JP2010181970A - Equipment for biometric identification, biometric identification device, biometric identification system, discrimination reference determination method, biometric identification method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a biometric identification device for improving precision of biometric identification. <P>SOLUTION: The main component of multi-dimensional correlation data related to correlation of a biological pattern acquired from a plurality of biological samples are analyzed. The first main component axis of correlation data distribution related to the same biological sample and the first component axis of correlation data distribution related to different biological samples are determined. One-dimensional data related to each distribution are generated by projecting the multi-dimensional correlation data on the first main component axis of each distribution. A Mahalanobis distance from the center of distribution of each distribution to each point of one-dimensional data is calculated on the first main component axis of each distribution. A determination reference value related to the Mahalanobis distance for discriminating the same biological sample from the different biological samples is determined on the basis of the Mahalanobis distance of each distribution calculated regarding the plurality of biological samples. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a biometric authentication device, a biometric authentication device, a biometric authentication system, a discrimination criterion determination method, a biometric authentication method, and a program.

  In recent years, with the progress of the information society, the value and importance of information held by individuals are rapidly increasing. Under these circumstances, biometric authentication technology (biomatrix technology) has attracted a great deal of attention as a method for realizing robust information management. Biometric authentication is to identify a person or another person using a characteristic part (hereinafter referred to as a biological part) of a human body (biological body). For example, since fingerprints are different between different biometrics, the fingerprints can be used for biometric authentication. Similar to fingerprints, human voiceprints, face shapes, hand shapes, iris patterns, vein patterns, and the like have different characteristics among different living bodies. Therefore, it is possible to specify an individual by using these feature amounts for biometric authentication, and to perform authentication processing, search processing, and the like.

  Thus, in order to specify an individual using biometric authentication, or to perform an authentication process, a search process, or the like, it is necessary to perform a comparison process of feature amounts acquired from a biological part. Therefore, it is acquired in the form of data (for example, image data, audio data, three-dimensional coordinate data, iris code, etc.) that can compare feature quantities (for example, fingerprints, voiceprints, vein patterns, etc.) of living body parts. Next, “reference data” registered in advance by the person in such a format is compared with “input data” input at the time of the authentication operation by some method, and similarity is measured. Then, identification of individuals, authentication processing, and the like are performed based on the similarity obtained as a result of the comparison.

  Regarding biometric authentication, Patent Document 1 below determines whether a biometric pattern detected by a biometric sensor is a living body or a non-living body before performing personal authentication using a biometric pattern. Techniques to do this are disclosed. In particular, this document discloses a technique for discriminating between a living body and a non-living body by capturing a unique statistical tendency found in a living body pattern. For example, the blood vessel pattern of a living body tends to align in a certain direction. With regard to this tendency, this document discloses a method for discriminating a biological pattern and a non-biological pattern based on the angular distribution and distribution intensity of each line segment forming the blood vessel pattern and eliminating the pseudo blood vessel pattern or the like according to the discrimination result. Has been proposed. By using such a method, biometric authentication can be performed with higher accuracy.

JP 2008-102780 A

  The technique described in the above document refers to a boundary for discriminating a biological pattern and a non-biological pattern from a distribution of shape values represented by two or more index values with reference to two or more index values related to the shape value of the biological pattern. It relates to a technique for determining a value. The technology also relates to a technology for discriminating between a living body and a non-living body using the boundary value. In particular, in this technique, the spread of the distribution is taken into consideration, and the boundary value is expressed by the Mahalanobis distance from the distribution center. However, this technology does not consider the directionality of the distribution spread.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to determine the reference value used for the person identification by taking into account the directionality of the distribution spread. It is an object of the present invention to provide a new and improved biometric authentication device, biometric authentication device, biometric authentication system, discrimination criterion determination method, biometric authentication method, and program capable of further improving the accuracy of authentication.

