Some charming youngsters from Cornell have created a fairly simple and effective face matching system using a webcam, a little LCD read-out, and a tiny Atmel ATmega644 8-bit microcontroller running a set of Eigenface tests on the face in question. The system is 88% accurate with no false positives. It is almost completely self-contained and is small and simple enough to add to a front door lock or other device where case real estate comes at a premium.
The project, by Brian Harding and Cat Jubinski, calculates “standardized face ingredients” based on a sample of 40 volunteer faces. The system can then compare these “keys” with the face in question and overlay the ones that match the features of the accessor’s face. For example, my face could consist of parts from face one, five, and thirty while yours could consist of parts from any other set of keys. Therefore, the eigenfaces that define my face will be distinctly different from yours.
Why is this important? Well, face recognition is historically very expensive. This small, microprocessor controlled system requires far fewer resources than your average face scanner and is far simpler to implement. You could also add hundreds of eigenfaces to improve the key security. It’s also cool to say the word “eigenfaces.”