U.S. patent application number 16/297093 was filed with the patent office on 2019-07-04 for feature-based matcher for distorted fingerprint matching.
This patent application is currently assigned to MorphoTrack, LLC. The applicant listed for this patent is MorphoTrack, LLC. Invention is credited to Hui Chen, Peter Zhen-Ping Lo.
Application Number | 20190205590 16/297093 |
Document ID | / |
Family ID | 58691105 |
Filed Date | 2019-07-04 |
![](/patent/app/20190205590/US20190205590A1-20190704-D00000.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00001.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00002.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00003.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00004.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00005.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00006.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00007.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00008.png)
![](/patent/app/20190205590/US20190205590A1-20190704-D00009.png)
United States Patent
Application |
20190205590 |
Kind Code |
A1 |
Lo; Peter Zhen-Ping ; et
al. |
July 4, 2019 |
Feature-Based Matcher for Distorted Fingerprint Matching
Abstract
In one aspect, methods that use a novel representation, referred
to as an octant feature vector (OFV), are used for matching
distorted fingerprints. For instance, a feature-based matcher for
distorted fingerprint matching may use a two-step local and global
matching scheme to compare a set of feature vectors that are
derived from minutiae of a reference fingerprint and a search
fingerprint. The relative geometric relationships between the
reference minutia and nearest minutiae may be derived and encoded
into a feature vector based on orientation difference. The OFV is
invariant to the rigid transformations and is insensitive to
nonlinear distortions since the relative geometric relationships
are independent from the rigid transformation.
Inventors: |
Lo; Peter Zhen-Ping;
(Mission Viejo, CA) ; Chen; Hui; (Foothill Ranch,
CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
MorphoTrack, LLC |
Anaheim |
CA |
US |
|
|
Assignee: |
MorphoTrack, LLC
Anaheim
CA
|
Family ID: |
58691105 |
Appl. No.: |
16/297093 |
Filed: |
March 8, 2019 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15609293 |
May 31, 2017 |
10255477 |
|
|
16297093 |
|
|
|
|
14942156 |
Nov 16, 2015 |
9690973 |
|
|
15609293 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06K 9/00067 20130101;
G06K 9/00046 20130101; G06K 9/0008 20130101; G06K 9/001 20130101;
G06T 7/0002 20130101 |
International
Class: |
G06K 9/00 20060101
G06K009/00; G06T 7/00 20060101 G06T007/00 |
Claims
1. A method performed by one or more computers, the method
comprising: identifying a geometrically aligned region between a
search fingerprint image and a reference fingerprint image;
determining a set of globally aligned mated minutia pairs based on
the geometrically aligned region between the search fingerprint
image and the reference fingerprint image; for each mated minutia
pair included in the set of globally aligned mated minutia pairs:
computing a first score representing a feature difference between
(i) a search minutia and (ii) reference minutiae included in a set
of reference minutiae that are identified as being closest
neighboring minutiae to the search minutia in the geometrically
aligned region, and computing a second score representing a ratio
of a number of paired mated minutiae in the geometrically aligned
region included in the set of reference minutiae relative to a
number of paired unmated minutiae in the geometrically aligned
region included in the set of reference minutiae; computing a match
similarity score between the search fingerprint image and the
reference fingerprint image based at least on the first score and
the second score; and providing the match similarity score for
output.
2. The method of claim 1, wherein identifying the geometrically
aligned region between the search fingerprint image and the
reference fingerprint image comprises: computing, for each search
minutia included in the set of globally aligned mated minutia
pairs, a local similarity score based at least on search octant
feature vectors of the search minutia and reference octant feature
vectors of the set of reference minutiae; and determining a
rotation angle between the search fingerprint image and the
reference fingerprint image based on the local similarity
scores.
3. The method of claim 2, further comprising: determining, based on
the local similarity scores, a set of local closest matched minutia
pairs; computing sets of rotation parameters based on the set of
local closest matched minutia pairs.
4. The method of claim 3, wherein computing the sets of rotation
parameters comprises identifying angles corresponding to a set of
top bins of an angle offset histogram of the set of local closest
matched minutiae pairs.
5. The method of claim 1, wherein each mated minutia pair included
in the set of globally aligned mated minutia pairs comprises a
search minutia and a corresponding reference minutiae included in
the geometrically aligned region that is determined to have
geometrically consistent features to the search minutia.
6. The method of claim 1, wherein: the search fingerprint image
comprises a distorted fingerprint; and the match similarity score
is used to determine a fingerprint match between the search
fingerprint image and the reference fingerprint image.
7. The method of claim 1, further comprising: comparing, for each
mated minutia pair included in the set of globally aligned mated
minutia pairs, features of a search minutia of a mated minutia pair
and features of a corresponding reference minutia of the mated
minutia pair; based on comparing the features of the search minutia
of the mated minutia pair and the features of the corresponding
reference minutia of the mated minutia pair: identifying a set of
rotation parameters for the geometrically aligned region, and
identifying a set of translation parameters for the geometrically
aligned region; and aligning the search minutiae to the reference
minutiae included in the set of globally aligned mated minutia
pairs based at least on the set of rotation parameters and the set
of translation parameters.
8. A system comprising: one or more computers; and one or more
storage devices storing instructions that are operable, when
executed by one or more computers, to cause the one or more
computers to perform operations comprising: identifying a
geometrically aligned region between a search fingerprint image and
a reference fingerprint image; determining a set of globally
aligned mated minutia pairs based on the geometrically aligned
region between the search fingerprint image and the reference
fingerprint image; for each mated minutia pair included in the set
of globally aligned mated minutia pairs: computing a first score
representing a feature difference between (i) a search minutia and
(ii) reference minutiae included in a set of reference minutiae
that are identified as being closest neighboring minutiae to the
search minutia in the geometrically aligned region, and computing a
second score representing a ratio of a number of paired mated
minutiae in the geometrically aligned region included in the set of
reference minutiae relative to a number of paired unmated minutiae
in the geometrically aligned region included in the set of
reference minutiae; computing a match similarity score between the
search fingerprint image and the reference fingerprint image based
at least on the first score and the second score; and providing the
match similarity score for output.
9. The system of claim 8, wherein identifying the geometrically
aligned region between the search fingerprint image and the
reference fingerprint image comprises: computing, for each search
minutia included in the set of globally aligned mated minutia
pairs, a local similarity score based at least on search octant
feature vectors of the search minutia and reference octant feature
vectors of the set of reference minutiae; and determining a
rotation angle between the search fingerprint image and the
reference fingerprint image based on the local similarity
scores.
10. The system of claim 9, wherein the operations further comprise:
determining, based on the local similarity scores, a set of local
closest matched minutia pairs; and computing sets of rotation
parameters based on the set of local closest matched minutia
pairs.
11. The system of claim 10, wherein computing the sets of rotation
parameters comprises identifying angles corresponding to a set of
top bins of an angle offset histogram of the set of local closest
matched minutiae pairs.
12. The system of claim 8, wherein each mated minutia pair included
in the set of globally aligned mated minutia pairs comprises a
search minutia and a corresponding reference minutiae included in
the geometrically aligned region that is determined to have
geometrically consistent features to the search minutia.
13. The system of claim 8, wherein: the search fingerprint image
comprises a distorted fingerprint; and the match similarity score
is used to determine a fingerprint match between the search
fingerprint image and the reference fingerprint image.
14. The system of claim 8, wherein the operations further comprise:
comparing, for each mated minutia pair included in the set of
globally aligned mated minutia pairs, features of a search minutia
of a mated minutia pair and features of a corresponding reference
minutia of the mated minutia pair; based on comparing the features
of the search minutia of the mated minutia pair and the features of
the corresponding reference minutia of the mated minutia pair:
identifying a set of rotation parameters for the geometrically
aligned region, and identifying a set of translation parameters for
the geometrically aligned region; and aligning the search minutiae
to the reference minutiae included in the set of globally aligned
mated minutia pairs based at least on the set of rotation
parameters and the set of translation parameters.
15. A non-transitory machine-readable data storage device that
stores instructions that, when executed by a data processing
apparatus, cause the data processing apparatus to perform
operations comprising: identifying a geometrically aligned region
between a search fingerprint image and a reference fingerprint
image; determining a set of globally aligned mated minutia pairs
based on the geometrically aligned region between the search
fingerprint image and the reference fingerprint image; for each
mated minutia pair included in the set of globally aligned mated
minutia pairs: computing a first score representing a feature
difference between (i) a search minutia and (ii) reference minutiae
included in a set of reference minutiae that are identified as
being closest neighboring minutiae to the search minutia in the
geometrically aligned region, and computing a second score
representing a ratio of a number of paired mated minutiae in the
geometrically aligned region included in the set of reference
minutiae relative to a number of paired unmated minutiae in the
geometrically aligned region included in the set of reference
minutiae; computing a match similarity score between the search
fingerprint image and the reference fingerprint image based at
least on the first score and the second score; and providing the
match similarity score for output.
