vii ii iii

advertisement
vii
TABLE OF CONTENTS
CHAPTER
1
2
TITLE
PAGE
DECLARATION
ii
DEDICATION
iii
ACKNOWLEDGEMENTS
iv
ABSTRACT
v
ABSTRAK
vi
TABLE OF CONTENTS
vii
LIST OF TABLES
x
LIST OF FIGURES
xi
LIST OF ABBREVIATIONS
xvi
LIST OF SYMBOLS
xvii
LIST OF APPENDICES
xix
INTRODUCTION
1
1.1
Introduction
1
1.2
Background of Research
2
1.3
Problem Statements
10
1.4
Research Goal
11
1.5
Objectives of Study
11
1.6
Scope of the Study
12
1.7
List of Research Contributions
13
1.8
Significance of the Study
14
1.9
Research Framework
15
1.10 Thesis Outline
16
LITERATURE REVIEW
18
2.1
18
Introduction
viii
2.2
2.3
2.4
2.5
Fingerprint Segmentation
19
2.2.1
Grey-scale Statistical Features
21
2.2.2
Local Directionality Features
22
2.2.3
Coherence Features
24
Fingerprint Enhancement
26
2.3.1
Normalization
28
2.3.2
Gaussian Filter
29
2.3.3
Directional Filter
31
Orientation Field Estimation
34
2.4.1
Orientation Field Estimation Using Gradient
35
2.4.2
Model-based Approach for Orientation Field
Estimation
38
Singular Point Detection
40
2.5.1
Poincaré Index Methods
40
2.5.2
Partitioning-based Methods
43
2.5.3
Methods Based on Local Characteristics
of the Orientation Field
2.6
3
44
Fingerprint Classification
46
2.6.1
Rule-based Approaches
47
2.6.2
Structure-based Approaches
50
METHODOLOGY
53
3.1
Introduction
53
3.2
Foreground Extraction
56
3.2.1
Grey-scale Normalization
56
3.2.2
Proposed Foreground Extraction Method
59
3.2.3
Noise Areas Identification and Marking
Using Gradient Coherence
65
3.3
Enhancement of Grey-level
68
3.4
Fingerprint Orientation Field
73
3.5
Orientation Field Enhancement
75
3.6
Region of Singular Point Detection
78
3.7
Singular Point Detection
82
3.8
A Two-tier Fingerprint Classification Technique
84
ix
3.9
4
3.8.1
The Classification Rule
85
3.8.2
Structure Shape Analysis of Orientation Fields
86
3.8.2.1
Examine True Cores
87
3.8.2.2
Examine True Deltas
92
Summary
95
EXPERIMENTAL RESULT AND DISCUSSION
97
4.1
Introduction
97
4.2
Dataset
97
4.3
Experimental Set-up
102
4.4
Experimental Results
104
4.4.1
Performance Measure of Fingerprint
Segmentation
4.4.2
104
Performance Measure of Noise Removal
and Ridge Pattern Enhancement for
Orientation Field Estimation
4.4.3
Performance Measurement of Singular
Point Detection
4.4.4
5
131
Performance Measurement of Fingerprint
Classification
4.5
119
Summary
143
155
CONCLUSION
157
5.1
Introduction
157
5.2
Contribution
160
5.3
Future Work
162
REFERENCES
164
Appendices A-G
172- 243
x
LIST OF TABLES
TABLE NO.
TITLE
2.1
Classification based on singularities
2.2
Summary of the fingerprint classification literature
review
4.1
46
52
Classes of fingerprint NIST Special Database 14
(f000001 to f027000 filenames)
4.2
PAGE
100
Scan type and sex of fingerprints NIST Special
Database 14 (f000001 to f027000 filenames)
101
4.3
Performance measures
103
4.4
Performance results of various fingerprint
segmentation methods from singular points detections
points of view
114
4.5
Comparison results of singular points detection
143
4.6
A confusion matrix of the experimental results of
the proposed fingerprint classification
145
4.7
Ambiguous fingerprints off the misclassified prints
145
4.8
Accuracy rates of the Cappelli and Maltoni’s
classification methods using 27,000 fingerprints
(f0000001 to f0027000) of DB14 (Source:
Cappelli and Maltoni (2009))
145
xi
LIST OF FIGURES
FIGURE NO.
