U.S. patent application number 12/172003 was filed with the patent office on 2008-12-04 for personal authentication apparatus and personal authentication method.
Invention is credited to Miwako DOI.
Application Number | 20080298645 12/172003 |
Document ID | / |
Family ID | 32025240 |
Filed Date | 2008-12-04 |
United States Patent
Application |
20080298645 |
Kind Code |
A1 |
DOI; Miwako |
December 4, 2008 |
PERSONAL AUTHENTICATION APPARATUS AND PERSONAL AUTHENTICATION
METHOD
Abstract
A personal authentication apparatus including a facial region
extraction which extracts an image of a facial region of a person
obtained from image sensing input unit, a guide unit which guides
motion of the person of interest, a feature amount extraction unit
which extracts the feature amount of a face from the image of the
facial region extracted by the facial region extraction unit while
the motion is guided by the guide unit, a dictionary registration
unit which registers the feature amount extracted by the feature
amount extraction unit as a feature amount of the person of
interest, and an authentication unit which authenticates the person
of interest in accordance with the similarity between the feature
amount extracted by the feature amount extraction unit, and a
feature amount which is registered by the dictionary registration
unit.
Inventors: |
DOI; Miwako; (Kawasaki-shi,
JP) |
Correspondence
Address: |
OBLON, SPIVAK, MCCLELLAND MAIER & NEUSTADT, P.C.
1940 DUKE STREET
ALEXANDRIA
VA
22314
US
|
Family ID: |
32025240 |
Appl. No.: |
12/172003 |
Filed: |
July 11, 2008 |
Current U.S.
Class: |
382/118 |
Current CPC
Class: |
G06F 21/32 20130101;
G06K 9/00221 20130101 |
Class at
Publication: |
382/118 |
International
Class: |
G06K 9/78 20060101
G06K009/78 |
Foreign Application Data
Date |
Code |
Application Number |
Sep 27, 2002 |
JP |
2002-282443 |
Claims
1. A personal authentication apparatus comprising: an image sensing
input unit configured to sense a person and input a first image at
registration and a second image at aunthentication; a facial region
extraction unit configured to extract first and second facial
images from the first and second images input by the image sensing
input unit; a guide unit configured to display a guidance of an
image sensing condition of the person at registration; a feature
amount extraction unit configured to extract first and second
feature amounts of the first and second facial images; a dictionary
registration unit configured to register the first feature amount;
and an authenticating unit configured to perform authentication by
determining a similarity between the first feature amount
registered by the dictionary registration unit and the second
feature amount.
2. The apparatus according to claim 1, further comprising: an image
sensing condition extraction unit configured to extract an image
sensing condition of the person at registration on the basis of a
size or luminance of the first facial image, and wherein the guide
unit displays the guidance so that the image sensing condition
extracted by the image sensing condition extraction unit falls
within a predetermined value range.
3. The apparatus according to claim 1, wherein the guide unit
displays a character which relaxes the person and guides a movement
of the person.
4. The apparatus according to claim 1, wherein the guide unit
displays the first facial image and a guide mark to which the
person is to be moved.
5. A personal authentication method comprising: sensing a person
and inputting a first image at registration and a second image at
authentication; extracting first and second facial images from the
first and second images input by the inputting; displaying a
guidance of an image sensing condition of the person at
registration; extracting first and second feature amounts of the
first and second facial images; registering the first feature
amount; and performing authentication by determining a similarity
between the first feature amount registered by the registering and
the second feature amount.
6. The method according to claim 5, further comprising: extracting
an image sensing condition of the person at registration on the
basis of a size or luminance of the first facial image, and
displaying the guidance so that the image sensing condition
extracted by the extracting falls within a predetermined value
range.
7. The method according to claim 5, further comprising: displaying
a character which relaxes the person and guides a movement of the
person.
8. The method according to claim 5, further comprising: displaying
the first facial image and a guide mark to which the person is to
be moved.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation Application of, and
claims the benefit of priority under 35 U.S.C. .sctn. 120 from,
U.S. application Ser. No. 10/462,620, filed Jun. 17, 2003, which
claims the benefit of priority under 35 U.S.C. .sctn. 119 from
Japanese Patent Application No. 2002-282443, filed Sep. 27, 2002.
