You are on page 1of 20

Chonnam National University

Electronics And Computer Engineering Faculty


Gait Authentication On Mobile Phone


Using Biometric Cryptosystem And Fuzzy
Commitment Scheme
Contents

1 Introductions

2 System architecture

2
Introductions
Security for identification, authorization or authentication:
Knowledge: password
Token: ID card
Biometric: fingerprint, iris, gait
Advantages of biometric:
No need to remember as password or reserve something as ID card
Authentication using gait:
Advantages:
Gait data can be collected implicitly.
Its difficult to mimic gait data.
Challenges and risks when using gait for:
Less discriminative
Much noise

3
Introductions (cont)
Current sensor-based gait recognition approaches:
Pattern recognition and machine learning
Gait template or extracted features are stored in unconcealed
form.
Critical vulnerability: device is stolen or malware infected,
attacker can obtain gait template Attacker can utilize
compromised template to access other sensitive devices.
Impossible to use hash for protecting gait template:
Hash do not tolerance any error bit.

4
Introductions (cont)
Main contribution of the paper:
Apply security and privacy for gait authentication on mobile
phone.
Determine appropriate parameters to construct an effective gait-
based biometric cryptosystem.
Analyze security strength of proposed system.

5
Contents

1 Introductions

2 System architecture

6
System architecture
Gait Feature
signal Extraction

Extract Key Generate


Binariza-
Reliable Encoding random
tion
bit (BCH) key m

Storage
Hash
Helper data Secured h(m)
function

Extract
Binariza-
Reliable Matching Decision
tion
bit

Gait Feature Decoding Hash


signal Extraction (BCH) function

Overall architecture 7
System architecture (cont)
Gait biometrics acquisition:
Gait signals based on Wearable Sensor Technology (WST)
Mobile used: Google Netxus One
Signal: built-in three sensor: gravity, acceleration and motion
Signal: combined of gravity acceleration, ground reaction force,
and inertial acceleration
Acquired data:

aX, aY, aY: acceleration value in X, Y, Z dimensions

8
System architecture (cont)
Gait biometrics acquisition:
Gait signals based on Wearable Sensor Technology (WST)
Mobile used: Google Netxus One
Signal: built-in three sensor: gravity, acceleration and motion
Signal: combined of gravity acceleration, ground reaction force,
and inertial acceleration
Acquired data:

aX, aY, aY: acceleration value in X, Y, Z dimensions

9
System architecture (cont)
Gait signal preprocessing and gait cycle extraction:
Raw signals: contents much noise multi-level wavelet
decomposition (Db6) is applied to eliminate noise.
Gait cycles extraction: cycle-based segmentation algorithm.
Real-valued gait template extraction:
Gait cycles are normalized to ns samples:

T is formed by concatenating m consecutive gait cycles:

10
System architecture (cont)
Real-valued gait template extraction (cont):
Let nt = mns , so:

11
System architecture (cont)
Real-valued gait template extraction (cont):
An interpolation method is adopted to resample T to an
appropriate size for binding with a cryptographic key.
Let y be the timestamp, so the acceleration samples with
timestamp are as:

Spline interpolation f(y) is applied to simulate the continuity of


T such that
The spince interpolation function is used is a cubic polynomial.

12
System architecture (cont)
Real-valued gait template extraction (cont):
An interpolation method is adopted to resample T to an
appropriate size for binding with a cryptographic key.
Let y be the timestamp, so the acceleration samples with
timestamp are as:

Spline interpolation f(y) is applied to each component ((aX)i, yi),


(aY)i, yi), (aZ)i, yi)) to simulate the continuity of T such that
The spince interpolation function is used is a cubic polynomial.

13
System architecture (cont)
Real-valued gait template extraction (cont):
Assume that size of gate template T is n x 3 after interpolation.
T is represent in vector (1 x 3n) by concatenating three
dimensions X, Y, Z as:

14
System architecture (cont)
Gait template binarization:
Denote: the number of users is N, the number of real-valued gait
template of each user is M
Denote:

is the jth real-valued template of user u

So, each real value template:

15
System architecture (cont)
Gait template binarization (cont):
The mean over intra-class variability of user u as:

The mean over all gait templates of all users is:

For each user u, the binary template (u) is calculated as:

16
System architecture (cont)
Reliable bit extraction:
The reliable bit is extracted by determine the index of reliable
bits in (u).
The reliability of a bit i(u) in (u) is calculated by Gaussian Error
function as:

The variance of ith component of gait template of user u:

17
System architecture (cont)
Reliable bit extraction:
Denote the index vector of components is:

The first nc (nc < nr) components of i(u) will be used to extract the
final binary template (u)

18
System architecture (cont)
Key binding scheme:
A key m of length k is generated randomly.
m is encoded by BCH(nc, k, t) to the code word c of length nc
The hash code h(m) of m is stored and used for authentication
user

is calculated by bind c to reliable bits


Helper data includes i and is also stored to extract reliable bit
in authentication step.

After having i, u and h(m), all other data and gait template will
be discarded.

19
Thank you very much!

You might also like