  In order to solve the above-described problem, according to one aspect of the present invention, principal component analysis is performed on correlation data of a plurality of dimensions related to correlation of biological patterns acquired from a plurality of biological samples, and correlation data distribution of the same biological sample is analyzed. A principal component analysis unit that determines a first principal component axis and a first principal component axis of a correlation data distribution relating to different biological samples; and a first of each distribution in which the multi-dimensional correlation data is determined by the principal component analysis unit. A one-dimensional data generation unit that generates one-dimensional data relating to each distribution by projecting onto the principal component axis, and from the distribution center of each distribution to each point of the one-dimensional data on the first principal component axis of each distribution A Mahalanobis distance calculation unit that calculates the Mahalanobis distance of the distribution, and a Mahalanobis distance of each distribution calculated by the Mahalanobis distance calculation unit for the plurality of biological samples. Can, and a determination reference value determining unit for determining a discriminant reference value for the Mahalanobis distance for discriminating between said different biological samples and the same biological sample, biometric authentication apparatus is provided.

  In addition, the biometric authentication device may include a first correlation value related to the feature point arrangement of each biometric pattern and a slope of a line segment connecting the feature points of each biometric pattern with respect to the biometric pattern acquired from the plurality of biometric samples. There may be further provided a sample data storage unit in which the second correlation value is recorded. In this case, the principal component analysis unit performs principal component analysis on the set of the first correlation value and the second correlation value recorded in the sample data storage unit as the multi-dimensional correlation data.

  The one-dimensional data generation unit is configured to perform coordinate conversion on a coordinate system having the first principal component axis as one coordinate axis for the multi-dimensional correlation data, and to perform coordinate conversion by the coordinate conversion unit. And a first principal component extraction unit that extracts a component of the first principal component axis and generates one-dimensional data from the multi-dimensional correlation data subjected to.

  The biological pattern may be any one or a combination of a vein pattern, a fingerprint pattern, and an iris pattern.

  In order to solve the above-mentioned problem, according to another aspect of the present invention, a biometric authentication sensor that acquires a biometric pattern from a predetermined biometric part, a registered pattern storage unit in which a registered biometric pattern is recorded, The first principal component axis of the correlation data distribution related to the same biological sample and the correlation data related to different biological samples determined by principal component analysis of the first dimension data of multiple dimensions related to the correlation of biological patterns acquired from a plurality of biological samples. Each of the distributions is projected onto the first principal component axis of the distribution by projecting multi-dimensional second correlation data relating to the correlation between the biometric pattern acquired by the biometric authentication sensor and the biometric pattern recorded in the registered pattern storage unit. A one-dimensional data generation unit for generating one-dimensional data relating to the distribution, and the first order from the distribution center of each distribution on the first principal component axis of each distribution A Mahalanobis distance calculation unit for calculating a Mahalanobis distance to one-dimensional data generated by the data generation unit; a biometric pattern acquired by the biometric sensor based on the Mahalanobis distance calculated by the Mahalanobis distance calculation unit; and the registration There is provided a biometric authentication device including a similarity determination unit that determines similarity between biometric patterns recorded in a pattern storage unit.

  In order to solve the above problem, according to another aspect of the present invention, the first biological data of a plurality of dimensions related to the correlation of biological patterns acquired from a plurality of biological samples is subjected to principal component analysis, and the same biological sample is obtained. A principal component analysis unit for determining a first principal component axis of the correlation data distribution relating to the correlation data distribution and a first principal component axis of the correlation data distribution relating to different biological samples, and the principal component analysis unit determining the first correlation data of the plurality of dimensions A first one-dimensional data generation unit that projects the first principal component axis of each distribution generated to generate first one-dimensional data related to each distribution, and the respective ones on the first principal component axis of each distribution A first Mahalanobis distance calculator for calculating a first Mahalanobis distance from each distribution center of the distribution to each point of the first one-dimensional data; and the first Mahalanobis distance with respect to the plurality of biological samples. A discriminant reference value determining unit for determining a discriminant reference value for the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the first Mahalanobis distance of each distribution calculated in the output unit; A biometric authentication device having a biometric authentication sensor that acquires a biometric pattern from a predetermined biometric part, a registered pattern storage unit that stores a registered biometric pattern, and a correlation data distribution of the same biosample A plurality of correlations between a biometric pattern acquired by the biometric sensor and a biometric pattern recorded in the registered pattern storage unit with respect to a main component axis and a first principal component axis of a correlation data distribution related to the different biological samples. Second one-dimensional data for generating second one-dimensional data relating to each distribution by projecting the second-dimensional correlation data. A generating unit; a second Mahalanobis distance calculating unit that calculates a second Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the second one-dimensional data; Based on the second Mahalanobis distance calculated by the second Mahalanobis distance calculation unit and the discrimination reference value determined by the discrimination reference value determination unit, the biometric pattern acquired by the biometric authentication sensor and the registered pattern storage unit There is provided a biometric authentication system including: a biometric authentication device having a similarity determination unit that determines similarity between recorded biometric patterns.