TITLE
PAGE
1.1
Example of five classes
3
1.2
Sample of ambiguous prints
4
1.3
Rejected prints
5
1.4
Dry, wet, and tainted prints
6
1.5
(a) Three fingerprints of the same class that have very
different characteristic (large-intra-class variability).
(Source: NIST Special Database 14)
7
(b) Three fingerprints belonging to different classes that
have similar appearance (small-inter-class variability).
(Source: NIST Special Database 14)
7
1.6
Ridges and valleys structure (Right-loop class)
8
1.7
Core and delta on a fingerprint image
9
1.8
Research framework flowchart
17
2.1
A sample of fingerprint segmentation results;
(a) Fingerprint image, (b) background regions,
foreground regions, and noise patches of the fingerprint
image
2.2
Quality of fingerprint image; (a) good, (b) broken/cut,
(c) low contrast, (d) dry, (e) wet, and (f) stain
2.3
20
28
Convolution masks; (a) 3× 3 Gaussian smooth filter,
(b) 5 × 5 Gaussian smooth filter
30
2.4
Oriented window and x-signature (Hong et al. 1998).
32
2.5
Eight DMF templates (Wu et al., 2004)
33
2.6
A site consists of 3× 3 blocks with the target block V
xii
in the centre and four overlapping neighbourhood
D1 , D 2 , D3 , D 4 (Wang et al. 2005)
2.7
34
Ridge flow direction at pixel (i, j ) and angle θ xy (i, j )
indicating the direction of flow with respect to the
horizontal axis
2.8
The labelling of neighbourhood pixels used to explain
the Sobel operator
2.9
35
36
Orientation image partitioning with the MASK
approach (Cappelli et al., 1999). The intersections
between region border lines denote fingerprint
singularities
2.10
The mask for capturing the centre point
(Hsieh et al. (2005))
2.11
45
Samples after flow-like tracing (Left-loop, Right-loop
and Tented-arch) (Wang and Xie, 2004)
3.1
44
The candidate regions for singular points
(Park et al., 2006)
2.12
43
51
On Automatic Fingerprint Classification Scheme
(AFCS): Proposed AFCS complements generic AFCS
(Note: The dashed rectangle denotes the proposed
AFCS)
55
3.2
The fingerprint segmentation scheme
56
3.3
Grey-scale normalization process
57
3.4
Normalization process. (a) Input fingerprint image
(NIST DB 14: f0000023), (b) normalized fingerprint
image
59
3.5
Flowchart of foreground extraction
60
3.6
(a) Horizontal Sobel mask operator S x ( p, q ) ,
(b) Vertical Sobel mask operator S y ( p, q )
3.7
62
Result of the foreground extraction process using the
proposed segmentation method. (Note: the input image
is referred to normalized image in Figure 3.2(b) above)
65
xiii
3.8
Noise areas of the foreground are identified and
labelled. The white coloured blocks indicate the noise
areas. (Note: the input image is referred to the extracted
foreground image in Figure 3.7 above)
67
3.9
CCB gradient magnitude of size 5 × 5
70
3.10
Result of the enhancement process in the noise areas
(circles highlight the noise areas), (a) Orientation image
before enhancement, (b) orientation image after the
enhancement. (Note: the input is the foreground image
in Figure 3.8)
3.11
72
Result of orientation field estimations in the fingerprint
image
75
3.12
A 3× 3 CCB of orientation field
76
3.13
Orientation fields of fingerprint after enhancement
using minimum variance of orientation fields.
77
3.14
Set of four pre-defined block of arches.