16. The device of claim 15, wherein identifying the geometrically
aligned region between the search fingerprint image and the
reference fingerprint image comprises: computing, for each search
minutia included in the set of globally aligned mated minutia
pairs, a local similarity score based at least on search octant
feature vectors of the search minutia and reference octant feature
vectors of the set of reference minutiae; and determining a
rotation angle between the search fingerprint image and the
reference fingerprint image based on the local similarity
scores.
17. The device of claim 16, wherein the operations further
comprise: determining, based on the local similarity scores, a set
of local closest matched minutia pairs; computing sets of rotation
parameters based on the set of local closest matched minutia
pairs.
18. The device of claim 17, wherein computing the sets of rotation
parameters comprises identifying angles corresponding to a set of
top bins of an angle offset histogram of the set of local closest
matched minutiae pairs.
19. The device of claim 15, wherein each mated minutia pair
included in the set of globally aligned mated minutia pairs
comprises a search minutia and a corresponding reference minutiae
included in the geometrically aligned region that is determined to
have geometrically consistent features to the search minutia.
20. The device of claim 15, wherein: the search fingerprint image
comprises a distorted fingerprint; and the match similarity score
is used to determine a fingerprint match between the search
fingerprint image and the reference fingerprint image.
Description
FIELD
[0001] The present disclosure relates generally to fingerprint
identification systems.
BACKGROUND
[0002] Pattern matching systems such as ten-print or fingerprint
matching systems play a critical role in criminal and civil
applications. For example, fingerprint identification is often used
for identify and track suspects and in criminal investigations.
Similarly, fingerprint verification is used in in civil
applications to prevent fraud and support other security
processes.
SUMMARY
[0003] In general, the matching of distorted fingerprints pose
several challenges to fingerprint identification and matching
technologies. For instance, distortions can impact matching
accuracy because the distortions may change both the geometric
locations and orientations of individual minutiae.
[0004] Accordingly, one innovative aspect of the subject matter
described throughout this specification involves methods that use a
novel representation, referred to as an octant feature vector
(OFV), for matching distorted fingerprints. For instance, a
feature-based matcher for distorted fingerprint matching may use a
two-step local and global matching scheme to compare a set of
feature vectors that are derived from minutiae of a reference
fingerprint and a search fingerprint. The relative geometric
relationships between the reference minutia and nearest minutiae
may be derived and encoded into a feature vector based on
orientation difference. The OFV is invariant to the rigid
transformations and is insensitive to nonlinear distortions since
the relative geometric relationships are independent from the rigid
transformation. The OFVs that are generated for individual minutiae
may be compared to compute a similarity between the reference and
search fingerprint.
[0005] Implementations may include one or more of the following
features. For example, a method for matching distorted
fingerprints, said method implemented by an automatic fingerprint
identification system including a processor, a memory coupled to
the processor, an interface to a fingerprint scanning device, and a
sensor associated with the fingerprint scanning device that
indicates a fingerprint match. The method may include: receiving
(i) a plurality of search octant feature vectors associated with a
plurality of search minutiae extracted from a search fingerprint,
and (ii) a plurality of reference octant feature vectors associated
with a plurality of reference minutiae extracted from a reference
fingerprint, where the search fingerprint includes one or more
non-overlapping sectors that each include a subset of the plurality
of search minutiae, and each of the plurality of search octant
feature vectors includes a set of relative features that represent
a difference between a particular search minutia and a plurality of
particular reference minutiae that are identified as closest
neighboring minutiae within a particular sector that includes the
particular search minutiae, and each of the plurality of reference
octant feature vectors includes a set of relative features that
represent a difference between a particular reference minutia and a
plurality of particular search minutiae that are identified as
closest neighboring minutiae within a particular sector that
includes the particular search minutiae;
[0006] The computer-implemented method may include: computing, for
each of the plurality of search minutiae, a first score that (i)
represents the difference between the particular search octant
feature associated with the particular search minutia and the
plurality of reference octant feature vectors associated with a
plurality of particular reference minutiae that are identified as
closest neighboring minutiae within the particular sector that
includes the particular search minutiae, and (ii) is adjusted based
at least on determining that at least one of the plurality of
particular reference minutiae that are identified as the closest
neighboring minutiae within the particular sector that includes the
particular search minutia has drifted to another of the one or more
sectors due to distortion within the search fingerprint.
[0007] The computer-implemented method may include: computing, for
each of the plurality of search minutiae, a second score that
represents a ratio indicating a number of paired mated minutiae
relative to a number of paired unmated minutiae all of the one or
more sectors, where: the paired mated minutiae represent the
particular search minutia that have the plurality of particular
reference minutiae that are identified as closest neighboring
minutiae within each of the one or more sectors, and the paired
unmated minutiae represent the particular search minutia that do
not have at least one particular reference minutia that is
identified as the closest neighboring minutiae within each of the
one or more sectors.
[0008] The computer-implemented method may include: computing, for
each of the plurality of search minutiae, a local similarity score
based at least on combining the first score and the second score;
determining, based at least on the computed local similarity scores
for each of the plurality of search minutiae, (i) a geometrically
aligned region between the search fingerprint and the reference
fingerprint, and (ii) a rotation angle between the search
fingerprint and the reference fingerprint; and identifying a
plurality of globally aligned mated minutiae within the
geometrically aligned region, where the plurality of globally
aligned mated minutiae represent a set of particular search minutia
that have geometrically consistent features, determined based at
least on the rotation angle, to a plurality of particular reference
minutiae within the geometrically aligned region.
[0009] The computer-implemented method may include: computing, for
each of the plurality of globally aligned mated minutiae, an
additional first score that represents the difference between a
particular search octant feature associated with the particular
search minutia within the geometrically aligned region and a
plurality of reference octant feature vectors associated with the
plurality of particular reference minutiae within the geometrically
aligned region; and computing, for each of the plurality of
globally aligned mated minutiae, an additional second score that
represents a ratio indicating a number of paired globally aligned
mated minutiae relative to a number of globally aligned unmated
minutiae within geometrically aligned region.
[0010] The computer-implemented method may include: computing (i) a
third score based at least on combining the respective first scores
for each of the plurality of search minutiae, and the respective
additional first scores for each of the plurality of globally
aligned mated minutiae, (ii) a fourth score based at least on
combining the respective second scores for each of the plurality of
search minutiae, and the respective additional second scores for
each of the plurality of globally aligned mated minutiae; computing
a match similarity score between the search fingerprint and the
reference fingerprint based at least on combining the third score
and the fourth score; and providing the match similarity score for
output to the automatic fingerprint identification system.
[0011] Other versions include corresponding systems, and computer
programs, configured to perform the actions of the methods encoded
on computer storage devices.
[0012] One or more implementations may include the following
optional features. For example, in some implementations, combining
the third score and the fourth score includes adding the values of
the respective scores.
[0013] In some implementations, identifying the plurality of
globally aligned mated minutiae within the geometrically aligned
region includes: obtaining a plurality of local closest matched
minutia pairs based on a plurality of top matched local similarity
scores; and computing a plurality sets of rotation parameters based
on the plurality of the local closest matched minutia pairs.
[0014] In some implementations, identifying the plurality of
globally aligned mated minutiae within the geometrically aligned
region includes: computing (i) a set of rotation parameters, and
(ii) a set of translation parameters, based on the plurality of
globally aligned mated minutiae within the geometrically aligned
region; and aligning the search minutiae to the reference minutiae
based at least on set of rotation parameters and the set of
translation parameters.
[0015] In some implementations, computing the plurality sets of
rotation parameters includes: identifying a plurality of angles
corresponding to a plurality of top bins of an angle offset
histogram of the plurality of the local closest matched minutiae
pairs.
[0016] In some implementations, prior to identifying the plurality
of angles, generating the angle offset histogram of the plurality
local matched minutia pairs using a smoothing process.
[0017] In some implementations, identifying the plurality of
globally aligned mated minutiae within the geometrically aligned
region includes: applying a two-stage pruning procedure to perform
local and global pairing operations on the local matched minutia
pairs.
