JP4553644B2 - Biometric authentication device - Google Patents

Description

  The present invention relates to a biometric information authentication device, and more specifically, is larger (wider) than an input feature point list created by combining an input feature point list extracted from an input biometric information image and a plurality of feature point lists. The present invention relates to a biometric information authentication apparatus capable of appropriately collating with one registered feature point list and performing authentication determination as to whether or not they are identical biometric information.

  In recent years, there has been an interest in the importance of security by identity verification, such as performing qualification authentication when entering or leaving a building or room. There are various methods such as authentication using a password as a method for identity verification. As one of the methods, an apparatus that performs authentication using unique biometric information is often used.

  For example, a fingerprint sensor having an area smaller than a finger is on the market for the purpose of mounting a biometric information authentication device such as a fingerprint on a mobile device. These fingerprint sensors are suitable for mounting because they are space-saving and inexpensive, but they are easily displaced because they are smaller than the fingers. If there is a difference in the finger placement position at the time of registration and collation, the fingerprints (patterns and feature points) of the two do not match, and there is a problem that the collation accuracy tends to be lowered.

  That is, the position where the finger is placed with respect to the fingerprint sensor is shifted, and data including a fingerprint of a part that is not registered at the time of collation may be input, so that even the person himself is rejected.

In a fingerprint authentication device, a misalignment is usually prevented by providing a finger guide or the like in the fingerprint input unit, but it is difficult to prevent it completely.
Therefore, in order to deal with the positional deviation, fingerprints are input multiple times during registration, and these are combined to register a fingerprint in a wider area than the fingerprint reading area of the sensor, and this registered fingerprint and an input fingerprint in a narrower area are registered. Is disclosed (see Patent Document 1).

JP 2003-67750 A

  In patent document 1, when the image area of collation fingerprint data is contained in all the image areas of registration fingerprint data, collation with all the image areas of collation fingerprint data and the image area corresponding to the collation fingerprint data in registration fingerprint data is carried out. Do. That is, the registration fingerprint and the input fingerprint are aligned, an area corresponding to the input fingerprint is extracted from the registration fingerprint, and collation is performed based on the number of matching feature points. Therefore, when the feature points are aligned, the feature points to be determined can be obtained from the registered fingerprints corresponding to the entire input fingerprint, so that the problem of misalignment at the time of collation is solved, and legitimate users are collated. It is possible to reduce the rejection rate of the person rejected by.

However, when the image area corresponding to the collation fingerprint data in the registered fingerprint data is used for the collation as it is, the inclination of the input fingerprint relative to the registration fingerprint is large or small (the degree of alignment). In order to make the same determination), not only the registrant's input fingerprint but also the registrant's input fingerprint is matched (coordinated) and collated by extremely large rotation and vertical and horizontal movement. May be.
As a result, the probability of correctly accepting the registrant increases, but the probability of erroneously accepting an unregistered person also increases, which may degrade the total verification accuracy.

In addition, if the matching execution condition is that the image area of the verification fingerprint data is included in the image area of the registered fingerprint data, the verification is performed when the feature points of the verification fingerprint data are detected in the area protruding from the registration fingerprint data. Not executed.
As a result, the probability of correctly rejecting another person who has not registered increases, but the probability of rejecting the registrant by mistake increases, and the total matching accuracy may be degraded.

  The present invention has been made to solve the above-described problems, and when comparing input feature point data (verification fingerprint data) and registered feature point data (registered fingerprint data) having a larger area, both are compared. An object of the present invention is to provide a body information authenticating apparatus that appropriately compares and collates and increases the total collation accuracy.

  In view of the above-described object, an object of the present invention is to provide biometric information input means for inputting biometric information in a predetermined area, storage means for registering biometric information in a wider area than the reading area of the biometric information input means, and registered biometric information. And a biometric information authenticating device configured to collate the input biometric information with each other and determine the identity, the collation means includes a position of the registered biometric information area and the input biometric information area including the rotation of the coordinate axis. Corresponding region extraction means for obtaining corresponding regions which are overlapping regions by matching, and the corresponding region increases as the relative angle between the coordinate axis of the registered biometric information region and the coordinate axis of the input biometric information region when the alignment is completed A collation area extracting means for extracting a collation area so as to reduce overlap with biometric information, and a biometric information collating means for collating the collation area portion in the registered biometric information with the input biometric information. And to provide a biometric information authentication device according to claim.

