Dosung Ahn (Visiting Scholar)
Office: EA 723B
Phone: (215) 204-3160
Fax: (215) 204-5960
E-mail: dosung at temple dot edu
■
Security of Biometric Information
■
Fusion of Thermal Infrared and Visible Images
■ Security of Biometric Information
Motivation, [Why to protect Biometrics?]
Biometric information
is permanently associated with the user, cannot be replace or cancel. These are
not secret and can be recorded and/or misused without owner's agreement. If we
do nothing about the weakness of biometric information security before its
spread out, privacy is in danger because once compromised, will never
compromises.
Objectives
As a result of
standardization in biometrics, privacy awareness is growing as the use of
biometrics becomes more widespread with well defined standards. It means a
stolen template and its minutiae information could be use to imitate the
original image. If stored template does not store any information of the owner,
for example, randomized template incapacitates an attacker for misuse. It is
strongly required that design an irreversible transformation of fingerprint
template; make matching algorithms for transformed template for privacy
benefits.
Possible Methods
Add audit information
to raw biometric data (e.g. steganography)
Stored template does
not store any information of the owner which has capability of matching them
(e.g. noninvertible transform)
Embed
purpose/expiration on raw data/template (e.g. cancellable)
Reinforce management
effort with legislation
Innovations
We propose new approach
to hide position of basic minutiae features using transformation parameter
clustering. We transform group of minutia and use geometrical properties to
match with them. Transformed new descriptor has noninvertible property and it can
be match without alignment. Performance criteria are proof of noninvertibility, loss of discriminant
power, computational complexity and memory changes.
Figure 1. Execution
screen dump of developed program (C DLL with Matlab)
■ Fusion of Thermal Infrared and Visible Images
Motivation, [Difficulties with Single Modalities]
Visible imaging
Measure the light
reflected from the object (0.4 – 0.7 mm)
Sensitive to the
variations in ambient illumination
Not operable in poor
lighting conditions
Cannot handle disguised
faces
Thermal infrared imaging
Measure the heat energy
radiated from the object (8 – 12 mm)
Less sensitive to
illumination changes
Work in low
illumination conditions or in darkness
Reveals anatomical
information useful to detect disguised faces
Subject to change of
body temperatures caused by physical exercise or ambient temperatures
Objectives
Enhanced security for critical
facilities
• Biometric Identification based on face recognition
Access control
Verification of
authorized person
Effective with frontal
and profile images
Do not interrupt user
activities
No need for
participants’ cooperation or knowledge
• Surveillance
Monitoring of
high-value assets
Identification of
unusual activities
Low-light environment
Perimeter
Figure 2. Thermal infrared band image examples.
Sample video clip: Fusion.wmv (1,702,597 Bytes)
■ Steganography
Motivation
Steganography is the field of
writing hidden messages. Hiding a secret message within a cover image in such a
way that unintended recipient cannot realizes the presence or contents of the
hidden message. If biometric data can be hided in the cover image and recovered
biometric data can be use to feature extraction module or matching module. The
safety of biometric data could be increases for the malicious attack. One
problem is how to keep the information integrity from the modification of cover
image, e.g. crop, resize, image enhancement, lossy
compression.
Objectives
Research and develop
the steganography algorithms for biometrics
information that are preserve biometrics feature and characteristics for
further use, e.g. feature extraction and/or matching during various
modification to cover image. Added to that, biometric raw data can considered
as the cover image.
Common approaches
How it hide/unhide:
(Optionally compressed) Encrypt
Message and put it in Cover – usually symmetric key system. Removes
encryption header to hide encryption method.
LSB insertion: If cover is 8 bit
image, boost 8 bit image to 24 bit image, or color reduction.
Masking and filtering: Masking is more robust
than LSB insertion with respect to compression.
Algorithms and
transformations: DCT, FFT, Wavelet. Patchwork (redundant pattern encoding), spread
spectrum methods.
Available Scenarios
Table 1. Scenario combinations and considerations to use of steganography for biometrics.
|
Scenario |
1 |
2 |
3 |
4 |
|
Cover
type |
Natural
scene or any kind of picture |
Biometric
raw data |
||
|
Message
type |
Biometric
raw data |
Biometric
Template |
Transaction
info - time stamp, id, … |
Biometric
raw data/template |
|
Main
purpose |
Hide
image |
Hide
template |
Challenge/
Response |
Multi-modal |
|
Maximum
size of cover |
Does
not matter |
Image
specifications are defined by biometric system or related standard |
||
|
Format
of cover |
Depend
on stego algorithm |
|||
|
Stego algorithm selection |
Does
not matter |
Restricted
by format of biometrics system |
||
|
Maximum
load |
Does
not matter |
Limited,
cover image size defined by biometrics system |
||
|
Signal
changes of stego image |
Allowed
within visible inspection |
Ruled
by repeatability of extracted features |
||
|
Size
changes of stego image |
Does
not matter |
Generally,
not allowed |
||
|
To lossy compression
of stego image |
Superfluous
job |
Many
standards use compression (face: JPEG2000, fingerprint: WSQ) |
||
|
Recursive
embedding |
Impractical |
|||
|
Place
of this event |
Client |
Sensor
– at the time of aquire |
Client |
|
(a) (b) (c)
Figure 3. Stego result
analysis. (a) Original cover image, (b) message image, (c) Difference of
(a) with stego image (equalized)