[0018] In some implementations, the local and global pairing
operations are performed using a plurality of rotations and the
plurality of top closest matched minutia pairs to remove: a subset
of the plurality of globally aligned mated minutiae that are
identified to not be the local closest matched pairs, a subset of
the plurality of globally aligned mated minutiae with lower local
similarity scores within a same index, or a subset of the plurality
of globally aligned mated minutiae with transformed pairs for the
corresponding rotations that do not yield the highest cumulative
local similarity scores.
[0019] In some implementations, the match similarity score is used
to determine a fingerprint match between the search fingerprint and
the reference fingerprint, where the search fingerprint is a
distorted fingerprint.
[0020] In some implementations, the computer-implemented method
includes determining, within a particular fingerprint
identification operation, a match between the search fingerprint
and the reference fingerprint based at least on the value of the
match similarity score.
[0021] The details of one or more implementations are set forth in
the accompanying drawings and the description below. Other
potential features and advantages will become apparent from the
description, the drawings, and the claims.
BRIEF DESCRIPTION OF THE DRAWINGS
[0022] FIG. 1A is a block diagram of an exemplary automatic
fingerprint identification system.
[0023] FIG. 1B is a block diagram of an exemplary feature
extraction process.
[0024] FIG. 2 is an exemplary illustration of geometric
relationships between a reference minutia and a neighboring
minutia.
[0025] FIG. 3 is a graphical illustration of the relationships
represented in an exemplary octant feature vector (OFV).
[0026] FIG. 4 is an exemplary process of generating an octant
feature vector (OFV).
[0027] FIG. 5 is an exemplary process of calculating similarity
between two minutiae.
[0028] FIG. 6 is an exemplary alignment process for a pair of
fingerprints.
[0029] FIG. 7 is an exemplary minutiae matching process for a pair
of fingerprints.
[0030] FIG. 8 is an exemplary process of distorted fingerprint
matching.
[0031] In the drawings, like reference numbers represent
corresponding parts throughout.
DETAILED DESCRIPTION
[0032] In general, the matching of distorted fingerprints pose
several challenges to fingerprint identification and matching
technologies. For instance, distortions can impact matching
accuracy because the distortions may change both the geometric
locations and orientations of individual minutiae.
[0033] System Architecture
[0034] FIG. 1 is a block diagram of an exemplary automatic
fingerprint identification system 100. Briefly, the automatic
fingerprint identification system 100 may include a computing
device including a memory device 110, a processor 115, a
presentation interface 120, a user input interface 130, and a
communication interface 135. The automatic fingerprint
identification system 100 may be configured to facilitate and
implement the methods described through this specification. In
addition, the automatic fingerprint identification system 100 may
incorporate any suitable computer architecture that enables
operations of the system described throughout this
specification.
[0035] The processor 115 may be operatively coupled to memory
device 110 for executing instructions. In some implementations,
executable instructions are stored in the memory device 110. For
instance, the automatic fingerprint identification system 100 may
be configurable to perform one or more operations described by
programming the processor 115. For example, the processor 115 may
be programmed by encoding an operation as one or more executable
instructions and providing the executable instructions in the
memory device 110. The processor 115 may include one or more
processing units, e.g., without limitation, in a multi-core
configuration.
[0036] The memory device 110 may be one or more devices that enable
storage and retrieval of information such as executable
instructions and/or other data. The memory device 110 may include
one or more tangible, non-transitory computer-readable media, such
as, without limitation, random access memory (RAM), dynamic random
access memory (DRAM), static random access memory (SRAM), a solid
state disk, a hard disk, read-only memory (ROM), erasable
programmable ROM (EPROM), electrically erasable programmable ROM
(EEPROM), and/or non-volatile RAM (NVRAM) memory. The above memory
types are exemplary only, and are thus not limiting as to the types
of memory usable for storage of a computer program.
[0037] The memory device 110 may be configured to store a variety
of data including, for example, matching algorithms, scoring
algorithms, scoring thresholds, perturbation algorithms, fusion
algorithms, virtual minutiae generation algorithms, minutiae
overlap analysis algorithms, and/or virtual minutiae analysis
algorithms. In addition, the memory device 110 may be configured to
store any suitable data to facilitate the methods described
throughout this specification.
[0038] The presentation interface 120 may be coupled to processor
115. For instance, the presentation interface 120 may present
information, such as a user interface showing data related to
fingerprint matching, to a user 102. For example, the presentation
interface 120 may include a display adapter (not shown) that may be
coupled to a display device (not shown), such as a cathode ray tube
(CRT), a liquid crystal display (LCD), an organic LED (OLED)
display, and/or a hand-held device with a display. In some
implementations, the presentation interface 120 includes one or
more display devices. In addition, or alternatively, the
presentation interface 120 may include an audio output device (not
shown), e.g., an audio adapter and/or a speaker.
[0039] The user input interface 130 may be coupled to the processor
115 and receives input from the user 102. The user input interface
130 may include, for example, a keyboard, a pointing device, a
mouse, a stylus, and/or a touch sensitive panel, e.g., a touch pad
or a touch screen. A single component, such as a touch screen, may
function as both a display device of the presentation interface 120
and the user input interface 130.
[0040] In some implementations, the user input interface 130 may
represent a fingerprint scanning device that is used to capture and
record fingerprints associated with a subject (e.g., a human
individual) from a physical scan of a finger, or alternately, from
a scan of a latent print. In addition, the user input interface 130
may be used to create a plurality of reference records.
[0041] A communication interface 135 may be coupled to the
processor 115 and configured to be coupled in communication with
one or more other devices such as, for example, another computing
system (not shown), scanners, cameras, and other devices that may
be used to provide biometric information such as fingerprints to
the automatic fingerprint identification system 100. Such biometric
systems and devices may be used to scan previously captured
fingerprints or other image data or to capture live fingerprints
from subjects. The communication interface 135 may include, for
example, a wired network adapter, a wireless network adapter, a
mobile telecommunications adapter, a serial communication adapter,
and/or a parallel communication adapter. The communication
interface 135 may receive data from and/or transmit data to one or
more remote devices. The communication interface 135 may be also be
web-enabled for remote communications, for example, with a remote
desktop computer (not shown).
[0042] The presentation interface 120 and/or the communication
interface 135 may both be capable of providing information suitable
for use with the methods described throughout this specification,
e.g., to the user 102 or to another device. In this regard, the
presentation interface 120 and the communication interface 135 may
be used to as output devices. In other instances, the user input
interface 130 and the communication interface 135 may be capable of
receiving information suitable for use with the methods described
throughout this specification, and may be used as input
devices.
[0043] The processor 115 and/or the memory device 110 may also be
operatively coupled to the database 150. The database 150 may be
any computer-operated hardware suitable for storing and/or
retrieving data, such as, for example, pre-processed fingerprints,
processed fingerprints, normalized fingerprints, extracted
features, extracted and processed feature vectors such as octant
feature vectors (OFVs), threshold values, virtual minutiae lists,
minutiae lists, matching algorithms, scoring algorithms, scoring
thresholds, perturbation algorithms, fusion algorithms, virtual
minutiae generation algorithms, minutiae overlap analysis
algorithms, and virtual minutiae analysis algorithms.
[0044] The database 150 may be integrated into the automatic
fingerprint identification system 100. For example, the automatic
fingerprint identification system 100 may include one or more hard
disk drives that represent the database 150. In addition, for
example, the database 150 may include multiple storage units such
as hard disks and/or solid state disks in a redundant array of
inexpensive disks (RAID) configuration. In some instances, the
database 150 may include a storage area network (SAN), a network
attached storage (NAS) system, and/or cloud-based storage.
Alternatively, the database 150 may be external to the automatic
fingerprint identification system 100 and may be accessed by a
storage interface (not shown). For instance, the database 150 may
be used to store various versions of reference records including
associated minutiae, octant feature vectors (OFVs) and associated
data related to reference records.
[0045] Feature Extraction
[0046] In general, feature extraction describes the process by
which the automatic fingerprint detection system 100 extracts a
list of minutiae from each of reference fingerprint, and the search
fingerprint. As described, a "minutiae" represent major features of
a fingerprint, which are used in comparisons of the reference
fingerprint to the search fingerprint to determine a fingerprint
match. For example, common types of minutiae may include, for
example, a ridge ending, a ridge bifurcation, a short ridge, an
island, a ridge enclosure, a spur, a crossover or bridge, a delta,
or a core.
