Biometric Authentication via

erosjellyΑσφάλεια

23 Φεβ 2014 (πριν από 3 χρόνια και 5 μήνες)

90 εμφανίσεις

Randomized Radon Transforms for
Biometric Authentication via

Fingerprint Hashing

2007 ACM Digital Rights Management Workshop

Alexandria, VA (USA)

October 29, 2007


Mariusz

H.
Jakubowski

Ramarathnam

Venkatesan

Microsoft Research

2007 ACM Digital Rights Management Workshop

October 29, 2007

2

Introduction


Biometrics: “What you are”


Measurements over bodily features (e.g., fingerprints)


Applications for security and convenience



Biometric hashing


One
-
way extraction of information from biometric data


Human identifiers for DRM authentication



Goals of our work:


New method for fingerprint hashing


Applications to strengthen and streamline DRM
security

2007 ACM Digital Rights Management Workshop

October 29, 2007

3

Overview


Introduction


Fingerprint hashing


Experimental results


Conclusion

Fingerprint hashing via Radon transform

2007 ACM Digital Rights Management Workshop

October 29, 2007

4

Fingerprint Hashing

Conversion of fingerprints to one
-
way hashes

for authentication applications




Fingerprint hash
: An irreversible compressed
representation of fingerprint data, extracted according
to a secret key.



Basic procedure:


Compute various metrics over a fingerprint image and
combine these into a hash vector.


Apply error correction and other methods to increase
hash robustness.

2007 ACM Digital Rights Management Workshop

October 29, 2007

5

Radon Transform


Standard: (
x,y
)


(
θ
, ρ), where
θ
and ρ denote angles and
distances of lines.


Line at angle
θ

and distance ρ from origin will result in high value of
transform coefficient (
θ
,

ρ).

Hash

transform: This line
-
based metric is replaced by a custom metric.

R(
θ
,

ρ)

Original image

2007 ACM Digital Rights Management Workshop

October 29, 2007

6

Randomizing the Transform


Standard:


Exhaustively enumerate all lines.


Typical metric: Compute projections of lines onto
image.


Randomized:


Generate a pseudorandom sequence of lines, using a
secret hashing key
.


Simpler metric: Compute
crossing counts

of lines with
image (i.e., number of times each line crosses or
grazes fingerprint curves).



Randomized transform leads to hashing scheme.

2007 ACM Digital Rights Management Workshop

October 29, 2007

7

Fingerprint Hashing: Example

Scanned fingerprint

Metric:
Crossing count
with
random lines and curves

2007 ACM Digital Rights Management Workshop

October 29, 2007

8

Fingerprint Hashing: Example

Scanned fingerprint

Metric:
Crossing count
with
random lines and curves

Cleaned
fingerprint

o

Generic clean
-
up: Filters, thresholds, etc.

o

Specialized methods:
VeriFinger

(
Neurotechnologija
, Inc.)

2007 ACM Digital Rights Management Workshop

October 29, 2007

9

Fingerprint Hashing: Example

Scanned fingerprint

5 random lines

Metric:
Crossing count
with
random lines and curves

Cleaned fingerprint

2007 ACM Digital Rights Management Workshop

October 29, 2007

10

Fingerprint Hashing: Example

Scanned fingerprint

25 21 24 25 25

5 random lines

Metric:
Crossing count
with
random lines and curves

Cleaned fingerprint

2007 ACM Digital Rights Management Workshop

October 29, 2007

11

Fingerprint Hashing: Example

Scanned fingerprint

25 21 24 25 25

22 17 21 23 23

22 22 27 24 25

14 23 25 27 25

5 random lines

15 random lines

Metric:
Crossing count
with
random lines and curves

Cleaned fingerprint

2007 ACM Digital Rights Management Workshop

October 29, 2007

12

Fingerprint Hashing: Example

Scanned fingerprint

25 21 24 25 25

22 17 21 23 23

22 22 27 24 25

14 23 25 27 25

5 random lines

15 random lines

Metric:
Crossing count
with
random lines and curves

10 random curves

Cleaned fingerprint

3 24 44 27 32

8 16 24 37 31

Hashes (crossing counts)

2007 ACM Digital Rights Management Workshop

October 29, 2007

13

Some Metrics for Hashing


Counts of crossings with lines and curves


Curvatures of fingerprint lines within random regions


Numbers and types of minutiae contained in random regions
(e.g., rectangles)

7 6 0 1 2 2

2007 ACM Digital Rights Management Workshop

October 29, 2007

14

Hash Properties


Secret key or password used to determine
metric types and parameters


Controllable length and security (e.g., 64,
128, or 256 bits)


Resistance against minor scanner
distortions and noise

2007 ACM Digital Rights Management Workshop

October 29, 2007

15

Fingerprint Authentication


Standard authentication
: Compare fingerprint scans
against stored “correct” fingerprints.


