A new computer classification technology, created by Pennsylvania State University researchers, actually exceeds humans at correctly identifying a person’s gender. Using nose, mouth, eye and voice cues, the system is correct nearly 100 percent of the time, while humans consistently score in the low 90s. "It’s automatic and nearly foolproof," says Rajeev Sharma, the research team’s leader and a Penn State associate professor of computer science and engineering.

The system is based on support vector machine (SVM) software, a sophisticated pattern-recognition technology used for a variety of difficult separation-oriented tasks, such as scanning tissue cell samples for abnormalities. Sharma and his researchers developed SVM software designed expressly for face and voice recognition. "It seemed like a natural extension of the technology," he says.

To test its creation, the team linked the software to a camera and examined 1,755 thumbnail facial images that were cropped to show only eyes, noses and mouths. A separate audio-enabled SVM was trained on voice samples that included just fractions of a second of voice data. A manager program then fused the results and made the final gender classifications. "It’s the combination of the two sources that makes the recognition rate so high," says Sharma.

Sharma says the system can easily be applied to security situations in which it’s critical to determine an individual’s sex. For example, the technology could be used to signal an alert whenever an unauthorized individual tries to enter a monitored rest room, fitting room or dormitory. The technology also promises to help marketers. "It could be used to see how many men versus women sit behind the wheel of a particular car or purchase a certain brand of soap," says Sharma. "It takes the ambiguity out of sexual identification."