[0047] FIG. 1B is a block diagram of an exemplary feature
extraction process 150. As shown, after receiving an input
fingerprint 104, the automatic fingerprint detection system 100
initially identifies a set of features 112 within the fingerprint,
generates a list of minutiae 114, and extracts a set of feature
vectors 116. For instance, the automatic fingerprint detection
system 100 may generate a list of minutia 114 for each of the
reference fingerprint (or "reference record") and a search
fingerprint (or "search record").
[0048] In some implementations, the feature vectors 116 may be
described using feature vector that is represented by
Mf.sub.i=(x.sub.i,y.sub.i,.theta..sub.i). As described, the feature
vector Mf.sub.i includes a minutia location that is defined by
coordinate geometry such as (x.sub.i,y.sub.i), and a minutiae
direction that is defined by the angle .theta..sub.i.di-elect
cons.[0,2.pi.]. In other examples, further minutiae characteristics
such as, quality, ridge frequency, and ridge curvature may also be
used to describe feature vector Mf.sub.i. The extracted feature
vectors may be used to generate octant feature vectors (OFVs) for
each of identified minutia within the search and reference
fingerprints.
[0049] Octant Feature Vector (OFV) Overview
[0050] The automatic fingerprint identification system 100 may
compare the search and reference records based on initially
generating feature vectors associated with minutiae that are
extracted from the search and reference records, respectively. For
instance, as described throughout this specification, in some
implementations, octant feature vectors (OFVs) may be used to as
feature vectors that define attributes of the extracted minutiae.
However, in other implementations, other minutiae descriptors may
be used.
[0051] OFVs encode geometric relationships between reference
minutiae and the nearest neighboring minutiae to the reference
minutiae in a particular sector (referred to as the "octant
neighborhood") of the octant. Each sector of the octant used in an
OFV spans 45 degrees of a fingerprint region. The nearest
neighboring minutiae may be assigned to one sector of the octant
based on their orientation difference. The geometric relationship
between a reference minutia and its nearest minutia in each octant
sector may be described by relative features including, for
example, distance between the minutiae and the orientation
difference between minutiae. The representation achieved by the use
of an OFV is invariant to transformation. In addition, this
representation is insensitive to a nonlinear distortion because the
relative features are independent from any transformation.
[0052] Pairs of reference minutiae and nearest neighboring minutiae
may be identified as "mated minutiae pairs." The mated minutiae
pairs in a reference record and a search record may be identified
by comparing the respective OFVs of minutiae extracted from the
reference record and the search record. The transformation
parameters may be estimated by comparing attributes of the
corresponding mated minutiae. For example, the transformation
parameters may indicate the degree to which the search record has
been transformed (e.g., perturbed or twisted) as relative to a
particular reference record. In other examples, the transformation
parameters may be applied to verify that, for a particular pair of
a reference record and a search record (a "potential matched
fingerprint pair"), mated minutiae pairs exhibit corresponding
degrees of transformation. Based on the amount of corresponding
mated minutiae pairs in each potential matched fingerprint pair,
and the consistency of the transformation, a similarity score may
be assigned. In some implementations, the pair of potential matched
fingerprint pairs with the highest similarity score may be
determined as a candidate matched fingerprint pair.
[0053] The automatic fingerprint identification system 100 may
calculate an OFV for each minutia that encodes the geometric
relationships between the reference minutia and its nearest
minutiae in each sector of the octant. For instance, the automatic
fingerprint identification system 100 may define eight octant
sectors and assigns the nearest minutiae to one sector of the
octant based on the location of each minutiae within the sectors.
The geometric relationship between a reference minutia and its
nearest minutia in each octant sector is represented by the
relative features. For example, in some implementations, the OFV
encodes the distance, the orientation difference, and the ridge
count difference between the reference feature and the nearest
neighbor features. Because the minutia orientation can flexibly
change up to 45.degree. due to the octant sector approach, relative
features are independent from any transformation.
[0054] The automatic fingerprint identification system 100 may use
the OFVs to determine the number of possible corresponding minutiae
pairs. Specifically, the automatic fingerprint identification
system 100 may evaluate the similarity between two respective OFVs
associated with the search record and the file record. The
automatic fingerprint identification system 100 may identify all
possible local matching areas of the compared fingerprints by
comparing the OFVs. The automatic fingerprint identification system
100 may also an individual similarity score for each of the mated
OFV pairs.
[0055] The automatic fingerprint identification system 100 may
clusters all OFVs of the matched areas with similar transformation
effects (e.g., rotation and transposition) into an associated
similar bin. Note that the precision of the clusters of the bins
(e.g., the variance of the similar rotations within each bin) is a
proxy for the precision of this phase. Automatic fingerprint
identification system 100 therefore uses bins with higher numbers
of matched OFVs (e.g., clusters with the highest counts of OFVs)
for the first phase global alignment.
[0056] The automatic fingerprint identification system 100 may use
the location and angle of each selected bin as the parameters of a
reference point (an "anchor point") to perform a global alignment
procedure. More specifically, the automatic fingerprint
identification system 100 may identify the global alignment based
on the bins that include the greatest number of the global mated
minutiae pairs, and the location and angle associated with each of
those bins. Based on the number of global paired minutiae found and
the total of individual similarity scores calculated for the
corresponding OFVs within the bin or bins, the automatic
fingerprint identification system 100 may identify the
transformations (e.g., the rotations of the features) with the best
alignment.
[0057] In a second phase, the automatic fingerprint identification
system 100 performs a more precise pairing using the
transformations with the best alignment to obtain a final set of
the globally aligned minutiae pairs. In this phase, automatic
fingerprint identification system 100 performs a pruning procedure
to find geometrically consistent minutiae pairs with tolerance of
distortion for each aligned minutiae set that factors in the local
and global geometrical index consistency. By performing such
alignment globally and locally, automatic fingerprint
identification system 100 determines the best set of global aligned
minutiae pairs. Automatic fingerprint identification system 100
uses the associated mini-scores of the global aligned pairs to
calculate the global similarity score. Furthermore, automatic
fingerprint identification system 100 factors in a set of absolute
features of the minutiae, including the quality, ridge frequency,
and the curvatures in the computation of the final similarity
score.
[0058] OFV Generation
[0059] The automatic fingerprint identification system 100 may
generate an octant feature vector (OFV) for each minutia of the
features extracted. Specifically, as described above, the automatic
fingerprint identification system 100 may generate OFVs encoding
the distance and the orientation difference between the reference
minutiae and the nearest neighbor in each of eight octant sectors.
Alternately, the automatic fingerprint identification system 100
may generate feature vectors with different numbers of sectors.
[0060] FIG. 2 is an exemplary illustration 200 of geometric
relationships between a reference minutia 210 and a neighboring
minutia 220. The geometric relationships may be used to construct a
rotation and translation invariant feature vector that includes
relative attributes (d.sub.ij, .alpha..sub.ij, .beta..sub.ij)
between the reference minutia 210 and the neighboring minutia
220.
[0061] As depicted in FIG. 2, the automatic fingerprint
identification system 100 may compute a Euclidean distance 230
between the reference minutia 210 and the neighboring minutia 220,
a minimum rotation angle 240 for the neighboring minutia, and a
minimum rotation angle 250 for the reference minutia 210. In
addition, the automatic fingerprint identification system may
compute a ridge count 260 across the reference minutia 210 and the
neighboring minutia 220.
[0062] Specifically, the rotation and translation invariant feature
vector may be represented as vector 1 (represented below). In some
implementations, an OFV is created to describe the geometric
relationship between. Further, M.sub.i represents a reference
minutia and M.sub.j represents a nearest neighbor minutiae in one
of the octant sectors. The OFV for each sector may be described in
the given vector from vector 1:
[0063] Vector 1: (d.sub.ij, .alpha..sub.ij, .beta..sub.ij), [0064]
where d.sub.ij denotes the Euclidean distance 230, [0065] where
.beta..sub.ij=.lamda.(.theta..sub.i,.theta..sub.j) denotes the
minimum rotation angle 240 required to rotate a line of direction
.theta..sub.i in a particular direction (e.g., counterclockwise in
FIG. 2) to make the line parallel with a line of direction
.theta..sub.j, and [0066] where
.theta..sub.ij=.lamda.(.theta..sub.i, .angle.(M.sub.i,M.sub.j))
denotes the minimum rotation angle 250, where
.angle.(M.sub.i,M.sub.j) denotes the direction from the reference
minutia 210 to the neighboring minutia 220, and where .lamda.(a,b)
denotes the same meaning as defined in .alpha..sub.ij
[0067] Specifically, because each element calculated in the feature
vector is a relative measurement between the reference minutia 210
and the neighboring minutia 220, the feature vector is independent
from the rotation and translation of the fingerprint. Elements 230,
240, and 250 may be referred to as relative features and are used
to compute the similarity between pair of reference minutia 210 and
the neighboring minutia 220. In some implementations, other
minutiae features such as absolute features may additionally or
alternatively be used by the automatic fingerprint identification
system 100 to weight the computed similarity score between a pair
of mated minutiae that includes reference minutia 210 and the
neighboring minutia 220.
