💻 computer science

BlowLive: Blow-Based Multi-Factor Biometrics with Liveness Detection and Revocability

BlowLive is a robust multi-factor biometric framework that combines blow-acoustic and facial modalities with Doppler-based liveness detection and fuzzy extractor-based template protection to achieve high authentication accuracy while ensuring privacy, revocability, and resistance to spoofing attacks.

Original authors: Eyasu Getahun Chekole, Howard Halim, Daniël Reijsbergen, Jianying Zhou

Published 2026-07-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Eyasu Getahun Chekole, Howard Halim, Daniël Reijsbergen, Jianying Zhou

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a digital lock on your phone. Usually, you unlock it with a password (something you know) or a fingerprint (something you are). But passwords can be stolen, and fingerprints can be faked with a photo or a mold.

The paper "BlowLive" proposes a new, super-secure way to unlock your phone that combines two things: blowing into your phone and showing your face. Think of it as a "double-check" system that is hard to trick, easy to use, and can be reset if something goes wrong.

Here is how it works, broken down into simple concepts:

1. The Two Keys: A "Blow" and a "Face"

Most security systems use just one type of ID. BlowLive uses two, like a bank vault that needs two different keys to open.

The Behavioral Key (The Blow): When you blow into your phone's microphone, you aren't just making noise. You are creating a unique "wind signature." Just as no two people have the exact same handwriting, no two people blow air in the exact same way (speed, pressure, shape of lips). The system listens to this specific "wind song."
The Physiological Key (The Face): At the exact same moment you blow, the phone takes a picture of your face. This is the standard "something you are" check.

By combining these two, the system becomes much harder to fool than using just one.

2. The "Live" Check: Catching the Fake

One of the biggest problems with biometrics is that hackers can use recordings or AI-generated videos (deepfakes) to trick the system. BlowLive has a special "lie detector" built-in.

The Analogy: Imagine the phone is whispering a secret, high-pitched sound (ultrasonic) that you can't hear. When you blow air at the phone, that air moves the sound waves, changing their pitch slightly (like the Doppler effect you hear when an ambulance zooms past).
The Trick: If a hacker plays back a recording of your blow, the air isn't actually moving, so the pitch doesn't change the right way. The system detects this lack of "movement" and says, "This isn't a real person blowing; this is a recording." It's like a bouncer checking if a person is actually breathing or just a wax statue.

3. The "Magic Box" (Fuzzy Extractor): Protecting Your Secrets

Usually, biometric systems store a copy of your face or your voice pattern. If that database gets hacked, your face or voice is compromised forever—you can't change your face like you can change a password.

BlowLive uses a clever mathematical trick called a Fuzzy Extractor.

The Analogy: Think of your biometric data (your blow and face) as a messy, slightly different handwriting every time you write it. Instead of storing the messy handwriting, the system puts it into a "Magic Box."
How it works: The box takes your messy input and turns it into a perfect, unchangeable digital key (like a password). It also creates a "helper note" (public data) that helps the box recreate the key later.
The Benefit: The system never stores your actual face or blow pattern. It only stores the helper note. Even if a hacker steals the helper note, they can't reverse-engineer your face or blow pattern from it.

4. The "Reset Button": Revocability

What happens if a hacker steals your digital key? In normal systems, you are stuck because you can't change your face.

The BlowLive Solution: Because the "blow" is a behavior, you can learn to blow differently!
The Analogy: If your house key is stolen, you don't have to move to a new house. You just cut a new key. With BlowLive, if your digital identity is compromised, you simply "re-enroll" by blowing into the phone in a slightly new way. The system generates a brand new digital key and helper note. The old one is thrown away, and the hacker's stolen data becomes useless.

5. How Well Does It Work?

The researchers tested this with 50 real people. The results were impressive:

Accuracy: It recognized the right people almost perfectly (99.56% for the blow, 100% when combined with the face).
Liveness: It caught fake recordings 99.46% of the time.
Speed & Ease: It takes only a few seconds, requires no special equipment (just a regular phone), and is completely touch-free (hygienic).

Summary

BlowLive is a security system that asks you to blow and smile at the same time. It uses a special radar to make sure you are a real, live person and not a recording. It turns your unique actions into a secret code that protects your privacy, and if that code is ever stolen, you can simply "re-blow" to get a fresh, new code. It's a secure, private, and resettable way to prove who you are.

Technical Summary: BlowLive

Problem Statement

Biometric authentication systems face four fundamental limitations that hinder their reliability, security, and long-term viability:

Vulnerability to Spoofing: Existing physiological (e.g., face, fingerprint) and behavioral (e.g., voice, gait) biometrics are increasingly susceptible to sophisticated attacks, including deepfakes, synthetic data, and playback attacks, often due to unreliable liveness detection.
Privacy and Template Security: Biometric templates, particularly physiological ones, are immutable. If a template is leaked or compromised, it cannot be revoked or changed, leading to catastrophic privacy breaches.
Intra-User Variability: Behavioral biometrics often suffer from high intra-user variability and environmental sensitivity, resulting in accuracy degradation compared to physiological methods.
Lack of Revocability: Current systems struggle to provide mechanisms to revoke compromised biometric traits or derived cryptographic keys without requiring a complete change of the underlying biometric trait.

Furthermore, existing multi-factor biometric (MFB) schemes often rely on invasive methods, require external hardware, or fail to integrate behavioral and physiological modalities effectively while maintaining robust liveness detection.