78
3.15
The mechanics of BoI
80
3.16
Results of CRSP using the proposed method. (Source
image: NIST Database 14, f0000023). Note: Blue
squares indicate CRSP
3.17
82
Core and delta detections using the proposed methods
(NIST DB 14: F0000023). Note: White and yellow
squares indicate the core and delta, respectively
3.18
Types of the core point (a) Upper-core, (b) lower-core,
(c) right-core, and (d) left-core
3.19
84
88
Omni-directional orientation fields neighbouring the
core points: (a) Upper-core, (b) lower-core, (c) rightcore, and (d) left- core
3.20
90
Omni-directional orientation fields neighbouring a
delta. (a) Right-delta, and (b) left-delta
93
4.1
Tagged comment in WSQ compressed file
99
4.2
Scar fingerprint
100
xiv
4.3
Fingerprint with a cross-referenced classification or
ambiguous fingerprint
4.4
101
Fingerprint with extraneous objects: (a) Like
handwritten characters and (b) other artefacts common
to inked fingerprints
4.5
(a) - (e): Results of the segmentation process of good
quality fingerprints
4.6
109
(a) - (e): Results of the segmentation process of low
contrast fingerprints
4.9
107
(a) - (e): Results of the segmentation process of wet
fingerprints
4.8
106
(a) - (e): Results of the segmentation process of dry
fingerprints
4.7
102
111
(a) - (e): Results of the segmentation process of
fingerprints which contain ink stains and handwritten
characters
4.10
112
On segmentation results: Local mean versus the
proposed method. (a) Original fingerprint; (a1) outcome
of local mean method; (a2) outcome of proposed method
4.11
116
On segmentation results: Zang and Yan versus the
proposed method. (b) Original fingerprint; (b1) outcome
of Zang and Yan method; (b2) outcome of proposed
method
4.12
117
On segmentation results: Local variance of gradient
magnitude method Vs the proposed method. (c) Original
fingerprint; (c1) outcome of local variance of gradient
magnitude method; (c2) outcome of the proposed method
4.13
118
On segmentation results: Outcome of the proposed
segmentation method. (d) Original fingerprint;
(d1) outcome of the proposed method
4.14
(a) - (e): Some results of the enhancement process for
good quality fingerprints
4.15
118
121
(a) - (e): Some results of the enhancement process for
dry fingerprints
122
xv
4.16
(a) - (e): Some results of the enhancement process for
wet fingerprints
4.17
(a) - (e): Some results of the enhancement process for
cuts fingerprints
4.18
126
(a) - (e): Some results of the enhancement process for
bruises fingerprints
4.19
124
127
(a) - (e): Some results of the enhancement process for
Low contrast fingerprints
129
4.20
Artificial singular point regions
130
4.21
Mean square error results of both Wang and the proposed
methods
4.22
(a) - (e): Results of the singular points detection for good
fingerprints
4.23
140
(a) - (e): Results of the singular points detection for low
contrast fingerprints
4.28
138
(a) - (e): Results of the singular points detection for bruises
fingerprints
4.27
137
(a) - (e): Results of the singular points detection for cuts
fingerprints
4.26
135
(a) - (e): Results of the singular points detection for wet
fingerprints
4.25
133
(a) - (e): Results of the singular points detection for dry
fingerprints
4.24
131
142
Some examples of correctly classified prints of various
qualities viz. good, dry, wet, cut, bruise, low contrast,
and stain
148
4.29
Fake core (file name: f0002084)
149
4.30
(a) - (c): Fingerprints classes with one missing delta.
