WO2024095462A1 - Comparison score calculation method, comparison score calculation device, comparison score calculation system, and comparison score calculation program - Google Patents

Abstract

In the present invention, after a computer acquires coordinate information and feature value information for a comparison feature point, the computer converts the coordinates of one among the comparison feature point and a registered feature point stored in a storage unit on the basis of an angular difference calculated using the coordinate information of the comparison feature point and the coordinate information of the registered feature point. The computer uses the angular difference and the acquired feature value information for the comparison feature point to calculate a comparison score between the other among the comparison feature point and the registered feature point and the feature point resulting from converting the coordinates of the one among the comparison feature point and the registered feature point.

Description

Matching score calculation method, matching score calculation device, matching score calculation system, and matching score calculation program

The present invention relates to a matching score calculation method, a matching score calculation device, a matching score calculation system, and a matching score calculation program.

In biometric authentication and other fields, matching methods using feature points are commonly used. In biometric authentication, characteristic points are extracted from a biometric image as feature points, and matching is performed using feature amounts calculated from pixel values near the feature points. Biometric images include fingerprint images and palm vein images.

FIG. 1 shows examples of feature points and feature quantities in a biometric image. FIG. 1(a) shows examples of feature points. Curve 101 represents, for example, a fingerprint in a fingerprint image, or veins in a palm vein image. In this case, branch point 111, end point 112, and end point 113 of curve 101 are extracted as feature point A1, feature point A2, and feature point A3, respectively.

FIG. 1(b) shows an example of feature data generated from the biometric image of FIG. 1(a). The feature data includes coordinates (X, Y) indicating the position of each feature point and the feature amount of each feature point. The feature amount is represented by, for example, a vector.

In matching methods using feature points, matching processing is performed using target data and registered data. The target data is feature data generated from a biometric image of the target person, and the registered data is feature data generated from a biometric image of the enrolled person. In the matching processing, the coordinates (X, Y) of the feature points are used to compare the feature amounts of the matching feature points contained in the target data with the feature amounts of the registered feature points contained in the registered data, and a matching score indicating the comparison result is calculated.

In relation to biometric authentication, an authentication device that achieves highly accurate biometric authentication while reducing memory usage is known (see, for example, Patent Document 1). A data generation method that suppresses accuracy degradation in biometric authentication using partial data is also known (see, for example, Patent Document 2). A biometric image processing device that improves authentication accuracy in biometric authentication that is performed by binarizing feature amounts acquired from a biometric image is also known (see, for example, Patent Document 3).

JP 2006-39951 A International Publication No. 2022/074786 JP 2019-45969 A

If the biometric image input for biometric authentication contains posture variations, the accuracy of the matching score based on the comparison between the feature values of matching features and the feature values of registered features may decrease.

Note that this problem is not limited to feature values of feature points extracted from biometric images, but occurs when comparing feature values of various feature points.

In one aspect, the present invention aims to improve the accuracy of the matching score between matching features and registered features.

In one proposal, when the computer acquires the coordinate information and feature amount information of the matching feature point, it performs coordinate transformation of either the matching feature point or the registered feature point based on an angle difference calculated using the coordinate information of the matching feature point and the coordinate information of the registered feature point stored in the storage unit. Using the angle difference and the acquired feature amount information of the matching feature point, the computer calculates a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point, and a matching score between the other of the matching feature point and the registered feature point.

According to one aspect, it is possible to improve the accuracy of the matching score between matching features and registered features.

FIG. 2 is a diagram showing feature points and feature amounts in a biometric image. FIG. 13 is a diagram illustrating a normalization process for a fingerprint image. FIG. 13 is a diagram illustrating normalization processing for a palm vein image. FIG. 1 illustrates a matching process using alignment search. FIG. 13 is a diagram showing a correspondence relationship between positions of feature points. FIG. 13 is a diagram illustrating a comparison of feature amounts. FIG. 2 is a functional configuration diagram of a matching score calculation device according to an embodiment. 13 is a flowchart of a matching score calculation process. FIG. 2 is a functional configuration diagram of an entrance/exit management system. 11 is a flowchart of a first biometric authentication process. 13 is a flowchart of a search process. 13 is a flowchart of a calculation process. FIG. 13 is a diagram showing a matching score using the Hamming distance. FIG. 1 illustrates encryption of binary data. FIG. 13 is a diagram illustrating the relationship between α(θ, n, n) and the rotation angle θ. 13 is a flowchart of a second biometric authentication process. FIG. 2 is a functional configuration diagram of a flapper gate management system. FIG. 2 is a hardware configuration diagram of an information processing device.

The following describes the embodiment in detail with reference to the drawings.

Biometric images input for biometric authentication often contain posture variations. Posture variations, such as rotation or positional shift, occur due to unstable posture of the biometric image input by the person to be matched or the user who is registered. For this reason, in biometric authentication, normalization processing is performed to correct for rotation, positional shift, etc.

FIG. 2 shows an example of normalization processing for a fingerprint image. Fingerprint image 201 is converted into fingerprint image 202 by performing corrections such as rotation or positional shift based on information such as the outline of the finger or the center coordinates of the fingerprint in fingerprint image 201.

FIG. 3 shows an example of normalization processing for a palm vein image. By applying normalization processing to palm vein image 301, palm vein image 301 is converted into palm vein image 302.

However, normalization processing can only perform approximate correction, not complete correction, so posture variations such as rotational errors often remain. In the case of biometric authentication, rotational correction is applied to both the enrollment data and the data to be matched, so the sum of the rotational errors of both affects the matching score. For example, if the enrollment data contains a rotational error of +3° and the data to be matched contains a rotational error of -3°, the matching process compares two pieces of data with an angle error of 6°.

Therefore, in the matching process, a registration search is performed to absorb the rotation error. In the registration search, a coordinate transformation T is applied to the data to be matched, and the data to be matched after the coordinate transformation is compared with the registered data.

Figure 4 shows an example of matching processing using alignment search. Matching target data 402 is generated by applying coordinate transformation T to matching target data 401, and matching target data 402 is compared with registered data 403. Coordinate transformation T includes rotation, translation, etc. Coordinate transformation T can be calculated by various methods. For example, a brute force search that tries multiple coordinate transformations in sequence may be used, or coordinate transformation T may be calculated from the correspondence between multiple feature points.

By applying this type of coordinate transformation T to the coordinates of feature points contained in the data to be matched, the effects of rotation error can be absorbed. On the other hand, if the feature amounts contained in the data to be matched are extracted from a biometric image that has rotation error, the corresponding error will be included in the feature amount, reducing the accuracy of the matching score.

Figure 5 shows an example of the correspondence relationship between the positions of feature points in a biometric image. Positions 512-i (i = 1 to 3) in palm vein image 502 of the person being matched correspond to positions 511-i in palm vein image 501 of the registrant. The person being matched is the same person as the registrant. In this case, in biometric authentication, the feature amount at position 512-i in palm vein image 502 is compared with the feature amount at position 511-i in palm vein image 501.