The entire contents of each of the above applications are
incorporated herein by reference.
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] The present invention relates to a personal authentication
apparatus and personal authentication method, which can implement
registration and a personal authentication method with high
reproducibility.
[0004] 2. Description of the Related Art
[0005] In recent years, interests and demands about the security
technique are growing. There are some personal authentication
methods that assure security. In a personal authentication system
that uses a magnetic card such as a credit card or the like, or a
contact type IC card with a built-in IC chip, a user must insert
such card into a reader. When the user has his or her hands full
with some pieces of baggage, the operation for inserting the card
into the reader is troublesome and very inconvenient. In a personal
authentication system that uses a non-contact type IC card
represented by a commuter pass ticket examination system using a
wireless communication, the user need not manually insert the card
into a reader unlike the system that uses the contact type IC card.
However, if the user loses his or her card, such card may be
illicitly used as in the contact type. Also, the user must always
carry the IC card.
[0006] By contrast, a personal authentication system that uses
biometric information (biometrics) such as a fingerprint, iris,
voice, face, and the like is known. In this system, the iris
pattern of the user is registered in an authentication apparatus in
advance, and is verified upon authentication. This authentication
system can assure authentication precision 10 times as high as
fingerprints. However, in order to assure high authentication
precision, the eye must be irradiated with auxiliary light, and the
user must bring his or her face into contact with an authentication
apparatus. Hence, the system forces the user to take given
authentication actions, and cannot assure user's hygiene. For this
reason, such authentication system is used for only some limited
users who require very high security. Recently, a non-contact
authentication system which authenticates the user by sensing an
image of the user's eye using a camera is available. However, in
case of such non-contact system, since the image sensing condition
of the user by the camera is unstable, sufficiently high
authentication precision cannot be assured.
[0007] In a system that uses user's fingerprint information, the
user touches a reader with his or her finger to sense its
fingerprint, and can be authenticated by matching feature points.
This system is unsusceptible to physical growth and aging as in the
iris pattern. However, since the user's skin touches the contact
surface of a detection device with his or her finger, the contact
surface is contaminated with fat and sweat of the hand, and the
precision deteriorates during use. Since the users directly touch
the detection surface with fingers, some users may hate to use such
system in terms of hygiene. If the hand of a person to be
authenticated is dry, his or her finger cannot well contact the
detection surface, and a fingerprint cannot be satisfactorily
read.
[0008] In a personal authentication system using user's
voice/utterance information, the authentication precision depends
on user's physical conditions. For example, even the same person
may often have lower voice reproducibility (e.g., a person may have
a hoarse voice due to cold or hangover). For this reason, speaker
recognition has a problem with its authentication precision, and
has not been developed to a practical level.
[0009] In a personal authentication system that uses user's facial
information, the user need not directly physically touch an
authentication apparatus, and the user's facial image which is
sensed by a camera need only be analyzed to authenticate that user.
Therefore, compared to other systems, the load on the user can be
lightened, and such system can be relatively easily used to
open/close a gate. Such personal authentication system using user's
facial information is described in, e.g., Jpn. Pat. KOKAI
Publication Nos. 9-251534 and 11-175718.
[0010] In order to improve the authentication precision of the
authentication system that utilizes a facial image, a facial image
with a large information size must be sensed, and pixels equal to
or larger than a predetermined value in number are required.
However, when the height of the user is relatively higher than the
camera position for image sensing or when the standing position of
the user is far from the camera, a facial image to be sensed is
small, the number of pixels of the facial image is also small and,
hence, a given image information size cannot be assured, thus
impairing the recognition precision. If the standing position of
the user or the illumination condition of the image sensing site is
different from that upon registration, i.e., the image sensing
conditions between registration and authentication are largely
different, the obtained image information varies, and personal
authentication consequently fails.
