Skip to main content

Liveness Verification System

Overview​

The Liveness Verification System by Facia ensures the authenticity of a selfie image or video by performing checks against various attacks such as replay attacks, face mask attacks, and paper attacks. The system employs multiple detection models and an ensemble model to calculate a final liveness score, determining whether the input represents a live and genuine human face.

Liveness Detection Offered By Facia​

Currently, Facia provides two types of liveness detection:

Active Liveness Detection​

Active Liveness Detection involves the use of specific challenges or stimuli that require a real-time, dynamic response from the user. These challenges are designed to be perceptible to a live person but challenging for automated or fraudulent attempts to replicate. Examples include asking the user to blink, smile, or perform a specific facial movement during the authentication process. The system analyses the response to these challenges to ensure the presence of a live and attentive user, enhancing the security of biometric verification processes.

Passive Liveness Detection​

Passive Liveness Detection relies on the continuous monitoring of various biometric indicators without requiring explicit user interaction or response to specific challenges. This method assesses the inherent characteristics of the biometric data being captured, such as facial features, to discern signs of vitality. Passive liveness detection systems employ advanced algorithms to analyse the natural variations and patterns in the biometric data, ensuring that it originates from a living, present individual. This approach is less intrusive and seamless for users, as it does not necessitate specific actions during the authentication process.

Liveness Verification Process​

1. Start​

The process begins with an image (Quick Liveness) or a video selfie (Detail Liveness) submitted for verification.

2. Extract Frames​

For a video selfie (Detail Liveness), the video is broken down into individual frames. For an image (Quick Liveness), the single image is used as the frame. The frames are then processed through all services in order to check whether the input sample is live or not.

3. Frame Processing​

In frame processing, the input request undergoes analysis by various detection services to identify potential spoof attempts or confirm a real identity. The following attacks are checked during the processing of a request:

Paper Mask​

A paper mask attack is when a fraudster holds a printed photo of another person's face in front of the camera, or wears it as a mask (sometimes with the eyes or mouth cut out), to impersonate that person. The system detects signs of paper such as flat surfaces, print texture, edges and the lack of natural skin detail.

Replay​

A replay attack is when a fraudster holds a digital screen, such as a phone, tablet or monitor, in front of the camera while it plays a pre-recorded video or shows a photo of another person's face. The system detects signs of a screen such as moiré patterns, screen glare, bezels and pixel artefacts.

Silicone Mask​

A silicone mask attack is when a fraudster wears a realistic 3D mask made of silicone, latex or rubber that imitates another person's face. The system detects signs of a mask such as unnatural skin texture, reflections, mask edges around the eyes, mouth and hairline, and the lack of natural facial movement.

Screenshot​

"Screenshot" is an image capture of the visible content on a computer screen or display at a particular moment in time. In the context of cybersecurity, screenshots may be used for documentation, analysis, or to illustrate specific issues during security assessments.

Deepfake​

"Deepfake" is a synthetic media production technique that uses artificial intelligence (AI) to create or alter audio and video content, making it appear as though it is real and unaltered. Deepfakes can be used for malicious purposes, such as spreading disinformation or creating convincing fake identities.

Evasion​

"Evasion" refers to the techniques employed by attackers to avoid detection or analysis by security systems. This may involve altering the characteristics of malicious code or network traffic to evade traditional security measures and intrusion detection systems.

3D Animated Face​

"3D animated face" refers to the creation of a three-dimensional, computer-generated representation of a human face, possibly for malicious purposes like impersonation or creating deceptive content.

4. Ensemble Model​

The outputs from all models/services are collected, and an ensemble model is executed to provide a final liveness score. This score informs the decision regarding spoofing and liveness.

Conclusion​

This concludes the technical documentation for the Liveness Verification System by Facia. Developers can integrate these systems into applications requiring robust liveness verification for enhanced security and protection against fraudulent impersonation.

Facial Liveness Accuracy at Facia​

  • 0%FAR@<1%FRR