SpoofSense Face
The most dangerous spoof never touches your camera. Virtual cameras, emulators and intercepted streams inject deepfakes directly into the verification flow — invisible to liveness checks that only watch the lens. SpoofSense flags injected media passively, on every frame, in the same API call as liveness and deepfake screening.
The basics
In a biometric injection attack, the fraudster never shows anything to a camera. Instead, they insert fake media — usually a deepfake — directly into the software capture pipeline, using a virtual camera, a device emulator, a tampered app, or by intercepting the network request itself. The verification system receives what looks like a normal camera feed showing a live, genuine face.
This is what makes injection the blind spot of traditional anti-spoofing. Presentation attack detection examines the face in the frame; injection attacks compromise how the frame got there. A verification stack that only runs PAD will confidently approve a deepfake delivered through a virtual camera.
Injection attack detection (IAD) closes that gap by verifying the integrity of the capture itself — and it now has its own standard, CEN/TS 18099, dedicated to biometric data injection attack detection.
Know the difference
| Presentation attack | Injection attack | |
|---|---|---|
| Where the fake enters | In front of a real camera | Inside the software pipeline — no camera involved |
| Typical instruments | Printed photos, screens, 2D/3D masks | Virtual cameras, emulators, intercepted APIs |
| Defended by | Presentation attack detection (PAD) | Injection attack detection (IAD) |
| Governing standard | ISO/IEC 30107-3 | CEN/TS 18099 |
| Typical payload | Physical artifact of a face | Deepfake or replayed digital video |
Attack vectors
Software cameras (OBS-style tools and browser plugins) that replay or stream synthetic video as if it came from real hardware.
Device emulators and tampered mobile apps that fabricate the entire capture environment, sensor data included.
Man-in-the-middle manipulation that swaps genuine capture payloads for fraudulent media after the camera, before the check.
HDMI-to-USB capture devices that present an external video source to the operating system as a physical webcam.
How it works
A genuine capture carries the signature of physical hardware. Injection breaks that signature in ways software can detect.
01
Virtual camera drivers, emulators and tampered capture environments expose signatures real devices don't produce.
02
Injected media carries image statistics and capture metadata inconsistent with a physical sensor's output.
03
Injection signals are fused with liveness and deepfake screening into one decision per frame — one API call, no user friction.
Defense in depth
No single check stops modern identity fraud. SpoofSense layers certified passive liveness (iBeta Level 1 & 2, ISO/IEC 30107-3), deepfake detection and injection screening on every frame — and DocLive extends the same protection to ID documents, where injection attacks target document capture.
FAQ
A digital injection attack bypasses the camera entirely: instead of presenting a fake face to the lens, the attacker feeds pre-made or synthetic video directly into the software capture pipeline — using virtual cameras, emulators, modified apps or network interception — so the verification system receives media that no camera ever captured.
A presentation attack shows something fake to a real camera — a printed photo, a screen replay, a mask. An injection attack skips the camera and inserts fake media into the data stream itself. Presentation attack detection (PAD) cannot see injection, because from the software's view a perfectly normal 'camera feed' arrived.
Injection attack detection (IAD) verifies the integrity of the capture pipeline. Injected streams leave traces genuine captures don't: virtual camera and driver signatures, emulator fingerprints, inconsistent capture metadata, and image statistics that don't match a physical sensor. SpoofSense screens for these signals passively on every frame.
Because deepfakes made them worthwhile. A convincing synthetic face is easy to generate but hard to physically present to a camera — so fraudsters inject it instead. Virtual camera software is free, and off-the-shelf tools now bundle deepfake generation with injection delivery.
Yes — CEN/TS 18099, 'Biometric data injection attack detection,' is the first standard dedicated to IAD, covering injection attack instruments, detection systems and test methodology. It complements ISO/IEC 30107-3, which governs presentation attack detection.
Yes. Liveness defends the camera, deepfake detection defends the image, and injection attack detection defends the pipeline. An attacker only needs one unguarded layer. SpoofSense runs all three checks together on every frame, in a single API call.