[0011] If the user creates another facial expression upon sensing a
facial image, the obtained facial pattern changes. Therefore, in
order to improve the precision of personal authentication, various
facial expressions must be registered upon registering a dictionary
of a given user. Upon registering various facial expressions, if
the user is strained, his or her expression looks stern. As a
result, expected facial expressions cannot be registered, and a
facial expression upon authentication becomes largely different
from that upon registration, thus disturbing improvement in
recognition precision. Conversely, if the system asks for user's
cooperation to sense various facial expressions upon registering a
dictionary, some users overreact, and a facial expression upon
overreaction becomes largely different from that upon
authentication.
BRIEF SUMMARY OF THE INVENTION
[0012] It is an object of the present invention to provide a
personal authentication apparatus and personal authentication
method, which can improve the authentication precision by reducing
the load on the user.
[0013] According to the first aspect of the present invention, a
personal authentication apparatus comprises: facial region
extraction unit configured to extract an image of a facial region
of a person obtained from image sensing input unit; guide unit
configured to guide motion of the person of interest; feature
amount extraction unit configured to extract a feature amount of a
face from the image of the facial region extracted by the facial
region extraction unit while the motion is guided by the guide
unit; dictionary registration unit configured to register the
feature amount extracted by the feature amount extraction unit as a
feature amount of the person of interest; and a unit configured to
authenticate the person of interest in accordance with a similarity
between the feature amount extracted by the feature amount
extraction unit, and a feature amount registered by the dictionary
registration unit.
[0014] According to the second aspect of the present invention, a
personal authentication method comprises: extracting an image of a
facial region of a person obtained from image sensing input means;
guiding motion of the person of interest; extracting a feature
amount of a face from the extracted facial region while the motion
is guided; and authenticating the person of interest in accordance
with a similarity between the extracted feature amount extracted by
the feature amount extraction means, and a feature amount of the
person of interest which is registered in advance.
BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING
[0015] FIG. 1 is a schematic diagram of the first embodiment of the
present invention;
[0016] FIGS. 2A through 2D are explanatory views of facial region
extraction in the first embodiment;
[0017] FIG. 3 is an explanatory view of a normalized pattern and
feature vector in the first embodiment;
[0018] FIG. 4 is a recognition flow chart in the first
embodiment;
[0019] FIG. 5 is an explanatory view of a similarity in the first
embodiment;
[0020] FIGS. 6A through 6C show an example of an interface upon
registration in the first embodiment;
[0021] FIGS. 7A through 7C show an example of an interface upon
authentication in the first embodiment;
[0022] FIGS. 8A and 8B show another example of an interface upon
registration in the first embodiment;
[0023] FIGS. 9A and 9B show another example of an interface upon
authentication in the first embodiment;
[0024] FIGS. 10A through 10E show an example of an interface upon
registration in the second embodiment;
[0025] FIGS. 11A through 11E show an example of an interface upon
authentication in the second embodiment;
[0026] FIGS. 12A through 12E show another example of an interface
upon authentication in the second embodiment; and
[0027] FIG. 13 is a flow chart showing the process of an image
sensing condition extraction means and guide means in the first and
second embodiments.
DETAILED DESCRIPTION OF THE INVENTION
[0028] Preferred embodiments of the present invention will be
described hereinafter with reference to the accompanying
drawings.
[0029] The first embodiment of the present invention will be
described first.
[0030] FIG. 1 shows a schematic arrangement of the first
embodiment. An image sensing input unit 1 inputs a facial image of
a user (person to be authenticated) upon registration and
authentication, and comprises a CCD camera used to acquire a moving
image, still image, or the like, lighting equipment used to light
up an image sensing space of the user, and the like. An analog
image signal sensed by a CCD or CMOS camera is converted into a
digital signal by an analog/digital conversion unit such as an
image input board, and the digital image signal is stored in an
image memory. The image memory may be mounted on the image input
board, or may use an internal memory or external storage device of
a computer corresponding to an information management unit 9.
[0031] A facial region extraction unit 2 detects a facial image
region or head image region from an image of the person to be
authenticated, which is stored in the image memory. There are some
facial region extraction methods. For example, when a sensed image
is a color image, a method using color information is available.
