As CNNs boost facial biometric accuracy, Australia researches child recognition
DST Group, Australia’s Defense Science and Technology agency, has established a research program to work on identifying and classifying children more accurately with biometric facial recognition to help fight child trafficking, after the completion of a research project which showed major challenges, but also potential for improvement.
Australia’s Science Channel reports that Dr. Dana Michalski of DST conducted a study in which 120 experienced practitioners, reported to be the largest number ever assembled for a study of the kind, to manually compare 23,760 different image pairs, in addition to biometric algorithms. Children from newborns to 17 years old were considered, with images taken 10 years apart.
When the initial image was taken at a young age, or the age gap to the second image increased, the number of correct matches, unsurprisingly, decreased. The agency’s biometric team has now developed its own algorithm which is being trained with deep learning techniques to improve its performance.
Michalski says facial recognition practitioners should be trained on the changes that take place in children’s faces, and is currently developing an age estimation algorithm.
“Great gains in accuracy”
Facial recognition may soon be powerful enough to match children over time, as NIST’s Patrick Grother recently told Federal News Network that convolutional neural networks (CNNs) are behind major improvements in biometric performance.
“There have been great gains in accuracy over the last five years,” according to Grother, due to the ability of CNNs to compensate for image quality issues such as face angle or poor lighting.
“If you were to take a mugshot of me … and search it against a database of 12 million people, with quite high certainty, you would find a prior photo of me,” Grother says.
Customs and Border Protection (CBP) Deputy Executive Director of Entry/Exit Transformation Dan Tanciar also tells Federal News Network that the agency plans to expand its biometric facial recognition checks at land and sea borders from three locations in Arizona across the southern border during 2019. CBP has caught 124 people using fraudulent travel documents this year, according to the report.
According to research from Visiongain, the market for facial recognition technology will grow to $4.6 billion this year.