  In order to solve the above problem, according to another aspect of the present invention, principal component analysis is performed on correlation data of a plurality of dimensions related to correlation of biological patterns acquired from a plurality of biological samples, and correlations regarding the same biological sample are performed. A principal component analysis step of determining a first principal component axis of the data distribution and a first principal component axis of the correlation data distribution relating to different biological samples, and each distribution in which the multi-dimensional correlation data is determined in the principal component analysis step A one-dimensional data generation step for generating one-dimensional data relating to each distribution by projecting onto the first principal component axis of the distribution, and from the distribution center of each distribution on the first principal component axis of each distribution, The Mahalanobis distance calculation step for calculating the Mahalanobis distance to each point and the Mahalanobis distance calculation step for the plurality of biological samples are calculated. And a discrimination reference value determining step for determining a discrimination reference value relating to the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the Mahalanobis distance of each distribution. Provided.

  In order to solve the above-described problem, according to another aspect of the present invention, the same is determined by principal component analysis of multi-dimensional first correlation data relating to correlation of biological patterns acquired from a plurality of biological samples. Acquired using a biometric authentication sensor for acquiring a biological pattern from a predetermined biological site with respect to the first principal component axis of the correlation data distribution related to the biological sample and the first principal component axis of the correlation data distribution related to different biological samples. A one-dimensional data generation step for projecting multi-dimensional second correlation data relating to the correlation between the biometric pattern and the registered biometric pattern to generate one-dimensional data relating to each distribution, and a first principal component axis of each distribution The Mahalano distance calculating the Mahalanobis distance from the distribution center of each distribution to the one-dimensional data generated in the one-dimensional data generation step. Similarity for determining similarity between the biometric pattern acquired using the biometric authentication sensor and the registered biometric pattern based on the Mahalanobis distance calculated in the mahalanobis distance calculating step A biometric authentication method including a determination step.

  In order to solve the above problem, according to another aspect of the present invention, the first biological data of a plurality of dimensions related to the correlation of biological patterns acquired from a plurality of biological samples is subjected to principal component analysis, and the same biological sample is obtained. A principal component analysis step for determining a first principal component axis of the correlation data distribution relating to the correlation data distribution and a first principal component axis of the correlation data distribution relating to different biological samples, and determining the first correlation data of the plurality of dimensions in the principal component analysis step. A first one-dimensional data generation step of generating first one-dimensional data relating to each distribution by projecting onto the first principal component axis of each distribution, and each of the distributions on the first principal component axis of each distribution A first Mahalanobis distance calculating step of calculating a first Mahalanobis distance from each distribution center of the distribution to each point of the first one-dimensional data; A discriminant criterion for determining a discriminant criterion value for the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the first Mahalanobis distance of each distribution calculated in the first Mahalanobis distance calculating step. In order to acquire a biological pattern from a predetermined biological part with respect to the value determining step, the first principal component axis of the correlation data distribution related to the same biological sample, and the first principal component axis of the correlation data distribution related to the different biological sample A second one for generating second one-dimensional data relating to each of the distributions by projecting a plurality of second-dimensional correlation data relating to the correlation between the biometric pattern acquired using the biometric authentication sensor and the registered biometric pattern. Dimensional data generation step, and from the distribution center of each distribution on the first principal component axis of each distribution, A second Mahalanobis distance calculating step for calculating a second Mahalanobis distance to the dimension data, a second Mahalanobis distance calculated in the second Mahalanobis distance calculating step, and a determination determined in the determination reference value determining step There is provided a biometric authentication method including a similarity determination step of determining a similarity between a biometric pattern acquired using the biometric authentication sensor based on a reference value and the registered biometric pattern.

