U.S. patent application number 11/941019 was filed with the patent office on 2010-11-18 for daubechies wavelet transform of iris image data for use with iris recognition system.
This patent application is currently assigned to SENGA ADVISORS, LLC. Invention is credited to SEONG-WON CHO.
Application Number | 20100290676 11/941019 |
Document ID | / |
Family ID | 19706518 |
Filed Date | 2010-11-18 |
United States Patent
Application |
20100290676 |
Kind Code |
A1 |
CHO; SEONG-WON |
November 18, 2010 |
DAUBECHIES WAVELET TRANSFORM OF IRIS IMAGE DATA FOR USE WITH IRIS
RECOGNITION SYSTEM
Abstract
Disclosed is a method of recognizing human iris using Daubechies
wavelet transform, wherein the dimensions of characteristic vectors
are reduced by extracting iris features from inputted iris image
signals through the Daubechies wavelet transform, binary
characteristic vectors are generated by applying quantization
functions to the extracted characteristic values so that utility of
human iris recognition can be improved since storage capacity arid
processing time thereof can be improved since storage capacity
characteristic vectors, and a measurement process suitable for the
low capacity characteristic vectors is employed when measuring
vectors and previously registered characteristic vectors.
Inventors: |
CHO; SEONG-WON; (SEOUL,
KR) |
Correspondence
Address: |
KNOBBE MARTENS OLSON & BEAR LLP
2040 MAIN STREET, FOURTEENTH FLOOR
IRVINE
CA
92614
US
|
Assignee: |
SENGA ADVISORS, LLC
BOSTON
MA
|
Family ID: |
19706518 |
Appl. No.: |
11/941019 |
Filed: |
November 15, 2007 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
10656885 |
Sep 5, 2003 |
7302087 |
|
|
11941019 |
|
|
|
|
PCT/KR01/01303 |
Jul 31, 2001 |
|
|
|
10656885 |
|
|
|
|
Current U.S.
Class: |
382/117 |
Current CPC
Class: |
G06K 9/00597
20130101 |
Class at
Publication: |
382/117 |
International
Class: |
G06K 9/00 20060101
G06K009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Mar 6, 2001 |
KR |
2001/11440 |
Claims
1. A method of processing iris image data, comprising: providing
data of an iris image for processing; processing the iris image
data so as to provide a reduced iris image data, wherein processing
includes conducting a Daubechies wavelet transform multiple times,
wherein the reduced iris image data has a smaller size than the
iris image data and has a smaller amount of high frequency
components than the iris image data; creating a characteristic
vector of the iris image using the reduced image data; providing a
reference characteristic vector of iris image of a preregistered
person; and determining whether the iris image is associated with
the preregistered person, using the characteristic vector and the
reference characteristic vector.
2. The method of claim 1, wherein processing comprises: computing
an inner product of the reference characteristic vector and the
characteristic vector of the iris image; comparing the inner
product against a predetermined threshold value; and determining
that the iris image is associated with the predetermined person
when the inner product is greater than the predetermined threshold
value.
3. The method of claim 1, wherein creating the characteristic
vector uses quantized pixel values of the reduced iris image
data.
4. The method of claim 3, wherein the quantized pixel values
comprise at least two positive values and at least two negative
values.
5. The method of claim 4, wherein the quantized pixel values
comprise one of the at least two positive values has the same
absolute value as one of the at least two negative values.
6. The method of claim 4, wherein the quantized pixel values
comprise a first positive value and a second positive value,
wherein the second positive value is greater than two times of the
first positive value.
7. The method of claim 1, wherein conducting each Daubechies
wavelet transform produces a plurality data representing
transformed images, wherein processing further comprises selecting
one from the plurality of transformed image data.
8. The method of claim 7, wherein the reduced iris image data is
one of the plurality of transformed image data created in the last
Daubechies wavelet transform.
9. The method of claim 7, wherein creating the characteristic
vector uses results of the Daubechies wavelet transforms in
addition to the reduced iris image data.
10. The method of claim 9, wherein creating the characteristic
vector uses at least one non-selected transformed image data in
addition to the reduced iris image data.