FIG. 6 shows an example of a comparison of features. Image 601 is an image of a region corresponding to position 511-i in palm vein image 501 in FIG. 5, and image 602 is an image of a region corresponding to position 512-i in palm vein image 502. Image 601 has no rotation error, and image 602 has a rotation error.

Feature point 621 is extracted from region 611 of image 601, and feature vector 641 representing the feature amount of feature point 621 is generated using image data 631 of region 611. Also, feature point 622 is extracted from region 612 of image 602, and feature vector 642 representing the feature amount of feature point 622 is generated using image data 632 of region 612.

In this case, even though the person being matched is the same person as the registrant and feature point 622 corresponds to feature point 621, feature vector 642 differs from feature vector 641, and the accuracy of the matching score decreases. As a result, the person being matched is more likely to be determined to be a person different from the registrant, and the authentication accuracy decreases.

It is also possible to correct the angle difference θ by calculating the angle difference θ from the correspondence between multiple feature points included in the data to be matched and multiple feature points included in the registered data, then rotating the biometric image of the person to be matched by -θ and re-extracting the feature amounts. However, this method has many problems in terms of processing time, memory consumption, and security.

First, the processing time increases because the features must be extracted again. In particular, when building a biometric authentication system with a client-server configuration, the features are generally extracted by the client, but the angle difference θ is calculated by a matching process on the server. Therefore, after the angle difference θ is sent from the server to the client, the features are extracted again by the client. This adds communication time, further increasing the processing time. Also, the authentication protocol becomes very complicated.

The client will retain the biometric images in memory from the time it acquires the biometric image of each person to be matched until it receives the angle difference θ sent by the server, which will increase memory consumption. In particular, in an operational system that opens and closes a flapper gate based on the authentication result, it is desirable for multiple users to be able to pass through the flapper gate quickly, so memory consumption will increase further depending on the number of users passing through.

Furthermore, from a security standpoint, an operational system that stores biometric images of subjects for matching in the client's memory for long periods of time is not a desirable configuration.

FIG. 7 shows an example of the functional configuration of a matching score calculation device according to an embodiment. The matching score calculation device 701 in FIG. 7 includes a storage unit 711, a coordinate conversion unit 712, and a matching score calculation unit 713. The storage unit 711 stores coordinate information of registered feature points. The coordinate conversion unit 712 and the matching score calculation unit 713 perform a matching score calculation process.

FIG. 8 is a flowchart showing an example of the matching score calculation process performed by the matching score calculation device 701 in FIG. 7. First, when the coordinate information and feature amount information of the matching feature point are acquired, the coordinate transformation unit 712 performs coordinate transformation of either the matching feature point or the registered feature point based on the angle difference calculated using the coordinate information of the matching feature point and the coordinate information of the registered feature point (step 801).

Next, the matching score calculation unit 713 uses the angle difference and the feature amount information of the acquired matching feature point to calculate a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point, and a matching score between the other of the matching feature point and the registered feature point (step 802).

The matching score calculation device 701 in FIG. 7 can improve the accuracy of the matching score between the matching feature points and the registered feature points.

FIG. 9 shows an example of the functional configuration of an entrance/exit management system including the matching score calculation device 701 of FIG. 7. The entrance/exit management system of FIG. 9 includes a matching score calculation device 901 and a control device 902. The matching score calculation device 901 corresponds to the matching score calculation device 701 of FIG. 7.

The matching score calculation device 901 calculates a matching score using matching target data generated from a biometric image of the person to be matched, performs biometric authentication on the person to be matched using the matching score, and transmits the authentication result to the control device 902.

The control device 902 controls the opening and closing of the door based on the authentication result. For example, if the authentication result indicates success, the control device 902 controls the door to open, and if the authentication result indicates failure, the control device 902 controls the door to close.

The matching score calculation device 901 includes a biometric image acquisition unit 911, a feature extraction unit 912, a coordinate transformation calculation unit 913, a coordinate transformation unit 914, a matching score calculation unit 915, a communication unit 916, and a memory unit 917. The coordinate transformation unit 914, the matching score calculation unit 915, and the memory unit 917 correspond to the coordinate transformation unit 712, the matching score calculation unit 713, and the memory unit 711 in FIG. 7, respectively.

The memory unit 917 stores a registration template 921 and fitting information 922. The registration template 921 includes registration data for each of a plurality of enrollees, and the enrollment data for each enrollee includes the coordinates and feature vectors of each of a plurality of feature points. Each feature point included in the enrollment data is an example of a registered feature point, and is a feature point extracted from a biometric image of each enrollee. The coordinates of each feature point included in the enrollment data is an example of coordinate information of the registered feature point.

The fitting information 922 is information for calculating correction parameters for the feature quantities from the angle difference θ between the data to be matched and the registered data. The correction parameters will be described later.

The biometric image acquisition unit 911 acquires a biometric image 923 of the person to be matched and stores it in the memory unit 917. The biometric image 923 is, for example, a fingerprint image, a palm vein image, a palm print image, or a face image. The biometric image acquisition unit 911 is, for example, a fingerprint sensor, a vein sensor, or an image sensor. By using a palm vein image as the biometric image 923, palm vein authentication can be performed.

The feature extraction unit 912 extracts multiple feature points from the biometric image 923 and calculates a feature vector from pixel values near each feature point. The feature extraction unit 912 then generates matching target data 924 including the coordinates and feature vectors of each of the multiple feature points and stores it in the storage unit 917. Each feature point extracted from the biometric image 923 is an example of a matching feature point, the coordinates of each feature point are an example of coordinate information of a matching feature point, and the feature vector of each feature point is an example of feature amount information of a matching feature point.

The feature extraction unit 912 can calculate feature vectors, for example, using a transformation based on a basis for which orthonormality holds. Examples of such transformations include Principal Component Analysis (PCA), discrete cosine transform (DCT), and fast Fourier transform (FFT).

The coordinate transformation calculation unit 913 calculates a coordinate transformation T for matching the matching target data 924 with the registered data of each registered person in the registered template 921. The coordinate transformation T includes, for example, a rotation angle θ and a parallel translation amount (ΔX, ΔY) in the X direction and Y direction. The rotation angle θ corresponds to an angle difference calculated using the coordinate information of the matching feature point and the coordinate information of the registered feature point.

The coordinate transformation unit 914 uses coordinate transformation T to transform the coordinates of each feature point contained in the matching target data 924.

The matching score calculation unit 915 corrects the feature vector of each feature point included in the matching target data 924 using the rotation angle θ included in the coordinate transformation T and the fitting information 922. Next, the matching score calculation unit 915 calculates a matching score 925 between the matching target data 924 and the registered data using the coordinates of each feature point after the coordinate transformation, the feature vector of each feature point after the correction, and the registered data of each registered person included in the registered template 921. Then, the matching score calculation unit 915 stores the matching score 925 in the memory unit 917.