According to the present invention, the collation accuracy is improved as follows.
The larger the rotation angle when registering the registered biometric information and the input biometric information, the smaller the overlap between the corresponding area and the collation area. Therefore, when collating both biometric information, the rotation angle when aligning is large. It is possible to reduce the probability of erroneously accepting input biometric information of other people who tend to be. Further, in the case of the person's biological information, the angle becomes small, so that the collation accuracy with respect to the person's biological information can be maintained.

In addition, by rotating the reference point included in the corresponding area (preferably the centroid of the corresponding area or the centroid of the matching feature point) by the rotation angle, the overlap between the corresponding area and the matching area is reduced at an appropriate rate. Can be made. Therefore, the balance between the probability of accepting the person and the probability of not accepting the other person can be maintained, and the overall matching accuracy can be improved.
Furthermore, by correcting so that the collation area does not protrude from the entire area of the registered biometric information, the above-described effects can be obtained without exception.

  The present invention can also be applied to biological information other than fingerprints. For example, blood vessel patterns (finger, back of hand, palm, retina, etc.), palm print, ear shape, and the like have end points and branch points similar to fingerprints, and thus the same authentication process can be applied. Therefore, high-accuracy authentication can be realized even in an authentication apparatus using such biometric information.

Hereinafter, an example will be described by taking a fingerprint as an example of biometric information.
FIG. 1 is a diagram showing the configuration of the present invention. Each part of FIG. 1 will be schematically described.

  The operation / display unit 100 performs input of a registrant's ID number, an operation state, and guidance. The operation / display unit 100 includes a touch panel display and a speaker. A numeric keypad and a liquid crystal display may be used instead of the touch panel display.

  The fingerprint input unit 110 outputs a digital image of a fingerprint when the registrant places his / her finger. The fingerprint input unit 110 is an LED that illuminates the finger, an optical fingerprint sensor that modularizes a CCD camera that captures the finger and converts it into a digital signal, and a finger guide member that allows the user to correctly recognize the finger placement position. Composed. There are other types of fingerprint sensors such as a capacitance type and a pressure type, but the present invention does not limit the type of fingerprint sensor. The shape of the digital image area (sensor area) of the fingerprint is a rectangle.

The storage unit 120 is configured by a memory device such as a flash memory, and stores a registered feature point list and a registered feature point distribution area, which will be described later, in association with a registrant's ID number.
The two-dimensional space in which the registered feature point list is distributed is wider than the two-dimensional space of the fingerprint sensor of the fingerprint input unit 110 (hereinafter referred to as a sensor region).

  FIG. 2 is a diagram schematically showing a registered feature point list. The triangle in the figure represents a feature point to be described later, the center of gravity of the triangle represents the position of the feature point, and the direction from the center of gravity of the triangle to the point corresponding to the minimum vertex angle represents the direction of the feature point. A rectangle in FIG. 2A represents a sensor region, and shows a state in which feature points obtained by four fingerprint inputs are associated with each other. The registered feature point list can be obtained by synthesizing the feature point list obtained by multiple times of fingerprint input as described above, and (b) is a feature point list synthesized from (a).

  In addition, information indicating a region where feature points are distributed (hereinafter referred to as a registered feature point distribution region) is stored in association with the registered feature point list. Ideally, the registered feature point distribution area should be a polygonal area 210. However, since the processing becomes complicated, it is approximated by a rectangular area 220 in reality, and a pair of these diagonal coordinates is stored.

  The above description is based on the premise that the same fingerprint input unit 110 is used at the time of registration and at the time of authentication, but registration is performed using a fingerprint input unit having a sensor area wider than a finger, and fingerprint input of a sensor area smaller than the finger is performed. Even if it authenticates using a part, the effect by this invention is not impaired.

The processing unit 130 is configured by a CPU or the like that can input and output digital signals and store and execute a program for processing the digital signals.
Each of the control unit 140, the feature extraction unit 150, and the collation unit 160 described below is included in the processing unit 130 and is realized by the program.