[0068] FIG. 3 is a graphical illustration of the relationships
represented in an exemplary octant feature vector (OFV) 300. The
OFV 300 may be generated for a reference minutia 310, which
corresponds to the reference minutia 210 as shown in FIG. 2. As
shown, the OFV 300 represents relationships between the reference
minutiae 310 and its nearest neighboring minutiae 322, 332, 342,
352, 362, 372, 382, and 392 in sectors 320, 330, 340, 350, 360,
370, 380, and 390, respectively. However, the graphical
illustration of OFV 300 does not depict the details of the
geographic relationships, which are described within respect to
FIG. 2. Although FIG. 3 indicates a neighboring minutia within each
octant sector, in some instances, there may be no neighboring
minutiae within a particular sector. In such instances, the OFV for
the particular sector without a neighboring minutia is set to zero.
Otherwise, because the neighboring minutiae 322, 332, 342, 352,
362, 372, 382, and 392 may not overlap with reference minutiae 310,
the OFV is greater than zero.
[0069] FIG. 4 is an exemplary process 400 of generating an octant
feature vector (OFV). Briefly, the process 400 may include
identifying a plurality of minutiae from the input fingerprint
image (410), selecting a particular minutia from the plurality of
minutiae (420), defining a set of octant sectors for the plurality
of minutiae (430), assigning each of the plurality of minutiae to
an octant sector (440), identifying a neighboring minutiae to the
particular minutia for each octant sector (450), and generating an
octant feature vector for the particular minutia (460).
[0070] In more detail, the process 400 may include the process may
include identifying a plurality of minutiae from the input
fingerprint image (410). For instance, the automatic fingerprint
identification system 100 may receive the input fingerprint 402 and
generate a list of minutiae 412 using the techniques described
previously with respect to FIG. 1B.
[0071] The process 400 may include selecting a particular minutia
from the plurality of minutiae (420). For instance, the automatic
fingerprint identification system 100 may select a particular
minutiae within the list of minutiae 412.
[0072] The process 400 may include defining a set of octant sectors
for the plurality of minutiae (430). For instance, the automatic
fingerprint identification system 100 may generate a set of octant
sectors 432 that include individual octant sectors k.sub.0 to
k.sub.7 as shown in FIG. 4. The set of octant sectors 432 may be
generated in reference to the particular minutia that is selected
in step 420.
[0073] The process 400 may include assigning each of the plurality
of minutiae to an octant sector (440). For instance, the automatic
fingerprint identification system 100 may assign each of the
plurality of minutiae from the list of minutiae 412 into
corresponding octant sectors within the set of octant sectors 432.
The assigned minutiae may be associated with the corresponding
octant sectors in a list 442 that includes the number of minutiae
that are identified within each individual octant sector. For
example, as shown in FIG. 4, the exemplary octant sector k.sub.1
has no identified minutiae, whereas the exemplary k.sub.6 includes
two identified minutiae within the octant sector. The graphical
illustration 444 represents the locations of the plurality of
minutiae, relative to the particular selected minutia, M.sub.i,
within the individual octant sectors.
[0074] The process 400 may include identifying a neighboring
minutiae to the particular minutia for each octant sector (450).
For instance, the automatic fingerprint identification system 100
may identify, from all the neighboring minutiae within each octant
sector, the neighboring minutia that is the closest neighboring
minutia based on the distance between each neighboring minutia and
the particular selected minutia, M.sub.i. For example, for the
octant sector k.sub.6, the automatic fingerprint identification
system 100 may determine that the minutia, M.sub.6 is the closest
neighboring minutia based on the distance between M.sub.i and
M.sub.6. The closest neighboring minutiae for all of the octant
sectors may be aggregated within a list of closest neighboring
minutiae 452 that identifies each of the closest neighboring
minutiae.
[0075] The process 400 may include generating an octant feature
vector for the particular minutia (460). For instance, the
automatic fingerprint identification system 100 may generate an
octant feature vector 462, based on the list of closest neighboring
minutiae 452, which includes a set of relative features such as the
Euclidean distance 230, the minimum rotation angle 240, and the
minimum rotation angle 250 as described previously with respect to
FIG. 2.
[0076] As described above with respect to FIGS. 2-4, OFVs for
minutiae may be used to characterize local relationships with
neighboring minutiae, which are invariant to the rotation and
translation of the fingerprint that includes the minutiae. The OFVs
are also insensitive to distortion, since the nearest neighboring
minutiae are assigned to multiple octant sectors in various
directions, thereby allowing flexibility of orientation of up to
45.degree.. In this regard, the OFVs of minutiae within a
fingerprint may be compared against the OFVs of minutiae within
another fingerprint (e.g., a search fingerprint) to determine a
potential match between the two fingerprints. Descriptions of the
general fingerprint matching process, and the OFV matching process
are provided below.
[0077] Fingerprint Identification and Matching
[0078] In general, the automatic fingerprint identification system
100 may perform fingerprint identification and matching in two
stages: (1) an enrollment stage, and (2) an
identification/verification stage.
[0079] In the enrollment stage, an individual (or a "registrant")
has their fingerprints and personal information enrolled. The
registrant may be an individual manually providing their
fingerprints for scanning or, alternately, an individual whose
fingerprints were obtained by other means. In some examples,
registrants may enroll fingerprints using latent prints, libraries
of fingerprints, and any other suitable repositories and sources of
fingerprints. As described, the process of "enrolling" and other
related terms refer to providing biometric information (e.g.,
fingerprints) to an identification system (e.g., the automatic
fingerprint identification system 100).
[0080] The automatic fingerprint identification 100 system may
extract features such as minutiae from fingerprints. As described,
"features" and related terms refer to characteristics of biometric
information (e.g., fingerprints) that may be used in matching,
verification, and identification processes. The automatic
fingerprint identification system 100 may create a reference record
using the personal information and the extracted features, and save
the reference record into the database 150 for subsequent
fingerprint matching, verification, and identification
processes.
[0081] In some implementations, the automatic fingerprint
identification system 100 may contain millions of reference
records. As a result, by enrolling a plurality of registrants (and
their associated fingerprints and personal information), the
automatic fingerprint identification system 100 may create and
store a library of reference records that may be used for
comparison to search records. The library may be stored at the
database 150 associated.
[0082] In the identification stage, the automatic fingerprint
identification system 100 may use the extracted features and
personal information to generate a record known as a "search
record". The search record represents a source fingerprint for
which identification is sought. For example, in criminal
investigations, a search record may be retrieved from a latent
print at a crime scene. The automatic fingerprint identification
may compare the search record with the enrolled reference records
in the database 150. For example, during a search procedure, a
search record may be compared against the reference records stored
in the database 150. In such an example, the features of the search
record may be compared to the features of each of the plurality of
reference records. For instance, minutiae extracted from the search
record may be compared to minutiae extracted from each of the
plurality of reference records.
[0083] As described, a "similarity score" is a measurement of the
similarity of the fingerprint features (e.g., minutiae) between the
search record and each reference record, represented as a numerical
value to degree of similarity. For instance, in some
implementations, the values of the similarity score may range from
0.0 to 1.0, where a higher magnitude represents a greater degree of
similarity between the search record and the reference record.
[0084] The automatic fingerprint identification system 100 may
compute individual similarity scores for each comparison of
features (e.g., minutiae), and aggregate similarity scores (or
"final similarity scores") between the search record to each of the
plurality of reference records. In this regard, the automatic
fingerprint identification system 100 may generate similarity
scores of varying levels of specificity throughout the matching
process of the search record and the plurality of reference
records.
[0085] The automatic fingerprint identification system 100 may also
sort each of the individual similarity scores based on the value of
the respective similarity scores of individual features. For
instance, the automatic identification system 100 may compute
individual similarity scores between respective minutiae between
the search fingerprint and the reference fingerprint, and sort the
individual similarity scores by their respective values.
[0086] A higher final similarity score indicates a greater overall
similarity between the search record and a reference record while a
lower final similarity score indicates a lesser over similarity
between the search record and a reference record. Therefore, the
match (e.g., the relationship between the search record and a
reference record) with the highest final similarity score is the
match with the greatest relationship (based on minutiae comparison)
between the search record and the reference record.
