Iris authentication apparatus

The present invention relates to an iris authentication apparatus which authenticates an individual from his/her iris which adjusts the amount of light entering the eye.

Conventionally, as an iris authentication apparatus of this type, an iris authentication apparatus like that disclosed in WO 94/09446 (reference 1) has been used. This iris authentication apparatus cuts an iris portion in the form of a plurality of arcuated portions (see FIG. 23: arcs SO) from a sensed eye image (eyeball image), and extracts the features of an iris pattern by using a bandpass filter such as a Gabor filter, thereby generating a one-dimensional code (iris code) for each arc SO. Such iris codes are generated for a registration eyeball image and a collation eyeball image, respectively. The Hamming distance between the iris codes is obtained, and the authentic person is discriminated from others on the basis of the Hamming distance.

Note that the iris is a diaphragm which is located between the cornea and eye lens of the eyeball, has a pupil in the center, and adjusts the amount of light entering the eyeball. The sclera (the white of the eye) is located outside the iris. A pattern is formed in the iris after about seven or eight months of pregnancy and becomes stable about two years after birth. The pattern is kept unchanged throughout lifetime. The left and right eyes of even the same person have different iris patterns, which are hardly generically influenced. In addition, since the iris is an internal organ, the iris is robust against damage and allows authentication in a noncontact manner.

According to the iris authentication apparatus disclosed in reference 1 described above, however, since an iris portion is cut out in the form of a plurality of arcuated portions from an eyeball image and processing is performed for each arc SO, if a cutting error concerning arcuated areas (an error in the radial direction, in particular) occurs, the iris position where feature extraction should be performed by a Gabor filter varies depending on whether a registration eyeball image or a collation eyeball image is to be processed. As a consequence, accurate authentication cannot be performed.

It is an object of the present invention to provide an iris authentication apparatus which can accurately perform authentication even if a slight error occurs in cutting out areas from an eyeball image.

In order to achieve the above object, the present invention comprises an iris area extraction means for extracting an iris area from a sensed eyeball image, a registration pattern generating means for generating a registration iris pattern image by performing polar coordinate transformation of an image of the iris area extracted from a registration eyeball image, a collation pattern generating means for generating a collation iris pattern image by performing polar coordinate transformation of an image of the iris area extracted from a collation eyeball image, and a collation means for collating the registration iris pattern image with the collation iris pattern image on the basis of the correlation between them.

According to the present invention, an image of an iris area extracted from a registration eyeball image is polar-coordinate-transformed into a registration iris pattern image (normalized image), an image of an iris area extracted from a collation eyeball image is polar-coordinate-transformed into a collation iris pattern image (normalized image), and the registration iris pattern image and the collation iris pattern image are collated with each other on the basis of the correlation between them.

In the present invention, a cutting error concerning an iris area may appear as translation, enlargement/reduction, distortion, or the like in an image after polar coordinate transformation (normalized image). However, the cutting error concerning the iris area is very small as compared with the entire area. Therefore, the cutting error concerning the iris area has little influence on a registration iris pattern image and collation iris pattern image after polar coordinate transformation, and accurate authentication can be performed.

According to the present invention, the iris area extraction means detects, for example, the boundary between the pupil and iris of an eyeball image as the inner boundary of the iris by elliptic approximation, detects the boundary between the iris and sclera of the eyeball image as the outer boundary of the iris by circular approximation, and extracts an area sandwiched between the detected inner and outer boundaries of the iris as an iris area.

Actually sensed eyeball images vary depending on the biometric features of individuals and imaging environments, and hence the boundary between the pupil and the iris (the inner boundary of the iris) is not necessarily a perfect circle. If the boundary between the pupil and the iris is detected on the assumption that the boundary is a perfect circle, correct authentication may not be performed. In the present invention, since the boundary between the pupil and the iris is detected on the assumption that the boundary is an ellipse, even if the boundary is not a perfect circle, correct authentication can be performed, thereby improving authentication accuracy.

Note that the boundary between the iris and the sclera (the outer boundary of the iris) often becomes a circle. In this case, either circular approximation or elliptic approximation may be used.

According to the present invention, the registration iris pattern image generating means performs polar coordinate transformation of an image of an iris area extracted from a registration eyeball image with reference to the center of the inner boundary of the iris detected from the registration eyeball image, and the collation iris pattern image generating means performs polar coordinate transformation of an image of an iris area extracted from a collation eyeball image with reference to the center of the inner boundary of the iris detected from the collation eyeball image. In the present invention, the center of the ellipse which is detected from the inner boundary of the iris indicates the accurate center of the pupil. In the present invention, polar coordinate transformation of an image of an iris area is performed with reference to this accurate center of the pupil.

According to the present invention, the collation means cuts out one or more rectangular areas from a registration iris pattern image and a collation iris pattern image so as not to include any non-iris portion, obtains a correlation value between rectangular areas common to the registration iris pattern image and the collation iris pattern image, and collates the registration eyeball image and the collation eyeball image on the basis of the correlation value.

In a sensed eyeball image, an image of an iris area may include an eyelid or eyelash or reflection of light may glare on the image. In this case, such images appear as non-iris portions in the registration iris pattern image and collation iris pattern image generated by polar coordinate transformation.

In the present invention, one or more rectangular areas are cut out from a registration iris pattern image and a collation iris pattern image so as not to include such non-iris portions. The correlation value between rectangular areas (common areas) common to the registration iris pattern image and the collation iris pattern image is obtained, and the two images are collated on the basis of the correlation value.

If there are few non-iris portions, since large areas can be ensured as a registration iris pattern image and a collation iris pattern image, the number of rectangular areas may be one. If there are many non-iris portions, large areas may not be ensured as a registration iris pattern image and a collation iris pattern image. In such a case, for example, widths may be determined stepwise from the upper end of an iris pattern image, and rectangular areas may be cut out for the respective widths. The correlation value between areas common to these rectangular areas may be obtained.

