What is a presentation attack?
A presentation attack is an attempt to interfere with a biometric system by presenting an artifact or manipulated biometric characteristic to its capture device. The attacker may try to impersonate another person, avoid being recognized, or create an identity record under false information.
In face-based identity verification, common presentation attacks include:
- A printed photograph held in front of the camera
- A face image displayed on a phone, tablet, or monitor
- A replayed or manipulated video
- A deepfake video shown to the camera
- A two-dimensional or three-dimensional mask
- Makeup, prosthetics, or other physical modifications intended to mislead the system
Presentation attacks occur at the point where the biometric sample is captured. This distinguishes them from attacks that bypass the physical sensor and introduce data directly into the processing pipeline.
How does a presentation attack work?
A biometric system normally captures a sample, such as a facial image, and compares it with a trusted reference. An attacker attempts to replace or manipulate that sample before the comparison takes place.
For example, someone may display a stolen portrait on another device and hold it in front of the verification camera. If the system performs only face matching, the displayed image may resemble the reference closely enough to produce a match. The system therefore needs an additional control that assesses whether the sample comes from a real person who is physically present.
How are presentation attacks detected?
Presentation attack detection (PAD) is the automated process of determining whether an attack is taking place at the biometric capture device. PAD may analyze physical characteristics, image texture, depth, reflections, movement, or the person’s response to an instruction. NIST describes liveness detection as one category of presentation attack detection methods.
Face liveness checks are commonly divided into two approaches:
- Active liveness detection: The user is asked to perform an action, such as turning their head or blinking. The system assesses whether the response is consistent with a live capture.
- Passive liveness detection: The system analyzes the facial image or video without requiring an additional action from the user.
Document liveness detection follows a different process. It may examine whether a document behaves like a physical object, identify artifacts associated with screens or printouts, and analyze dynamic security features such as holograms or optically variable elements.
PAD does not replace biometric comparison. Liveness detection assesses the capture, while face matching determines whether the captured person resembles the trusted reference. A remote identity verification process commonly needs both.
How is a presentation attack different from an injection attack?
A presentation attack is delivered to the physical capture device. Examples include holding a printed photograph, screen, or mask in front of a camera.
An injection attack bypasses or manipulates the normal capture process by inserting fabricated images or video directly into the digital pipeline. This may involve a virtual camera, an emulator, manipulated application traffic, or a compromised device.
The same synthetic content can be used through either route. A deepfake displayed on another device and shown to the camera is a presentation attack. The same deepfake inserted directly as a camera feed is an injection attack.
ISO/IEC 30107 focuses on attacks performed at the biometric capture device, while injection attacks require separate controls for capture-source and pipeline integrity.
How can Regula help detect presentation attacks?
Regula Face SDK provides active and passive liveness detection for face-based identity verification. It analyzes the capture to help determine whether a real person is present instead of a photograph, screen replay, video, or mask. Regula Face SDK has passed iBeta PAD Level 1 and Level 2 testing under ISO/IEC 30107-3.
Regula Document Reader SDK supports document liveness detection during remote verification. It can analyze dynamic security features, document geometry, and signs that an identity document is being presented as a screenshot, electronic display, or physical copy rather than as the expected original document.
These checks can be combined with document authenticity verification and
1:1 face matching. Controls designed to protect the capture source and data pipeline should also be used to address injection attacks, which fall outside the scope of biometric PAD.