  In order to solve the above problem, according to another aspect of the present invention, principal component analysis is performed on correlation data of a plurality of dimensions related to correlation of biological patterns acquired from a plurality of biological samples, and correlations regarding the same biological sample are performed. A principal component analysis function for determining a first principal component axis of data distribution and a first principal component axis of correlation data distribution relating to different biological samples, and each distribution in which the multi-dimensional correlation data is determined by the principal component analysis function A one-dimensional data generation function for generating one-dimensional data relating to each distribution by projecting onto the first principal component axis of the distribution, and the distribution of the one-dimensional data from the distribution center of each distribution on the first principal component axis of each distribution Mahalanobis distance calculation function for calculating the Mahalanobis distance to each point, and the Maha Nobis distance calculation function for the plurality of biological samples calculated by the Mahalanobis distance calculation function. Based on Nobis distance, a program for implementing a determination reference value determining function of determining the discrimination reference value for the Mahalanobis distance for determining said it said same biological sample different biological samples to the computer is provided.

  In order to solve the above-described problem, according to another aspect of the present invention, the same is determined by principal component analysis of multi-dimensional first correlation data relating to correlation of biological patterns acquired from a plurality of biological samples. Acquired using a biometric authentication sensor for acquiring a biological pattern from a predetermined biological site with respect to the first principal component axis of the correlation data distribution related to the biological sample and the first principal component axis of the correlation data distribution related to different biological samples. A one-dimensional data generation function for projecting multi-dimensional second correlation data relating to the correlation between the biometric pattern and the registered biometric pattern to generate one-dimensional data relating to each distribution, and a first principal component axis of each distribution The Mahalanobis distance for calculating the Mahalanobis distance from the distribution center of each distribution to the one-dimensional data generated by the one-dimensional data generation function. A similarity determination function for determining a similarity between a biometric pattern acquired using the biometric sensor and the registered biometric pattern based on the Mahalanobis distance calculated by the output function and the Mahalanobis distance calculation function A program for causing a computer to realize the above is provided.

  In order to solve the above problem, according to another aspect of the present invention, a computer-readable recording medium in which the above program is recorded can be provided.

  As described above, according to the present invention, it is possible to further improve the accuracy of identity authentication by determining the reference value used for identity determination in consideration of the directionality of the distribution spread.

It is explanatory drawing which shows one structural example of the biometrics authentication system which concerns on one Embodiment of this invention. It is explanatory drawing which shows the example of 1 structure of the biometric sensor which concerns on the embodiment. It is explanatory drawing which shows the function structural example of the pre-processing apparatus which concerns on the same embodiment. It is explanatory drawing which shows the whole flow of the pre-processing which concerns on the same embodiment. It is explanatory drawing which shows the flow of a part of pre-processing which concerns on the embodiment. It is explanatory drawing which shows an example of the principal distribution and others distribution calculated in the pre-processing which concerns on the embodiment. It is explanatory drawing which shows the flow of a part of pre-processing which concerns on the embodiment. It is a conceptual diagram for demonstrating the principal component analysis method which concerns on the same embodiment. It is explanatory drawing which shows the flow of a part of pre-processing which concerns on the embodiment. It is explanatory drawing which shows an example of the threshold straight line determined in the pre-processing which concerns on the embodiment. It is explanatory drawing which shows the function structural example of the biometrics apparatus which concerns on the embodiment. It is explanatory drawing which shows the flow of the authentication process which concerns on the same embodiment. It is explanatory drawing which shows the whole flow of the pre-processing and authentication process in the biometric authentication system which concerns on the embodiment. It is explanatory drawing which shows notionally the whole flow of the pre-process in the biometrics system which concerns on the embodiment, and an authentication process. It is explanatory drawing which shows the result of having calculated Mahalanobis distance with respect to original data. It is explanatory drawing which shows the data distribution after performing coordinate transformation by principal component analysis. It is explanatory drawing which shows the result of having calculated the Mahalanobis distance with respect to the data distribution after performing coordinate transformation by principal component analysis. It is explanatory drawing which shows the two-dimensional plot result of the Mahalanobis distance calculated with respect to the data after the coordinate transformation by a principal component analysis. It is explanatory drawing which shows the hardware structural example of the apparatus which concerns on the same embodiment.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

