Skip to content

Latest commit

 

History

History
85 lines (81 loc) · 5.66 KB

1017.md

File metadata and controls

85 lines (81 loc) · 5.66 KB

精读

Summary

  • Basic concepts: Acoustic vibration has different harmonics and intermodulation under different acoustic emulation. The attenuations at different frequencies is different when going through different multipath.
  • Hand-surface vibration vibration responses: uniqueness, contributed by physiological characteristics of human hands, and nonlinearity, whose complexity prevents attackers from predicting the response protocol.

Details

  • Four kinds of attacks
    • zero-effort attack
    • imopersonation
    • raw signal replay attack
    • synthesis attack
  • Challenge desing
    • requirments: distinguishable (among users); distinguishable (among challenges) + unpredictable
    • chirp + sinusoidal wave
  • Feature processing
    • alignment and segment + bandpass filtering
    • Ceptral feature (MFCC) + Statistical feature of ceptral feature (mean, variance,...)
  • Classification (OC-kNN, an instance-based classifier)
    • mapminmax --> weight --> distance dj --> threshold estimation
  • Evaluation
    • Hardware: Surface (copper plate lying on a ploymer foam pad), Stimuli (loudspeaker), Receiver (two contact microphones (accelerometers))
    • data collection: 15 subjects, three data collection sesssions (intra/inte-days), fixed hand shape for consistent alignment
    • train/test:
      • intra-day: with 30min apart
      • inter-day: with 5 days later to collect 3rd session as the testset while the formar two session as the trainset.
    • Metrics: FNR (usibility), FPR (security), EER (FNR==FPR)

Strength

  • Solve the problem: human biometrics are non-resilient by introducing a challenge-response biometric authentication
  • A new perspective of human biometrics: a dynamic view (challenge-response)

Weakness

  • the position of hands/speaker/receiver
  • the sound level of the speaker
  • authentication time
  • when a new user entrolls. the thresholds should be re-calculated? Or the threshold only correlates with the baseline response without hand contact?
  • the evaluation about the three kinds of synthesis attack is not convictive.

Inspiration

Nonlinear vibra-response --> nonlinear mm-response

  1. mmWave-based Palmprint (Tag-based)
  • device: IWR1642 or 120GHz radar
  • Goal:
    • authentication
    • recover pixel-level palmprint
  1. mmWave-based Fingerprint (Tag-based)
  • device: 120GHz radar
  1. challenge-response auth in hardware fingerprint?

泛读

Vocie Controllable System

  • Voice capture + Speech recognition + Cmd execution
  • Activation Stage ("Hey Google") + Recognition Stage (NLP)

Active attacks

Passive attacks (Covert listening deivce)

Defense

  • Synthetic sounds: white noise/random sounds to pollute the side-channel, or dedicated sounds to cheat the attacker.
  • Masking sound: should be evaluated against Independent Component Analysis (ICA) attack.
  • Usability discussion: addtional components, extra action of the user (degradation on user experience)

summary

VibWrite enables finger inputs on ubiquitous surfaces to authenticate both the password and the user. It integrates passcode, behavioral and physiological characteristics and surface dependecy together to achieve the authentication.

Solved problem

  1. Enable authentication on ubiquitous surfaces with a low-cost hardware solution (vibration motor and receiver)
  2. Enable both password-auth and user-auth at the same time. (Extract user-dependent feautures from the vibration)

Main idea

  1. A challenge-response solution
  2. Extract user behavioral and physiological feature from the vibration (in the frequency domain).
  3. cm-level location discrimination, unique features are embedded in a user's finger pressing at different lcoations on a solid surface