Scientists used different ways, including those above, to show photos to participants in the facial recognition study. During the study, participants were shown two photos and had to decide in a short amount of time whether the individual in each was the same person.
Analysts with training and experience are much better at the forensic art of facial recognition than either computer algorithms or other people, according to a new study co-authored by a UT Dallas professor.
Dr. Alice O’Toole, Aage and Margareta Møller Professor in the School of Behavioral and Brain Sciences, said forensics are key to investigations and court cases. She said studies such as this one are important when considering the veracity of evidence in the investigations.
“Each forensics lab has its own standards and procedures,” O’Toole said. “But they typically don’t share information about the type of testing they conduct during their analysis. This research is one of the first that has attempted to evaluate the capabilities of a diverse group of facial recognition experts.”
O’Toole said that with the number of recent DNA-based exonerations in the U.S., there has been a push to ensure that forensic tools, such as fingerprints, hair samples and facial recognition, are used in the most effective ways possible. In response to these exonerations, the National Academy of Sciences convened a panel in 2009 to investigate the state of forensic sciences.
“This study was a first attempt to do a systematic exam of face recognition expertise within an international group of professionals,” O’Toole said.
Dr. Alice O'Toole
Tests were performed at a conference of the Facial Identification Scientific Working Group, which includes facial recognition experts from around the world. Two groups were tested at the conference: experts who do facial recognition work every day and support staff members who don’t perform the task regularly. A group of undergraduate students also was tested separately, and the photos also were run through a computer algorithm.
Participants were shown two photos and had to decide whether the individual in each was the same person. Decisions had to be made in a short period of time.
According to O’Toole, the expert group clearly performed the best. But the study also showed that all the human groups outperformed the computer algorithms on some of the more difficult recognition tasks.
“At present, computer programs are set only to look at faces, while the people were able to take in other body cues, such as body size, hair and shoulders,” she said. “That made a big difference.”
O’Toole said the study supports the notion that the experts in investigative agencies are indeed more skilled at facial recognition than people with no training and experience.
“I think it is a reassuring note that they’re better than the rest of us,” she said, noting that the professionals would likely do even better if they had their tools, labs and more time.
She said the message from this project is that there is more work to be done.
“We eventually would like to develop a test that these professional agencies can use in house for education and certification.”
The lead author of the study was Dr. David White from the University of New South Wales in Sydney. The authors also included UT Dallas doctoral students Matthew Hill and Amanda Hahn, as well as P. Jonathon Phillips of the National Institute of Standards and Technology. The work appeared this month in The Proceedings of the Royal Society, B.