150
4.31
A sample of large-intra-class variation prints
152
4.32
Small-inter-class variations
154
4.33
Poor quality fingerprints
155
xvi
LIST OF ABBREVIATIONS
FBI
-
Federal Bureau Investigation
AFIS
-
Automatic Fingerprint Identification System
AFCS
-
Automatic Fingerprint Classification System
NIST
-
National Institute of Standards and Technology
NN
-
Neural Network
HMM
-
Hidden Markov Model
2D
-
Two dimensional
DMF
-
Directional Median Filter
DB14
-
NIST Special Database 14
SVM
-
Support Vector Machine
FVC2002
-
Second International Competition for Fingerprint
Verification Algorithm
CCB
-
Cross Centre Block
BoI
-
Block of Interest
CB
-
Convergence Block
CRSP
-
Candidate Region Singular Point
GoB
-
Group of Block
CoB
-
Combination of Block
NCIC
-
National Crime Information Centre
WSQ
-
Wavelet Scalar Quantisation
MC
-
Miss Rate of Cores
MD
-
Miss Rate of Deltas
FC
-
False alarm rate of Cores
FD
-
False alarm rate of Deltas
MSE
-
Mean Square Error
xvii
LIST OF SYMBOLS
I (m, n)
-
Intensity value of the pixel at the m-th row and n-th column in the
fingerprint image
W ×H
-
Size of fingerprint image
Mg
-
Global mean value of fingerprint image
Vg
-
Global variance value of fingerprint image
N (m, n)
-
Intensity value of the pixel at the m-th row and n-th column in the
normalized fingerprint image
Mg 0
-
Desired mean value for determine normalization
Vg 0
-
Desired variance value for determine normalization
Mn
-
Global mean value of normalized fingerprint image
B×B
-
Size of block in the fingerprint image
Mb(i, j )
-
Local mean value of block (i, j )
Gx (m, n)
-
Gradient of pixel (m, n) in horizontal direction
G y (m, n)
-
Gradient of pixel (m, n) in vertical direction
Sx
-
Horizontal Sobel mask operator
Sy
-
Vertical Sobel mask operator
Gr (m, n)
-
Gradient magnitude of pixel (m, n)
Gth
-
Threshold value for gradient
c
-
Threshold factor for gradient
Mgr (i, j )
-
Local mean value of gradient magnitude of block (i, j )
Vgr (i, j )
-
Local variance value of gradient magnitude of block (i, j )
Coh(i, j )
-
Coherence value of block (i, j )
Vx (i, j )
-
Vector gradient x-direction of block (i, j )
Vy (i, j )
-
Vector gradient y-direction of block (i, j )
xviii
Vz (i, j )
-
Vector gradient resultant direction of block (i, j )
D
-
Set of gradient magnitude in CCB
Di
-
Subset i-th of gradient magnitude in CCB
Mgm Di
-
Mean value of subset i-th of gradient magnitude in CCB
Vgm Di
-
Variance value of subset i-th of gradient magnitude in CCB
P
-
Set of normalized fingerprint image in CCB
Pi
-
Subset i-th of normalized fingerprint image in CCB
Mi Pi
-
Mean value of subset i-th of normalized fingerprint image in CCB
Q
-
Set of gradient angle in CCB
Qi
-
Subset i-th of gradient angle in CCB
Ma Qi
-
Mean value of subset i-th of gradient angle in CCB
Va Qi
-
Variance value of subset i-th of gradient angle in CCB
Gs( p, q)
-
Gaussian filtering mask
θ (i, j )
-
Gradient angle of orientation field
δ (k )
-
Closed-curve angle
PI
-
Poincaré index in continuous vector
P(i, j )
-
Poincaré index in discrete vector
Nc
-
Number of cores
Nd
-
Number of deltas
xix
LIST OF APPENDICES
APPENDIX
TITLE
A
Scars Fingerprint Image
B
Results of Fingerprint Segmentation using Local
Mean Method
C
G
195
207
Results of Fingerprint Segmentation using
Proposed Method
F
183
Results of Fingerprint Segmentation using Local
Variance of Gradient Magnitude
E
172
Results of Fingerprint Segmentation using a
Combination of Local Mean and Coherence
D
PAGE
219
Results of Singular Points Detection using Poincaré
Index Method
231
List of Publications
243
Download