Fingerprint Verification

advertisement
Fingerprint Identification
BIOM 426
Instructor: Natalia A. Schmid
January 21, 2004
1
Introduction
Applications:
- law enforcement
- access to computer, network, bank-machine, car, home
- security applications (US Visit)
January 21, 2004
2
Introduction
Factors in favor of fingerprint applications:
• small and inexpensive capture devices (about 100 USD);
• fast computing hardware;
• recognition rate meets the needs of many applications (about 1 sec);
• increasing number of networks and Internet transactions;
• awareness of the need for ease-of-use as an important component of reliable
security
• well accepted by public
January 21, 2004
3
History
• Use of fingerprints for identification since 7000 to 6000 BC by ancient
Assyrians and Chinese (prints on pottery, clay, bricks).
• Fingerprinting of criminals for identification ~ Babylon around 1792-1750 BC.
January 21, 2004
4
History
• In the mid-1800's two facts were established: (i) no two fingerprints have
the same ridge pattern and (ii) fingerprint pattern have good permanence.
• Use of fingerprints for criminal identification in Argentina in 1892.
• Henry's fingerprint classification system was introduced in 1897.
• Computer processing began in 1960s with introduction of computer
hardware.
• Since 1980s fingerprints are used in non-criminal applications (due to
personal computers and optical scanners).
• Personal use ~ due to introduction of inexpensive capture devises and
reliable matching algorithms.
January 21, 2004
5
Feature Types
The lines that flow in various patterns
across fingerprints are called ridges
and the space between ridges are
valleys.
Fingerprint features (associated with
some matching algorithm):
• ridge pattern - global pattern
matching;
1 and 2 are endings;
3 is bifurcation
• minutiae (ridge ending and ridge
bifurcation)
- minutiae matching;
- attributes: type, (x,y)location, orientation
January 21, 2004
6
Feature Types
• core and delta are used for
classification or as landmarks;
- core is a center of pattern
- delta is a point where three
patterns deviate;
• pore location - the finest level of
detail
Required resolution:
1000 dpi
January 21, 2004
7
Block Diagram
Database 1
Off-line processing
No
Database 2
Fingerprint
Scanner
Quality
Check
Yes
Image
Enhancement
Feature
Extraction
Classification
Database 5
On-line processing
No
Yes
Fingerprint
Scanner
Quality
Check
Image
Enhancement
Feature
Extraction
Classification
Minutia
Matching
Matching
Score
January 21, 2004
8
Image processing
Goal: to obtain the best quality leading to the best match result.
Steps:
- image noise reduction and enhancement,
- segmentation,
- singularity detection,
- manutiae detection, and
- matching.
Image specifications:
- 8-bit gray scale (256 levels);
- 500 dpi resolution;
- (1-by-1) inch size.
January 21, 2004
9
Image Enhancement
Noise in the fingerprint image is due to:
dry or wet skin, dirt, cut, worn, noise of the
capture device.
Orientation Field
Two image enhancement operations:
(i) the adaptive matched filter
(enhances ridges oriented in
the same direction as those
I
in the same locality) ;
(ii) adaptive thresholding
(binarization: im2bw; graythresh).
Estimation of orientation field (gradient
method, slit-sums, etc.).
Local adaptive thresholding can be used
(images with different contrast).
Binarized Image
January 21, 2004
10
Image Enhancement
Thinning reduces ridge width to a single pixel
(Matlab: bwmorph)
Preserves connectivity and minimizes the
number of artifacts, e.g. erroneous
bifurcations.
Conclusions:
Image processing is time consuming.
Thinned Image
However, the results of all subsequent
operations depend on the quality of image as
captured and processed at this
stage.
January 21, 2004
11
Other image enhancement methods
Image can be divided into windows. Local ridge orientation is found for
each window.
• Spatial or frequency domain processing.
• D. Maio and D. Maltoni proposed an algorithm that traces ridges and
detect minutiae using grayscale image.
• Multi-resolution approach (multiple window sizes).
January 21, 2004
12
Feature Extraction
Singularity and Core Detection:
- Poincare index
- local histogram method
- irregularity operator
- multi-resolution approach
January 21, 2004
13
Feature Extraction
Endings have one black pixel in 8-neighborhood.
Bifurcations have more than 2 black pixels in 8-neighborhood.
Noise and previous processing steps produce extraneous minutiae. They can
be reduced by a thresholding method.
Example:
- bifurcation with short branch is a spur;
- two endings on a short line is line due to noise;
- two endings closely opposing is a broken ridge;
- endings at the boundary is due to projection;
January 21, 2004
14
Feature Extraction
Each minutia is described by:
- minutia type,
- (x,y)-location,
- minutia direction.
Minutia template - minutia with all its attributes.
Number of minutiae: from 10 to 100.
Type
1 bit
Location (each x and y)
9 bits
Direction
8 bits
Then 100 features require 2700 bits.