11. The method of claim 10, wherein the characteristic vector
comprises substantially more information representing the reduced
iris image data than information representing the at least one
non-selected transformed image data.
12. The method of claim 7, wherein each of the plurality of
transformed image data is classified one of HH, HL, LH and LL,
wherein HH represents high frequency components in a first
direction and a second direction in the transformed image, the
first and second directions being perpendicular to each other,
wherein HL represents a high frequency component in the first
direction and a low frequency component in the second direction,
wherein LH represents a low frequency component in the first
direction and a high frequency component in the second direction,
and wherein LL represents low frequency components in the first and
second directions, wherein LL is selected among HH, HL, LH and LL
for following Daubechies wavelet transform.
13. The method of claim 12, wherein an average value of the piece
of transformed image data classified as HH is included in the
characteristic vector.
14. The method of claim 13, wherein a total number of the
Daubechies wavelet transform is N, wherein the characteristic
vector comprises an N-1 number of values representing the HH data
pieces.
15. The method of claim 1, wherein the number of the multiple times
is from 2 to 7.
16. The method of claim 1, wherein the number of the plurality of
times is from 4.
17. An iris image data processing apparatus, comprising at least
one integrated circuit programmed to perform the method of claim
1.
18. The apparatus of claim 17, wherein processing comprises:
computing an inner product of the reference characteristic vector
and the characteristic vector of the iris image; comparing the
inner product against a predetermined threshold value; and
determining that the iris image is associated with the
predetermined person when the inner product is greater than the
predetermined threshold value.
19. The apparatus of claim 17, wherein creating the characteristic
vector uses quantized pixel values of the reduced iris image data,
and wherein the quantized pixel values comprise at least two
positive values and at least two negative values.
20. The apparatus of claim 17, wherein conducting each Daubechies
wavelet transform produces a plurality data representing
transformed images, wherein processing further comprises selecting
one from the plurality of transformed image data, and wherein the
reduced iris image data is one of the plurality of transformed
image data created in the last Daubechies wavelet transform.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation application of
application Ser. No. 10/656,885, which is a continuation
application under 35 U.S.C. .sctn.365 (c) claiming the benefit of
the filing date of PCT Application No. PCT/KR01/01303 designating
the United States, filed Jul. 31, 2001. The PCT Application was
published in English as WO 02/071317 A1 on Sep. 12, 2002, and
claims the benefit of the earlier filing date of Korean Patent
Application No. 2001/11440, filed Mar. 6, 2001. The contents of the
Korean Patent Application No. 2001/11440, the international
application No. PCT/KR01/01303 including the publication WO
02/071317 A1, and application Ser. No. 10/656,885 are incorporated
herein by reference in their entirety.
BACKGROUND
[0002] 1. Field
[0003] The present disclosure relates to a method of processing
iris image data, and more particularly, to a method of determining
whether an iris image matches the preregistered iris image.
[0004] 2. Discussion of the Related Technology
[0005] An iris recognition system is an apparatus for performing
identification of each individual by differentiating iris patterns
of the pupil of an eye, which are unique for each individual. It
has superior identification accuracy and excellent security as
compared with other biometric method using voice and fingerprints
from each individual. A human iris is the region between a pupil
and a white sclera of an eye, and iris recognition is a technique
for performing identification of each individual based on
information that is obtained from an analysis of the iris patterns
which are different in each individual.
[0006] In general, it is a core technology to efficiently acquire
unique characteristic information from input images in the field of
an applied technology for performing identification of each
individual by utilizing the characteristic information of the human
body. A wavelet transform is used to extract characteristics of the
iris images, and it is a kind of technique of analyzing signals in
multiresolution mode. The wavelet transform is a mathematical
theory of formulating a model for signals, systems and a series of
processes by using specifically selected signals. These signals are
referred to as little waves or wavelets. Recently, the wavelet
transform is widely employed in the field of signal and image
processing since it has a fast rate as compared with a signal
processing algorithm based on the Fourier transform, and it can
efficiently accomplish signal localization in time and frequency
domains.