FIG. 1 is a block diagram showing the arrangement of an iris authentication apparatus according to an embodiment of the present invention;

FIG. 2 is a flowchart for explaining iris registering operation in the iris authentication apparatus shown in FIG. 1 ;

FIGS. 3A to 3D are views showing registration original images to explain iris registering operation in the iris authentication apparatus shown in FIG. 1 ;

FIG. 4 is a flowchart for explaining how a pupil center is estimated in the iris authentication apparatus shown in FIG. 1 ;

FIG. 5 is a flowchart for explaining boundary detection by elliptic approximation of the pupil and the iris in the iris authentication apparatus shown in FIG. 1 ;

FIG. 6 is a flowchart for explaining boundary detection by circular approximation of the iris and the sclera in the iris authentication apparatus shown in FIG. 1 ;

FIGS. 7A and 7B are views for explaining the normalization (polar coordinate transformation) of an iris area in the iris authentication apparatus shown in FIG. 1 ;

FIGS. 8A and 8B are views showing photographs on a display which indicate a registration original image and an iris pattern image after coordinate transformation (normalized image);

FIG. 9 is a flowchart for explaining boundary detection by elliptic approximation of the iris and eyelid in the iris authentication apparatus shown in FIG. 1 ;

FIGS. 10A to 10E are flowcharts for explaining iris collating operation in the iris authentication apparatus shown in FIG. 1 ;

FIGS. 11A to 11C are views showing photographs on the display which respectively indicate an iris pattern image, an expanded iris pattern image, and a registration iris pattern image to explain boundary detection by elliptic approximation of the iris and eyelid;

FIGS. 12A and 12B are views showing photographs on the display which respectively indicate a collation original image and an iris pattern image after coordinate transformation (normalized image);

FIGS. 13A to 13D are views showing photographs on the display which respectively indicate a registration iris pattern image and a collation iris pattern image before and after unnecessary areas are cut off;

FIGS. 14A to 14D are views for explaining displacement correction for a registration iris pattern image and a collation iris pattern image from which unnecessary areas are cut off;

FIGS. 15A and 15B are views showing photographs on the display which respectively indicate effective frequency bands in images of frequency domains of a registration image and collation image when they are collated with each other by BLPOC;

FIG. 16 is a view showing a photograph on the display which indicates a case wherein segmented images are cut out from a registration iris pattern image;

FIG. 17 is a view showing a photograph on the display which indicates a case wherein segmented images are cut out from a collation iris pattern image;

FIG. 18 is a functional block diagram of a CPU in FIG. 1 ; and

FIG. 19 is a view showing a graph on a display which indicates arcuated areas cut out from the iris portion of an eyeball image in a conventional iris authentication apparatus.

FIG. 1 shows an iris authentication apparatus according to an embodiment of the present invention. Referring to FIG. 1 , reference numeral 10 denotes an operation unit; and 20 , a control unit. The operation unit 10 includes a ten-key pad 10 - 1 , display (LCD) 10 - 2 , camera (CCD camera) 10 - 3 , and the like.

The control unit 20 comprises a CPU 20 - 1 , ROM 20 - 2 , RAM 20 - 3 , hard disk (HD) 20 - 4 , frame memory (FM) 20 - 5 , and external connection unit (I/F) 20 - 6 . A registration program and collation program are stored as programs unique to this embodiment in the ROM 20 - 2 .

[Registration of Iris]

An iris of a user is registered in this iris authentication apparatus in the following manner. Before operation, the user inputs an ID number assigned to him/her by using the ten-key pad 10 - 1 ( FIG. 2 : step S 101 ), and stands in front of the camera 10 - 3 . The camera 10 - 3 senses an eyeball image of the user, and sends the sensed eyeball image as a registration original image F(n 1 , n 2 ) (see FIG. 3A ) to the control unit 20 . Note that in the registration original image F(n 1 , n 2 ), reference numeral 1 denotes a pupil; 2 , an iris; and 3 , a sclera.

[Estimation of Pupil Center]

The CPU 20 - 1 of the control unit 20 loads the registration original image F(n 1 , n 2 ) from the operation unit 10 through the frame memory 20 - 5 (step S 102 ), and estimates a pupil center (Cp 1 , Cp 2 ) from the loaded registration original image F(n 1 , n 2 ) (step S 103 ). The pupil center (Cp 1 , Cp 2 ) is estimated as follows:

(1) The histogram of the registration original image F(n 1 , n 2 ) is obtained ( FIG. 4 : step S 201 ). (2) The registration original image F(n 1 , n 2 ) is binarized by using the maximum luminance value of the histogram as a threshold (step S 202 ). In this case, the luminance values of pixels having luminance values equal to or less than the threshold are set to “0”, and the luminance values of other pixels are set to “1”. This processing makes it possible to obtain an image formed by extracting almost only the pupil from the registration original image F(n 1 , n 2 ).

In this case, the maximum luminance value of the histogram is set as a threshold. However, an image may be binarized by using a predetermined threshold. In this case, the value of a threshold varies depending on the database to be used, a threshold is experimentally

determined by using several images. (3) Projections of the binarized image in the n 1 direction (horizontal direction) and the n 2 direction (vertical direction) are obtained (step S 203 ). (4) The center of a portion where the values of the projections in the n 1 and n 2 directions become small is obtained from the projections and is set as a pupil center (Cp 1 , Cp 2 ) (step S 204 ).

In this case, the pupil center (Cp 1 , Cp 2 ) is obtained from the projections in the n 1 and n 2 directions. However, the center of gravity of the image binarized in step S 202 may be calculated and set as a pupil center (Cp 1 , Cp 2 ).

[Boundary Detection by Elliptic Approximation of Pupil and Iris]

The CPU 20 - 1 detects the boundary between the pupil and the iris (the inner boundary of the iris: indicated by S 1 in FIG. 3A ) is detected from the loaded registration original image F(n 1 , n 2 ) (step S 104 ).

The loaded registration original image F(n 1 , n 2 ) varies depending on the biometric features of the individual and imaging environments, and hence the inner boundary S 1 of the iris is not necessarily a perfect circle (it is more often the case that the inner boundary is not a perfect circle). If the inner boundary S 1 of the iris is detected on the assumption that the boundary is a perfect circle, correct authentication may not be performed.

In this embodiment, therefore, the inner boundary S 1 of the iris is detected by elliptic approximation on the assumption that the boundary is an ellipse. This makes it possible to perform correct authentication and improve the authentication accuracy even if the boundary between the pupil and the iris is not a perfect circle. The inner boundary S 1 of the iris is detected by elliptic approximation in the following manner.