Next, the matching score calculation unit 915 performs biometric authentication on the person to be matched using the matching score 925 and generates an authentication result. The communication unit 916 transmits the authentication result to the control device 902.

Next, a method for calculating the coordinate transformation T will be described. The coordinate transformation calculation unit 913 calculates the coordinate transformation T from the correspondence between multiple feature points included in the matching target data 924 and multiple feature points included in the registered data of each registered person.

First, the coordinate transformation calculation unit 913 generates feature points pairs that represent combinations of each feature point included in the matching target data 924 and each feature point included in the registered data. Next, for each feature points pair, the coordinate transformation calculation unit 913 calculates a feature points score between the feature vector of the feature point included in the matching target data 924 and the feature vector of the feature point included in the registered data. Then, the coordinate transformation calculation unit 913 selects the top X (X is an integer of 2 or more) feature points pairs in order from the best feature points score.

The feature point score can be the similarity or dissimilarity of two feature vectors. The similarity can be the inner product of two feature vectors, and the dissimilarity can be the inter-vector distance of two feature vectors. When the feature point score represents the similarity of two feature vectors, the larger the feature point score, the better the value. On the other hand, when the feature point score represents the dissimilarity of two feature vectors, the smaller the feature point score, the better the value.

Next, the coordinate transformation calculation unit 913 selects two feature points pairs from the X feature points pairs, and uses the coordinates of each feature point included in the two selected feature points pairs to calculate a coordinate transformation T. Since the total number of combinations of two feature points pairs _that can be selected from the X feature points pairs is _XC2 , _XC2 coordinate transformations Ti (i= _{1 to XC2} ) are found.

In calculating the coordinate transformation Ti, it is assumed that one of the two selected feature point pairs includes a combination of registered feature point P1 and matching feature point Q1, and the other includes a combination of registered feature point P2 and matching feature point Q2. The coordinates (X, Y) of each feature point are written as follows:

P1 (x1, y1)
Q1 (u1, v1)
P2 (x2, y2)
Q2 (u2, v2)

The coordinate transformation calculation unit 913 uses the least squares method to find the coordinate transformation Ti that minimizes the distance between P1 and Q1 and the distance between P2 and Q2 after the coordinate transformation.

The coordinate transformation Ti, which includes the rotation angle θ and the translation amount (ΔX, ΔY), is expressed by the following formula:

(x', y') represents the coordinates after (x, y) is transformed. When equation (1) is expressed in homogeneous form, it becomes the following equation.

If we impose the condition that Q1 and Q2 after coordinate transformation match P1 and P2, respectively, we obtain the following equation.

Combining equations (3) and (4), we obtain the following equation.

A B = C (5)

The coordinate transformation calculation unit 913 calculates θ, ΔX, and ΔY that satisfy the formula (5) by the least squares method. To find the matrix A that minimizes the error by the least squares method, the pseudo inverse matrix B ⁻¹ of the matrix B is found, and the matrix A is calculated by the following formula.

A = CB ^-1 (9)

Next, a method for calculating the feature vector after correction of each feature point will be described. In the following description, the feature vector for image vector I(θ) representing a biometric image rotated by a rotation angle θ is represented by x = (x(1), x(2), ..., x(N)) (N is an integer equal to or greater than 2). On the other hand, the feature vector for image vector I representing an unrotated image is represented by x' = (x'(1), x'(2), ..., x'(N)). The image represented by image vector I corresponds to the image represented by image vector I(θ) rotated by -θ.

When feature vectors are calculated using PCA, feature extraction for I is expressed by the following formula:

I = Σx'(i)P(i) (10)

x'(i) (i = 1 to N) represents the i-th element of feature vector x, and P(i) represents the i-th basis vector (principal component vector). Σ represents the sum for i = 1 to N. In the operation of the access control system, N is set to an appropriate integer. The right-hand side of equation (10) represents a linear combination of the N basis vectors P(i), and x'(i) represents a coefficient.

Next, if the rotation process with a rotation angle θ is expressed as a matrix R(θ), I(θ) is expressed by the following formula.

I(θ) = R(θ)I (11)

On the other hand, feature extraction for I(θ) is expressed by the following formula:

I(θ)＝Σx(i)P(i)　　　(12)

I(θ) is rotated by θ, while P(i) is not. For this reason, x(i) contains a rotation error. However, if feature extraction can be performed using basis vector P(θ, i) obtained by rotating P(i) by the same θ as I(θ), then the same results can be obtained as with feature extraction for unrotated I. Feature extraction for I(θ) using P(θ, i) is expressed by the following equation.

I(θ)＝Σx(i)P(i)＝Σx'(i)P(θ,i)　　(13)

In equation (13), x'(i) is the i-th element of feature vector x' extracted using P(θ, i). By calculating the matching score using x'(i), the rotation error contained in x(i) can be cancelled. Hereinafter, feature vector x' may be referred to as corrected feature vector x'. Corrected feature vector x' is an example of corrected feature amount information.

Multiplying equation (13) by P(θ,n) gives the following equation.

(Σx(i)P(i))P(θ,n)
= (Σx′(i)P(θ,i))P(θ,n) (14)

Since P(θ, i) is an orthonormal basis vector, the following equation holds:

P(θ,i)P(θ,n)＝δ(i,n) (15)

δ(i,n) is the Kronecker δ. Therefore, when i=n, δ(i,n)=1, and when i≠n, δ(i,n)=0. Applying equation (15) to equation (14), we obtain the following equation.

x'(n) = (Σx(i)P(i))P(θ,n) (16)

Next, feature extraction for P(θ,n) using P(i) is expressed by the following formula:

P(θ,n)＝Σα(θ,n,i)P(i)　　　(17)

α(θ,n,i) is the i-th element of the feature vector α(θ,n) obtained by performing feature extraction on P(θ,n). α(θ,n,i) is a parameter that indicates how P(i) changes when it is rotated, and has a unique value for each P(i). α(θ,n,i) corresponds to the correction parameter for the feature. Substituting equation (17) into equation (16), the following equation is obtained.

x'(n)
= (Σx(i)P(i)) (Σα(θ,n,i)P(i)) (18)

Since P(i) is an orthonormal basis vector, the following equation holds:

P(i)P(n) = δ(i,n) (19)

By applying equation (19) to equation (18), we obtain the following equation.

x'(n) = Σx(i)α(θ,n,i) (20)

Therefore, x'(n) is expressed as the inner product of feature vector x and feature vector α(θ,n). α(θ,n) is the feature vector of P(θ,n) and can be calculated in advance. α(θ,n) can be obtained numerically by fitting it as a function of θ. α(θ,n,i) is expressed, for example, by the following equation.