[0007] On the other hand, the images, which are obtained by
extracting only iris patterns from the iris images acquired by
image acquisition equipment and normalizing the patterns at a
450.times.60 size, are used to extract characteristic values
through the wavelet transform. Further, a Harr wavelet transform
has been widely used in iris recognition, image processing and the
like. However, Harr wavelet functions have disadvantages in that
the characteristic values are discontinuously and rapidly changed
and that high resolution of the images cannot be obtained in a case
where the images are again decompressed after they have been
compressed. On the contrary, since Daubechies wavelet functions are
continuous functions, the disadvantages of the Harr wavelet
functions that the values thereof are discontinuously and rapidly
changed can be avoided, and the characteristic values can be
extracted more accurately and delicately. Therefore, in a case
where the images are to be again decompressed after they have been
compressed by using the Daubechies wavelet transform, the images
can be restored in high resolution nearer to the original images
than when the Harr wavelet transform is used. Since the Daubechies
wavelet functions are more complicated than the Han wavelet
functions, there is a disadvantage in that more arithmetic quantity
may be needed. However, it can be easily overcome by the recent
advent of ultrahigh speed microprocessors.
[0008] There is also an advantage in that the Daubechies wavelet
transform can obtain fine characteristic values in the process of
performing the wavelet transform for extracting the characteristic
values. That is, if the Daubechies wavelet transform is used,
expression of the iris features can be made in a low capacity of
data and extraction of the features can be made accurately.
[0009] Methods of extracting the characteristic values and forming
the characteristic vectors by using Gabor transform been mainly
used in the iris recognition field. However, the characteristic
vectors generated by these methods are formed to have 256 or more
dimensions, and they have at least 256 bytes even though it is
assumed that one byte is assigned to one dimension. Thus, there is
a problem in that practicability and efficiency can be reduced when
it is used in the field where low capacity information is needed.
Accordingly, it is necessary to develop a method of forming the low
capacity characteristic vectors wherein processing, storage,
transfer, search, and the like of the pattern information can be
efficiently made. In addition, since a simple method of measuring a
distance such as a Hamming distance (HD) between two characteristic
vectors (characteristic vectors relevant to the input pattern and
stored reference characteristic vectors) is used for pattern
classification in U.S. Pat. No. 5,291,560, there are disadvantages
in that formation of the reference characteristic vectors through
generalization of the pattern information cannot be easily made and
information characteristics of each dimension of the characteristic
vectors cannot be properly reflected.
[0010] That is, in the method of using the Hamming distance in
order to verify the two characteristic vectors generated in the
form of binary vectors, bit values assigned according to respective
dimensions are compared with each other. If they are identical to
each other, 0 is given; and if they are different from each other,
1 is given. Then, a value divided by the total number of the
dimensions is obtained as a final result. The method is simple and
useful in discriminating a degree of similarity between the
characteristic vectors consisted of binary codes. When the Hamming
distance is used, the comparison result of all the bits becomes 0
if identical data are compared with each other. Thus, the result
approaching to 0 implies that the data belong to the persons
themselves. If the data do indeed belong to the person, the
probability of a degree of similarity will be 0.5. Thus, upon
comparison with the other person's data, it is understood that the
values converge around 0.5. Accordingly, a proper limit set between
0 and 0.5 will be a boundary for differentiating the data of the
persons themselves from the other person's data. The Hamming
distance (HD) is excellent in performance thereof in a case where
the information is obtained from the extracted iris features by
subdividing the data, but it is not suitable when low capacity data
is to be used.
[0011] The foregoing discussion in the background section is to
provide general background information and does not constitute an
admission of prior art.
SUMMARY
[0012] One aspect of the invention provides a method of processing
iris image data, which comprise: providing data of an iris image
for processing; processing the iris image data so as to provide a
reduced iris image data, wherein processing includes conducting a
Daubechies wavelet transform multiple times, wherein the reduced
iris image data has a smaller size than the iris image data and has
a smaller amount of high frequency components than the iris image
data; creating a characteristic vector of the iris image using the
reduced image data; providing a reference characteristic vector of
iris image of a preregistered person; and determining whether the
iris image is associated with the preregistered person, using the
characteristic vector and the reference characteristic vector.