(1) The inner boundary S 1 of the iris is assumed to be an ellipse, and the center coordinates of the boundary, the length of the elliptical axis in the n 1 direction, and the elliptical axis in the n 2 direction are respectively represented by (Cp 1 , Cp 2 ), l 1 , and l 2 . The center coordinates (Cp 1 , Cp 2 ), the length l 1 of the elliptical axis in the n 1 direction, and the length l 2 of the ellipse in the n 2 direction are initialized ( FIG. 5 : step S 301 ). In this case, as the center coordinates (Cp 1 , Cp 2 ), the center (Cp 1 , Cp 2 ) of the pupil estimated in step S 103 is set as an initial value. As the length l 1 of the elliptical axis in the n 1 direction and the length l 2 of the elliptical axis in the n 2 direction, predetermined values are set as initial values. (2) The line integral of the luminance values of N pixels on the circumference of the ellipse determined by Cp 1 , Cp 2 , l 1 , and l 2 in the registration original image F(n 1 , n 2 ) is obtained (step S 302 ). (3) One of the values Cp 1 , Cp 2 , l 1 , and l 2 is changed so as to generate all combinations thereof by “moving Cp 1 by m 2 pixels within the range of ±m 1 pixels”, “moving Cp 2 by m 2 pixels within the range of ±m 1 pixels”, “increasing l 1 by k 1 within the range from l 1 min to l 1 max”, or “increasing l 2 by k 2 within the range from l 2 min to l 2 max” (step S 303 ). (4) The line integral of the luminance values of N pixels on the circumference of the ellipse determined by Cp 1 , Cp 2 , l 1 , and l 2 after the above change in the registration original image F(n 1 , n 2 ) is obtained (step S 304 ). (5) The difference between the line integral before change which is obtained in step S 302 and the line integral after change which is obtained in step S 304 is obtained (step S 305 ). (6) It is checked whether the differences between line integrals before change and line integrals after change are obtained with respect to all the combinations of Cp 1 , Cp 2 , l 1 , and l 2 (step S 306 ). (7) By repeating steps S 303 to S 306 , the differences between the line integrals before change and the line integrals after change are obtained with respect to all the combinations of Cp 1 , Cp 2 , l 1 , and l 2 (see FIG. 3B ). (8) Among all the combinations, Cp 1 , Cp 2 , l 1 , and l 2 with which the line integral difference is maximized are obtained (step S 307 ). The ellipse determined by the obtained values Cp 1 , Cp 2 , l 1 , and l 2 is set as the inner boundary S 1 of the iris.

Assume that in this detection of the inner boundary S 1 of the iris, N=128, m 1 =3, m 2 =1, l 1 min =30, l 1 max=70, k 1 =1, l 2 min=l1−8, l 2 max=70, and k 2 =1.

In this detection of the inner boundary S 1 of the iris, more accurate boundary detection can be performed by regarding the inner boundary S 1 of the iris as an affine transformation of a circle and taking rotation into consideration as well instead of simply assuming the boundary as an ellipse. That is, elliptic approximation may be performed by using five parameters, i.e., adding a rotation angle to the center coordinates in the n 1 direction, the center coordinates in the n 2 direction, the length of the elliptical axis in the n 1 direction, and the length of the elliptical axis in the n 2 direction, and using affine transformation or the like.

[Boundary Detection by Circular Approximation of Iris and Sclera]

The CPU 20 - 1 detects the boundary between the iris and the sclera (the outer boundary of the iris: indicated by S 2 in FIG. 3A ) from the loaded registration original image F(n 1 , n 2 ) (step S 105 ). In many case, the outer boundary S 2 of the iris becomes circular (close to a perfect circle). In this embodiment, therefore, the outer boundary S 2 of the iris is detected by circular approximation on the assumption that the outer boundary S 2 of the iris is a circle. The outer boundary S 2 of the iris is detected by circular approximation in the following manner. Note that the outer boundary S 2 of the iris may be detected by elliptic approximation like the inner boundary S 1 .

(1) The outer boundary S 2 of the iris is assumed to be a circle, and the center coordinates and radius of the circle are respectively represented by (Ci 1 , Ci 2 ) and r. The center coordinates (Ci 1 , Ci 2 ) and the radius r are initialized ( FIG. 6 : step S 401 ).

In this case, as the center coordinates (Ci 1 , Ci 2 ), the center (Cp 1 , Cp 2 ) of the inner boundary S 1 of the iris obtained in step S 104 is set as an initial value. As the radius r, a predetermined value is set as an initial value.

(2) The line integral of the luminance values of N pixels (left: N/2 pixels, right: N/2 pixels) on left and right arcs (3π/8 [rad]) of the circle determined by Ci 1 , Ci 2 , and r in the registration original image F(n 1 , n 2 ) is obtained (step S 402 ). (3) One of the values Ci 1 , Ci 2 , and r is changed so as to generate all combinations thereof by “moving Ci 1 by m 2 pixels within the range of ±m 1 pixels”, “moving Ci 2 by m 2 pixels within the range of ±m 1 pixels”, or “increasing r by k within the range from rmin to rmax” (step S 403 ). (4) The line integral of the luminance values of N pixels (left: N/2 pixels, right: N/2 pixels) on the left and right arcs (3π/8 [rad]) of the circle determined by Ci 1 , Ci 2 , and r after the above change in the registration original image F(n 1 , n 2 ) is obtained (step S 404 ). (5) The difference between the line integral before change which is obtained in step S 402 and the line integral after change which is obtained in step S 404 is obtained (step S 405 ). (6) It is checked whether the differences between line integrals before change and line integrals after change are obtained with respect to all the combinations of Ci 1 , Ci 2 , and r (step S 406 ). (7) By repeating steps S 403 to S 406 , the differences between the line integrals before change and the line integrals after change are obtained with respect to all the combinations of Ci 1 , Ci 2 , and r (see FIG. 3C ). (8) Among all the combinations, Ci 1 , Ci 2 , and r with which the line integral difference is maximized are obtained (step S 407 ). The circle determined by the obtained values Ci 1 , Ci 2 , and r is set as the outer boundary S 2 of the iris.

Assume that in this detection of the outer boundary S 2 of the iris, N=100, m 1 =5, m 2 =1, rmin=l+10, rmax=l+100, and k=1. Note, however, that l is set to a larger one of the values of l 1 and l 2 which are obtained in detection of the inner boundary S 1 of the iris.

[Extraction of Iris Area]

The CPU 20 - 1 then extracts an area S 3 (see FIG. 3D ), as an iris area, which is sandwiched between the inner boundary S 1 of the iris which is detected in step S 104 and the outer boundary S 2 of the iris which is detected in step S 105 (step S 106 ).

[Normalization (Polar Coordinate Transformation) of Iris Area]

The radius of the pupil changes with a change in brightness. For this reason, even the iris pattern of the same person changes depending on the timing of imaging. The size of an iris image changes with a change in the distance between the camera and the object.

In this embodiment, a change in iris pattern is regarded as enlargement/reduction in the radial direction, and the image of the iris area S 3 is normalized by being polar-coordinate-transformed onto a plane of p 1 ×p 2 (step S 107 ).

Note that only the lower half (semicircle) of the iris area S 3 is polar-coordinate-transformed, instead of the entire iris area S 3 , with reference to the center (Cp 1 , Cp 2 ) of the inner boundary S 1 of the iris, i.e., the center (Cp 1 , Cp 2 ) of the pupil, in order to avoid the influences of the eyelid and eyelash.

Polar coordinate transformation in this case will be described with reference to FIGS. 7A and 7B . FIG. 7A shows an image of the iris area S 3 before polar coordinate transformation. FIG. 7B shows an image (normalized image: to be referred to as an iris pattern image hereinafter) of the iris area S 3 after polar coordinate transformation. In the image before polar coordinate transformation, the lower half of the iris area S 3 with the center (Cp 1 , Cp 2 ) of the pupil being the origin is segmented into 256 parts in the circumferential direction (θ direction), and segmented into 128 parts in the radial direction (R direction). The luminances of the respective pixels obtained by this segmentation are plotted along an abscissa θ and an ordinate R to obtain an iris pattern image f(p 1 , p 2 ).

