As described in a recently discovered patent, Apple’s method would sense when a user is approaching the device — for example, if it’s seated in a dock, and the user walks toward it. The device would then use its image processor to execute facial recognition to unlock the device, all with low battery penalties. If the device is used for business applications, higher security levels could even be set.
Apple’s method would use a “weighted difference” map instead of a computationally “expensive” method called correlation mapping, according to Patently Apple, which first reported on the patent.
Here’s a few ways it could work:
The device’s front-facing camera would capture an image of the user. Rather than analyzing the entire face, the face-detection system would look at just “high information portion” areas like the eyes, nose, and mouth. These areas would be matched with a reference image.
The two images could also be normalized, a process that adjusts the pixel values in a photo. The two normalized images could be subtracted, and a score (called a weight) would be assigned to how closely they match. The high information areas (like the eyes) would be assigned a heavy weight, for example, and a lower weight would be given to less relevant, identifiable parts of the face.