identifying faces, and while the accuracy fell with time 鈥� dropping to 73 per cent after three years 鈥� it is still higher than any other available matching software. The usual accuracy rates for young children, aged zero to four, after just six months.
鈥淭his is a very good paper that could potentially be used to identify missing children,鈥� says Amarjot聽Singh at the University of Cambridge, who was not involved in the research.
However, Matthew Guzdial, from the聽Georgia Institute of Technology, thinks the team should have looked at larger gaps of time between photographs, as sex trafficked children 鈥� one of the highest population of missing children 鈥� usually survive for seven years.
This is Deb鈥檚 next goal, in fact. 鈥淲e aim to further the state-of-the-art in child face recognition by further increasing the time gap,鈥� he says. He and his teams also hope to develop an Android app that could be used in developing countries like China and India, where trafficking is an 鈥榚pidemic鈥�.
搁别蹿别谤别苍肠别:听