In this embodiment, the center (Cp 1 , Cp 2 ) of the pupil is accurately obtained as the center (Cp 1 , Cp 2 ) of the inner boundary S 1 of the iris, and the image of the iris area S 3 is polar-coordinate-transformed with reference to the accurate center (Cp 1 , Cp 2 ) of the pupil. Therefore, accurate authentication can be performed.

[Boundary Detection by Elliptic Approximation of Iris and Eyelid and Masking for Eyelid Area]

In this embodiment, since the lower half of the iris area S 3 is used for collation, the influences of the upper eyelid and upper eyelash can be avoided. If, however, the lower eyelid overlaps the iris area S 3 , accurate collation may not be performed due to the influence of the lower eyelid.

FIG. 8A shows the registration original image F(n 1 , n 2 ) when the lower eyelid overlaps the iris area S 3 . FIG. 8B shows the iris pattern image f(p 1 , p 2 ) when the lower eyelid overlaps the iris area S 3 . In this case, an eyelid area S 4 appears in the iris pattern image f(p 1 , p 2 ). In the eyelid area S 4 , the boundary (indicated by S 5 in FIG. 8B ) with the iris can be regarded as part of an ellipse.

In this embodiment, therefore, a boundary S 5 (to be referred to as an eyelid boundary hereinafter) between the eyelid area S 4 and the iris is assumed to be an ellipse, and is detected by elliptic approximation (step S 108 ). The eyelid area S 4 is specified by the eyelid boundary S 5 , and the eyelid area S 4 is masked (step S 109 ). The eyelid boundary S 5 is detected by elliptic approximation, and the eyelid area S 4 is masked in the following manner.

(1) The iris pattern image f(p 1 , p 2 ) is enlarged twice ( FIG. 9 : step S 501 ; see FIGS. 11A and 11B ). (2) The lower end of the expanded iris pattern image f(p 1 , p 2 ) is set to a center coordinate Cq 2 in the p 2 direction, and a center coordinate Cq 1 in the p 1 direction, a length l 1 of the elliptical axis in the p 1 direction, and a length l 2 of the elliptical axis in the p 2 direction are initialized (step S 502 ). In this case, as the center coordinate Cq 1 , the center of the expanded iris pattern image f(p 1 , p 2 ) in the p 1 direction is set as an initial value. As the lengths l 1 and l 2 of the ellipse axes in the p 1 and p 2 directions, predetermined values are set as initial values, respectively. (3) The line integral of the luminance values of N pixels on the upper arc (3π/4 [rad]) of the ellipse determined by Cq 1 , l 1 , and l 2 in the expanded iris pattern image f(p 1 , p 2 ) is obtained (step S 503 ). (4) One of the values Cq 1 , l 1 , and l 2 is changed so as to generate all combinations thereof by “moving Cq 1 by m 2 pixels within the range of ±m 1 pixels”, “increasing l 1 by k 1 within the range from l 1 min to l1 max”, or “increasing l 2 by k 2 within the range from l 2 min to l 2 max” (step S 504 ). (5) The line integral of the luminance values of N pixels on the upper arc (3π/4 [rad]) of the ellipse determined by Cq 1 , l 1 , and l 2 after the above change in the expanded iris pattern image f(p 1 , p 2 ) is obtained (step S 505 ). (6) The difference between the line integral before change which is obtained in step S 503 and the line integral after change which is obtained in step S 505 is obtained (step S 506 ). (7) It is checked whether the differences between line integrals before change and line integrals after change are obtained with respect to all the combinations of Cq 1 , l 1 , and l 2 (step S 507 ). (8) By repeating steps S 504 to S 507 , the differences between the line integrals before change and the line integrals after change are obtained with respect to all the combinations of Cq 1 , l 1 , and l 2 . (9) Among all the combinations, Cq 1 , l 1 , and l 2 with which the line integral difference is maximized are obtained (step S 508 ). (10) It is checked whether the maximum line integral difference obtained in step S 508 is equal to or more than a predetermined value d (step S 509 ). (11) If the maximum line integral difference is not equal to or more than the predetermined value d (NO in step S 509 ), it is determined that no eyelid area exists (step S 510 ), and the iris pattern image f(p 1 , p 2 ) before expansion is set as a registration iris pattern image (step S 511 ). (12) If the maximum line integral difference is equal to or more than the predetermined value d (YES in step S 509 ), it is determined that an eyelid area exist. In this case, the elliptic arc determined by Cq 1 , l 1 , and l 2 obtained in step S 508 is set as the eyelid boundary S 5 , and the area surrounded by the eyelid boundary S 5 and the lower end of the iris pattern image f(p 1 , p 2 ) before expansion is detected as the eyelid area S 4 . Masking is then executed for the detected eyelid area S 4 (step S 512 ; see FIG. 10C ). In this case, all the luminance values of the pixels in the detected eyelid area S 4 are set to “0”, and the iris pattern image f(p 1 , p 2 ) before expansion, which is obtained by masking the eyelid area S 4 , is set as a registration iris pattern image (step S 513 ; see FIG. 11C ).

In this detection of the eyelid boundary S 5 , for example, N=128, m 1 =50, m 2 = 3 , l 1 min=50, l 1 max=150, k 1 =3, l 2 min=N2/4, l 2 max=3·N2/4, k 2 =3, and d=2700. Assume that N 1 and N 2 represent the image size (N 1 ×N 2 ) of the iris pattern image f(p 1 , p 2 ).

[Registration of Registration Iris Pattern Image]

The CPU 20 - 1 extracts the iris area S 3 from the registration original image F(n 1 , n 2 ), performs polar coordinate transformation of the image of the extracted iris area S 3 , and executes masking, as needed. The CPU 20 - 1 then stores the registration iris pattern image f(p 1 , p 2 ) obtained in this manner in the hard disk 20 - 4 in correspondence with the ID number input in step S 101 (step S 110 ).

[Iris Collation]

In this iris authentication apparatus, iris collation is performed as follows. During operation, the user inputs the ID number assigned to him/her by using the ten-key pad 10 - 1 (step S 601 in FIG. 10A ) and stands in front of the camera 10 - 3 . The camera 10 - 3 senses an eyeball image of the user and sends the sensed eyeball image as a collation original image G(n 1 , n 2 ) (see FIG. 12A ) to the control unit 20 .