The feature extraction unit 150 processes a digital image of a fingerprint and extracts feature points. Here, the feature points mean the end points of the fingerprint ridges and the branch points that are the ridges.
FIG. 3 is a diagram illustrating an example of position coordinates and directions of feature points of biological information such as fingerprints. In FIG. 3, 300, 310, and 320 are end points, and 330 is a branch point.

The feature point has data elements of position (x, y), declination (hereinafter referred to as direction parameter) θ corresponding to the direction, and reliability W as individual information.
The position (x _T , y _T ) of the end point 300 is illustrated in FIG. 3A, and the direction parameter θ _T of the end point 300 is illustrated in FIG. 3B.

The direction of the feature point is the direction in which the neighboring ridges flow. If the point is a branch point, the direction is a branch destination direction. The direction parameter θ is set clockwise from the x-axis direction.
The reliability W of the feature point is obtained as a value proportional to the magnitude of the direction vector obtained by adding the respective direction unit vectors of neighboring feature points including the feature point. This is a value reflecting the degree of consistency between the feature points and the neighboring feature points, and takes a maximum value when the neighboring feature point directions including the feature points are all in the same direction. The reliability W is an index representing the reliability of the direction of each feature point and the reliability of existence, and the calculation method is not limited to this.
A plurality of feature points (x _i , y _i , θ _i , W _i ) are extracted from one fingerprint image, and these are output as a feature point list. Thus, one feature point list has a coordinate system and has coordinate axes. This axis is defined by the input system of the fingerprint input unit 110. Similar to the feature point list of the input fingerprint, the feature point list of the registered fingerprint has a coordinate axis.

The collation unit 160 is a feature point list (hereinafter referred to as an input feature point list) obtained by processing the fingerprint image of the person to be authenticated (person who is going to receive authentication) input from the fingerprint input unit 110 at the time of authentication by the feature extraction unit 150. ) And a registered feature point list read from the storage unit 120 are compared to calculate a similarity described later, and is compared with a verification threshold value to determine whether or not to give authentication as a registrant.
That is, if the degree of similarity is equal to or higher than the collation threshold, the user is identified as a registrant and authentication is given to the person to be authenticated, otherwise authentication is not given.

FIG. 4 is a diagram illustrating how the registered feature point list for verification is cut out.
FIG. 4 schematically shows a state of matching, in which triangles represent feature points, and rectangular areas 410 and 430 represent sensor areas of the fingerprint sensor. Among the feature points, black triangles are input feature points, and the input feature point list is a list of these positions (x _i , y _i ), direction parameters θ _i and reliability W _i . White triangles are registered feature points, and a list of similar feature points is a registered feature point list. The registered feature point distribution area 220 indicates a range in which registered feature points are distributed.

Hereinafter, with reference to FIG. 4, the collation will be described focusing on the process of extracting feature points for collation that is a feature of the present invention.
At the time of collation, a process called alignment is performed in order to cope with a deviation (position deviation) of the finger placement position at the time of registration and authentication. The alignment is a coordinate conversion performed so that the registered feature point list and the input feature point list are best matched. In the example of FIG. 4B, the input feature point list (rectangular region 410) is aligned by coordinate transformation that slightly tilts.
By this alignment, the coordinate axis of the registered feature point list (registered feature point distribution region 220) and the coordinate axis of the input feature point list (rectangular region 410) have a relative angle. This relative angle is an angle obtained by rotating the rectangular area 410 during alignment.

After the registration, the similarity described later is calculated. At this time, instead of using the entire registered feature point distribution area 220, an area having the same area as the sensor area as indicated by the rectangular area 420 is used. Then, the registered feature points are cut out and the similarity is calculated. In this way, the feature points distributed in the same area are compared equally. The cut-out feature point list is called a matching registered feature point list.
In the present invention, the region for cutting out the registered feature point list for matching is directly opposed to the coordinate axis of the registered feature point list.

Here, the similarity used in the collation is defined. In general, the similarity is normalized based on the number of feature points and the reliability as follows.
The similarity S is calculated, for example, as follows based on the matching results and reliability of the feature points in the registered feature point list and the input feature point list.

Here, N is the number of white triangles in the registered registered feature point list, M is the number of black triangles in the input feature point list, Wi is the reliability of the i-th feature point in the registered registered feature point list, and Ci. Is the association result (with pair: 1, no pair: 0), Wj is the reliability of the jth feature point in the input feature point list, and Cj is the association result (with pair: 1, without pair: 0) It is.
If defined in this way, even if some feature points coincide by chance with another person's fingerprint, the degree of similarity does not become unreasonably high.