[0013] In the foregoing method, processing may comprise computing
an inner product of the reference characteristic vector and the
characteristic vector of the iris image; comparing the inner
product against a predetermined threshold value; and determining
that the iris image is associated with the predetermined person
when the inner product is greater than the predetermined threshold
value. Creating the characteristic vector may use quantized pixel
values of the reduced iris image data. The quantized pixel values
may comprise at least two positive values and at least two negative
values. The quantized pixel values may comprise one of the at least
two positive values has the same absolute value as one of the at
least two negative values. The quantized pixel values may comprise
a first positive value and a second positive value, wherein the
second positive value is greater than two times of the first
positive value.
[0014] Still in the foregoing method, conducting each Daubechies
wavelet transform may produce a plurality data representing
transformed images, wherein processing may further comprise
selecting one from the plurality of transformed image data. The
reduced iris image data may be one of the plurality of transformed
image data created in the last Daubechies wavelet transform.
Creating the characteristic vector may use results of the
Daubechies wavelet transforms in addition to the reduced iris image
data. Creating the characteristic vector may use at least one
non-selected transformed image data in addition to the reduced iris
image data. The characteristic vector may comprise substantially
more information representing the reduced iris image data than
information representing the at least one non-selected transformed
image data.
[0015] Yet in the foregoing method, each of the plurality of
transformed image data may be classified one of HH, HL, LH and LL,
wherein HH represents high frequency components in a first
direction and a second direction in the transformed image, the
first and second directions being perpendicular to each other,
wherein HL represents a high frequency component in the first
direction and a low frequency component in the second direction,
wherein LH represents a low frequency component in the first
direction and a high frequency component in the second direction,
and wherein LL represents low frequency components in the first and
second directions, wherein LL is selected among HH, HL, LH and LL
for following Daubechies wavelet transform. An average value of the
piece of transformed image data classified as HH may be included in
the characteristic vector. A total number of the Daubechies wavelet
transform is N, wherein the characteristic vector may comprise an
N-1 number of values representing the HH data pieces. The number of
the multiple times may be from 2 to 7. The number of the plurality
of times may be from 4.
[0016] Another aspect of the invention provides an iris image data
processing apparatus, comprising at least one integrated circuit
programmed to perform the foregoing method.
[0017] In the foregoing apparatus, processing may comprise:
computing an inner product of the reference characteristic vector
and the characteristic vector of the iris image; comparing the
inner product against a predetermined threshold value; and
determining that the iris image is associated with the
predetermined person when the inner product is greater than the
predetermined threshold value. Creating the characteristic vector
may use quantized pixel values of the reduced iris image data, and
wherein the quantized pixel values may comprise at least two
positive values and at least two negative values. Conducting each
Daubechies wavelet transform may produce a plurality data
representing transformed images, wherein processing may further
comprise selecting one from the plurality of transformed image
data, and wherein the reduced iris image data is one of the
plurality of transformed image data created in the last Daubechies
wavelet transform.
BRIEF DESCRIPTION OF THE DRAWINGS
[0018] FIG. 1 is a view showing the constitution of image
acquisition equipment used for performing an iris recognition
method according to an embodiment of the present invention.
[0019] FIG. 2 is a flowchart showing procedures for verifying an
iris image according to an embodiment of the present invention.
[0020] FIG. 3 is a flowchart showing procedures for multi-dividing
the iris image through Daubechies wavelet transform according to an
embodiment of the present invention.
[0021] FIG. 4 shows an example of multi-dividing the iris image
through the Daubechies wavelet transform.
[0022] FIG. 5 is a flowchart showing procedures for forming a
characteristic vector of the iris image based on data acquired from
the procedures of multi-dividing the iris image according to an
embodiment of the present invention.
[0023] FIG. 6a shows a distribution example of characteristic
values of the extracted iris image.
[0024] FIG. 6b shows a quantization function for generating binary
characteristic vector from the distribution example of FIG. 6a.
[0025] FIG. 7 is a flowchart showing procedures for determining
user authenticity through a similarity measurement between the
characteristic vectors.
DETAILED DESCRIPTION OF EMBODIMENTS
[0026] Hereinafter, various embodiments of the present invention
will be explained in detail with reference to the accompanying
drawings.