January 21, 2004
15
Matching
Method 1:
- Pick a minutia in one of templates.
- Compare a graph formed by its neighborhood against all possible
neighborhoods in the second template.
(distance between minutiae and their orientations)
Use a distance measure to calculate similarity.
Result is a match score.
Method 2:
Align fingerprints using landmarks (core and delta).
Core and delta can be found using Poincare index or using estimated orientation
flow.
January 21, 2004
16
Matching
Method 3: Sort minutiae in some order. Then compare ordered vectors.
Method 4: Use other features to describe minutiae (e.g. length and curvature
of ridge).
Method 5: Matching on the basis of overall ridge pattern (correlation,
global matching, image multiplication). Translate one image over another
and perform multiplication at each pixel. Find the sum. Sum is the highest
when images match.
Method 6: Perform correlation matching in frequency domain. Perform
2-D FFT; multiply two transformed images; sum multiplied values.
Correlation matching is less tolerable to noise and non-linear transformation.
Problems: translational, rotational freedom (depend on landmarks).
January 21, 2004
17
Evaluation
Measure of performance?
In stochastic estimation and detection, a typical measure is the average
probability of error or, for a binary case, ROC curve.
There is no good stochastic model.
Outcomes are: match or no match.
Given a large database of labeled templates.
Test the system.
Count the number of erroneous decisions.
January 21, 2004
18
Evaluation
Fingerprint images are very noisy.
January 21, 2004
19
Evaluation
Matching Score 
Number of minutia pairs that match
Total number of minutia pairs
• Two fingerprint from two different
individuals may produce a high Matching
Score (an error);
• Two fingerprints from the same
individual may produce a low Matching
Score (an error)
January 21, 2004
20
Evaluation
There are two types of error:
FAR = ratio of number of instances of
pairs of different fingerprints found to
(erroneously) match to total number
of match attempts.
FRR = ratio of number of instances of
pairs of same fingerprint are found not
to match to total number of match
attempts.
January 21, 2004
21
Evaluation
Receiver Operating Curve (ROC)
January 21, 2004
22
Image Capture Devices
Analog-to-Digital converter
Reading
device
Responsible for communicating
with external devices
Secure identification system requirements:
• protection/encription (secure identification system)
• discard fake fingerprints
Additional Issues: storage for large AFIS; compression methods.
January 21, 2004
23
Fingerprint Images: Resolution
Number of dots or pixels per inch.
Minimum resolution
for extracting minutiae
(correlation techniques)
Minimum resolution for FBIcompliant scanners
January 21, 2004
24
Image Capture Devices
Optical: based on frastrated total internal reflection (FTIR)
Size:
6 x 3 x 6 in. in 1970s
3 x 1 x 1 in. mid-1990s
Cost drop: $1500 - $100
Solid-state sensors:
- capacitive,
- pressure sensitive,
- temperature sensitive
Size:
1 x 1 in. (small)
Resolution: 500 dpi
Ultrasonic scanning: high quality images
January 21, 2004
25
Image Capture Devices
Fingerprint sensors can be embedded in a variety of devices
for user recognition purposes.
January 21, 2004
26
Image Capture Devices
a) a live-scan FTIR-based optical scanner;
b) a live-scan capacitive scanner;
c) a live-scan piezoelectric scanner;
d) a live-scan thermal scanner;
e) an off-line inked impression;
f) a latent fingerprint
January 21, 2004
27
Available Databases
1. NIST special databases
http://www.itl.nist.gov/iaui/894.03/databases/defs/vip_dbases.html
2. Fingerprint Verification Competition (FVC2000, FVC2002)
http://bias.csr.unibo.it/fvc2000/
3. FBI database (>200 million fingerprints)
4. East Shore Technologies
http://www.east-shore.com/data.html
January 21, 2004
28
References
1. D. Maltoni, et al., Handbook of Fingerprint Recognition, Springer, New York,
2003.
2. A. Jain, et al., Biometrics: Personal Identification in Networked Society,
Ch. 2, pp. 43-64, Kluwer Acad. Pub., 1999.
3. “An Identity Authentication System Using Fingerprints,” Proceedings of the
IEEE, vol. 85, no. 9, 1997, pp. 1365-1388.
4. L. Hong, Y. Wan, and A. Jain, “Fingerprint Image Enhancement: Algorithm and
Performance Evaluation,” IEEE Tans. on PAMI, vol. 20, no. 8, 1998, pp. 777-789.
5. K. Karu and A.K. Jain, “Fingerprint Classification,” Pattern Recognition, Vol.
29, No. 3, pp. 389-404, 1996.
6. A.K. Jain, L. Hong and R. Bolle, “On-line Fingerprint Verification,” IEEE
Trans. on PAMI, Vol. 19, No. 4, pp. 302-314, 1997.
January 21, 2004
29
Download