[Generation of Collation Iris Pattern Image]

The CPU 20 - 1 of the control unit 20 loads the collation original image G(n 1 , n 2 ) from the operation unit 10 through the frame memory 20 - 5 (step S 602 ) and performs “estimation of pupil center” (step S 603 ), “boundary detection by elliptic approximation of pupil and iris” (step S 604 ), “boundary detection by circular approximation of iris and sclera” (step S 605 ), “extraction of iris area” (step S 606 ), “normalization (polar coordinate transformation) of iris area” (step S 607 ), “boundary detection by elliptic approximation of iris and eyelid” (step S 608 ), and “masking for eyelid area” (step S 609 ) with respect to the loaded collation original image G(n 1 , n 2 ) in the same manner at the time of iris registration to obtain the collation iris pattern image g(p 1 , p 2 ) (see FIG. 12B ).

[Read of Registration Iris Pattern Image]

On the basis of the ID number input in step S 601 , the CPU 20 - 1 reads out the registration iris pattern image f(p 1 , p 2 ) stored in the hard disk 20 - 4 in correspondence with the ID number (step S 610 ). Note that in this embodiment, since a 1:1 collation scheme is used, it is necessary to input an ID number. According to a 1:N collation scheme (a scheme of performing collation with all registration iris pattern images), there is no need to input an ID number.

[Check on Unnecessary Area]

Reflection of light sometimes glares on the registration iris pattern image f(p 1 , p 2 ) or collation iris pattern image g(p 1 , p 2 ). An eyelid area by which light is reflected or which is masked needs to be cut off as an unnecessary area because it hinders correct collation. The CPU 20 - 1 cuts off an eyelid area by which light is reflected or which is masked from the registration iris pattern image f(p 1 , p 2 ) or collation iris pattern image g(p 1 , p 2 ) in the following manner.

First of all, the reflection of light on the registration iris pattern image f(p 1 , p 2 ) and collation iris pattern image g(p 1 , p 2 ) is discriminated on the basis of luminance values ( FIG. 10B : step S 611 ).

Widths Wf and Wg of unnecessary areas are obtained on the registration iris pattern image f(p 1 , p 2 ) and collation iris pattern image g(p 1 , p 2 ) such that a masked eyelid area and reflection of light are not included (step S 612 ; see FIGS. 13A and 13B ). A larger one of Wf and Wg is set as a width w of an unnecessary area to be cut off (step S 613 ).

It is then checked whether the width w of this unnecessary area is larger than a predetermined threshold wm (step S 614 ). If the width w is equal to or less than wm (NO in step S 614 ), it is determined that a wide area can be ensured as a common area, and the flow advances to step S 615 . If the width w of the unnecessary area is larger than wm (YES in step S 614 ), it is determined that a wide area cannot be ensured as a common area, and the flow advances to processing in step S 627 and subsequent steps ( FIG. 10C ). The common area will be described later.

[Collation between Registration Iris Pattern Image and Collation Iris Pattern Image]

If the width w of the unnecessary area is equal to or less than wm (NO in step S 614 ), i.e., it is determined that a wide area can be ensured as a common area used for collation, the CPU 20 - 1 immediately starts collation between the registration iris pattern image f(p 1 , p 2 ) and the collation iris pattern image g(p 1 , p 2 ). This collation is performed in the following manner.

[Cutting-Off of Unnecessary Area]

In collating the registration iris pattern image f(p 1 , p 2 ) with the collation iris pattern image g(p 1 , p 2 ), the CPU 20 - 1 cuts off images with the width w of the unnecessary area from the registration iris pattern image f(p 1 , p 2 ) and the collation iris pattern image g(p 1 , p 2 ), thereby obtaining a registration iris pattern image fc(p 1 , p 2 ) and a collation iris pattern image gc(p 1 , p 2 ) (step S 615 ; see FIGS. 13C and 13D ).

[Displacement Correction]

A displacement may occur between the registration iris pattern image f(p 1 , p 2 ) and the collation iris pattern image g(p 1 , p 2 ) due to, for example, the rotation of the head of the person or the motion of the eyeball even if the images obtained by imaging the same person. For this reason, in this embodiment, displacement correction is performed for the registration iris pattern image fc(p 1 , p 2 ) and collation iris pattern image gc(p 1 , p 2 ) from which the unnecessary areas are cut off.

This displacement correction is performed by obtaining a POC function between the registration iris pattern image fc(p 1 , p 2 ) and the collation iris pattern image gc(p 1 , p 2 ) by a phase-only correlation (POC) method and obtaining the displacement (π1, π2) of a correlation peak.

The phase-only correlation method is described in detail in Japanese Patent Laid-Open No. 10-63847 and K. Takita et al., “High-accuracy subpixel image registration based on phase-only correction”, ICICE Trans. Fundamentals, Vol. E86-A, No. 8, pp. 1925-1934, August 2003. According to the phase-only correlation method, two-dimensional discrete Fourier transform is performed for a collation image (two-dimensional image) to generate a collation Fourier image. This collation Fourier image is combined with a registration Fourier image of a registration image which has been generated by performing similar processing. Performing processing for this combined Fourier image to set its amplitude to 1, and two-dimensional discrete Fourier transform is performed to obtain correlation data. This correlation data (POC function) comprises only a phase with its amplitude being set to 1 in a frequency space. This data can be basically regarded as data obtained by convoluting the collation image and the registration image, and represents the correlation between the collation image and the registration image.

In step S 615 , the CPU 20 - 1 cuts off images from the registration iris pattern image fc(p 1 , p 2 ) and collation iris pattern image gc(p 1 , p 2 ) by π1 and π2 pixels in the p 1 direction (horizontal direction) and the p 2 direction (vertical direction) to obtain a registration iris pattern image f′(p 1 , p 2 ) and a collation iris pattern image g′(p 1 , p 2 ) (see FIGS. 14C and 14D ). With this operation, common areas between the registration iris pattern image fc(p 1 , p 2 ) and the collation iris pattern image gc(p 1 , p 2 ) are extracted as f′(p 1 , p 2 ) and g′(p 1 , p 2 ).

[Collation]

The CPU 20 - 1 sets the extracted registration iris pattern image f′(p 1 , p 2 ) and the collation iris pattern image g′(p 1 , p 2 ) as a registration image and a collation image, respectively, and collates the registration image f′(p 1 , p 2 ) with the collation image g′(p 1 , p 2 ) by a band-limited phase-only correlation (BLPOC) method (step S 617 ). Note that BLPOC is described in detail in K. Ito et al., “A Fingerprint matching algorithm using phase-only correlation”, IEICE Trans. Fundamentals, Vol. E87-A, No. 3, pp. 682-691, March 2004 and Ayumi Morita et al., “A Study of a Fingerprint Matching Algorithm Based on the Combination of Phase-Only Correlation and Feature-Based Matching”, SICE Tohoku Chapter, the 219th workshop, material No. 219-5, November 2004. According to BLPOC, a registration image and a collation image are collated with each other by the phase-only correlation (POC) method using only an effective frequency band.

