Researchers find multiracial facial recognition system offers greater accuracy
The University of Surrey has developed a multiracial facial recognition system that delivers more accurate results than other systems typically used.
As detailed in Pattern Recognition journal, the 3D morphing face model ‘learns’ from different racial faces to more effectively identify people in 2D pictures – even if an individual’s appearance is affected by their pose, expression, lighting or poor image resolution.
Though many facial recognition systems already adapt 3D models to 2D faces found in pictures, most systems use the same model for different races and ignore key differences.
Researchers at Surrey’s Centre for Vision, Speech and Signal Processing (CVSSP) found that the use of multiracial 3D face models yields more accurate results when attempting to recognize people.
The study also found that the team’s aging effect technology – which is used to identify people after a long period of time has passed – is far more accurate when you use a model that is taught to learn different races.
“It’s safe to say that facial recognition technology is slowly becoming more prevalent in our daily lives,” Dr Zhenhua Feng from CVSSP, lead author of the paper, said. “We need to make sure it’s as accurate as possible, so people can trust the technology. We have found that our model that understands black, white and Asian faces is far more accurate at recognising 2D faces than the typical all-in-one models used today.”
Dr Feng recently received a European Biometric Industry Award for his work involving facial landmark localization.
He is part of a team at CVSSP that is working on a £6 million (US$7.9 million) project for the Engineering and Physical Sciences Research Council to bring facial recognition to the masses across the country.
“We believe that facial recognition technology will be a force for good,” Professor Josef Kittler, Distinguished Professor at the University of Surrey and founder of CVSSP, said. “It will help us protect our possessions, provide better security for our data and keep us safe from harm. However, the matter of accuracy is something we all have to be mindful of and that is what we are working on improving at CVSSP. Dr Feng’s project and the wider work we are doing at the Centre is focused on improving the accuracy of facial recognition technology, even in extreme cases where the resolution of the corresponding image is compromised, or in cases where people may try to trick a system.”
In March, the House Committee on Oversight and Government Regulation held a bipartisan hearing to review the use of facial recognition technology by law enforcement agencies, particularly the policies of the FBI. At the time, the Center on Privacy and Technology at Georgetown Law recommended that the FBI simply test its system for racial bias.