More specifically, the sensed color image is converted from an RGB
color space which is specified by three components Red, Green, and
Blue into an HSV color space which specified by three components
Hue (color appearance, hue), Saturation (saturation), and Value
(lightness). The converted image is segmented by region
segmentation into a facial image region, head hair region, and the
like using color information such as hue, saturation, and the like.
Then, a facial region is detected from the segmented partial
regions using a region growing method or the like. In the region
growing method, a target figure (region) is extracted by combining
pixels having similar natures around an appropriately designated
pixel (start pixel) in turn (or by expanding a region of that
region to surrounding pixels). Details of the region growing method
are described in, e.g., Junichiro Toriwaki "Three-dimensional
Digital Image Processing", Jul. 5, 2002, Shokodo.
[0032] As another method of obtaining a facial region, a template
for facial detection, which is prepared in advance, is moved in an
image to calculate correlation values at respective positions. A
region with the highest correlation value is determined to be a
region with a high coincidence with the template, i.e., a facial
region in the image. In still another method, distances or
similarities may be calculated by an Eigenface method or subspace
method in place of the correlation values to extract a region with
the minimum distance or maximum similarity. In yet another method,
near infrared light may be projected in addition to the CCD camera
and a region corresponding to a face is extracted based on the
reflected light. The present invention can adopt any of the
aforementioned method or other methods.
[0033] A facial component detection unit 3 detects facial
components such as eyes, a nose, a mouth, and the like from the
image of the facial region. For example, the eye positions are
detected from the image of the facial region extracted by the
facial region extraction unit 2. As the detection method, a method
based on pattern matching as in the above extraction method, a
method described in a reference (Kazuhiro Fukui & Osamu
Yamaguchi, "Facial Feature Point Extraction by Combination of Shape
Extraction and Pattern Matching", IEICE Journal, Vol. J80-D-II, No.
8, pp. 2170-2177 (1997)), and the like may be used. In this
embodiment, any of the aforementioned method or other methods may
be used.
[0034] A feature amount extraction unit 4 extracts an image feature
amount required for personal authentication from an input image.
Based on the position of the facial region detected by the facial
region extraction unit 2 and those of the facial components
detected by the facial component detection unit 3, a region having
a given size and shape is clipped. Density information of the
clipped image is used as feature information. At least two
components are selected from the facial components detected by the
facial component detection unit 3. If a line segment that connects
these two components falls within the range of the facial region
extracted in advance at a given ratio, it is converted into an m
(pixels).times.n (pixels) region (m and n are integers equal to or
larger than 1), which is used as a normalized pattern.
[0035] FIGS. 2A through 2D show an extraction example when the two
eyes are selected as facial components in the facial region
extraction method of the personal authentication apparatus. In FIG.
2A, a black rectangle which indicates the position of the facial
region extracted by the facial region extraction means 2 is
superimposed on the facial image of a person to be authenticated,
which is sensed by the image sensing input means 1, and black cross
lines indicating the positions of the facial components (eyes,
nasal cavities, and mouth edges) extracted by the facial component
detection unit 3 are further superimposed. FIG. 2B illustrates an
image of the facial image. On such image of the facial region, let
V1 be a vector which connects from the right eye to the left eye,
and has a value corresponding to the distance from the right eye to
the left eye, C be the central point of that vector, and V2 be a
vector which is headed from C to the middle point between the two
nasal cavities, as shown in FIG. 2C. If the distances from central
point C of vector V1 to the respective components have a given
ratio, (e.g., the ratio between the sizes of V1 and V2 falls within
a predetermined range), it is determined that the region is a
facial region which includes the two eyes and nasal cavities.
Density pixel information is generated from that facial image, thus
obtaining density pixel matrix information of m pixels.times.n
pixels, as shown in FIG. 2D. The density pixel matrix information
pattern shown in FIG. 2D will be referred to as a normalized
pattern hereinafter.
