Introduction
How does it work in implementations?
Face recognition steps may be spoofed by malicious users, by using photos, videos, masks and other methods of faking that person in front of a webcam (or mobile phone cam) is the one it claims to be. This brought the need for liveness detection, which is a process of determining if the person in front of the camera is a real person or a spoof.
Presentation attack detection (PAD) technologies can utilize active (explicit) or passive (implicit) detection methods. Typically those are associated with facial recognition but liveness can also be applied to voice recognition (and other biometric methods) to distinguish present speakers from audio recordings. Here, we are talking only about face recognition and liveness detection.
Topic of liveness detection is a separate scientific field, and we will not touch the details of it here, but will present only MachineSense differentiators as compared to other solutions.
Most of the competition is offering liveness detection by asking the user to perform some kind of action in front of the camera. This is
what we refer to as "Explicit liveness detection".
User will, typically, in such a case, be asked to move his head, to blink, to smile, to move his lips, to move his eyes, etc.
In most cases, we are seeing methods involving, for example: Asking to move head left/right and up/down (or a combination of).
MachineSense also offers such liveness detection methods, which would be comparable to the ones mentioned.
However, MachineSense also offers the "Implicit liveness detection" method, where the user is not asked to perform any action, but the liveness is detected by the system, based on the analysis of the selfie image(s), in a completely automated manner. This is done by implementing special machine-learning methods and trained models, which are able to detect if the selfie is a real person or a spoof.
MachineSense liveness detection works already on the level of a single-image selfie, but will -for more precise results- also support the use of multiple images from selfie-feed, when available and when those are sent to our API-server.
Final result and benefit for your users is that they do not need to perform any actions in front of the camera, but just to
take a selfie. This is much more convenient for the user.
Whoever tried at least once the "move your head left/right and up/down" method, knows how annoying it is. It is also not very
user-friendly, and is not very good for user-experience.
This implicit and efortless liveness detection is a USP of MachineSense.
Inclusion of liveness detection is done on the parametric level, in all the face-API related methods. Thus, when initiating either face-enrolment, face-verification or any type of KYC method, you can specify that liveness-detection should be performed.
This inclusion is done by setting the liveness parameter to true and adding a few other parameters,
such as number of images to be used for liveness detection, or the liveness detection method to be used (implicit or explicit, and -if explicit-
which particular flavor to use).
For details, see API calls of any method in the Face-API.