Android Face Recognition: How to Cheat and How to Improve it

Sergei Terekhov

Written by Sergei Terekhov | Wednesday, 12 March 2014

The new version of the mobile OS Google Android 4.0 Ice Cream Sandwich has a function, which seems to be interesting at first glance, - unlocking phone using face recognition of its owner. You can read its detailed description in one of my previous blog posts Presentation Android 4.0 Innovations and improvements. Part 1. In this article, I’ll consider how to cheat Android face recognition; and how it can be improved.
So, by hearing about Face Unlock, a serious counter-argument immediately crosses the mind. Won’t it be enough to show such system a photo of its owner and to get access to the device in this way?
That’s what the author of popular Android add-on called CyanogenMod suggested in public. In response he was told that, well, no, it would be too easy. However, it turned out that this photography method actually works, and absolutely flawlessly. The owner of SoyaCincau.com web site, invited to the Hong Kong presentation of Samsung Galaxy Nexus, tried to adjust the software on recognizing his face. Then, he took a photo with Samsung Galaxy Note, brought its screen to the camera at required distance; and the phone was immediately unlocked.

Herewith, a photo of device’s owner can be found in social networks very easily.
It turns out that reliability of Android face recognition and herewith Face Unlock is considerably inferior to gesture code and password. Does this mean that you shouldn’t use it? It’s not quite true - in fact, manipulations with taking photo still need to be made. In any case, Face Unlock is better than usual screen lock, and you don’t need to make any efforts for its use.
But how can Android face recognition and Face Unlock be improved? There are some nice ideas and suggestions.
Firstly, mobile software developers can teach the phone to recognize video and not just a photo. A 3D face model can be used (and also model of environment, to verify data from the accelerometer). There are similar functions in OpenCV or C3. Reference points can be taken from Android face recognition algorithm. But there is a significant complexity for implementation of this idea - weak phone cameras.

Secondly, to strengthen Face Unlock protection, the combination “face recognition + simple graphical key” can be used for user’s identification. But if the camera goes kaput, there should be a possibility to enter a long numeric password. Then, it will be necessary to have a photo + to know key or long password to access the device. It shows to me as a very reasonable option.
Thirdly, the idea of video identifying using a 3D model can be developed further. The phone can be made ask for unpredictable actions, which a potential malefactor probably won’t have in video recording. That is, the phone identifies its owner, but requests to close his left eye, to turn right, to smile etc for unlocking. By analyzing video changes and comparing those with expected changes in the model, this way can provide a very high degree of reliability.
And even this issue can't do without Apple’s meddling. At the end of December 2011, the U.S. Patent Office has announced a patent application called Low Threshold Face Recognition from the Apple Company. Apple Experts offered to create a face recognition function in mobile devices and laptops. Owner’s face recognition will be performed using camera on the phone’s front panel. Eyes, nose, ears, mouth, skin color, etc will be used as indicators. Distinguishing feature is that the camera is constantly in face recognition mode, even if the screen is off. The patent application also states that a certain number of users can be stored in device’s memory.
The answer of Google is announcement of patenting a new type of screen unlocking. Probably, this method will be integrated into the new operating system Android 5.0 Jelly Bean, which is expected by mid-year.
Anf finally, a sweet video how Face Unlock should work on Galaxy Nexus:

comments powered by Disqus