[0036] In this normalized pattern, the density values of elements
(pixels) of an m (pixels).times.n (pixels) matrix line up, as shown
in the left figure of FIG. 3. When such matrix is converted into
vector expression, the matrix is expressed by an
(m.times.n)-dimensional vector, as shown in the right figure of
FIG. 3. This feature vector Nk (k indicates the number of
normalized patterns obtained for an identical person) is used in
the subsequent processes.
[0037] A feature amount used in person recognition is a subspace
obtained by lowering the number of data dimensions of an
orthonormal vector, which is obtained by calculating a correlation
matrix of feature vector Nk and then calculating an K-L expansion
of that matrix. Note that correlation matrix C is given by:
C = 1 r k = 1 r N k N k T ##EQU00001##
where r is the number of normalized patterns acquired for an
identical person. By diagonalizing C, principal components
(eigenvectors) are obtained. M out of these eigenvectors in
descending order of eigenvalue are used as a subspace. This
subspace is used as a personal authentication dictionary.
[0038] Referring back to FIG. 1, a dictionary registration unit 5
registers the feature amount extracted by the feature amount
extraction unit 4 together with index information such as the ID
number of the person of interest, spatial space (eigenvalues,
eigenvectors, the number of dimensions, the number of sample data),
and the like.
[0039] An authentication unit 6 compares the feature amount
registered in this dictionary and that extracted from the sensed
facial image, and collates their similarity. FIG. 4 is a flow chart
showing that process. When a person to be authenticated appears in
front of the personal authentication apparatus of the present
invention, an authentication procedure starts (step S1), and an
image of the person to be authenticated is sensed and input to the
authentication apparatus (step S2). A facial region is extracted
from the input image by the aforementioned method (step S3), and an
image feature amount required for personal authentication is
extracted from the extracted facial region of the input image (step
S4). In this way, an authentication data acquisition process is
repeated (steps S5 and S6) until a predetermined number of (n)
normalized patterns suited to verification are obtained. After the
predetermined number of (n) normalized patterns of the person to be
authenticated are obtained, pattern matching is made with a
dictionary facial image of the person to be authenticated, which is
registered in advance, by a mutual subspace method (step S7). If a
predetermined similarity is obtained, the person to be
authenticated is identified to be the person he or she claims to
be; otherwise, the person to be authenticated is identified not to
be the person he or she claims to be.
[0040] Note that the similarity is defined by distances and vector
angles make in an M-dimensional subspace specified by a feature
amount, as shown in FIG. 5. In FIG. 5, assume that data of person A
having a feature amount expressed by "pattern 1" and that of person
B having a feature amount expressed by "pattern 2" are registered
in an (N.times.N)-dimensional space. Data of a person to be
authenticated (a vector indicated by a black bold line in FIG. 5)
is input. The differences (distances) between a vector indicating
the data of the person to be authenticated and those expressed by
patterns 1 and 2, and the like are calculated. Let len1 be the
distance between the vector which indicates the data of the person
to be authenticated, and that expressed by pattern 1, len2 be the
distance between the vector which indicates the data of the person
to be authenticated, and that expressed by pattern 2, .theta.1 be
the angle the vector which indicates the data of the person to be
authenticated makes with the vector expressed by pattern 1, and
.theta.2 be the angle the vector indicating data of the person to
be authenticated makes with the vector expressed by pattern 2. As
can be seen from FIG. 5, since len1 is smaller than len2, and
.theta.1 is smaller than .theta.2, a similarity between the person
to be authenticated and person A is larger than that between the
person to be authenticated and person B. In this manner, upon
comparing the feature amount of the person to be authenticated with
those of persons A and B, it is determined that the feature vector
of the already registered person (person A and B) which has a
smaller distance from and makes a smaller angle with that of the
person to be authenticated has a higher similarity to that person,
and that person is identical to the person to be authenticated,
thus outputting a verification result.
