© Jain, 2004
Fingerprint Matching TechniquesFingerprint Matching Techniques
• Minutiae-based
• Uses location, orientation, and minutia type
• Point pattern matching problem
• Hard decision is made on the correspondence
• Correlation-based
• Spatial correlation of template and query
• Sensitive to rigid and non-linear transformation
• Computationally expensive
• Ridge Feature-based
• Orientation and frequency of ridges, ridge shape,
texture information, etc. are used
• Suffers from low discriminative ability
© Jain, 2004
Texture-based Fingerprint
Representation
Texture-based Fingerprint
Representation
(c) Fourier spectrum of (a)(a) Ridges in local region (b) Ridge directions in (a)
The ridge pattern in a fingerprint may be viewed as an oriented
texture pattern having a fixed dominant spatial frequency and
orientation in a local neighborhood. The frequency is due to the
inter-ridge spacing present in the fingerprint and the
orientation is due to the ridge flow pattern
© Jain, 2004
Filtered ImagesFiltered Images
An input fingerprint image is filtered using 8 Gabor filters
all having the same frequency but different orientations
(0
o
, 22.5
o
, 45
o
, 67.5
o
, 90
o
, 112.5
o
, 135
o
, 157.5
o
)
© Jain, 2004
Texture-based RepresentationTexture-based Representation
Compute variance of
each cell
(a) Filtered image (b) Square tessellation (c) Feature values
© Jain, 2004
Ridge Feature MapRidge Feature Map
The filtered images are examined using a square tessellation
and the variance of pixel intensities in every cell is used as a
feature value
The ridge feature map is a fixed-length feature vector
*Ross et al, “A Hybrid Fingerprint Matcher”, Pattern Recognition, Vol. 36, July 2003
© Jain, 2004
Performance of Hybrid Matcher
(Minutiae & Texture)
Performance of Hybrid Matcher
(Minutiae & Texture)
© Jain, 2004
Correlation-based Matching Algorithm
• Normalized cross-correlation of regions around
corresponding minutia points is used to describe the
quality of a minutia match
• Gray level values around the minutiae retain most
of the local information; hence, this method
determines the degree of minutia match accurately
• Procrustes analysis of corresponding ridge curves is
used to estimate rotations and displacements
• Gabor filter-bank based technique is used for
fingerprint enhancement and segmentation
• Since correlation is done locally, the method is
reasonably resistant to non-linear deformations
© Jain, 2004
Fingerprint Matching using Local Correlation
Correlation
Minutia
Extraction
Estimation of Rotation and Translation
using Ridge Correspondences
Query MinutiaeQuery Image
Template Image Template Minutiae
Regions in Template
around Template Minutiae
Regions in Rotated Query Image around
Transformed Template Minutiae
Matching
Score
© Jain, 2004
Performance of Hybrid Matcher
(Minutiae and Local Correlation)
Performance of Hybrid Matcher
(Minutiae and Local Correlation)
MSU-VERIDICOM fingerprint database; 160 users and 4
impressions/user
© Jain, 2004
Performance of Hybrid Matcher
(Minutiae, Texture & Local Correlation)
Performance of Hybrid Matcher
(Minutiae, Texture & Local Correlation)
© Jain, 2004
FVC 2004 ResultsFVC 2004 Results
Algorithm EER(%) Avg Enroll
Time (sec)
Avg Match
Time (sec)
Avg Model
Size (KB)
Bioscrypt
Inc.
2.07 0.08 1.48
2.07
0.67
0.71
1.19
24
Sonda Ltd 2.10 2.07 1.3
Chinese
Academy of
Sciences
2.30 0.35 16.4
Gevarius 2.45 0.69 2.0
Jan Lunter 2.90 1.01 3.1
• Database:
– DB1: optical sensor "V300" by CrossMatch
– DB2: optical sensor "U.are.U 4000" by Digital Persona
– DB3: thermal sweeping sensor "FingerChip FCD4B14CB" by Atmel
– DB4: synthetic fingerprints