[About the flow of explanation]
Here, the flow of explanation regarding the embodiment of the present invention described below will be briefly described. First, the configuration of the biometric authentication system 10 according to the embodiment will be described with reference to FIG. Next, a configuration of a vein authentication sensor which is an example of the biometric sensor 100 according to the embodiment will be described with reference to FIG. Next, a functional configuration of the determination criterion determination device 110 according to the embodiment will be described with reference to FIG.

  Next, with reference to FIGS. 4 to 10, a method for determining a discrimination reference value used for personal authentication according to the embodiment will be described. First, referring to FIG. 4, an overall processing flow regarding the method will be described. Next, a correlation value calculation method will be described with reference to FIGS. Next, a principal component axis calculation method will be described with reference to FIGS. Next, a method for calculating the Mahalanobis distance will be described with reference to FIGS. In this description, a method for determining the discrimination reference value will be described with reference to FIG.

  Next, the functional configuration of the biometric authentication device 130 according to the embodiment will be described with reference to FIG. Next, a flow of biometric authentication processing according to the embodiment will be described with reference to FIG. Next, an overall processing flow including determination criterion value determination processing and authentication processing according to the embodiment will be described with reference to FIGS. 13 and 14. Next, effects obtained by applying the technique according to the embodiment will be described with reference to FIGS. Next, a hardware configuration example of the discrimination criterion determination device 110 and the biometric authentication device 130 according to the embodiment will be described.

(Description item)
1: Configuration of Biometric Authentication System 1-1: Configuration of Biometric Authentication Sensor 100 1-2: Functional Configuration of Discrimination Criteria Determination Device 110 1-3: Functional Configuration of Biometric Authentication Device 130 1-4: According to Biometric Authentication System 10 Overall Process Flow 2: Effect Obtained by Applying Biometric Authentication System 3: Hardware Configuration of Discrimination Criteria Determination Device 110 and Biometric Authentication Device 130

<Embodiment>
An embodiment of the present invention will be described. In this embodiment, in order to realize more accurate biometric authentication, a method for determining a reference value capable of discriminating between a person and another person with higher accuracy is proposed. In addition, a method for performing biometric authentication using the discrimination reference value determined by the same method is also proposed. By applying these methods, biometric authentication can be performed with high accuracy.

[1: Configuration of biometric authentication system 10]
First, the configuration of the biometric authentication system 10 according to the present embodiment will be described with reference to FIG. FIG. 1 is an explanatory diagram illustrating a configuration example of a biometric authentication system 10 according to the present embodiment.

  As shown in FIG. 1, the biometric authentication system 10 mainly includes a biometric authentication sensor 100, a discrimination criterion determining device 110, and a biometric authentication device 130. The biometric sensor 100 is a means for acquiring biometric information from a biological part. Further, the discrimination criterion determination device 110 is a means for determining a discrimination reference value for discriminating whether or not two pieces of biological information are of the same living body. Then, the biometric authentication device 130 collates the biometric information (reference data) registered in advance for each user with the biometric information (input data) input when performing biometric authentication based on the discrimination reference value, and the input data It is a means to specify the user corresponding to. The discrimination criterion determination device 110 is an example of a biometric authentication device.

  In the biometric authentication system 10, first, biometric information is acquired from a large number of biological samples using the biometric authentication sensor 100. In the following description, the biological information acquired from the biological sample at this stage will be referred to as sample data. These sample data are input to the discrimination criterion determining device 110. However, it is assumed that the biological sample (input user) corresponding to each sample data is known, and each sample data and information on each biological sample are recorded in association with each other in the discrimination criterion determination device 110. Based on the large number of sample data, the discrimination criterion determination device 110 determines a discrimination criterion value.