[0041] When the facial image of a person to be authenticated is
registered in a dictionary, the person to be authenticated normally
inputs his or her ID number and stands at a position relatively
near the image sensing input unit 1 to sense his or her face and to
register the sensed image. By contrast, upon personal
authentication, if the need for inputting the ID number is
obviated, the person to be authenticated may undergo an
authentication process at a position which is not so near the
personal authentication apparatus. When the image sensing
conditions of the person to be authenticated are largely different
upon image registration and authentication, the feature amount of a
face used by the authentication unit 6 becomes considerably
different from that used in the dictionary registration unit 5 even
for an identical person, and the person cannot often be recognized
as a person he or she claims to be.
[0042] That is, when the standing position of the person to be
authenticated upon registration is largely different from that upon
authentication, the size of a person to be sensed and that of a
facial region of the person contained in the sensed image are
different. More specifically, an image sensed near the image
sensing means upon registration includes a relatively small facial
region. To prevent this, the size of the extracted facial region
can be controlled to fall within a given range.
[0043] Also, the irradiation condition of light coming through a
window largely varies depending on the hours (e.g., morning,
daytime, evening, and the like) of the day. Also, outside light
coming from the window also largely change depending on seasons. If
outside light is too strong, a facial image sensed under such
condition blurs by halation, and a facial region cannot be clipped
from such image. To prevent this problem, the average luminance
value of the extracted facial image can be controlled to fall
within a given range.
[0044] In order to solve the aforementioned problems, it is
effective to add an image sensing condition extraction unit 8 and
guide unit 7. The image sensing condition extraction unit 8
extracts image sensing conditions which include standing positions
upon registration and authentication and the like, and has a
function of checking if the size and the average luminance value of
the facial region extracted by the facial region extraction unit 2
fall within predetermined ranges. The guide unit 7 guides the
person to be authenticated in accordance with the extracted image
sensing conditions, so as to attain the same image sensing
conditions upon dictionary registration and authentication.
[0045] FIGS. 6A through 6C show an example of an interface used
upon registering a facial image of a person to be authenticated in
a dictionary when the person to be authenticated undergoes personal
authentication while he or she stands near the personal
authentication apparatus which is equipped in front of him or her.
When the person to be authenticated stands near the personal
authentication apparatus, as shown in FIG. 6A, the image of this
person is displayed on a monitor, and the image sensing condition
extraction means 8 calculates the size of a facial region to be
extracted from the personal image input from the image input means
and the average luminance value of pixels included in the facial
region. When the size of the facial region to be extracted becomes
larger than a predetermined size, or when the average luminance
value of pixels becomes larger than a predetermined threshold
value, it is determined that the person to be authenticated falls
within an image sensing range. Then, as shown in FIG. 6B, an
elliptic frame is superimposed near the facial region, and a bleep
tone is generated to inform the person to be authenticated of the
start of registration. In this case, the ID number and the like
required to register this person are input prior to registration
using a number input unit such as a ten-key pad and the like. Note
that messages such as "registration start" and the like for the
person to be authenticated are displayed on an upper portion of the
screen, so that the person to be authenticated can look up not to
hide his or her forehead with hair upon sensing a facial image (see
FIGS. 6B and 6C).
[0046] After a predetermined number of images required for
dictionary registration are acquired, bleep tones that inform the
person of the end of registration are produced, and a message that
advises accordingly is displayed, as shown in FIG. 6C. At that
time, normalized patterns are extracted from the predetermined
number of input images, an N-dimensional feature vector is
generated, and a subspace is calculated and is registered in the
dictionary by the dictionary registration unit 5.
[0047] Upon authenticating a person, as shown in FIGS. 7A through
7C, the person to be authenticated proceeds to an authentication
procedure while the image sensing condition extraction unit 8
monitors the facial image acquisition condition of the person. As
shown in FIG. 7A, when the person to be authenticated approaches
the personal authentication apparatus, the facial image acquisition
condition of the person to be authenticated is monitored (e.g., the
size of the facial region and the average luminance value of pixels
are checked) by the same operation as that upon registration, and
it is checked if the conditions such as the position, posture, and
the like of the person to be authenticated match those for image
sensing. The personal authentication procedure does not start
before the person to be authenticated reaches a facial image
sensing position. When the person to be authenticated has reached
the facial image sensing position, a bleep tone is generated, and
an elliptic frame is superimposed on the facial image of the person
to be authenticated displayed on the monitor screen together with a
message "authentication starts", thus starting the authentication
procedure. After a predetermined number of authentication images
are acquired, bleep tones are generated to inform the person of the
end of the authentication procedure, and a message of the
authentication result is displayed on the monitor screen.