Hash
-
based authentication
: Compare hashes of
scanned fingerprints with stored “correct” hashes.



Benefits of hashes:


Actual fingerprints need not be stored for comparison.


Stolen hashes do not reveal or compromise entire
fingerprints.


Key
-
derived hashes bind passwords and fingerprints
tightly.


Short hash length allows usage in network protocols, Web
services, etc.

2007 ACM Digital Rights Management Workshop

October 29, 2007

16

Experiments

Original fingerprint

Hash: 28 19 21 23 22

2007 ACM Digital Rights Management Workshop

October 29, 2007

17

Experiments

Original fingerprint

Hash: 28 19 21 23 22

Distorted fingerprint

Hash:

29 19 20 23 22

Difference:
1 0
-
1 0 0

o

StirMark

distortions used

o

Approximation of real
-
life scanner distortions

2007 ACM Digital Rights Management Workshop

October 29, 2007

18

Experiments

Original fingerprint

Hash: 28 19 21 23 22

Distorted fingerprint

Hash:

29 19 20 23 22

Difference:
1 0
-
1 0 0

Different hash key

Hash :

20 26 28 21 17

Difference:
-
8 7 7
-
2
-
5

2007 ACM Digital Rights Management Workshop

October 29, 2007

19

Experiments

Original fingerprint

Hash: 28 19 21 23 22

Different fingerprint #1

Hash:

38 17 24 34 28

Difference:

10
-
2 3 11 6

Distorted fingerprint

Hash:

29 19 20 23 22

Difference:
1 0
-
1 0 0

Different hash key

Hash :

20 26 28 21 17

Difference:
-
8 7 7
-
2
-
5

2007 ACM Digital Rights Management Workshop

October 29, 2007

20

Experiments

Original fingerprint

Hash: 28 19 21 23 22

Different fingerprint #1

Hash:

38 17 24 34 28

Difference:

10
-
2 3 11 6

Different fingerprint #2

Hash:

19 26 18 24 23

Difference:

-
9 7
-
3 1 1

Distorted fingerprint

Hash:

29 19 20 23 22

Difference:
1 0
-
1 0 0

Different hash key

Hash :

20 26 28 21 17

Difference:
-
8 7 7
-
2
-
5

2007 ACM Digital Rights Management Workshop

October 29, 2007

21

Experimental Results

0
5
10
15
20
25
0
10
20
30
40
50
60
70
80
90
Fingerprint Number
Distance
Distances between each fingerprint and its distorted version

Distances between each fingerprint and other distinct fingerprints

5 random lines

2007 ACM Digital Rights Management Workshop

October 29, 2007

22

Experimental Results

0
5
10
15
20
25
0
10
20
30
40
50
60
70
80
90
Fingerprint Number
Distance
0
5
10
15
20
25
0
100
200
300
400
500
600
Fingerprint Number
Distance
Distances between each fingerprint and its distorted version

Distances between each fingerprint and other distinct fingerprints

5 random lines

50 random lines

2007 ACM Digital Rights Management Workshop

October 29, 2007

23

Experimental Results

0
5
10
15
20
25
0
100
200
300
400
500
600
Fingerprint Number
Distance
Distances between each fingerprint and its distorted version

Distances between each fingerprint and other distinct fingerprints

50 random lines

200 random lines

(diminishing returns)

0
5
10
15
20
25
0
500
1000
1500
2000
2500
Fingerprint Number
Distance
2007 ACM Digital Rights Management Workshop

October 29, 2007

24

Conclusion


Contributions


Methodology to extract fingerprint entropy


Applications in biometric authentication


Address “too many passwords” problem


Augment password
-
based schemes



Future work


Handling scanner distortions


Naturally robust metrics


Better error correction


Explicit fingerprint synchronization


Applications to other biometric data


Retinal blood vessels


Vein patterns on hands