Biometric Liveness Detection Accuracy in Low-Quality Conditions

Biometric verification increasingly happens outside controlled environments. A customer may open a bank account from an older smartphone, recover an account in a dim room, or complete identity verification over an unstable mobile network. The conditions are rarely perfect.

Poor lighting can hide facial details. Low-resolution cameras can reduce biometric data quality. Motion blur and compression can interfere with capture. Weak connectivity can interrupt data transfer and slow verification.

At the same time, identity verification systems must identify increasingly sophisticated attacks. The FBI 2025 Internet Crime Report reported that investment scam complaints involving a reported AI nexus resulted in more than $632 million in losses during 2025. The agency also documented the use of AI-generated videos, voices and images to support impersonation and financial fraud.

This makes biometric liveness detection accuracy more than a laboratory metric. Verification systems need to distinguish genuine users from fraudulent attempts across the devices, networks and environments people actually use. Performance under imperfect conditions is therefore becoming essential to secure remote onboarding at scale.

Why Biometric Liveness Detection Accuracy Matters

Face matching and liveness detection tackle different aspects of biometric verification. Face matching identifies whether the captured face matches the person’s identity. Liveness detection determines if the sample is from a real person, with a real face, not a photo, replay, mask or other presentation attack.

The False Acceptance Rate (FAR) indicates the percentage of times that an unauthorised person is accepted, and the False Rejection Rate (FRR) indicates the percentage of times a valid user is rejected. For presentation attack detection, other metrics like Attack Presentation Classification Error Rate (APCER) and Bona Fide Presentation Classification Error Rate (BPCER) are more appropriate.

ISO/IEC 30107-3 specifies the approaches to assessing presentation attack detection. It targets attacks at the biometric capture point.

How Poor Capture Conditions Affect Biometric Verification

The device, camera, lighting, network or user behaviour can be a cause of poor-quality verification.

Poor Lighting

Facial information may be obscured or poorly lit in photos, may form strong shadows or may be overexposed. NIST says poor images lead to false negatives, and that inadequate lighting, underexposure, and overexposure contribute to false negatives. This matters for liveness detection in poor lighting conditions, as in real-world applications.

Low-Resolution Cameras

The lower-resolution cameras provide less visual information for biometric analysis. Budget smartphones and older devices can also add more image noise, particularly in low light. This can lead to higher failure rates in obtaining sufficient usable biometric information or in accepting a user as legitimate.

Motion Blur and Compression

Blurred frames can result from movement or unstable positioning. Image and video compression can also remove fine visual details. If the initial capture is already poor, significant compression can further reduce the information available to biometric systems.

Liveness Detection Accuracy Benchmarks: Low-Light, Low-Bandwidth, Low-End Devices

You cannot determine how biometric verification will operate in production by the headline accuracy value. Organisations need to apply liveness detection systems to the conditions customers may face.

The more interesting question isn’t so much, “What is the accuracy rate? It is, “How is performance affected if conditions worsen?”

According to Facia’s newest DeepLiveness update, the false accept rate (FAR) is 0.06%, and the false reject rate (FRR) is 0.3%. Facia also claims zero APCER in the presentation attack scenarios it tested in its iBeta Level 2 testing.

These are helpful reference metrics, but should not be offered as specific low-light / low-resolution performance results unless measured under such circumstances.

Liveness Detection in Poor Lighting

One of the most common uncontrolled variables in remote verification is lighting. Users may confirm themselves indoors in the dark, under artificial light, or in rooms with a darker side to their face.

A good capture process should identify poor image quality early. The system can guide the user to fix the lighting, adjust their face position, or re-capture the image so the final decision is more accurate.

Biometric Verification on Low-End Devices

Biometric verification on low-end devices matters for organisations that serve a wide range of markets. Limiting tests to high-end smartphones can create an unrealistic picture of production performance.

Customers can use older operating systems, lower-resolution front cameras, entry-level Android phones or those with limited processing power. Testing should include a variety of high-end, mid-range, and entry-level smartphones, as well as appropriate older phones, if available.

The goal is to determine whether any device categories are more likely to be rejected, process more slowly, or trigger a recapture request. Efficient SDKs, quality assessment, and capture guidance will help keep performance consistent across a wider range of hardware.

Why Emerging Markets Need a Different Testing Approach

Device-inclusive testing becomes even more important in mobile-first markets. The GSMA Mobile Report reports that around 4.7 billion people, or 58% of the global population, use mobile internet on their own device.

Mobile is particularly important across low- and middle-income countries. A customer may complete KYC on an older smartphone using an inconsistent 3G connection while also dealing with poor lighting.

That shouldn’t mean security standards weaken. It means verification technology needs to perform reliably across more varied environments. Organisations serving mobile-first markets should test lower-specification cameras, limited bandwidth, fluctuating connectivity, and different processing capabilities.

Does Weak Internet Reduce Liveness Accuracy?

Evaluate weak connectivity differently from poor camera quality. A slow mobile connection does not automatically make a biometric algorithm less accurate.

Limited bandwidth can affect how quickly biometric information reaches the verification system and whether the session completes successfully. Slow uploads, missing frames, timeouts, and repeated attempts can increase verification time and lead to abandonment.

Businesses testing low-bandwidth environments should therefore examine completion rate, latency, retry behaviour, and biometric performance together.

How Businesses Should Test Liveness Detection Before Deployment

Testing should recreate actual customer conditions rather than ideal scenarios. Organisations should evaluate genuine users and relevant attack types across different devices, camera qualities, lighting environments, and network profiles.

Assess FAR, FRR, APCER, and BPCER alongside completion rate, verification time, capture retries, and abandonment. Testing should continue after deployment because devices, operating systems, customer behaviour, and fraud techniques change.

Production monitoring can reveal where performance declines and where capture guidance, technical optimisation, or additional controls may be required.

How Facia Supports Reliable Liveness Detection in Real-World Conditions

Real-world identity verification rarely happens under ideal conditions. Users may verify themselves through older smartphones, low-resolution cameras, poor lighting, or unstable networks while fraudsters simultaneously look for weaknesses in the same digital journey.

Facia helps businesses manage these challenges through DeepLiveness, face verification, and deepfake detection designed for remote identity workflows. Its published FAR and FRR figures provide measurable biometric performance indicators, while iBeta Level 2 testing demonstrates resistance to the presentation attack scenarios included in that evaluation.

Reliable identity verification, however, is not defined by a single benchmark. Organisations need technology that can maintain security without creating unnecessary friction when devices, networks, and capture environments vary.

By combining biometric analysis with liveness and deepfake detection, Facia helps businesses build verification journeys designed for fraud prevention and a strong user experience.

The real measure of biometric security is not performance when everything is perfect. It is the ability to verify a genuine person confidently when conditions are far from ideal.

See how Facia DeepLiveness performs where real-world verification gets difficult. Book a demo today.