[0087] Minutiae and OFV Matching
[0088] In general, the OFVs of minutiae may be compared between two
fingerprints to determine a potential match between a reference
fingerprint and a search fingerprint. The automatic fingerprint
identification system 100 may compare the OFVs of corresponding
minutiae from the reference fingerprint and the search fingerprint
to compute an individual similarity score that reflects a
confidence that the particular reference minutiae corresponds to
the particular search minutiae that is being compared to. The
automatic fingerprint identification system 100 may then compute
aggregate similarity scores, between a list of reference minutiae
and a list of search minutiae, based the values of the individual
similarity scores for each minutiae. For instance, as described
more particularly below, various types of aggregation techniques
may be used to determine the aggregate similarity scores between
the reference fingerprint and the search fingerprint.
[0089] FIGS. 5-7 generally describe different processes that may be
used to during fingerprint identification and matching procedures.
For instance, FIG. 5 illustrates an exemplary process of
calculating an individual similarity score between a reference
minutia and a search minutia. FIG. 6 illustrates an exemplary
alignment process between two fingerprints using extracted minutiae
from the two fingerprints, and FIG. 7 illustrates an exemplary
minutiae matching technique that may be employed after the
alignment procedure represented in FIG. 7. As described with
respect to FIG. 8, the processes represented in FIGS. 5-7 may be
used in conjunction during an exemplary distorted fingerprint
matching process.
[0090] Referring to FIG. 5, a similarity determination process 500
may be used to compute an individual similarity score between a
reference minutia 502a from a reference fingerprint, and a search
minutia 502b from a search fingerprint. A reference OFV 504a and a
search OFV 504b may be generated for the reference minutia 502a and
the search minutia 504b, respectively, using the techniques
described with respect to FIG. 4. As shown, the search and
reference OFVs 504a and 504b include individual octant sectors
k.sub.0 to k.sub.7 as described as illustrated in FIG. 3. Each of
the reference OFV 504a and the search OFV 504b may include
parameters such as, for example, the Euclidian distance 230, the
minimum rotation angle 240, and the minimum rotation angle 250 as
described with respect to FIG. 2. As shown in FIG. 5, exemplary
parameters 506a and 506b may represent the parameters for octant
sector k.sub.0 of the reference OFV 502a and the search OFV 502b,
respectively.
[0091] The similarity score determination may be performed by an
OFV comparison module 520, which may be a software module or
component of the automatic fingerprint identification system 100.
In general, the similarity score calculation includes four steps.
Initially, the Euclidian distance values of a particular octant
sector may be evaluated (522). Corresponding sectors between the
reference OFV 504a and the similarity OFV 504b may then be compared
(524). A similarity score between a particular octant sector within
the reference OFV 504a and its corresponding octant sector in the
search OFV 504b may then be computed (526). Finally, the similarity
scores for the between other octant sectors of the reference OFV
504a and the search OFV 504b may then be computed and combined to
generate the final similarity score between the reference OFV 504a
and the search OFV 504b (528).
[0092] With respect to step 522, the similarity module 520 may
initially determine if the Euclidian distance values within the
parameters 506a and 506b are non-zero values. For instance, as
shown in decision point 522a, the Euclidean distance associated
with the octant sector of the reference OFV 504a, d.sub.RO, is
initially be evaluated. If this value is equal to zero, then the
similarity score for the octant sector k.sub.0 is set to zero.
Alternatively, if the value of d.sub.RO is greater than zero, then
the similarity module 520 proceeds to decision point 522b, where
the Euclidean distance associated with the octant sector of the
reference OFV 504a, d.sub.SO, is evaluated. If the value of
d.sub.SO is not greater than zero, then the OFV similarity module
520 evaluates the value of the Euclidean distance d.sub.S1, which
is included in an adjacent sector k.sub.1 to the octant sector
k.sub.0 within the reference OFV 504b. Although FIG. 5 represents
only one of the adjacent sectors being selected, because each
octant sector includes two adjacent octant sectors as shown in FIG.
3, in other implementations, octant sector k.sub.7 may also be
evaluated. If the value of the Euclidean distance within the
adjacent octant sector is not greater than zero, then the
similarity module 520 sets the value of individual similarity score
S.sub.RS01, between the octant sector k.sub.0 of the reference OFV
504a and the octant sector k.sub.1 of the reference OFV 504b, to
zero.
[0093] Alternatively, if the either the value of Euclidean distance
d.sub.RO within the octant sector k.sub.0 of the reference OFV
504a, or the Euclidean distance d.sub.S1 within the adjacent octant
sector k.sub.1 of the search OFV 504b is determined to be a
non-zero value within the decision points 522b and 522c,
respectively, then the similarity module proceeds to step 524.
[0094] In some instances, a particular octant vector may include
zero corresponding minutiae within the search OFV 504b due to
localized distortions within the search fingerprint. In such
instances, where the corresponding minutiae may have drifted to an
adjacent octant sector, the similarity module 520 may alternatively
compare the features of the octant vector of the reference OFV 504a
to a corresponding adjacent octant vector of the search OFV 504b as
shown in step 524b.
[0095] If proceeding through decision point 522b, the similarity
module 520 may proceed to step 524a where the corresponding octant
sectors between the reference OFV 504a and the search OFV 504b are
compared. If proceeding through decision point 522c, the similarity
module 620 may proceed to step 524b where the octant sector k.sub.0
of the reference OFV 504a is compared to the corresponding adjacent
octant sector k.sub.1 the search OFV 504b. During either process,
the similarity module 520 may compute the difference between the
parameters that are included within each octant sector of the
respective OFVs. For instance, as shown, the difference between the
Euclidean distances 230, .DELTA.d, the difference between the
minimum rotation angles 240, .DELTA..alpha., and the difference
between the minimum rotation angles 250, .DELTA..beta., may be
computed. Since these parameters represent geometric relationships
between pairs of minutiae, the differences between them represent
distance and orientation differences between the reference and
search minutiae with respect to particular octant sectors.
[0096] In some implementations, dynamic threshold values for the
computed feature differences may be used to handle nonlinear
distortions within the search fingerprint in order to find mated
minutiae between the search and reference fingerprint. For
instance, the values of the dynamic thresholds may be adjusted to
larger or smaller values to adjust the sensitivity of the mated
minutiae determination process. For example, if the value of the
threshold for the Euclidean distance is set to a higher value, than
more minutiae within a particular octant sector may be determined
to be a neighboring minutia to a reference minutiae based on the
distance being lower than the threshold value. Likewise, if the
threshold is set to a lower value, then a smaller number of
minutiae within the particular octant sector may be determined to
be neighboring minutia based on the distance to the reference
minutia being greater than the threshold value.
[0097] After either comparing the corresponding octant sectors in
step 524a or comparing the corresponding adjacent octant sectors in
step 524b, the similarity module 520 may then compute an individual
similarity score between the respective octant sectors in steps
526a and 526b, respectively. For instance, the similarity score may
be computed based on the values of the feature differences as
computed in steps 524a and 524b. For instance, the similarity score
may represent the feature differences and indicate minutiae that
are likely to be distorted minutiae. For example, if the feature
differences between a reference minutiae and corresponding search
minutia within a particular octant sector are close the dynamic
threshold values, the similarity module 520 may identify the
corresponding search minutia as a distortion candidate.
[0098] After computing the similarity score for either the
corresponding octant sectors, or the corresponding adjacent sectors
in steps 526a and 526b, respectively, the similarity module 520 may
repeat the steps 522-526 for all of the other octant sectors
included within the reference OFV 504a and the search OFV 504b. For
instance, the similarity module 520 may iteratively execute the
steps 522-526 until the similarity scores between each
corresponding octant sector and each corresponding adjacent octant
sector are computed for the reference OFV 504a and the search OFV
504b.
[0099] The similarity module may then combine the respective
similarity scores for each corresponding octant sectors and/or the
corresponding adjacent octant sectors to generate a final
similarity score between the reference minutia and the
corresponding search minutia. This final similarity score is also
referred to as the "individual similarity score" between
corresponding minutiae within the search and reference fingerprints
as described in other sections of this specification. The
individual similarity score indicates a strength of the local
matching of the corresponding OFV.
[0100] In some implementations, the particular aggregation
technique used by the similarity module 522 to generate the final
similarity score (or the "individual similarity score") may vary.