[0027] FIG. 1 shows the constitution of image acquisition equipment
for use in a method of recognizing a human iris according to an
embodiment of the present invention. Referring to FIG. 1, the
constitution of the iris image acquisition equipment will be
explained. The image acquisition equipment for use in the method of
recognizing the human iris according an embodiment of to the
present invention comprises a halogen lamp 11 for illuminating the
iris in order to acquire clear iris patterns, a CCD camera 13 for
photographing an eye 10 of a user through a lens 12, a frame
grabber 14 connected to the CCD camera 12 for acquiring an iris
image, and a monitor 15 for showing the image, which are currently
inputted to the camera, to the user so that acquisition of correct
images and positioning convenience of the user can be obtained when
images are acquired.
[0028] According to the constitution of the image acquisition
equipment, the CCD camera is used to acquire the image, and iris
recognition is made through a pattern analysis of iridial folds.
However, in a case where the iris image is acquired indoors by
using an ordinary illuminator, it is difficult to extract desired
pattern information since the iris image is generally gloomy.
Additional illuminators therefore are used so that the information
on the iris image cannot be lost. In such a case, loss of the iris
pattern information and deterioration of recognition capability due
to reflective light should be prevented, and proper illuminators
are utilized so that a clear iris pattern can be obtained. In one
embodiment of the present invention, the halogen lamp 11 having
strong floodlighting effects is used as a main illuminator so that
the iris pattern can be clearly shown. Further, as shown in FIG. 1,
the loss of the iris image information and eye fatigue of the user
can be avoided by placing the halogen lamp illuminators on the left
and right sides of the eye in order to cause the reflective light
from the lamp to be formed on outer portions of the iris
region.
[0029] FIG. 2 is a flowchart showing procedures for verifying the
iris image according to an embodiment of the present invention.
Referring to FIG. 2, an eye image is acquired through the image
acquisition equipment as mentioned above in step 200. In step 210,
images of the iris regions are extracted from the acquired eye
image-through pre-processing and transformed into a polar
coordinate system, and the transformed iris pattern is inputted to
a module for extracting the features. In step 220, the Daubechies
wavelet transform of the inputted iris pattern transformed into the
polar coordinate system is performed, and the features of the iris
regions are then extracted. The extracted features have real
numbers. In step 230, a binary characteristic vector is generated
by applying K-level quantization function to the extracted
features. In step 240, similarity between the generated
characteristic vector and previously registered data of the user is
measured. Through the similarity measurement, user authenticity is
determined and then verification results are shown.
[0030] In a case where the features of the iris regions are
extracted by performing the Daubechies wavelet transform as
described above, the Daubechies wavelet function having eight,
sixteen or more coefficients can extract more delicate
characteristic values than the Daubechies wavelet function having
four coefficients, even though the former is more complicate than
the latter. Although the Daubechies wavelet function having eight
or more coefficients has been used and tested in an embodiment of
the present invention, greater performance improvement was not
obtained from an embodiment of the present invention and arithmetic
quantity and processing time are increased, as compared with a case
where the Daubechies wavelet function having four coefficients has
been used and tested. Thus, the Daubechies wavelet function having
four coefficients has been used for extracting the characteristic
values.
[0031] FIG. 3 is a flowchart showing procedures for multi-dividing
the iris image by performing the Daubechies wavelet transform
according to an embodiment of the present invention, and FIG. 4
shows an image divided through the Daubechies wavelet transform.
Referring to FIGS. 3 and 4, in an embodiment of the present
invention, the Daubechies wavelets among various mother wavelets
are used to perform extraction of the iris image characteristics.
As shown in FIG. 4, when "L" and "H" are respectively used to
indicated low frequency and high frequency components, the term
"LL" means a component that has passed through a low-pass filter
(LPF) in all x and y directions whereas a term "HH" means a
component that has passed through a high-pass filter (HPF) in the x
and y directions. Furthermore, subscript numerals signify
image-dividing stages. For example, "LH.sub.2" means that the image
has passed through the low-pass filter in the x direction and
through the high-pass filter in the y direction during 2-stage
wavelet division.