  4A is an example of the input feature point list of the registrant, FIG. 4C is an example of the input feature point list of another person who is not the registrant, and FIG. 4B and FIG. 4D are the registered feature point list of these and the registrant. It is a state when the alignment is performed.

  In the case of another person (FIG. 4 (d)), since the fingerprints do not match with each other, the result of alignment often involves an extreme rotation compared to the case of the other person (FIG. 4 (b)). At this time, in the present invention, feature points included in a region (corresponding region) overlapping with the rectangular region 410 and the rectangular region 430 are not extracted from the registered feature point list but included in the rectangular region 420 or the rectangular region 450 (collation region). The registered feature point list for verification is generated by cutting out the feature points to be checked.

Each side of the dotted rectangle is made parallel to the coordinate system of the registered feature point list.
The coordinate system of the registered feature point list substantially matches the direction in which the registrant places his / her finger during registration.

  In the rectangular area 450 that is the collation area extracted in this way, the coincidence between the registered feature point and the input feature point other than the overlapping area between the area and the rectangular area 430 is not counted. Compared with the case where 430 is used as a cutout region, the number of matching feature points tends to decrease. Therefore, the similarity when the registered feature point is cut out from the rectangular area 450 and collated is lower than when the registered feature point is cut out from the rectangular area 430 and used for collation (in the case of another person). The same thing happens between the rectangular areas 410 and 420 (in the case of the person). However, in the case of the person himself / herself, since the relative angle is small and the overlapping region is wide, the number of matching feature points does not decrease so much and the decrease in similarity is small. That is, in general, a decrease in the degree of similarity is greater for others than for the principal, leading to improved collation accuracy.

  In other words, making the region for cutting out the registered feature point list for verification directly face the coordinate axis of the registered feature point list is equivalent to resetting the rotation of the translation and rotation performed in the alignment. .

In the first place, the direction of the finger with respect to the fingerprint sensor is kept substantially constant by the action of the finger guide of the fingerprint input unit 110, etc., so that the direction of the finger does not differ greatly between registration and verification, and a position with a large rotation Matching is not required.
Nevertheless, the alignment of another person's fingerprint is accompanied by a large rotation, not by correcting the misalignment, but by forcibly associating different fingerprint data. By resetting the amount that is forcibly associated in this way, the degree of similarity with respect to another person's fingerprint is greatly reduced by intentionally leaving an alignment mismatch intentionally.

  In order to obtain this effect, the collation area (420) has the above-mentioned relative angle so that the axis of the corresponding area overlapping the rectangular frame (410) of the input feature point list and the axis of the registered feature point list are parallel. Only the corresponding area is rotated and extracted. At this time, the collation area (420) faces the coordinate axis of the registered feature point list. Therefore, the larger the relative angle when aligned, the smaller the overlap between the corresponding area and the collation area.

4 (b) and 4 (d), feature points surrounded by circles represent pairs of input feature point black triangles and registered feature point white triangles that are substantially matched at the stage of alignment, and ★ marks indicate the features that constitute the pair. It represents the center of gravity of the point.
The rectangular area 420 is obtained by rotating the rectangular area 410 around the center of gravity ★. By having the center of gravity ★ as the center, many feature point pairs can be included in the rotated region.
In addition, you may rotate centering | focusing on the gravity center (x mark of FIG. 4) of the rectangular area | region 410. FIG. This also allows a relatively large number of feature point pairs to be included in the rotated region, and the overlapping region to be reduced at a certain rate by the rotation.

  Further, in the present embodiment, when the collation area extends beyond the registered feature point distribution area, the area after the translation is performed with respect to the coordinate axis of the registered feature point list so that the collation area is included in the registered feature point distribution area. Is obtained as a verification region.

  The collation region extracted by this parallel movement does not collapse the direct relationship with the coordinate system of the registered feature point list, and is kept in the size of the sensor region.

  In FIG. 4D, a rectangular area 440 obtained by rotating the rectangular area 430 protrudes from the registered feature point distribution area. In FIG. 4D, the rectangular area 450 is obtained as a collation area by translating the rectangular area 440, and the feature points of the registered feature point list included in this area are used for collation.

