Japanese police have turned to Australian scientists to help develop facial-recognition technology capable of distinguishing between people of non-Japanese extraction.
The police body said that it was having increasing difficulty dealing with criminal suspects who do not fit the typical Japanese description; the police force had trouble identifying suspects, and witnesses also had a hard time identifying individuals in a line-up. To help solve the problem, researchers at the National Research Institute of Police Science (NRIPS) and NEC in Japan developed a three-dimensional facial recognition system, Fiore.
However, facial recognition technology that is programmed for a particular set of characteristics can have problems identifying people whose features stray too far from the defined set used by the program. The problem is further exacerbated when the program attempts to correlate pictures of a persons' face taken from different angles.
In order to pilot the program on a non-Japanese population, NRIPS teamed up with the Oral Anatomy, Medicine & Surgery (OAMS) Unit at the School of Dental Science at the University of Melbourne.
"It was set up to be ideal for people with Japanese skin colour and shape of face," Sherie Blackwell, technical officer/forensic archivist at OAMS told ZDNet Australia. "One volunteer [we tested] was African-American, and the machine wouldn't recognise him because he's too dark."
The results of the test have been sent to NRIPS, and the next version of the machine should be able to recognise a wider range of ethnicities.
The machine takes 3D images by projecting stripes of white light onto a face to create a series of fringe patterns. Two CCD cameras are activated simultaneously to produce 16 images of the face, which are meshed together to produce a photo-realistic colour 3D image in under two seconds.
The operator then manually identifies between 14 and 18 standard anatomical points on the face, such as the corner of an eye or the edge of the lips. The program then compares the 3D image with a 2D image captured from a surveillance camera -- for example -- and rates the similarity of the faces using those points. Because it uses a 3D image it can handle images where only part of the face is shown. Extensive testing in Japan was used to determine a cut-off value for a positive match.
"There's always a chance of getting a false negative or a false positive, but the cut-off value tries to minimise that," said Blackwell. "They've used the technology in Japanese courts. Of 13 cases, two were a certain ID, four a probable ID, four a possible ID, and three were an exclusion. Exclusions are just as important as positive identifications."
In addition to the identification of criminals, the machine has been used to study facial swelling as a result of the removal of wisdom teeth, how the left and right side of the face express the same emotions in different physical ways and studying children who have syndromes that affect the facial appearance.
Blackwell spoke of the new technology at the 21st Science Forum hosted by the University of Technology in Sydney: "Death and Resurrection: The science of living, dying and reversing extinction".