α(θ,n,i)＝F(n,i,θ) (21)

F(n, i, θ) is a fitting function. In this case, parameters such as coefficients that define the function form of F(n, i, θ) are stored in the storage unit 917 as fitting information 922. As an example, if F(n, i, θ) is a quadratic function of θ, α(θ, n, i) is expressed by the following equation.

α(θ, n, i)
= p1(n,i) ^θ2 + p2(n,i)θ + p3(n,i) (22)

p1(n,i), p2(n,i), and p3(n,i) are fitting coefficients and are used as fitting information 922. Substituting equation (22) into equation (20), the following equation is obtained.

x'(n)
=Σx(i)(p1(n,i) ^θ2 +p2(n,i)θ+p3(n,i)) (23)

The corrected feature vector x' is a feature vector for the image vector I that represents an unrotated image, so the accuracy of the matching score can be improved by calculating the matching score using x'(n).

The matching score calculation unit 915 calculates α(θ, n, i) using the rotation angle θ included in the coordinate transformation T and the fitting information 922. The matching score calculation unit 915 then calculates a corrected feature vector x' using the feature vector x of each feature point included in the matching target data 924 and α(θ, n, i) according to formula (20).

Next, the matching score calculation unit 915 calculates a matching score using the corrected feature vector x' of each feature point included in the matching target data 924 and the feature vector of each feature point included in the registered data of each registrant. As the matching score, the similarity or dissimilarity of two feature vectors can be used. The similarity may be the inner product of two feature vectors, and the dissimilarity may be the vector distance between two feature vectors.

The matching score calculated using the corrected feature vector x' corresponds to the feature point obtained by transforming the coordinates of either the matching feature point or the registered feature point, and the matching score between the other of the matching feature point and the registered feature point.

FIG. 10 is a flowchart showing an example of the first biometric authentication process performed by the matching score calculation device 901 of FIG. 9. First, the biometric image acquisition unit 911 acquires a biometric image 923 of the person to be matched (step 1001). Then, the feature extraction unit 912 extracts a plurality of feature points from the biometric image 923, calculates a feature vector for each feature point, and generates matching target data 924 including the coordinates and feature vector for each feature point (step 1002).

Next, the coordinate transformation calculation unit 913 generates feature points pairs representing combinations of feature points included in the matching target data 924 and feature points included in the enrollment data of each enrolled person, and generates M (M is an integer equal to or greater than 1) combinations of two feature points pairs. When generating selectable combinations of two feature points pairs from X feature points pairs selected in order of best feature point score, M = _X C _2. Then, the coordinate transformation calculation unit 913 sets 1 to a control variable i (step 1003).

The coordinate transformation calculation unit 913 then selects the i-th combination from the M combinations, and calculates the i-th coordinate transformation Ti by the least squares method using the coordinates of each feature point included in the two feature point pairs corresponding to the selected combination (step 1004).The coordinate transformation calculation unit 913 then checks the validity of the coordinate transformation Ti (step 1005).

If the least squares residual in the calculation of the coordinate transformation Ti is smaller than a predetermined value, the coordinate transformation Ti is determined to be valid, and if the least squares residual is equal to or greater than the predetermined value, the coordinate transformation Ti is determined to be invalid.

If the coordinate transformation Ti is not valid (step 1005, NO), the coordinate transformation calculation unit 913 increments i by 1 (step 1014), and the matching score calculation device 901 repeats the processing from step 1004 onwards.

If the coordinate transformation Ti is valid (step 1005, YES), the coordinate transformation unit 914 uses the coordinate transformation Ti to transform the coordinates of each feature point included in the matching target data 924 (step 1006).

Next, the matching score calculation unit 915 uses the rotation angle θ included in the coordinate transformation Ti and the fitting information 922 to calculate the corrected feature vectors of each feature point included in the matching target data 924 (step 1007).

Next, the matching score calculation unit 915 performs a search process to match feature points included in the matching target data 924 with feature points included in the registered data (step 1008). Then, the matching score calculation unit 915 performs a calculation process to calculate a matching score Si for the coordinate transformation Ti from the matching scores of the two matched feature points (step 1009). The matching score Si is calculated for each registered user.

Next, the matching score calculation unit 915 updates the matching score 925 between the matching target data 924 and the registered data of each registered person (step 1010). At the start of the biometric authentication process, the matching score 925 of each registered person is set to an initial value. If the matching score Si calculated in step 1009 is better than the matching score 925, the matching score calculation unit 915 updates the matching score 925 by setting the matching score Si to the matching score 925.

If the matching score represents the similarity between two feature vectors, the larger the matching score, the better the value. On the other hand, if the matching score represents the dissimilarity between two feature vectors, the smaller the matching score, the better the value.

Next, the coordinate transformation calculation unit 913 increments i by 1 (step 1011) and compares i with M (step 1012). If i is equal to or smaller than M (step 1012, NO), the matching score calculation device 901 repeats the processes from step 1004 onwards.

If i is greater than M (step 1012, YES), the matching score calculation unit 915 performs biometric authentication on the person to be matched using the matching score 925 and generates an authentication result (step 1013).

In biometric authentication, the matching score calculation unit 915, for example, identifies registrants who satisfy a predetermined condition from among multiple registrants registered in the registration template 921. When the matching score represents the similarity, the predetermined condition may be that the matching score 925 is greater than a judgment threshold. When the matching score represents the dissimilarity, the predetermined condition may be that the matching score 925 is less than a judgment threshold.

If a registered person who satisfies the specified conditions is identified, the matching score calculation unit 915 generates an authentication result indicating successful authentication, and if a registered person who satisfies the specified conditions is not identified, it generates an authentication result indicating unsuccessful authentication. This makes it possible to determine whether or not the person to be matched is registered in the registration template 921. The communication unit 916 transmits the generated authentication result to the control device 902.

FIG. 11 is a flowchart showing an example of the search process in step 1008 of FIG. 10. First, the matching score calculation unit 915 initializes the feature point pair list (step 1101), sets the control variable i to 1 (step 1102), and sets the control variable j to 1 (step 1103).

Next, the matching score calculation unit 915 compares the distance D(i,j) between the ith (i=1 to N1) feature point included in the registered data and the jth (j=1 to N2) feature point included in the target data 924 with a threshold TH1 (step 1104). N1 represents the number of feature points included in the registered data, and N2 represents the number of feature points included in the target data 924.

If D(i,j) is greater than or equal to TH1 (step 1104, NO), the matching score calculation unit 915 increments j by 1 (step 1105) and compares j with N2 (step 1106). If j is less than or equal to N2 (step 1106, NO), the matching score calculation unit 915 repeats the processing from step 1104 onwards.

If j is greater than N2 (step 1106, YES), the matching score calculation unit 915 increments i by 1 (step 1107) and compares i with N1 (step 1108). If i is less than or equal to N1 (step 1108, NO), the matching score calculation unit 915 repeats the processing from step 1103 onwards.