  By doing in this way, normalization of similarity expressed by Equation 1 and Equation 2, correction of similarity disparity by cutting out feature points distributed over a certain area from the registered feature point list, registered features for verification The effect of improving collation accuracy by making the area for cutting out the point list directly face the coordinate axis of the registered feature point list can be achieved without exception.

The output unit 170 is a terminal connected to other devices such as an electric lock controller.
When authentication is given, an unlock signal flows here.

  FIG. 5 is a flowchart showing a collation process of a biometric information authentication apparatus such as a fingerprint according to the present invention. The verification process of this apparatus will be described with reference to FIG.

The touch panel display of the operation / display unit 100 is operated by the person to be authenticated (person who is going to receive authentication) and the ID number is input (S500).
The person to be authenticated places his / her finger on the fingerprint sensor constituting the fingerprint input unit 110 according to a guidance such as “Please input fingerprint”.

A fingerprint image digital signal is output from the fingerprint input unit 110 and input to the processing unit 130 (S505).
The control unit 140 activates the feature extraction unit 150 and outputs a fingerprint image to the feature extraction unit 150.

  The feature extraction unit 150 extracts a plurality of feature points from the fingerprint image, obtains information about each feature point, and generates an input feature point list (S510). FIG. 4A is a schematic diagram of an input feature point list. The input feature point list is a list in which the number of feature points is enumerated as a set of coordinates (x, y) representing the positions of feature points represented by black triangles, direction parameters θ, and values representing reliability. Stored as data.

The control unit 140 activates the matching unit 160 and inputs the input feature point list to the matching unit 160.
The matching unit 160 reads the registered feature point list associated with the ID number input from the storage unit 120, and calculates the similarity between the registered feature point list and the input feature point list as follows.

A matching feature point pair is searched from the registered feature point list and the input feature point list (S515). The process of verifying various combinations of the registered feature point white triangle and the input feature point black triangle is repeated to find a pair whose position and direction substantially match.
In addition, since the calculation cost required for the search becomes large, the feature points may be thinned out and executed. In this case, if a predetermined number of feature points are selected in descending order of reliability W, it is possible to prevent a decrease in accuracy due to thinning.

  The registered feature point list and the input feature point list are aligned based on the matching feature points obtained by the search in S515 (S520). This alignment is obtained in this way by a parallel movement in which the centroid of the input feature point black triangle and the centroid of the registered feature point white triangle constituting the coincident feature point group become the same point (★ in FIG. 4). The rotation is performed so that the input feature point black triangle and the registered feature point white triangle are closest to each other with the center of gravity ★ as the center. In this way, an alignment result as indicated by 410 in FIG. 4B is obtained. At this time, the coordinate system of the input feature point list is converted into the coordinate system of the registered feature point list. Further, the same coordinate conversion is performed on the coordinate value of the upper left corner and the coordinate value of the lower right corner of the rectangular area 410.

  Next, rotation is applied to the rectangular area 410 so that the long side of the rectangular area 410 is parallel to the y-axis of the registered feature point list around the center of gravity ★ (S525). That is, the rectangular area 410 is rotated by the relative angle generated by the above-described alignment so that the coordinate axes of the input feature point list before coordinate conversion and the registered feature point list are parallel. This is a process of further rotating the coordinate values of the corners of the rectangular area 410 described above. As a result, a rectangular area 420 in FIG. 4B is obtained.

It is determined whether the rectangular area 420 is included in the registered feature point distribution area 220 (S530). Since the corner coordinate value of the registered feature point distribution region 220 is stored in the storage unit 120, it may be read out and compared with the corner coordinate value of the rectangular region 420.
If it is included in the registered feature point distribution area 220 as in the rectangular area 420, this is used as a collation area, and the process proceeds to the next process. On the other hand, when not included in the registered feature point distribution area 220 as in the rectangular area 440, coordinate conversion for translation is performed on the rectangular area 440 to obtain the rectangular area 450 as a collation area (S535).

  Next, the registered feature points included in the rectangular area 420 (matching region) are cut out from the registered feature point list to generate a matching registered feature point list (S540). The coordinate value of each feature point in the registered feature point list may be compared with the coordinate value of the corner of the rectangular area 420.