Methodology: The BlowLive Framework

The authors propose BlowLive, a robust Multi-Factor Biometric (MFB) framework that integrates blow-acoustic signals (behavioral) and facial data (physiological) as complementary authentication factors. The system operates on a smartphone, utilizing the device's microphone and front-facing camera.

1. Data Acquisition and Liveness Detection

Acquisition: The system captures simultaneous blow-acoustic signals and facial images. To enable liveness detection, the phone emits an imperceptible 20 kHz ultrasonic tone.
Doppler Shift-Based Liveness Detection: A novel module analyzes the interaction between the emitted ultrasonic tone and the user's blowing airflow.
- Mechanism: Genuine blowing creates air turbulence and micro-motion, causing measurable Doppler shifts in the reflected ultrasonic signal. Pre-recorded or synthetic attacks lack these specific dynamic frequency shifts.
- Feature Extraction: The system extracts a hybrid score based on:
  - GFCC Energy: Spectral energy of the blowing signal.
  - Doppler RMS Envelope: Temporal magnitude of frequency deviations.
  - Doppler Max: The maximum absolute frequency shift.
- Decision: A weighted linear fusion of these features determines liveness, rejecting playback and deepfake attempts.

2. Feature Extraction and Fusion

Blow-Acoustic Modality:
- Feature Extraction: The system uses Gammatone Frequency Cepstral Coefficients (GFCC) to capture fine-grained spectral characteristics of the blow.
- Embedding: A Convolutional Neural Network (CNN) processes the GFCC sequences (static and delta coefficients) to produce discriminative 128-dimensional embeddings, removing temporal dependencies.
Facial Modality:
- Feature Extraction: A FaceNet model (Inception-ResNet v1 backbone) extracts 512-dimensional embeddings, which are projected to 128 dimensions.
Multimodal Fusion:
- Score-Level Fusion: Weighted sum of the blow-acoustic and facial similarity scores.
- Feature-Level Fusion: Concatenation of the two embedding vectors to form a unified multimodal representation.

3. Fuzzy Extractor (FE) Based Authentication and Revocability

Instead of storing raw biometric templates, BlowLive employs a Fuzzy Extractor scheme to derive stable cryptographic keys from noisy biometric inputs.

Process: The fused embeddings are binarized. A Fuzzy Extractor generates a cryptographic key ( $K$ ) and a public helper string ( $P$ ) using Bose-Chaudhuri-Hocquenghem (BCH) error-correcting codes.
Authentication: During login, a new embedding is binarized and processed with $P$ to reproduce $K$ . Access is granted if the Hamming Distance between the query and enrolled embeddings is within a threshold.
Revocability:
- Key Revocation: If a key is compromised, a new random secret can be sampled to generate a fresh key and helper data without changing the biometric input.
- Template Revocation: Since blow-acoustics are behavioral, users can adopt new blowing patterns to generate entirely new biometric identities, effectively revoking the old template. This dual-level revocability (key and template) addresses the immutability issue of physiological biometrics.

Key Contributions

Novel MFB Framework: Integration of blow-acoustic and facial biometrics, leveraging the behavioral distinctiveness of blowing patterns and the physiological stability of facial features.
High-Accuracy Behavioral Biometrics: Achievement of nearly 100% accuracy in the behavioral modality (blow-acoustics) through advanced GFCC spectral analysis and CNN-based embedding, overcoming the traditional instability of behavioral traits.
Doppler-Based Liveness Detection: A specialized mechanism designed specifically for non-speech acoustic signals (blowing), effectively distinguishing live airflow turbulence from playback and synthetic attacks.
Revocable and Secure Architecture: Implementation of a Fuzzy Extractor-based scheme that provides unlinkable, non-invertible, and revocable cryptographic keys and biometric templates.
Comprehensive Evaluation: Rigorous testing on a dataset of 50 participants covering diverse postures (sitting/standing) and attack vectors.

Experimental Results

The framework was evaluated using data from 50 participants:

Authentication Accuracy:
- Blow-Acoustics (Behavioral): 99.56% accuracy.
- Facial Recognition (Physiological): 100% accuracy.
- Score-Level Fusion: 99.95% accuracy.
- Feature-Level Fusion: 100% accuracy.
Liveness Detection:
- Achieved 99.46% accuracy using a per-user (dynamic) threshold, with a 0% False Acceptance Rate (FAR) and 1.08% False Rejection Rate (FRR).
Revocability: The system successfully demonstrated the ability to regenerate keys and templates without exposing original biometric data.
Performance: The system operates efficiently on standard smartphone hardware, with feature extraction and liveness detection taking approximately 200–235 ms, and key generation/reproduction taking less than 2 ms.

Significance and Claims

The paper claims that BlowLive advances the state-of-the-art in biometric authentication by addressing the "trilemma" of accuracy, security, and usability in a unified framework. Specifically:

It overcomes the accuracy limitations of behavioral biometrics, proving that with advanced spectral analysis, behavioral traits can match physiological accuracy.
It provides robust defense against modern threats (deepfakes, playback) via a novel Doppler shift mechanism tailored for non-speech acoustic inputs.
It solves the revocability crisis by enabling the renewal of both cryptographic keys and the underlying biometric templates (via new blowing patterns), a feature rarely found in physiological systems.
It maintains high usability and non-invasiveness, requiring only a simple blow and a facial snapshot on a standard smartphone, without additional hardware.

The authors conclude that BlowLive offers a practical path for deploying high-assurance biometric security in adversarial environments, suitable for applications like smartphone unlocking, IoT authentication, and secure access control.