If D(i, j) is smaller than TH1 (step 1104, YES), the matching score calculation unit 915 performs the process of step 1109. In step 1109, the matching score calculation unit 915 adds the combination of the i-th feature point included in the registered data and the j-th feature point included in the matching target data 924 to the feature pair list as a corresponding feature pair.

If i is greater than N1 (step 1106, YES), the matching score calculation unit 915 ends the process.

FIG. 12 is a flowchart showing an example of the calculation process in step 1009 of FIG. 10. First, the matching score calculation unit 915 sets the matching score Si for the coordinate transformation Ti to 0, which indicates that the matching is invalid (step 1201).

Next, the matching score calculation unit 915 compares the number of feature points pairs included in the feature points pair list with a threshold value TH2 (step 1202). If the number of feature points pairs is equal to or greater than TH2 (step 1202, NO), the matching score calculation unit 915 determines that the matching is valid.

In this case, the matching score calculation unit 915 calculates a matching score for each feature point pair using the corrected feature vector of the feature point included in the matching target data 924 and the feature vector of the feature point included in the registered data. Then, the matching score calculation unit 915 selects the top K matching scores (K is an integer equal to or greater than 1) in descending order of the best matching score, and sets the statistical value of these K matching scores to Si (step 1203). As the statistical value, the average, median, maximum value, minimum value, etc. are used.

If the number of feature point pairs is smaller than TH2 (step 1202, YES), the matching score calculation unit 915 determines that the matching is invalid and ends the process.

According to the entrance/exit management system of FIG. 9, even if the matching target data 924 to which the normalization process has been applied contains a rotation error, a matching score 925 equivalent to that obtained when no rotation error is contained is obtained, thereby improving the accuracy of the matching score 925.

In addition, after calculating the angle difference θ, there is no need to rotate the biometric image 923 of the person to be matched by -θ and re-extract the feature vector, which makes it possible to suppress increases in processing time and memory consumption, and achieves biometric authentication that is desirable from a security standpoint.

Incidentally, in biometric authentication, as described in Patent Document 3, the Hamming distance is sometimes used to speed up matching processing.

FIG. 13 shows an example of a matching score using the Hamming distance. FIG. 13(a) shows an example of binary data representing a feature. When using the Hamming distance, the feature 1301 is binarized in advance and converted into binary data 1302 consisting of 0s and 1s.

FIG. 13(b) shows an example of a matching score based on binary data. When comparing binary data 1302 contained in the registration data with binary data 1303 contained in the data to be matched, bits of the same digit in binary data 1302 and binary data 1303 are compared. In this case, the number of bits that differ between binary data 1302 and binary data 1303 is calculated as the Hamming distance and used as the matching score.

Specifically, the exclusive OR (XOR) of binary data 1302 and binary data 1303 is calculated for each bit, and the number of logical values "1" contained in the resulting binary data 1304, "9", is found as the Hamming distance.

The Hamming distance can be calculated quickly by using the bit operation function of the CPU (Central Processing Unit). The Hamming distance is particularly fast in CPUs for embedded devices that have poor floating-point arithmetic.

Furthermore, as described in Patent Document 2, features may be encrypted to improve security.

FIG. 14 shows an example of binary data encryption. Encrypted data 1411 is random number data used to encrypt the binary data. Encrypted binary data 1421 is generated by calculating the XOR of binary data 1401 included in the registration data and encrypted data 1411 for each bit. Encrypted binary data 1422 is generated by calculating the XOR of binary data 1402 included in the data to be matched and encrypted data 1411 for each bit.

Encrypted data 1411 is binary data of the same length as binary data 1401 and binary data 1402, and is generated based on AES (Advanced Encryption Standard) or the like.

Due to the characteristics of the XOR operation, if the Hamming distance between binary data 1401 and binary data 1402 is HD, the Hamming distance between encrypted binary data 1421 and encrypted binary data 1422 will also be HD. By using such an encryption method, it becomes possible to calculate the matching score while keeping the features encrypted.

Next, we will explain how to calculate the matching score when the feature vector is encrypted. In this case, since x(i) is encrypted, it is not appropriate to calculate x'(n) using equation (20). Therefore, the following approximation is applied to α(θ, n, i).

α(θ, n, n)≈1 (31)
α(θ, n, i)≒0 (n≠i) (32)

Equation (31) indicates that when n = i, α(θ, n, i) is close to 1. Equation (32) indicates that when n ≠ i, α(θ, n, i) is close to 0.

α(θ,n,n) represents the extent to which the basis vector P(θ,n) rotated by θ contains the original basis vector P(n). On the other hand, α(θ,n,i) represents the extent to which the basis vector P(θ,n) rotated by θ contains basis vectors P(i) other than the original basis vector P(n). For example, when θ is sufficiently small, such as in biometric authentication where normalization processing is applied, P(θ,n) is not considered to deviate significantly from P(n), and this approximation holds true.

By applying the approximation of equation (32) to equation (20) and assuming that α(θ, n, i) = 0 when n ≠ i, the following equation is obtained.

x'(n) = x(n)α(θ, n, n) (33)

α(θ,n,n) can also be considered as a coefficient that indicates the stability of x'(n) against rotation. For example, when α(θ,n,n) is a large value (close to 1), x'(n) is less susceptible to the effects of rotation. Conversely, when α(θ,n,n) is a small value (smaller than 1), x'(n) is more susceptible to the effects of rotation.

Figure 15 shows an example of the relationship between α(θ,n,n) and the rotation angle θ. Curve 1501 represents α(θ,n,n) that is not easily affected by rotation, and curve 1502 represents α(θ,n,n) that is easily affected by rotation. Whether α(θ,n,n) is easily affected by rotation varies depending on the value of n, and is determined by the characteristics of P(i).

Even if x(n) is encrypted, x'(n) can be calculated by using equation (33) instead of equation (20).

As an example, assume that the feature vector x included in the data to be matched and the feature vector included in the registered data are encrypted binary data, and the Hamming distance taking into account the weight is used as the matching score. In this case, the matching score calculation unit 915 calculates the matching score SD between the feature vector x and the feature vector included in the registered data using the following formula.

SD = Σw(i)d(i) (34)

d(i) represents the XOR of the i-th element x(i) of the feature vector x and the i-th element of the feature vector included in the registered data. w(i) represents the weight for d(i), and the right-hand side of equation (34) represents the weighted sum of d(i).

α(θ,i,i) is used as w(i). α(θ,i,i) is considered to be a coefficient that indicates the stability of x'(i) against rotation, so by using α(θ,i,i) as a weight for each d(i), the accuracy of the matching score SD can be improved.

In the following, the method of calculating the matching score using x'(n) in formula (20) may be referred to as the first method, and the method of calculating the matching score SD using formula (34) may be referred to as the second method.

Compared to the first method, the second method has a limited effect on improving accuracy, but has the advantage of being applicable in a wider range of applications. With the second method, the matching score SD can be calculated even if the feature vectors contained in the matching target data and registered data remain encrypted.

It is also possible to calculate the matching score SD using the normalized w(i). Normalizing w(i) allows for a stable calculation of the matching score SD. The weight w'(i) after normalization can be calculated, for example, by the following formula.

w'(i) = w(i) / (Σw(i) / N) (35)

According to w'(i) in equation (35), the sum of w'(1) to w'(N) is a constant value N, so the matching score SD can be prevented from becoming an abnormal value.

FIG. 16 is a flowchart showing an example of the second biometric authentication process performed by the matching score calculation device 901 in FIG. 9. In the biometric authentication process in FIG. 16, the second method is used, and the feature vectors included in the matching target data and the registered data are encrypted binary data.

The processing in steps 1601 to 1606, 1608, and 1610 to 1614 is similar to the processing in steps 1001 to 1006, 1008, and 1010 to 1014 in FIG. 10.

In step 1607, the matching score calculation unit 915 calculates α(θ, i, i) as the weight w(i) in equation (34). The matching score calculation unit 915 calculates α(θ, i, i) using the rotation angle θ included in the coordinate transformation Ti and the fitting information 922.

In step 1609, the matching score calculation unit 915 performs the calculation process of FIG. 12. In this case, in step 1203, the matching score calculation unit 915 calculates the matching score for each feature point pair using equation (34). Then, the matching score calculation unit 915 selects the top K matching scores in descending order of the best matching score, and sets the statistical value of these K matching scores to Si.

In formula (34), the XOR of each element of the encrypted binary data contained in the data to be matched and the registered data is used as d(i). However, instead of the XOR, another dissimilarity or similarity that can be calculated from the encrypted binary data may be used as d(i).

FIG. 17 shows an example of the functional configuration of a flapper gate management system including the matching score calculation device 701 of FIG. 7. The flapper gate management system of FIG. 17 includes a feature acquisition device 1701, a control device 1702, and a matching score calculation device 1703. The feature acquisition device 1701 is, for example, a client, and the matching score calculation device 1703 is, for example, a server. In the flapper gate management system, the feature vector of the person to be matched flows over a communication network, so the second method is used and the feature vector is encrypted.

The matching score calculation device 1703 corresponds to the matching score calculation device 701 in FIG. 7. The flapper gate management system is an example of a matching score calculation system.

The feature acquisition device 1701 is installed near the flapper gate, and includes a biometric image acquisition unit 1711, a feature extraction unit 1712, a communication unit 1713, and a storage unit 1714. The feature extraction unit 1712 is an example of an acquisition unit, and the communication unit 1713 is an example of a transmission unit.

The biometric image acquisition unit 1711 acquires a biometric image 1721 of the person to be matched and stores it in the storage unit 1714. The biometric image 1721 is, for example, a fingerprint image, a palm vein image, a palm print image, or a face image. The biometric image acquisition unit 1711 is, for example, a fingerprint sensor, a vein sensor, or an image sensor.

The feature extraction unit 1712 extracts multiple feature points from the biometric image 1721 and calculates a feature vector from pixel values in the vicinity of each feature point. For example, encrypted binary data is calculated as the feature vector. The feature extraction unit 1712 then generates matching target data 1722 including the coordinates and feature vectors of each of the multiple feature points, and stores the data in the memory unit 1714. The communication unit 1713 transmits the matching target data 1722 to the matching score calculation device 1703 via a communication network.

The matching score calculation device 1703 calculates a matching score using the matching target data 1722, performs biometric authentication on the person to be matched using the matching score, and transmits the authentication result to the control device 1702. The control device 1702 controls the opening and closing of the flapper gate based on the authentication result. For example, if the authentication result indicates success, the control device 1702 controls the opening of the flapper gate, and if the authentication result indicates failure, the control device 1702 controls the closing of the flapper gate.

The matching score calculation device 1703 includes a coordinate transformation calculation unit 1731, a coordinate transformation unit 1732, a matching score calculation unit 1733, a communication unit 1734, and a memory unit 1735. The coordinate transformation unit 1732, the matching score calculation unit 1733, and the memory unit 1735 correspond to the coordinate transformation unit 712, the matching score calculation unit 713, and the memory unit 711 in FIG. 7, respectively. The communication unit 1734 is an example of a receiving unit.

The storage unit 1735 stores a registration template 1741 and fitting information 1742. The registration template 1741 includes registration data for each of a plurality of registered persons, and the registration data for each of the registered persons includes the coordinates and feature vectors of each of a plurality of feature points. For example, encrypted binary data is used as the feature vector. The fitting information 1742 is similar to the fitting information 922 in FIG. 9.

The communication unit 1734 receives the matching target data 1722 from the feature acquisition device 1701, and the storage unit 1735 stores the matching target data 1722. The coordinate transformation calculation unit 1731 calculates a coordinate transformation T for matching the matching target data 1722 with the registered data of each registered person in the registered template 1741. The coordinate transformation T includes, for example, a rotation angle θ and a translation amount (ΔX, ΔY) in the X and Y directions.

The coordinate transformation unit 1732 uses coordinate transformation T to transform the coordinates of each feature point contained in the matching target data 1722.

The matching score calculation unit 1733 calculates α(θ, i, i) using the rotation angle θ included in the coordinate transformation T and the fitting information 1742. The matching score calculation unit 1733 then calculates the matching score 1743 between the matching target data 1722 and the registered data using α(θ, i, i) as the weight w(i) in equation (34).

At this time, the matching score calculation unit 1733 calculates the matching score 1743 using the coordinates after coordinate transformation of each feature point included in the matching target data 1722, the feature vectors of each feature point, and the enrollment data of each enrollee included in the enrollment template 1741. Then, the matching score calculation unit 1733 stores the matching score 1743 in the storage unit 1735.

Next, the matching score calculation unit 1733 performs biometric authentication on the person to be matched using the matching score 1743 and generates an authentication result. The communication unit 1734 transmits the authentication result to the control device 1702 via the communication network.

The processing performed by the feature acquisition device 1701 is similar to the processing in steps 1601 and 1602 in FIG. 16, and the processing performed by the matching score calculation device 1703 is similar to the processing in steps 1603 to 1614 in FIG. 16.

Next, a modified example of the matching score calculation process will be described. F(n, i, θ) in formula (21) does not have to be a quadratic function of θ, and may be a function of the absolute value of θ, |θ|. When F(n, i, θ) is a function of |θ|, α(θ, n, i) is expressed, for example, by the following formula.

α(θ,n,i)＝p4(n,i)｜θ｜+p5(n,i)　　　(36)

p4(n,i) and p5(n,i) are fitting coefficients. By using equation (36), the rotation of P(i) can be more appropriately expressed, which is expected to improve the accuracy of the matching score.

It should be noted that whether to use formula (22) or formula (36) may be switched for each P(i). For example, when i=1, fitting may be performed using formula (22), and when i=2, fitting may be performed using formula (36). For each P(i), fitting may be performed using either formula (22) or formula (36), whichever has the smaller fitting error.

Furthermore, it is possible to switch between the first method and the second method as appropriate. For example, when the angle difference θ is small, the second method may be used to calculate the matching score with sufficiently high accuracy. Therefore, by using the second method, which is simple to calculate, it is possible to speed up the processing. On the other hand, when the angle difference θ is large, the first method may be used to calculate the matching score with higher accuracy.

The matching score calculation device 901 in FIG. 9 or the matching score calculation device 1703 in FIG. 17 may convert the coordinates of the feature points included in the registered data of each registered person instead of the feature points included in the data to be matched. In this case, the matching score calculation device 901 or the matching score calculation device 1703 calculates the matching score using the coordinates after coordinate conversion of each feature point included in the registered data, the feature vectors of each feature point, and the data to be matched, in a calculation method similar to the first method or the second method.

The configuration of the matching score calculation device 701 in FIG. 7 is merely an example, and some of the components may be omitted or changed depending on the use or conditions of the matching score calculation device 701. The configurations of the entrance/exit management system in FIG. 9 and the flapper gate management system in FIG. 17 are merely an example, and some of the components may be omitted or changed depending on the use or conditions of the entrance/exit management system or the flapper gate management system.

The flowcharts in Figures 8, 10 to 12, and 16 are merely examples, and some processing may be omitted or changed depending on the configuration or conditions of the matching score calculation device 701, the entrance/exit management system, or the flapper gate management system.

The feature points and feature amounts shown in FIG. 1, the feature points shown in FIG. 5, and the feature amounts shown in FIG. 6, FIG. 13, and FIG. 14 are merely examples, and the feature points and feature amounts change depending on the biometric image. The normalization process shown in FIG. 2 and FIG. 3 is merely an example, and the normalization process changes depending on the biometric image. The matching process shown in FIG. 4 is merely an example, and the matching process changes depending on the data to be matched and the registered data.

The relationship between α(θ,n,n) and the rotation angle θ shown in Figure 15 is merely an example, and the relationship between α(θ,n,n) and the rotation angle θ changes depending on the characteristics of the basis vector P(i).

Equations (1) to (36) are merely examples, and the entrance/exit management system and the flapper gate management system may use other calculation formulas to perform biometric authentication processing.

FIG. 18 shows an example of the hardware configuration of an information processing device used as the matching score calculation device 701 in FIG. 7, the matching score calculation device 901 in FIG. 9, the feature acquisition device 1701 in FIG. 17, and the matching score calculation device 1703 in FIG. 17.

The information processing device of FIG. 18 includes a CPU 1801, a memory 1802, an input device 1803, an output device 1804, an auxiliary storage device 1805, a medium drive device 1806, and a network connection device 1807. These components are hardware and are connected to each other via a bus 1808. The biometric image acquisition unit 911 of FIG. 9 and the biometric image acquisition unit 1711 of FIG. 17 may be hardware sensors connected to the bus 1808.

Memory 1802 is, for example, a semiconductor memory such as a ROM (Read Only Memory), a RAM (Random Access Memory), or a flash memory, and stores programs and data used in processing. Memory 1802 may operate as memory unit 711 in FIG. 7, memory unit 917 in FIG. 9, memory unit 1714 in FIG. 17, or memory unit 1735 in FIG. 17.

The CPU 1801 (processor) operates as the coordinate conversion unit 712 and the matching score calculation unit 713 in FIG. 7 by, for example, executing a program using the memory 1802.

The CPU 1801 also operates as the feature extraction unit 912, the coordinate transformation calculation unit 913, the coordinate transformation unit 914, and the matching score calculation unit 915 in FIG. 9 by executing a program using the memory 1802.

The CPU 1801 also operates as the feature extraction unit 1712, the coordinate transformation calculation unit 1731, the coordinate transformation unit 1732, and the matching score calculation unit 1733 in FIG. 17 by executing a program using the memory 1802.

The input device 1803 is, for example, a keyboard, a pointing device, etc., and is used to input instructions or information from the operator. The output device 1804 is, for example, a display device, a printer, a speaker, etc., and is used to output inquiries to the operator or processing results. The processing results may be the matching score 925, the matching score 1743, or an authentication result.

The auxiliary storage device 1805 is, for example, a magnetic disk device, an optical disk device, a magneto-optical disk device, a tape device, etc. The auxiliary storage device 1805 may be a hard disk drive or an SSD (Solid State Drive). The information processing device can store programs and data in the auxiliary storage device 1805 and load them into the memory 1802 for use. The auxiliary storage device 1805 may operate as the memory unit 711 in FIG. 7, the memory unit 917 in FIG. 9, the memory unit 1714 in FIG. 17, or the memory unit 1735 in FIG. 17.

The medium drive device 1806 drives the portable recording medium 1809 and accesses the recorded contents. The portable recording medium 1809 is a memory device, a flexible disk, an optical disk, a magneto-optical disk, etc. The portable recording medium 1809 may be a CD-ROM (Compact Disk Read Only Memory), a DVD (Digital Versatile Disk), a USB (Universal Serial Bus) memory, etc. The operator can store programs and data in the portable recording medium 1809 and load them into the memory 1802 for use.

In this way, the computer-readable recording medium that stores the programs and data used in the processing is a physical (non-transitory) recording medium such as memory 1802, auxiliary storage device 1805, or portable recording medium 1809.

The network connection device 1807 is a communication interface circuit that is connected to a communication network such as a WAN (Wide Area Network) or a LAN (Local Area Network) and performs data conversion associated with communication. The information processing device receives programs and data from an external device via the network connection device 1807 and can load them into the memory 1802 for use. The network connection device 1807 may operate as the communication unit 916 in FIG. 9, the communication unit 1713 in FIG. 17, or the communication unit 1734 in FIG. 17.

Note that the information processing device does not need to include all of the components in FIG. 18, and some components may be omitted or modified depending on the application or conditions. For example, if an interface with an operator is not required, the input device 1803 and the output device 1804 may be omitted. If the information processing device does not use the portable recording medium 1809 or a communication network, the medium drive device 1806 or the network connection device 1807 may be omitted.

Although the disclosed embodiments and their advantages have been described in detail, those skilled in the art may make various modifications, additions, and omissions without departing from the scope of the present invention as expressly set forth in the claims.

Claims (16)

1. The computer
   when the coordinate information and feature amount information of a matching feature point are acquired, a coordinate transformation is performed on either the matching feature point or the registered feature point based on an angle difference calculated using the coordinate information of the matching feature point and coordinate information of a registered feature point stored in a storage unit;
   a process of calculating a feature point obtained by coordinate transformation of one of the matching feature point and the registered feature point, and a matching score between the other of the matching feature point and the registered feature point, using the angle difference and feature amount information of the acquired matching feature point.
2. The process of calculating the matching score includes:
   a process of correcting feature amount information of either the matching feature point or the registered feature point using the angle difference;
   calculating the matching score by using corrected feature amount information of either the matching feature point or the registered feature point;
   2. The method for calculating a match score according to claim 1, further comprising:
3. The process of calculating the matching score includes:
   A process of calculating a weight using the angle difference;
   calculating the match score using the weights;
   2. The method for calculating a match score according to claim 1, further comprising:
4. the matching feature points are feature points extracted from a biometric image of a person to be matched,
   the registered feature points are feature points extracted from biometric images of each of a plurality of enrolled persons;
   the process of converting coordinates of either the matching feature point or the registered feature point includes a process of converting coordinates of either the matching feature point or the registered feature point of each of the multiple enrolled persons based on an angle difference calculated using coordinate information of the matching feature point and coordinate information of the registered feature point of each of the multiple enrolled persons stored in the storage unit,
   the process of calculating a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point, and a matching score between the other of the matching feature point and the registered feature point includes a process of calculating a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point of each of the multiple enrolled persons, and a matching score between the matching feature point and the other of the registered feature points of each of the multiple enrolled persons,
   4. The method for calculating a matching score according to claim 1, further comprising: a step of identifying a registrant who satisfies a predetermined condition among the plurality of registrants, based on feature points obtained by coordinate transformation of either the matching feature points or the registered feature points of each of the plurality of registrants, and a matching score between the matching feature points and the other of the registered feature points of each of the plurality of registrants.
5. The matching score calculation method according to claim 4, characterized in that the biometric image of the person to be matched is a palm vein image of the person to be matched, and the biometric image of each of the multiple registered persons is a palm vein image of each of the multiple registered persons.
6. A storage unit that stores coordinate information of the registered feature points;
   a coordinate conversion unit that converts the coordinates of one of the matching feature point and the registered feature point based on an angle difference calculated using the coordinate information of the matching feature point and the coordinate information of the registered feature point when the coordinate information and feature amount information of the matching feature point are acquired;
   a matching score calculation unit that calculates a feature point obtained by performing coordinate transformation on one of the matching feature point and the registered feature point using the angle difference and feature amount information of the acquired matching feature point, and a matching score between the other of the matching feature point and the registered feature point;
   A matching score calculation device comprising:
7. The matching score calculation device according to claim 6, characterized in that the matching score calculation unit uses the angle difference to correct feature amount information of either the matching feature point or the registered feature point, and calculates the matching score using the corrected feature amount information of either the matching feature point or the registered feature point.
8. The matching score calculation device according to claim 6, characterized in that the matching score calculation unit calculates a weight using the angle difference and calculates the matching score using the weight.
9. the matching feature points are feature points extracted from a biometric image of a person to be matched,
   the registered feature points are feature points extracted from biometric images of each of a plurality of enrolled persons;
   the coordinate transformation unit transforms the coordinates of either the matching feature point or the registered feature point of each of the plurality of enrolled persons based on an angle difference calculated using coordinate information of the matching feature point and coordinate information of the registered feature point of each of the plurality of enrolled persons stored in the storage unit;
   9. The matching score calculation device according to claim 6, wherein the matching score calculation unit calculates a matching score between a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point of each of the multiple registrants and the other of the matching feature point and the registered feature point of each of the multiple registrants, and identifies a registrant who satisfies a predetermined condition from among the multiple registrants, based on the matching score between the feature point obtained by coordinate transformation of either the matching feature point or the registered feature point of each of the multiple registrants and the other of the registered feature points of each of the multiple registrants.
10. The matching score calculation device according to claim 9, characterized in that the biometric image of the person to be matched is a palm vein image of the person to be matched, and the biometric image of each of the multiple registered persons is a palm vein image of each of the multiple registered persons.
11. A matching score calculation system including a feature acquisition device and a matching score calculation device,
    The feature acquisition device includes:
    an acquisition unit that acquires coordinate information and feature amount information of a matching feature point;
    a transmission unit that transmits coordinate information and feature amount information of the matching feature points to the matching score calculation device,
    The matching score calculation device includes:
    a receiving unit that receives coordinate information and feature amount information of the matching feature points from the feature acquisition device;
    A storage unit that stores coordinate information of the registered feature points;
    a coordinate conversion unit that converts the coordinates of one of the matching feature point and the registered feature point based on an angle difference calculated using coordinate information of the matching feature point and coordinate information of the registered feature point;
    a matching score calculation unit that calculates a feature point obtained by performing coordinate transformation on one of the matching feature point and the registered feature point using the angle difference and feature amount information of the acquired matching feature point, and a matching score between the other of the matching feature point and the registered feature point.
    A matching score calculation system comprising:
12. when the coordinate information and feature amount information of a matching feature point are acquired, a coordinate transformation is performed on either the matching feature point or the registered feature point based on an angle difference calculated using the coordinate information of the matching feature point and coordinate information of a registered feature point stored in a storage unit;
    a matching score calculation program for causing a computer to execute a process of calculating a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point, and a matching score between the other of the matching feature point and the registered feature point, using the angle difference and feature amount information of the acquired matching feature point.
13. The process of calculating the matching score includes:
    a process of correcting feature amount information of either the matching feature point or the registered feature point using the angle difference;
    calculating the matching score by using corrected feature amount information of either the matching feature point or the registered feature point;
    13. The collation score calculation program according to claim 12, further comprising:
14. The process of calculating the matching score includes:
    A process of calculating a weight using the angle difference;
    calculating the match score using the weights;
    13. The collation score calculation program according to claim 12, further comprising:
15. the matching feature points are feature points extracted from a biometric image of a person to be matched,
    the registered feature points are feature points extracted from biometric images of each of a plurality of enrolled persons;
    the process of converting coordinates of either the matching feature point or the registered feature point includes a process of converting coordinates of either the matching feature point or the registered feature point of each of the multiple enrolled persons based on an angle difference calculated using coordinate information of the matching feature point and coordinate information of the registered feature point of each of the multiple enrolled persons stored in the storage unit,
    the process of calculating a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point, and a matching score between the other of the matching feature point and the registered feature point includes a process of calculating a feature point obtained by coordinate transformation of either the matching feature point or the registered feature point of each of the multiple enrolled persons, and a matching score between the matching feature point and the other of the registered feature points of each of the multiple enrolled persons,
    15. The matching score calculation program according to claim 12, further causing the computer to execute a process of identifying a registrant who satisfies a predetermined condition among the multiple registrants, based on feature points obtained by coordinate transformation of either the matching feature points or the registered feature points of each of the multiple registrants, and a matching score between the matching feature points and the other of the registered feature points of each of the multiple registrants.
16. The matching score calculation program according to claim 15, characterized in that the biometric image of the person to be matched is a palm vein image of the person to be matched, and the biometric image of each of the multiple registered persons is a palm vein image of each of the multiple registered persons.

Images