The similarity between the input feature point list and the registered feature point list for matching is calculated (S545). At this time, the similarity is calculated after the input feature point list and the matching feature point list are aligned again. By doing this, it is possible to relieve the matching feature point existing at a position slightly deviated from the overlapping region between the corresponding region and the matching region, and perform fine matching.
The similarity is compared with the verification threshold, and if the similarity is equal to or higher than the verification threshold, the control unit is notified that the authentication target is authenticated, and otherwise is notified to the control unit that the authentication is not performed ( S550).

The control unit 140 performs control according to the notification from the matching unit 160.
If the notification is to give authentication, an unlock signal is sent to the output unit 170, and the electric lock of the door is unlocked through the connected electric lock controller (S555). In addition, a guidance such as “OK to collate” is displayed on the operation / display unit 100 (S560).
On the other hand, if the notification indicates that authentication is not given, guidance such as “failure of collation” is displayed on the operation / display unit 100 (S560).

  In the above description, the collation is started by the user's ID number input, but the ID number input (S500) may be omitted and the collation may be started by the fingerprint input (S505). In this case, the fingerprint authentication apparatus executes the processing from S510 to S550 in order for all registered ID numbers, and gives authentication if the similarity with any registered fingerprint is equal to or greater than the collation threshold. Further, both the process of starting collation by inputting an ID number and the process of starting collation by inputting a fingerprint may be executed so that the user can select them.

  In the above description, since the shape of the sensor area of the fingerprint input unit 110 is rectangular, the corresponding area and the matching area are also rectangular. However, this shape is not limited to this, and any shape can be used as long as the region surely changes when rotated. Preferably, it is any one of a rectangle, a rhombus, and an ellipse that are symmetrical vertically and horizontally.

  In the above description, an embodiment of an authentication apparatus using a fingerprint as biometric information for identifying an individual has been described. However, the present invention can also be applied to biometric information other than fingerprints. For example, a blood vessel pattern (finger, back of hand, palm, retina, etc.), palm print, ear shape, and the like have end points and branch points as shown in FIG. Therefore, high-accuracy authentication can be realized even in an authentication apparatus using such biometric information.

It is a figure showing the structure of this invention. It is the figure which represented the registration feature point list typically. It is a figure showing the example of the position coordinate and direction of the feature point of biometric information, such as a fingerprint. It is a figure showing the mode of extraction of the registration feature point list for collation. It is a flowchart which shows the collation process of biometric information authentication apparatuses, such as a fingerprint, by this invention.

Explanation of symbols

100 Operation / Display Unit 110 Fingerprint Input Unit 120 Storage Unit 130 Processing Unit 140 Control Unit 150 Feature Extraction Unit 160 Collation Unit 170 Output Unit

Claims (6)

The biometric information input means for inputting the biometric information of the predetermined area, the storage means for registering biometric information in a larger area than the reading area of the biometric information input means, and the registered biometric information and the input biometric information are collated and the identity In a biometric information authentication device configured with a verification means for determining
The verification means includes
A corresponding area extracting means for obtaining a corresponding area that is an overlapping area by aligning the area of the registered biometric information and the area of the input biometric information, including rotation of coordinate axes;
A collation area that extracts a collation area so that the overlapping with the corresponding area decreases as the relative angle between the coordinate axis of the registered biometric information area and the coordinate axis of the input biometric information area when the alignment is completed Extraction means;
Biometric information collating means for collating the collation region portion in the registered biometric information with the input biometric information;
A biometric information authentication device comprising:   The biometric information authentication device according to claim 1, wherein the collation area extraction unit obtains a collation area by rotating the corresponding area by the relative angle around a reference point included in the corresponding area.   The biometric information authentication device according to claim 2, wherein the reference point is a center of gravity of the corresponding region.   The biometric information authentication device according to claim 2, wherein the reference point is a centroid of a feature point that coincides with a feature point of the registered biometric information and a feature point of the input biometric information.   5. The collation area extracting unit extracts the collation area by translating the corresponding area so that the collation area is completely included in the entire area of the registered biometric information. The biometric information authentication device described.   The biometric information authentication apparatus according to any one of claims 1 to 5, wherein the predetermined area is any one of a rectangle, a rhombus, and an ellipse.

Images