鈥淢Y TWIN did it鈥� might seem like the perfect defence, but it won鈥檛 work for much longer. A computer program that can pick out small differences in identical twins鈥� facial features as they age can distinguish them 90 per cent of the time.
The identical DNA, appearance and voices of twins can make life difficult for forensic investigators. In Canada last week, for instance, identical twins who had broken into a house together were , since police couldn鈥檛 determine which twin had committed an assault or used a weapon.
Although there can be some very small differences in DNA between identical twins (see 鈥�Picking the perp鈥�), developing tests to spot these changes is too expensive to be practical. Instead, the US Federal Bureau of Investigation funds a research project at in Twinsburg, Ohio, where researchers can test out various biometric technologies to distinguish the pairs.
Advertisement
Marios Savvides of in Pittsburgh, Pennsylvania, believes the key to distinguishing twins is to look at their behaviour, which is reflected in the face. 鈥淲hen you have an expression or smile, that鈥檚 when your brain comes into play and you exhibit asymmetry,鈥� he says.
Savvides collected photos of twins from a Twins Days event and found that in a side-by-side comparison, even twins with neutral facial expressions showed asymmetry. These asymmetries became easier to spot when Savvides鈥檚 group cut digital photos of both twins鈥� faces in half and turned each side into a mirror image of itself.
Next, they looked at how faces change over time. As people age, habits such as smoking plus weight change and common facial expressions alter the appearance of the face. Twins are no different, Savvides says, making older pairs of twins easier to distinguish than younger twins.
鈥淎s people age, experiences such as smoking alter the appearance of the face. Twins are no different鈥�
Savvides鈥檚 group had been developing an algorithm that determines the age of a person in a photo by examining facial lines and textures in nine places around the eyes and upper lip. Even young people have the beginnings of such lines, he says.
They have now adapted this algorithm to highlight the differences in age lines between identical twins. When tested on photos of 638 pairs of identical twins, the program was able to correctly distinguish between the twins 90 per cent of the time. If a security camera at a crime scene captures the face of a twin, such a computer algorithm should be able to match it with the correct twin, says Savvides. The research has been submitted to IEEE Transactions on Image Processing.
of the University of Notre Dame in Indiana, whose research group created the data set used by Savvides, says techniques to distinguish twins could help train forensic examiners in facial recognition. Because a twin-detecting algorithm needs to be precise, it could also lead to algorithms that are more accurate for all faces, he says.
Picking the perp
Face recognition techniques can鈥檛 solve all twin identification problems (see main story). Although he was the main suspect in his identical twin鈥檚 death in 2007, jurors couldn鈥檛 reach a verdict in the . Because their DNA is identical, any DNA evidence at the scene was deemed useless. Similarly, police in France are struggling to identify which of accused of multiple rapes to prosecute.
But there鈥檚 still a chance their DNA could help. In 2008, geneticists at the University of Alabama discovered that identical twins sometimes have different numbers of copies of a gene (). And in 2005 Mario Fraga of the Spanish National Cancer Research Center found that small epigenetic changes mean twins鈥� DNA is not always identical ().
Standard DNA tests wouldn鈥檛 pick up either of these changes, however, making them expensive and time-consuming to check. Niall Firth
This article appeared in print under the headline 鈥淒ouble trouble鈥�
