Not a member yet? Register for full benefits!

Apple Image Recognition Patent

On the 10th of May 2012, Apple corporation filed a new patent, building on earlier patents they have filed, and more importantly, turned into actual technologies. It focuses on improved machine vision and image recognition algorithms. Specifically, facial and other 3D object recognition from 2D data only and in any light, on minimum computational power devices.

The initial applications are obvious. This is a facial recognition system which does not attempt to recognise faces. Rather it looks for features such as corners and other spatial points of reference to extrapolate what the image should be. It then recreates the data in a 3D form, in the same manner as the system the virtual world Twinity uses to create its avatars.

Unlike Twinity, which requires a side-on photo and a front-on photo to draw a 3D head, the Apple patent relies on just a face-on picture or photo. This of course means its applications are partially limited when used for avatar construction it cannot extrapolate what the back of the head looks like, for example. However, for a security application this is not necessary.

Because of the way it works, it could take a snapshot from a user's webcam, or even their face from a video and extrapolate who they are using just the computational power of an iphone or ipod tablet. The ideal way to use this would perhaps be as a security lockout, demanding the right user's face to unlock the system. Unfortunately of course, an old digital photo could easily fool the system into thinking the user was really present, when they are not. So, perhaps the best use for this system would be as a mocap tool, transcribing the user's facial emotions and expression in real-time, onto their avatar's face


Apple patent describes a more secure face-recognition system

United States Patent Application US20120114251 (PDF)

Staff Comments


Untitled Document .