[0048] FIG. 13 is a flow chart showing the aforementioned
dictionary registration process and personal authentication
procedure. A facial region is extracted (step S12) from an input
personal image of the person to be authenticated (step S11) using
color information, a template, or the like. Upon registration, an
image of a facial image with an image size equal to or larger than
a predetermined size is extracted (step S13). If the average
luminance value of the image of the facial region is equal to or
larger than a predetermined value (step S14), a normalized pattern
of the person to be authenticated is acquired until a predetermined
number of normalized patterns are acquired. After that, the
personal authentication procedure is completed by determining a
similarity between the acquired normalized patterns and the already
registered normalized patterns (step S20).
[0049] In the aforementioned registration/authentication procedure,
when the person to be authenticated moves away from the image
sensing apparatus or falls outside the image sensing range by
loosing his or her balance after the registration/authentication
procedure has started, image acquisition is canceled. In this case,
a normalized pattern for registration or authentication is not
calculated. After that, when the person to be authenticated meets
the facial image acquisition conditions again, and is ready to
acquire an image of the facial region, the
registration/authentication process is repeated until a
predetermined number of normalized patterns are generated.
[0050] In the above embodiment, since the image sensing means is
set to locate the face of the person to be authenticated at nearly
the center of the acquired image, the person to be authenticated is
located in front of the personal authentication apparatus and
approaches it. However, when the image sensing unit of the personal
authentication apparatus is set on a wall in the neighborhood of an
entrance, a facial image is acquired while the person to be
authenticated obliquely looks in the image sensing unit. Hence, the
standing position of the person to be authenticated may deviate not
only in a direction to and from the image sensing device but also
in the right-and-left directions. FIGS. 8A and 8B and FIGS. 9A and
9B show a case wherein a normalized pattern can be extracted only
when the facial image is located at the central position of the
acquired image, so as to match the image sensing conditions of the
person to be authenticated in such environment.
[0051] In FIG. 8A, when the facial image of the person to be
authenticated is displayed on the monitor screen, a bleep tone is
generated to inform the person of the start of the registration
procedure, and a message that advises accordingly is displayed on
the monitor. In addition, the facial image of the person to be
authenticated, which is sensed by the image sensing unit is
displayed on the monitor, and a message and cross mark which guide
the position and posture of the person to be authenticated to
locate the facial image at the center of the screen are displayed.
When the facial image of the person to be authenticated is captured
at the center of the screen, acquisition of the facial image,
extraction of a feature amount, and registration of an
authentication image start. After a predetermined number of data
are acquired, bleep tones are generated, and a message that informs
the person of the end of registration is displayed at the same time
(see FIG. 8B).
[0052] Upon authenticating a person, as in registration, when the
person to be authenticated approaches the personal authentication
apparatus, the position and posture of the person to be
authenticated are guided so that the facial image of the person to
be authenticated is located at the center of the sensed image (see
FIG. 9A). When the facial image of the person to be authenticated
is captured at the center of the screen, an authentication image is
acquired, and the authentication procedure starts. After a
similarity with registered data is determined, an audible message
is generated, and the authentication result is displayed, thus
ending the authentication procedure (see FIG. 9B).
[0053] The second embodiment of the present invention will be
described below.
[0054] In the first embodiment, the facial image of the person to
be authenticated, which is sensed upon dictionary registration or
personal authentication, is displayed on the monitor, and the
person to be authenticated is guided based on the displayed
contents. However, many users may be strained when their facial
images are displayed on the monitor in practice. Especially, since
many users are strained upon dictionary registration, the facial
expression upon personal authentication becomes different from that
upon dictionary registration, and authentication often fails. Also,
when the facial expression changes largely, since the mouth and eye
positions apparently change, the feature vector changes, and
authentication often fails. On the other hand, a shadow is often
cast on a face due to the influences of hair style of the person to
be authenticated and illumination, and the pixel values of the
obtained image change largely due to the influence of illumination
and shadow, thus impairing the authentication precision.