[0032] In step 310, the inputted iris image is multi-divided by
using the Daubechies wavelet transform. Since the iris image is
considered as a two-dimensional signal in which one-dimensional
signals are arrayed in the x and y directions, quarterly divided
components of one image is extracted by passing through the LPF and
HPF in all x and y directions in order to analyze the iris image.
That is, one two-dimensional image signal is wavelet-transformed in
vertical and horizontal directions, and the image is divided into
four regions LL, LH, HL, and HH after the wavelet transform has
been performed once. At this time, through the Daubechies wavelet
transform, the signal is divided into a differential component
thereof that has passed through the high-pass filter, and an
average component that has passed through the low-pass filter
[0033] Alternatively, performance of the iris recognition system is
evaluated in view of two factors; a false acceptance rate (FAR) and
a false rejection rate (FRR). Here, the FAR means a probability
that entrance of unregistered persons (imposters) may be accepted
due to false recognition of unregistered persons as registered
ones, and the FRR means a probability that entrance of registered
persons (enrollees) is rejected due to false recognition of the
registered persons as unregistered ones. For reference, when the
method of recognizing the human iris using the Daubechies wavelet
transform according to an embodiment of the present invention is
employed, the FAR has been reduced from 5.5% to 3.07% and the FRR
has also been reduced from 5.0% to 2.25%, as compared with the
method of recognizing the human iris using the Harr wavelet
transform.
[0034] In step 320, a region HH including only high frequency
components in the x and y directions is extracted from the divided
iris image.
[0035] In step 330, after increasing the iterative number of times
of dividing the iris image, the processing step is completed when
the iterative number is greater than a predetermined number.
Alternatively, if the iterative number is lower than the
predetermined number, the information on the region HH is stored as
information for extracting the iris features in step 340.
[0036] Further, in step 350, a region LL comprising only low
frequency components in the x and y directions is extracted from
the multi-divided iris image. Since the extracted region LL
(corresponding to the image reduced in a fourth size as compared
with the previous image) includes major information on the iris
image, it is provided as an image to be newly processed so that the
wavelet transform can be again applied to the relevant region.
Thereafter, the Daubechies wavelet transform is repeatedly
performed from step 310.
[0037] On the other hand, in a case where the iris image is
transformed from the Cartesian coordinate system to polar
coordinate system, in order to avoid changes in the iris features
according to variations in the size of the pupil, the region
between the inner and outer boundaries of the iris is divided into
60 segments in the r direction and 450 segments in the 0 direction
by varying the angles by 0.8 degrees. Finally, the information on
the iris image is acquired and normalized as 450.times.60
(.theta..times.r) data. Then, if the acquired iris image is once
again wavelet-transformed, the characteristics of the 225.times.30
region HH.sub.1 of which size is reduced by half are obtained,
namely, the 225.times.30 information is used as a characteristic
vector. This information may be used as it is, but a process of
dividing the signals is repeatedly performed in order to reduce the
information size. Since the region LL includes major information on
the iris image, the characteristic values of further reduced
regions such as HH.sub.2, HH.sub.3 and are obtained by successively
applying the wavelet transform to respective relevant regions.
[0038] The iterative number, which is provided as a discriminating
criterion for repeatedly performing the wavelet transform, is set
as a proper value in consideration of loss of the information and
size of the characteristic vector. Therefore, in an embodiment of
the present invention, the region HH.sub.4 obtained by performing
the wavelet transform four times becomes a major characteristic
region, and values thereof are selected as components of the
characteristic vector. At this time, the region HH.sub.4 contains
the information having 84 (=28.times.3) data
[0039] FIG. 5 is a flowchart showing procedures for forming the
characteristic vector of the iris image by using the data acquired
from the process of multi-dividing the iris image according to an
embodiment of the present invention. Referring to FIG. 5, the
information on the n characteristic vector extracted from the above
process, i.e., the information on the regions HH.sub.1, HH.sub.2,
HH.sub.3, and HH.sub.4 is inputted in step 510. In step 520, in
order to acquire the characteristic information on the regions
HH.sub.1, HH.sub.2 and HH.sub.3 excluding the information on the
region HH.sub.4 obtained through the last wavelet transform among
the n characteristic vector, each average value of the regions
HH.sub.1, HH.sub.2 and HH.sub.3 is calculated and assigned one
dimension. In step 530, all the values of the final obtained region
HH.sub.4 are extracted as the characteristic values thereof. After
extraction of the characteristics of the iris image signals has
been completed, the characteristic vector is generated based on
these characteristics. A module for generating the characteristic
vector mainly performs the processes of extracting the
characteristic values in the form of real numbers and then
transforming them to binary codes consisting of 0 and 1.