In this embodiment, in images f′(k 1 , k 2 ) and g′(k 1 , k 2 ) in the frequency domains of the registration image f′(p 1 , p 2 ) and collation image g′(p 1 , p 2 ) (see FIGS. 15A and 15B ), h % (e.g., 50% (a/2)) of an image size (a×b) in the k 1 direction and v % (e.g., 25% (b/4)) of an image size (a×b) in the k 2 direction are set as effective frequency bands, and images in the effective frequency bands are cut out to be collated, thereby obtaining a BLPOC function.

The iris patterns of iris images cannot be assumed to be striped patterns like fingerprints, and vary depending on individuals. It is therefore difficult to specify the frequency band which a given iris has. In this embodiment, accurate authentication can be performed by determining effective frequency bands from image sizes (the sizes of common areas).

The CPU 20 - 1 obtains a collation score SC from the peak of the BLPOC function obtained by collation between the registration image f′(p 1 , p 2 ) and collation image g′(p 1 , p 2 ) by BLPOC (step S 618 ). The CPU 20 - 1 compares the collation score SC with a predetermined value sc 1 (step S 619 ). If the collation score SC is equal to or less than the predetermined value sc 1 (YES in step S 619 ), it is determined that the registration image f′(p 1 , p 2 ) does not match the collation image g′(p 1 , p 2 ), and the input eyeball image is not authentic (step S 620 ).

If the collation score SC is larger than the predetermined value sc 1 , the collation score SC is compared with a predetermined value sc 2 (sc 2 >sc 1 ) (step S 621 ). If the collation score SC is equal to or larger than the predetermined sc 2 (YES in step S 621 ), it is determined that the registration image f′(p 1 , p 2 ) matches the collation image g′(p 1 , p 2 ), and the input eyeball image is authentic (step S 622 ).

In this embodiment, a cutting error concerning the iris area S 3 may appear as translation, enlargement/reduction, distortion, or the like in the image after polar coordinate transformation (normalized image). However, the cutting error concerning the iris area S 3 is very small as compared with the entire area. Therefore, the cutting error concerning the iris area S 3 has little influence on the registration iris pattern image f(p 1 , p 2 ) and the collation iris pattern image g(p 1 , p 2 ) after polar coordinate transformation, and accurate authentication can be performed.

If collation score SC satisfies sc 1 <SC<sc 2 (NO in step S 619 ), the CPU 20 - 1 enlarges/reduces (bicubic interpolation) the collation image g′(p 1 , p 2 ) and collates the resultant image with the registration image f′(p 1 , p 2 ) by BLPOC (step S 623 ).

For example, the collation image g′(p 1 , p 2 ) is reduced by 3%, 5%, and 7% in the p 1 direction (horizontal direction), and each reduced collation image g′(p 1 , p 2 ) is collated with the registration f′(p 1 , p 2 ) by BLPOC. Likewise, the collation image g′(p 1 , p 2 ) is enlarged by 3%, 5%, and 7% in the p 1 direction (horizontal direction), and each enlarged collation image g′(p 1 , p 2 ) is collated with the registration image f′(p 1 , p 2 ) by BLPOC.

A collation score is then obtained from the peak of each BLPOC function (step S 624 ), and the maximum value of these collation scores is set as the collation score SC (step S 625 ). The collation score SC is compared with the predetermined value sc 2 (step S 626 ). If the collation score SC is equal to or more the predetermined value sc 2 (YES in step S 626 ), “match” is determined (step S 622 ). If the collation score SC is less than the predetermined value sc 2 (NO in step S 626 ), “mismatch” is determined (step S 623 ).

[Reason Why Collation is Performed in Consideration of Enlargement/Reduction]

In obtaining center coordinates C(Cp 1 , Cp 2 ) of the registration original image F(n 1 , n 2 ) or collation original image G(n 1 , n 2 ) and center coordinates C(Ci 1 , Ci 2 ) of the pupil and iris, if the positional relationship between the two sets of center coordinates shifts in the images of the same person, enlargement/reduction occurs in the p 1 direction at the time of polar coordinate transformation. This becomes a cause for a decrease in collation score SC of the same person.

If the collation score SC satisfies sc 1 <SC<sc 2 , collation is performed while the collation image g′(p 1 , p 2 ) is reduced or enlarged little by little in the p 1 direction. This makes it possible to cope with a case wherein the collation score SC does not become equal to or more than sc 2 due to a shift of the positional relationship of the center coordinates of the pupil or iris.

In this embodiment, in step S 618 , when the collation score SC is to be obtained from the peak of the BLPOC function between the registration image f′(p 1 , p 2 ) and the collation image g′(p 1 , p 2 ), only a correlation peak in a predetermined range near the center is used. In an environment in which the relationship between the vertical axis of the imaging apparatus and the actual vertical axis does not change, the vertical direction of an iris image does not greatly change. In this case, in an image obtained by polar coordinate transformation of an eyeball image, both the radial direction and the angular direction slightly shift, but do not greatly shift. When a correlation is obtained by using BLPOC under this condition, a peak appears almost near the center. If a peak appears at a position which greatly deviates from the center, the obtained peak cannot be said to be a correct result, and the authentication accuracy can be improved by performing collation upon removal of the peak.

[Segmentation of Iris Area]

If the width w of the unnecessary area is larger than wm (YES in step S 614 ), i.e., it is determined that a wide area cannot be ensured as a common area used for collation, the CPU 20 - 1 obtains pairs of six segmented images fcd 1 (p 1 , P 2 ) to fcd 6 (p 1 , P 2 ) and six segmented images gcd 1 (p 1 , P 2 ) to gcd 6 (p 1 , P 2 ) by cutting out rectangular areas from the left and right ends with three widths wd (wd 1 , wd 2 , wd 3 ) such that no unnecessary areas are included in both the registration iris pattern image f(p 1 , p 2 ) and the collation iris pattern image g(p 1 , p 2 ) ( FIG. 10C : step S 627 ).

FIG. 16 shows a case wherein the segmented images fcd 1 (p 1 , p 2 ) to fcd 6 (p 1 , p 2 ) are cut out from the registration iris pattern image f(p 1 , p 2 ). As shown in FIG. 16 , cut-out widths wd from the upper end of the registration iris pattern image f(p 1 , p 2 ) are set to three widths wd 1 , wd 2 , and wd 3 (wd 1 <wd 2 <wd 3 ), and rectangular areas are cut from the left and right ends with wd 1 , wd 2 , and wd 3 so as not to include any unnecessary area, thereby obtaining the segmented images fcd 1 (p 1 , p 2 ), fcd 2 (p 1 , p 2 ), fcd 3 (p 1 , p 2 ), fcd 4 (p 1 , p 2 ), fcd 5 (p 1 , p 2 ), and fcd 6 (p 1 , p 2 ). In this embodiment, wd 1 =55 pixels, wd 2 =75 pixels, and wd 3 =95 pixels.