[0055] In order to solve such problems, by registering the facial
image while moving the face of the person to be authenticated upon
registration and authentication, the authentication precision can
be improved. The second embodiment is an invention which is made to
solve the above problems. More specifically, the guide unit 7
displays a character on the monitor in place of the facial image of
the person to be authenticated, thereby guiding the person to be
authenticated.
[0056] FIGS. 10A through 10E are views for explaining the guide
sequence of the guide unit 7 upon dictionary registration. In FIG.
10A, when the dictionary registration procedure starts together
with generation of a bleep tone, a message and character used to
guide the motion of the face of the person to be authenticated are
displayed on the monitor together with a message that indicates the
start of registration. More specifically, the character is
displayed to make a round clockwise along the circumference of the
display region of the monitor. The person to be authenticated moves
his or her face to follow the movement of the character (FIGS. 10B
through 10D). During this interval, the personal authentication
apparatus senses different facial images of the person to be
authenticated, who looks up, down, and right, and left, extracts
feature amounts, and generates normalized patterns. Upon completion
of generation of a predetermined number of normalized patterns,
which are to be registered in a dictionary, a message "end of
registration" and a character are displayed, thus ending the
registration procedure.
[0057] In this embodiment, the image sensing condition extraction
unit 8 calculates an image sensing range on the basis of the size
and the average luminance value of the facial region as in the
first embodiment. That is, when the person to be authenticated
approaches the image sensing range, the image sensing range is
calculated. If it is determined that a facial image can be sensed,
a character created by computer graphics (CG) or the like is
displayed on the monitor in place of the facial image of the person
and the elliptic frame. The character may move about in the screen
until n normalized patterns are acquired in place of making a round
along the circumference of the screen. In this case, when the
apparatus guides the person to follow the motion of the character
by moving not only eyes but also the face, facial images free from
any nonuniformity against a change in illumination can be acquired.
Furthermore, when a bowing character is displayed upon completion
of the registration procedure, it can relax the person to be
authenticated, and facial image data of the person to be
authenticated can be acquired in a relatively relaxed state.
[0058] FIGS. 11A through 11E are views for explaining the guide
procedure of the guide unit upon personal authentication. The basic
procedure is the same as that upon dictionary registration. In FIG.
11A, when a personal authentication procedure starts, a character
makes a round along the circumference of the monitor to guide the
movement of the face of the person to be authenticated (FIGS. 11B
through 11D). After a predetermined number of normalized patterns
are acquired and an authentication result is obtained, a message
"end of authentication" and a character are displayed (FIG.
11E).
[0059] Upon completion of personal authentication, the number of
times the person to be authenticated has passed the door may be
presented, thus providing information that attracts the interest of
the person to be authenticated. When the information that attracts
the interest of the person to be authenticated is presented, the
face of the person to be authenticated can be closer to the
authentication apparatus, thereby further improving the
authentication precision. Alternatively, as shown in FIGS. 12A
through 12E, a plurality of different characters to be randomly
displayed may be prepared, and are used daily or randomly, thus
attracting the interest of the person to be authenticated. Also, as
shown in FIG. 12E, fortune-telling using a similarity upon
authentication (a better fortune-telling result can be obtained
with increasing similarity) may be displayed to attract the
interest of the person to be authenticated. With this arrangement,
since the person to be authenticated stands at a position near the
authentication apparatus and approaches his or her face to the
monitor screen to look into it, the image sensing conditions upon
registration and authentication can become stable, and a
predetermined number of normalized patterns can be acquired easily,
thus improving the authentication precision.
[0060] Additional advantages and modifications will readily occur
to those skilled in the art. Therefore, the invention in its
broader aspects is not limited to the specific details and
representative embodiments shown and described herein. Accordingly,
various modifications may be made without departing from the spirit
or scope of the general inventive concept as defined by the
appended claims and their equivalents.
* * * * *