[0040] However, in step 540, the N-1 characteristic values
extracted from step 520 and the M (the size of the final obtained
region HH) characteristic values extracted from step 530 are
combined and (M+N-1)-dimensional characteristic vector is
generated. That is, the total 87 data, which the 84 data of the
region HH.sub.4 and the 3 average data of the regions HH.sub.1,
HH.sub.2 and HH.sub.3 are combined, are used as a characteristic
vector in an embodiment of the present invention.
[0041] In step 550, the values of the previously obtained
characteristic vector, i.e., respective component values of the
characteristic vector expressed in the form of the real numbers are
quantized into binary values 0 or 1. In step 560, the resultant
(M+N-1)-bit characteristic vector is generated by the quantized
values. That is, according to an embodiment of the present
invention, the resultant 87-bit characteristic vector is
generated.
[0042] FIG. 6a shows a distribution example of the characteristic
values of the extracted iris image. When the values of the
87-dimensional characteristic vector are distributed according to
respective dimensions, the distribution roughly takes a shape of
FIG. 6a. The binary vector including all the dimensions is
generated by the following Equation 1.
f.sub.n=0 if f(n)<0
f.sub.n=1 if f(n)>0 [Equation 1]
[0043] where f(n) is a characteristic value of the n-th dimension
and f.sub.n is a value of the n-th characteristic vector.
[0044] When the 87-bit characteristic vector that is obtained by
assigning one bit to the total 87 dimensions are generated in order
to use a low capacity characteristic vector, improvement of the
recognition rate is limited to some extent since loss of the
information on the iris image is increased. Therefore, when
generating the characteristic vector, it is necessary to prevent
information loss while maintaining the minimum capacity of the
characteristic vector.
[0045] FIG. 6b shows a quantization function for generating a
binary characteristic vector from the distribution example of the
characteristic values shown in FIG. 6a The extracted
(M+N-1)-dimensional characteristic vector shown in FIG. 6a is
evenly distributed mostly between 1 and -1 in view of its
magnitude. Then, the binary vector is generated by applying the
K-level quantization function shown in FIG. 6a to the
characteristic vector. Since only signs of the characteristic
values are obtained through the process of Equation 1, it is
understood, that information on the magnitude has been discarded.
Thus, in order to accept the magnitude of the characteristic
vector, a 4-level quantization process was utilized in an
embodiment of the present invention.
[0046] As described above, in order to efficiently compare the
characteristic vector generated through the 4-level quantization
with the registered characteristic vector, the quantization levels
have the weights expressed in the following Equation 2.
f.sub.n=4 if f(n).gtoreq.0.5 (level 4)
f.sub.n=1 if 0.5>f(n).gtoreq.0 (level 3)
f.sub.n=-1 if 0>f(n)>-0.5 (level 2)
f.sub.n=-4 if f(n).ltoreq.-0.5 (level 1) [Equation 2]
[0047] where f.sub.n means an n-th dimension of the previously
registered characteristic vector f.sub.R of the user or the
characteristic vector f.sub.T of the user generated from the iris
image of the eye image of the user. Explanation of how to use the
weights expressed in Equation 2, is as follows.
[0048] In a case where the n-th dimensional characteristic value
f(n) is equal or more than 0.5 (level 4), the value of the i-th
dimension f.sub.Ri or f.sub.Ti is converted and assigned 4 if the
value is "11". In a case where the n-th dimensional characteristic
value f(n) is more than 0 and, less than 0.5 (level 3), the value
of the i-th dimension f.sub.Ri or f.sub.Ti is converted and
assigned 1 if the value is "10". In a case where the n-th
dimensional characteristic value f(n) is more than -0.5 and less
than 0 (level 2), the value of the i-th dimension f.sub.Ri or
f.sub.Ti, is converted and assigned -1 if the value is "01". In a
case where the n-th dimensional characteristic value f(n) is equal
or less than -0.5 (level 1), the value of the i-th dimension
f.sub.Ri or f.sub.Ti, is converted and assigned -4 if the value is
"00". This is due to the weights being applied to respective values
as expressed in Equation 2 as it is suitable for the following
verification method of an embodiment of the present invention.
