Google working on continuous real time authentication systems.

image

  •  
  •  
  •  
  •  
  •  
  •  
  •  
  •  
  •  

 

In the backdrop of Amazon and MasterCard announcing about using Selfie-based authentication systems, a new facial recognition feature that works on a continuous real time basis is being conceptualised by Google. It differs from normal facial recognition systems in that it does authentication on a continuous basis while the user is making a payment or conducting an online bank transaction. The Selfie pay systems that are currently being devised by the likes of Amazon and MasterCard makes use of facial recognition techniques wherein a user’s photo is taken using the front-facing camera to validate his/her identity. Moreover, they also incorporate what is known as a ‘Liveness’ Test by checking for a blink of the eye or the roll of the face to confirm the same.





The current selfie-based authentication system is found to be ineffectual especially if user transacts business over the Internet for longer durations-there is no mechanism to authenticate on real time basis during the course of the entire session. It is here that Google’s new real time system scores over current systems. Using the mobile phone sensors, the system will monitor a combination of a bio-signs such as fingerprint and keystrokes; if any mismatch is encountered at any point of time immediately access is blocked.

How it works?

The Google Researchers have devised a machine-learning algorithm that works by allowing the smart phone’s selfie-camera to operate in video mode and perform authentication by scanning the face of the user by breaking it down into parts. The technique, known as facial segment based face detector, first stores multiple images of the user in its database-this would constitute the training Phase. In the learning phase, the face is broken down into facial segments which might compose of a left eye, the mouth, the forehead, etc. FSFD then devises its own set of ‘probable faces’ with which to compare during the actual authentication procedure. Needless to say the efficiency of the system is much dependant on the learning ability of the system which will continue to evolve over time.




The Google researchers performed the test on 50 different iPhone users under fluctuating lighting conditions. They were able to successfully recognise partial and full faces irrespective of the pose or lighting conditions. The only scenario where it failed was when only a very small fragment of the face could be captured by the camera, and that too when it was accompanied by blur and odd angle positioning.

The FSFD technique is a good bet for face based continuous authentication because of its high rate of precision, and can be expected to be used in mobile phone authentications systems in the foreseeable future.