For example, in some instances, the final similarity score may be
computed based on adding the values of the similarity scores for
the corresponding octant sectors and the corresponding adjacent
octant sectors, and normalizing the sum by a sum of a total number
of possible mated minutiae for the reference minutia and a total of
number of possible mated minutiae for the search minutia. In this
regard, the final similarity score is weighted by considering the
number of mated minutiae and the total number of possible mated
minutiae.
[0101] FIG. 6 illustrates an exemplary alignment process 600
between a reference fingerprint and a search fingerprint. Briefly,
the process 600 may initially compare a list of reference OFVs 604a
associated with a list of reference minutiae 602a and a list of
search OFVs 604b associated with a list of search minutiae 604b,
and generate a list of all possible mated minutiae 612. A global
alignment module 620 may then perform a global alignment procedure
on the list of all possible mated minutia 612 to generate a
clustered list of all possible mated minutiae 622, and determine
two best alignment rotations 622 for the search fingerprint
relative to the reference fingerprint. A precision alignment module
may then use the two best rotations 624 perform a second alignment
procedure to generate a list that includes the best-aligned pair
632 for the plurality of bins, which are provided as outputs of the
alignment process.
[0102] As described previously with respect to FIG. 5, the OFVs of
corresponding minutiae within the reference fingerprint and the
search fingerprint may be compared by the OFV comparison module 610
to generate the list of all possible mated minutiae 612. As
described, "mated minutiae" refer to a pair of minutiae that
includes a particular reference minutia and a corresponding search
minutia based at least on the OFV comparison performed by the OFV
comparison module 610, and the value of the individual similarity
score between the two respective OFVs of the reference and search
minutiae. The individual similarity score indicates a strength of
the local matching of the corresponding OFV. In addition, the list
of all possible mated minutiae 612 includes all of the minutiae
within the octant sectors that are identified as neighboring
minutiae to a particular reference minutia and have a non-zero
similarity score, although additional mated minutiae may exist with
similarity score values equal to zero. Although as shown in the
FIG., the list of all possible minutiae 612 includes one search
minutia per reference minutia, in some instances, multiple mated
minutiae may exist within the list of all possible minutiae 612 for
a single reference minutia.
[0103] The global alignment module 620 performs a global alignment
process on the list of all possible mated minutiae 612, which
estimates a probable (or best rotation) alignment between the
reference fingerprint and the search fingerprint based on comparing
the angle offsets between the individual minutiae within the mated
minutiae. For instance, the global alignment module 620 may
initially compute an angle offset for each mated minutiae pair
based on the individual similarity scores between a particular
reference minutia and its corresponding search minutia.
[0104] Each of the mated minutiae within the list of all possible
mated minutiae 612 may then be grouped into a historical bin that
is associated with a particular angle offset range. For instance,
two mated minutiae pairs within the list of all possible mated
minutiae 612 may be grouped into the same historical bin if their
respective individual similarity scores indicate a similar angular
offset between the individual minutia within each mated minutiae
pair. In some instances, the number of historical bins for the list
of all possible mated minutiae 612 is used to estimate a
fingerprint quality score for the search fingerprint. For example,
if the quality of the search fingerprint is excellent, then the
angular offset among each of the mated minutiae within the list of
all possible mated minutiae 612 should be consistent, and majority
of the mated minutiae will be grouped into a single historical bin.
Alternatively, if the fingerprint quality is poor, then the number
of historical bins would increase, representing significant
variations between the angular offset values between the mated
minutiae within the list of all possible mated minutiae 612.
[0105] In addition to grouping the mated minutiae into a particular
historical bin, the global alignment module 620 may determine a set
of rigid transformation parameters, which indicate geometric
differences between the reference minutiae and the search minutiae
with similar angular offsets. The rigid transformation parameters
thus indicate a necessary rotation of the search fingerprint at
particular locations, represented by the locations of the minutiae,
in order to geometrically align the search fingerprint to the
reference fingerprint. Since the rigid transformation parameters
are computed for all possible mated minutiae, the necessary
rotation represents a global alignment between the reference
fingerprint and the search fingerprint. The global alignment module
620 may then generate a clustered list of all possible mated
minutiae, which groups the mated minutiae by the historical bin
based on the respective angle offsets, and includes a set of rigid
transformation parameters. In some implementations, the histogram
represented by the plurality of bins may be smoothened by a
Gaussian function.
[0106] The global alignment module 620 may use the clustered list
of all possible mated minutiae to determine two best rotations 624.
For instance, the two best rotations 624 may be determined by using
the rigid transformation parameters to calculate a set of alignment
rotations for the search fingerprint using each historical bin as a
reference point. Each alignment rotation may then be applied to the
search fingerprint to generate a plurality of transformed search
fingerprints that is individually mapped to each alignment
rotation. For example, in some instances, the number of alignment
rotations corresponds to the number of historical bins generated
for the list of all possible mated minutiae 612. In such instances,
the number of transformed search fingerprints generated corresponds
to the number of historical bins included in the cluster list of
all possible mated minutiae 622. Each set of transformed search
fingerprints may then be compared to the reference fingerprint to
determine the two best rotations 624. For example, as described
more particularly with respect to FIG. 7, each transformed search
fingerprint may be compared to the reference fingerprint using a
minutiae matching technique to determine which particular alignment
rotations generate the greatest number of correctly matched
minutiae between a particular transformed search fingerprint and
the reference fingerprint. The global alignment module 620 may then
extract the two best alignment rotations 624, which are then used
by the precision alignment module 630.
[0107] In some implementations, different matching constraints may
be used with the minutiae matching techniques to determine the two
best alignment rotations 624.
[0108] The precision alignment module 630 may then use the two best
alignment rotations 624 to perform a precision alignment process
that iteratively rotates individual minutiae within the search
fingerprint around the two best alignment rotations 632 several
times with small angle variations to obtain a more precise pairing
between individual search minutiae and their corresponding
reference minutiae. For example, in some, twelve rotations may be
used with two degree angle variations. The minutiae that are
associated with the precise pairing between the search fingerprint
and the reference fingerprint are determined to be the list of
best-aligned minutiae 632, which are provided for output by the
process 600. The list of best-aligned minutiae 632 represent
transformations of individual search minutiae within the search
fingerprint that most closely pair with the corresponding reference
minutiae of the reference fingerprint as a result of the global
alignment and the precision alignment processes.
[0109] FIG. 7 illustrates an exemplary minutiae matching process
700. The minutiae matching process 700 may be performed after the
fingerprint alignment process 600 as described in FIG. 6 to remove
false correspondences included within a list of aligned minutiae
that is outputted from alignment process. For instance,
fingerprints from two fingers of an individual may share local
structures, which can result in false correspondence minutiae
between a search fingerprint of one finger and a reference
fingerprint of another finger. To resolve this, the minutiae
matching process 700 includes a two-stage pruning process to remove
false correspondence minutiae pairs within a list of all possible
mated minutiae.
[0110] Briefly, the process 700 may include a local geometric
module 710 receiving a list of aligned minutiae 710, and generating
a modified list of all possible minutiae 712 that does not include
false correspondence minutiae. A global consistency pairing module
may then sort the modified list of all possible minutiae 712 by
values of the respective individual similarity scores to generate a
sorted modified list of all possible minutiae 722. The global
consistency pairing module 720 may remove minutiae pairs with
duplicate indexes 724, and group the list of mated minutiae based
on conducting a global geometric consistency evaluation to generate
a list of geometrically consistent groups 726, which are then
outputted with a top average similarity score from one of the
geometrically consistent groups.
[0111] Initially, the local geometric module 710 may select the
best-paired minutiae from the list of all possible mated minutiae.
For instance, after aligned the search fingerprint and the
reference fingerprint as described in FIG. 6, the local geometric
module 710 may scan the list of all possible minutiae and identify
the two minutiae pairs with the minimum orientation difference.
[0112] In some instances, the identification of the best-paired
minutiae may additionally be subject to satisfying a set of
constraints. For example, one constraint may be that the index of
the first pair is different from that of the first pair. In other
examples, the rigid transformation parameters of the two
best-paired minutiae pairs may be compared to threshold values to
ensure that the two identified pairs are geometrically
consistent.