Likewise, with respect to the collation iris pattern image g(p 1 , p 2 ), as shown in FIG. 17 , cut-out widths wd from the upper end of the collation iris pattern image g(p 1 , p 2 ) are set to three widths wd 1 , wd 2 , and wd 3 (wd 1 <wd 2 <wd 3 ), and rectangular areas are cut from the left and right ends with wd 1 , wd 2 , and wd 3 so as not to include any unnecessary area, thereby obtaining the segmented images gcd 1 (p 1 , p 2 ), gcd 2 (p 1 , p 2 ), gcd 3 (p 1 , p 2 ), gcd 4 (p 1 , p 2 ), gcd 5 (p 1 , p 2 ), and gcd 6 (p 1 , p 2 ).

The CPU 20 - 1 obtains common areas in the same manner as in step S 616 described above with respect to the pair of the segmented images fcd 1 (p 1 , p 2 ) and gcd 1 (p 1 , p 2 ), the pair of the segmented images fcd 2 (p 1 , p 2 ) and gcd 2 (p 1 , p 2 ), the pair of the segmented images fcd 3 (p 1 , p 2 ) and gcd 3 (p 1 , p 2 ), the pair of the segmented images fcd 4 (p 1 , p 2 ) and gcd 4 (p 1 , p 2 ), the pair of the segmented images fcd 5 (p 1 , p 2 ) and gcd 5 (p 1 , p 2 ), and the pair of the segmented images fcd 6 (p 1 , p 2 ) and gcd 6 (p 1 , p 2 ), and obtains segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) (step S 628 ).

It is checked whether the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) include any image with a width of less than 40 pixels in the horizontal direction (p 1 direction) (step S 629 ). If all the images have widths of 40 pixels or more (NO in step S 629 ), it is determined sufficient areas are ensured for all the common areas, and the flow advances to the processing in step S 630 and subsequent steps. If there is at least one image with a width of less than 40 pixels (YES in step S 629 ), it is determined that there is an image for which a sufficient area is not ensured as a common area, and the flow advances to the processing in step S 641 ( FIG. 10D ) and subsequent steps.

[When Widths of All Images in Horizontal Direction Are 40 Pixels or More]

If the widths of the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) in the horizontal direction are 40 pixels or more (NO in step S 629 ), the CPU 20 - 1 collates the pairs of the fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) with each other by BLPOC (step S 630 ).

A collation score is obtained from the peak of each BLPOC function (step S 631 ), and the average value of the collation scores is set as the collation score SC (step S 632 ). If the collation score SC is equal to or less than the predetermined value sc 1 (YES in step S 633 ), “mismatch” is determined (step S 634 ). If the collation score SC is equal to or more than the predetermined value sc 2 (sc 2 >sc 1 ) (YES in step S 635 ), “match” is determined (step S 636 ).

If the collation score SC satisfies sc 1 <SC<sc 2 (NO in step S 635 ), the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) are enlarged/reduced, and the resultant images are collated with the corresponding segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) by BLPOC as in step S 623 described above (step S 637 ).

Collation scores are obtained from the respective BLPOC functions (step S 638 ), and the maximum value of the collation scores is set as the collation score SC (step S 639 ). If the collation score SC is equal to or more than the predetermined value sc 2 (YES in step S 640 ), “match” is determined (step S 636 ). If the collation score SC is less than the predetermined value sc 2 (NO in step S 640 ), “mismatch” is determined (step S 634 ).

[When There is Image with Width of Less than 40 Pixels]

If the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) include an image with a width of less than 40 pixels in the horizontal direction (YES in step S 629 ), the CPU 20 - 1 checks whether all the images have widths of less than 40 pixels ( FIG. 10D : step S 641 ). If at least one image has a width equal to or more than 40 pixels (NO in step S 641 ), the flow advances to the processing in step S 642 and subsequent steps. If the widths of all the images are less than 40 pixels (YES in step S 641 ), the flow advances to the processing in step S 653 ( FIG. 10E ) and subsequent steps.

[When at Least One Image Has Width of 40 Pixels or More]

If the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) include at least one image with a width of 40 pixels or more in the horizontal direction (YES in step S 641 ), the CPU 20 - 1 collates the pairs of all the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) except for the image with a width of less than 40 pixels by BLPOC (step S 642 ).

Collation scores are obtained from the peaks of the respective BLPOC functions (step S 643 ), and the average value of the collation scores is set as the collation score SC (step S 644 ). If the collation score SC is equal to or less than the predetermined value sc 1 (YES in step S 645 ), “mismatch” is determined (step S 646 ). If the collation score SC is equal to or more than the predetermined value sc 2 (sc 2 >sc 1 ) (YES in step S 647 ), “match” is determined (step S 648 ).

If the collation score SC satisfies sc 1 <SC<sc 2 (NO in step S 647 ), the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) except for an image with a width of less than 40 pixels are enlarged/reduced, and the resultant images are collated with the corresponding segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) by BLPOC as in step S 623 described above (step S 649 ). Collation scores are obtained from the peaks of the respective BLPOC functions (step S 650 ), and the maximum value of the collation scores is set as the collation score SC (step S 651 ). If the collation score SC is equal to or more than the predetermined value sc 2 (YES in step S 652 ), “match” is determined (step S 648 ). If the collation score SC is less than the predetermined value sc 2 (NO in step S 652 ), “mismatch” is determined (step S 646 ).

[When All Images Have Widths of Not Less Than 40 Pixels]

If the widths of all the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) and segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) in the horizontal direction are less than 40 pixels (YES in step S 641 ), the CPU 20 - 1 sets a pair of one of the segmented registration images fcd 1 ′(p 1 , p 2 ) to fcd 6 ′(p 1 , p 2 ) which has the largest width in the horizontal direction and one of the segmented collation images gcd 1 ′(p 1 , p 2 ) to gcd 6 ′(p 1 , p 2 ) which has the largest width in the horizontal direction as fcdi′(p 1 , p 2 ) and gcdi′(p 1 , p 2 ), and collates the segmented registration image fcdi′(p 1 , p 2 ) and the segmented collation image gcdi′(p 1 , p 2 ) by BLPOC ( FIG. 10E : step S 653 ).

The collation score SC is obtained from the peak of the BLPOC function (step S 654 ). If the collation score SC is equal to or less than the predetermined value sc 1 (YES in step S 655 ), “mismatch” is determined (step S 656 ). If the collation score SC is equal to or more than the predetermined value sc 2 (sc 2 >sc 1 ) (YES in step S 657 ), “mismatch” is determined (step S 658 ).

If the collation score SC satisfies sc 1 <SC<sc 2 (NO in step S 657 ), the segmented collation image gcdi′(p 1 , p 2 ) is enlarged/reduced, and the resultant image is collated with the segmented registration image fcdi′(p 1 , p 2 ) by BLPOC as in step S 623 (step S 649 ). Collation scores are then obtained from the peaks of the respective BLPOC functions (step S 660 ), and the maximum value of the collation scores is set as the collation score SC (step S 661 ). If the collation score SC is equal to or more than the predetermined value sc 2 (YES in step S 662 ), “match” is determined (step S 658 ). If the collation score SC is less than the predetermined value sc 2 (NO in step S 662 ), “mismatch” is determined (step S 656 ).