[0049] FIG. 7 is a flowchart showing procedures for discriminating
the user authenticity through similarity measurement between the
characteristic vectors. Referring to FIG. 7, in step 710, the
characteristic vector f.sub.T of the user is generated from the
iris image of the eye image of the user. In step 720, the
previously registered characteristic vector f.sub.R of the user is
searched. In step 730, in order to measure the similarity between
the two characteristic vectors, the weights are assigned to the
characteristic vectors f.sub.R and f.sub.T depending on the value
of the binary characteristic vector based on Equation 2.
[0050] In step 740, an inner product or scalar product S of the two
characteristic vectors is calculated and the similarity is finally
measured. Among the measures generally used for determining
correlation between the registered characteristic vector f.sub.R
and the characteristic vector f.sub.T of the user, it is the inner
product S of the two characteristic vectors which indicate the most
direct association. That is, after the weights have been assigned
to the respective data of the characteristic vector in step 730,
the inner product S of the two characteristic vectors is used to
measure the similarity between the two vectors.
[0051] The following Equation 3 is used for calculating the inner
product of the two characteristic vectors.
S = i = 1 n f Ri f Ti = ( f R 1 f T 1 + f R 2 f T 2 + + f Rn f T n
) . [ Equation 3 ] ##EQU00001##
[0052] where f.sub.R is the characteristic vector of the user that
has been already registered, and f.sub.T is the characteristic
vector of the user that is generated from the iris image of the eye
of the user.
[0053] According to the above processes, one effect which can be
obtained by the quantization according to the sign of the
characteristic vector values as in the method in which the binary
vector is generated with respect to the values of the
characteristic vector extracted from the iris image according to
respective dimensions can be maintained. That is, like the Harming
distance, the difference between 0 and 1 can be expressed. In a
case where the two characteristic vectors have the same-signed
values with respect to the each dimension, positive values are
added to the inner product S of the two characteristic vectors.
Otherwise, negative values are added to the inner product S of the
two vectors. Consequently, the inner product S of the two
characteristic vectors increases if the two data belong to an
identical person, while the inner product S of the two
characteristic vectors decreases if the two data does not belong to
an identical person.
[0054] In step 750, the user authenticity is determined according
to the measured similarity obtained from the inner product S of the
two characteristic vectors. At this time, the determination of the
user authenticity based on the measured similarity depends on the
following Equation 4.
If S>C, then TRUE or else FALSE [Equation 4]
[0055] where C is a reference value for verifying the similarity
between the two characteristic vectors.
[0056] That is, if the inner product S of the two characteristic
vectors is equal or more than the verification reference value C,
the user is determined as an enrollee. Otherwise, the user is
determined as an imposter.
[0057] As described above, the method of recognizing the human iris
using the Daubechies wavelet transform according to an embodiment
of the present invention has an advantage that FAR and FRR can be
remarkably reduced as compared with the method using the Harr
wavelet transform, since the iris features are extracted from the
inputted iris image signals through the Daubechies wavelet
transform.
[0058] Furthermore, in order to verify the similarity between the
registered and extracted characteristic vectors f.sub.R and
f.sub.T, the inner product S of the two characteristic vectors is
calculated, and the user authenticity is determined based on the
measured similarity obtained by the calculated inner product S of
the two vectors. Therefore, there is provided a method of measuring
the similarity between the characteristic vectors wherein loss of
the information, which may be produced by forming the low capacity
characteristic vectors, can be minimized.
[0059] The foregoing is a mere embodiment for embodying the method
of recognizing the human iris using the Daubechies wavelet
transform according to an embodiment of the present invention. The
present invention is not limited to the embodiment described above.
A person skilled in the art can make various modifications and
changes to embodiments of the present invention without departing
from the technical spirit and scope of the present invention
defined by the appended claims.
* * * * *