[0113] After identifying the two best-paired minutiae pairs, the
local geometric module 710 may use the two best-paired minutiae
pairs as reference pairs to remove other minutiae pairs from the
list of all possible mated minutiae. In some instances, particular
pairs may be removed if they satisfy one or more removal criteria
based on the attributes of the two best paired minutiae pairs. For
example, one constraint may be that if the minutiae index of a
particular pair is the same as one of the best-paired minutiae
pairs, then that particular pair may be identified as a duplicate
within the list of all possible minutiae and removed as a false
correspondence. In another example, a rotational constraint may be
used to remove particular minutiae pairs that have a large
orientation difference compared to the two best-paired minutiae
pairs. In another example, distance constraints may be used to keep
each particular minutiae pair within the list of all possible mated
minutiae geometrically consistent with the two best-paired minutiae
pairs. The updated list of minutiae pairs that is generated is the
modified list of all possible mated minutiae 712.
[0114] After the modified list of all possible mated minutiae is
generated, the global consistency pairing module 720 may perform a
global consistency pairing operation on the modified list of all
possible mated minutiae 722 to generate the list of
geometrically-consistent groups 726 (or "globally aligned mated
minutiae"). For instance, the global consistency pairing module 720
may initially sort the list modified list of all possible mated
minutiae 712 by the similarity score, and then scan the list and
remove particular minutiae pairs 724 that have minutiae indexes
that are similar to the minutiae pairs with the highest similarity
scores in the sorted modified list of all possible mated minutiae
722.
[0115] The global consistency pairing module 720 may then
initialize a set of groups based on the number of reference
minutiae included within the list. For instance, a group may be
created for each reference minutia such that if there are multiple
minutiae pairs within the list of sorted modified list of all
possible minutiae 722 for a single reference minutiae, the multiple
minutiae pairs are included in the same group. In some instances,
the global consistency pairing module 720 may additionally check
the geometric consistency between each of the minutiae pairs within
the same group and remove minutiae pairs that are determined not be
geometrically consistent. The global consistency pairing module 720
may then compute an average similarity score for each group based
on aggregating the individual similarity scores associated with
each of the minutiae pairs within the group.
[0116] After computing the average similarity scores for each
group, the global consistency pairing module 720 may then compare
the average similarity scores between each group and select the
group that has the highest average similarity score and then
provide the list of minutiae that are included within the group for
output of the process 700 and include the top average similarity
score.
[0117] As describe above, FIGS. 5-7 illustrate processes that are
utilized by the automatic fingerprint identification system to
process, analyze, and match individual minutiae from a search
fingerprint to a reference fingerprint. As described in FIG. 8,
these processes may be utilized within a matching operation for a
distorted fingerprint to compute a final similarity score that
indicates a confidence of a fingerprint match between a search
fingerprint and a reference fingerprint.
[0118] Distorted Fingerprint Matching
[0119] FIG. 8 illustrates an exemplary distorted fingerprint
matching operation 800. The fingerprint matching operation 800 may
be used to match a search fingerprint that may be nonlinearly
distorted or may exhibit other types of errors that would like to
cause inaccurate results using contemporary fingerprint matching
technologies. The fingerprint matching operation 800 employs the
principles and techniques described previously in FIGS. 1-7 to
reduce errors resulting from distortion.
[0120] The fingerprint matching operation 800 may include
generating a list of invariant and distortion-tolerant reference
OFVs 802a for a list of reference minutiae 802a, and a list of
invariant and distortion-toleration search OFVs 804b for a list of
search minutiae 802a. For instance, as described in FIG. 4, the
list of reference OFVs 804a and the list of search OFVs 804b may be
generated by comparing each of the list of search minutiae 802a and
the list of reference 802b to its neighboring minutiae within a set
of octant sectors.
[0121] An OFV comparison module may compare the respective search
and reference OFVs 804a and 804b, respectively, and generate a list
of all possible mated minutiae 812. For instance, as described in
FIGS. 6-7, the list of all possible mated minutiae may include
individual pairs of search minutiae and corresponding reference
minutiae based on comparing the features included within the
respective reference OFVs included in the list of reference OFVs
804a and the respective search OFVs included in the list of search
OFVs 804b. The list of reference OFVs 804a and the list of search
OFVs 804b may include a set of absolute features that improve
fingerprint matching for distorted fingerprint matching. For
instance, the absolute features may include a direction, x-y
coordinates, a quality score, a ridge frequency, and a curvature.
In some implementations, the absolute features are included into
the calculation of individual similarity scores as described in
FIG. 5 as weights to factor in impacts of distortions on the search
fingerprint.
[0122] A global alignment module 820 may cluster all of the list of
all possible mated minutiae pairs into a plurality of histogram
bins, and subsequently perform a first global alignment procedure
on each historical bin to determine two best rotations 822 for the
list of all possible mated minutiae 812. For instance, as described
in FIG. 6, the global alignment module 820 may initially cluster
the list of all possible mated minutiae 812 based on a range of
rotation angles, determine the two best rotations 822 based on
comparing the each of the transformed search fingerprint under each
rotation angle to the reference fingerprint.
[0123] A precision alignment module 830 may use the two best
rotations 822 to identify a list of best-aligned minutiae 832
within the list of all possible mated minutiae. The precision
alignment module 830 may also perform a two-step pruning procedure
to generate a list of geometrically-consistent groups 834. For
instance, as described in FIG. 7, the two-step pruning procedure
may include performing a local geometric pairing operation by
removing false correspondence minutiae from the list of all
possible mated minutiae.
[0124] A similarity module 840 may compute an individual similarity
score for each mated minutiae pair within the list of
geometrically-consistent groups 834. For instance, as described in
FIG. 5, the individual similarity score may computed based on
aggregating single similarity scores between individual octant
sectors of the respective OFVs of the search and reference minutia
within a mated minutiae pair. The similarity score calculation may
additionally include wrights based on the absolute features
included within the respective OFVs for the search and reference
minutiae included within the list of all possible mated minutiae.
The list of individual similarity scores for all mated minutiae may
then be aggregated to compute a final similarity score between the
reference fingerprint and the search fingerprint. In some
instances, the final similarity score is computed based on
selecting the maximum value of the individual similarity score for
all of the mated minutiae.
[0125] In some implementations, a distortion flag may additionally
be set after the final similarity score is computed based on the
number of mated minutiae that are identified as distortion
minutiae. In some instances, the number of distorted minutiae may
be compared to a threshold value to determine whether the final
similarity score may be severely distorted. In such instances, a
distortion notification may be provided to an end-user of the
automatic fingerprint identification system 100 indicating that the
final similarity score may be distorted and requires manual
verification or re-enrollment of either the search or reference
fingerprints.
[0126] It should be understood that processor as used herein means
one or more processing units (e.g., in a multi-core configuration).
The term processing unit, as used herein, refers to
microprocessors, microcontrollers, reduced instruction set circuits
(RISC), application specific integrated circuits (ASIC), logic
circuits, and any other circuit or device capable of executing
instructions to perform functions described herein.
[0127] It should be understood that references to memory mean one
or more devices operable to enable information such as
processor-executable instructions and/or other data to be stored
and/or retrieved. Memory may include one or more computer readable
media, such as, without limitation, hard disk storage, optical
drive/disk storage, removable disk storage, flash memory,
non-volatile memory, ROM, EEPROM, random access memory (RAM), and
the like.
[0128] Additionally, it should be understood that communicatively
coupled components may be in communication through being integrated
on the same printed circuit board (PCB), in communication through a
bus, through shared memory, through a wired or wireless data
communication network, and/or other means of data communication.
Additionally, it should be understood that data communication
networks referred to herein may be implemented using Transport
Control Protocol/Internet Protocol (TCP/IP), User Datagram Protocol
(UDP), or the like, and the underlying connections may comprise
wired connections and corresponding protocols, for example,
Institute of Electrical and Electronics Engineers (IEEE) 802.3
and/or wireless connections and associated protocols, for example,
an IEEE 802.11 protocol, an IEEE 802.15 protocol, and/or an IEEE
802.16 protocol.
[0129] A technical effect of systems and methods described herein
includes at least one of: (a) increased accuracy in facial matching
systems; (b) reduction of false accept rate (FAR) in facial
matching; (c) increased speed of facial matching.
[0130] Although specific features of various embodiments of the
invention may be shown in some drawings and not in others, this is
for convenience only. In accordance with the principles of the
invention, any feature of a drawing may be referenced and/or
claimed in combination with any feature of any other drawing.
[0131] This written description uses examples to disclose the
invention, including the best mode, and also to enable any person
skilled in the art to practice the invention, including making and
using any devices or systems and performing any incorporated
methods. The patentable scope of the invention is defined by the
claims, and may include other examples that occur to those skilled
in the art. Such other examples are intended to be within the scope
of the claims if they have structural elements that do not differ
from the literal language of the claims, or if they include
equivalent structural elements with insubstantial differences from
the literal language of the claims.
* * * * *