The function blocks of the CPU 20 - 1 will be described next. As shown in FIG. 18 , the CPU 20 - 1 comprises an iris area extraction unit 31 which extracts iris areas from a sensed registration eyeball image and collation eyeball image, a registration iris pattern image generating unit 32 which, when the iris area extraction unit 31 extracts an iris area from a registration eyeball image, generates a registration iris pattern image by performing polar coordinate transformation of an image in the extracted iris area, a collation iris pattern image generating unit 33 which, when the iris area extraction unit 31 extracts an iris area from a collation eyeball image, generates a collation iris pattern image by performing polar coordinate transformation of an image in the extracted iris area, and a collation unit 34 which collates a registration iris pattern image output from the registration iris pattern image generating unit 32 and a collation iris pattern image output from the collation iris pattern image generating unit 33 on the basis of the correlation between them.

The iris area extraction unit 31 comprises an inner area detection unit 31 a which detects the boundary between the pupil and iris of an eyeball image as the inner boundary of the iris by elliptic approximation, an outer area detection means 31 b for detecting the boundary between the iris and sclera of an eyeball image as the outer boundary of the iris by either circular approximation or elliptic approximation, and an area extraction unit 31 c which extracts, as an iris area, an area sandwiched between the inner boundary of the iris output from the inner area detection unit 31 a and the outer boundary of the iris output from the outer area detection means 31 b.

The collation unit 34 comprises a rectangular area cutting unit 34 a which cuts out one or more rectangular areas from the registration iris pattern image output from the registration iris pattern image generating unit 32 and the collation iris pattern image output from the collation iris pattern image generating unit 33 so as not to include any non-iris portion, a correlation value calculation unit 34 b which obtains a correlation value between rectangular areas common to the registration iris pattern image output from the registration iris pattern image generating unit 32 and the collation iris pattern image output from the collation iris pattern image generating unit 33 , and a collation determination unit 34 c which collates two images on the basis of the correlation value output from the collation unit 34 .

The iris area extraction unit 31 executes the processing in steps S 102 to S 106 and in steps S 602 to S 606 . The registration iris pattern image generating unit 32 executes the processing in step S 107 . The collation iris pattern image generating unit 33 executes the processing in step S 607 . The collation unit 34 executes the processing in steps S 617 to S 619 .

In the iris area extraction unit 31 , the inner area detection unit 31 a executes the processing in step S 104 , the outer area detection means 31 b executes the processing in step S 105 , and the area extraction unit 31 c executes the processing in step S 106 .

In the collation unit 34 , the rectangular area cutting unit 34 a executes the processing in step S 627 , the correlation value calculation unit 34 b executes the processing in steps S 616 to S 618 , and the collation determination unit 34 c executes the processing in step S 619 .

In the above embodiment, the cut-out widths wd are determined stepwise from the upper end of the registration iris pattern image f(p 1 , p 2 ) or collation iris pattern image g(p 1 , p 2 ), and rectangular areas are cut out for the respective widths. The BLPOC functions between areas common to these rectangular areas are obtained. However, rectangular areas may be selected and cut out so as to have as large areas as possible.

In the above embodiment, if the collation score SC satisfies sc 1 <SC<sc 2 , the collation image g′(p 1 , p 2 ) is enlarged/reduced in the horizontal direction (the angular direction in the eyeball image), and the resultant image is collated with the registration image f′(p 1 , p 2 ). However, the collation image may be enlarged/reduced in the vertical direction (the radial direction in the eyeball image), and the resultant image may be collated with the registration image f′(p 1 , p 2 ). Alternatively, a combination of enlarging/reducing operation in the horizontal direction and that in the vertical direction may be employed.

If the centers of the iris and pupil are not properly extracted due to noise or the like, the corresponding error may appear as an error in the angular direction in the eyeball image. At this time, since these images are processed as two-dimensional images, enlarging/reducing in the angular direction makes it possible to easily reduce the influence of such error. This makes it possible to perform accurate authentication.

If the boundary of the iris cannot be properly detected due to noise or the like, the error may appear as an error in the radial direction in the eyeball image. At this time, since this image is processed as a two-dimensional image, enlarging/reducing in the radial direction makes it possible to easily reduce the influence of the error. In addition, this makes it possible to reduce the influence of the error in boundary detection. Therefore, accurate authentication can be performed.

In the above embodiment, the registration image f′(p 1 , p 2 ) and the collation image g′(p 1 , p 2 ) are collated with each other by the band-limited phase-only correlation (BLPOC) method. However, collation may be performed by the phase-only correlation (POC) method upon image reduction processing instead of band-limited processing. That is, two images may be collated by using “image reduction processing+phase-only correlation (POC) method” instead of “band-limited phase-only correlation (BLPOC) method”.

According to the present invention, an image of an iris area extracted from a registration eyeball image is polar-coordinate-transformation into a registration iris pattern image (normalized image), and an image of an iris area extracted by a collation eyeball image is polar-coordinate-transformed into a collation iris pattern image (normalized image). The registration iris pattern image is collated with the collation iris pattern image on the basis of the correlation between them. Even if a slight error occurs in cutting an iris area from the eyeball image, accurate authentication can be performed.

According to the present invention, the boundary between the pupil and iris of an eyeball image is detected as the inner boundary of the iris by elliptic approximation, and the boundary between the iris and the sclera of the eyeball image is detected as the outer boundary of the iris by either circular approximation or elliptic approximation. An area sandwiched between the detected inner and outer boundaries of the iris is extracted as an iris area. This makes it possible to perform proper authentication even if the boundary between the pupil and the iris is not a perfect circle, thereby improving the authentication accuracy.

In addition, according to the present invention, polar coordinate transformation is performed for an image of an iris area extracted from a registration eyeball image with reference to the center of the inner boundary of the iris which is detected from the registration eyeball image, and polar coordinate transformation is performed for an image of an iris area extracted from a collation eyeball image with reference to the center of the inner boundary of the iris which is detected from the collation eyeball image. With this operation, both the images of the iris areas for registration and collation are polar-coordinate-transformed with reference to the accurate center of the pupil, thereby performing accurate authentication.

Furthermore, according to the present invention, one or more rectangular areas are cut out from a registration iris pattern image and a collation iris pattern image so as not to include any non-iris portion, and the correlation value between rectangular areas common to the registration iris pattern image and collation iris pattern image is obtained, and the two images are collated with each other on the basis of the correlation value. This makes it possible to accurately perform authentication even if an eyelid or eyelash is included in an image of an iris area or reflection of light glares on the image.