  Information including the discrimination criterion value determined by the discrimination criterion determination device 110 (hereinafter, the identity determination parameter) is input to the biometric authentication device 130. In addition, reference data is input to the biometric authentication device 130 in advance using the biometric authentication sensor 100. When input data is input from the biometric authentication sensor 100 to the biometric authentication device 130, the biometric authentication device 130 collates each reference data with the input data using the input identity determination parameter. As a result of this collation processing, when reference data having a degree of similarity exceeding the criterion value is detected, the biometric authentication device 130 determines that the user corresponding to the reference data is the user corresponding to the input data.

  Thus, in biometric authentication, the accuracy of collation between input data and reference data affects the accuracy of biometric authentication. Therefore, in order to increase the authentication accuracy, it is necessary to devise a method for determining that the same user's biometric information is more accurately the same and determining that the different users' biometric information is more accurately different. Normally, in biometric authentication, a fingerprint image, a vein pattern image, or the like is input as biometric information, and threshold determination is performed by quantifying the degree of similarity of such image data.

  Then, the identity of the two image data to be compared is determined according to the result of the threshold determination. At this time, if the distribution of the values quantified between the result of comparing the biometric patterns of the same user and the result of comparing the biometric patterns of different users cannot be clearly distinguished, an erroneous determination result is obtained by the threshold determination. The probability of being increased. As a result, the authentication accuracy is degraded. Therefore, in this embodiment, a method for quantifying biological information is proposed so that both distributions are clearly distinguished. This quantification method is realized by the discrimination criterion determination device 110 and the biometric authentication device 130 described later.

  Hereinafter, the biometric authentication sensor 100, the discrimination criterion determining device 110, and the biometric authentication device 130 included in the biometric authentication system 10 will be described in more detail.

(1-1: Configuration of Biometric Sensor 100)
First, the biometric sensor 100 will be described. In the biometric authentication system 10, the user inputs biometric information acquired from a predetermined biometric site using the biometric authentication sensor 100 and performs biometric authentication. As biometric information, for example, a biometric pattern that appears in a vein, fingerprint, iris, face, voiceprint, or the like is used. In the present embodiment, any biological pattern including a characteristic shape value unique to each user can be used.

  The biometric authentication sensor 100 is a unit that acquires a biometric pattern from a biometric part when the user places or places a predetermined biometric part in the sensing area. The sensing area referred to here is an area suitable for acquiring a biological pattern from a placed biological part. For example, a sensing area is provided between the light irradiation unit and the imaging unit. In this case, sensing is performed so that light is efficiently emitted toward the living body part placed in the sensing area, and light that is transmitted through the living body part or scattered inside the living body part is efficiently incident on the imaging unit. The position of the area is set.

(About vein authentication sensor)
Here, referring to FIG. 2, a vein authentication sensor is taken as an example of the biometric authentication sensor 100, and the configuration and operation of the vein authentication sensor will be specifically described. FIG. 2 is an explanatory diagram showing the main configuration and operation of the vein authentication sensor. FIG. 2 also shows a captured image (A) of the living body part (finger FG) acquired by the vein authentication sensor and a vein image (B) obtained by performing image processing on the captured image (A). It is published.

  As shown in FIG. 2, the vein authentication sensor mainly includes a near-infrared light irradiation light source (LED 102) and an image sensor 104. The vein authentication sensor images a living body part (for example, a finger FG) placed on a sensing area (such as a gantry), and generates a captured image of a vein existing inside the living body part (vein image (B)). However, the LED is an abbreviation for Light Emitting Diode. In the example of FIG. 2, the LED is used as the near-infrared light irradiation light source, but the application range of the present embodiment is not limited to this, and a light source other than the LED may be used.

  The LED 102 is a means for irradiating a living body part placed in the sensing area with near infrared light in a predetermined wavelength band. Near-infrared light has a characteristic of being easily absorbed by hemoglobin (reduced hemoglobin) in blood while being highly permeable to body tissues. Therefore, when near-infrared light is irradiated to a living body part (for example, a finger, palm, back of hand, etc.), veins distributed inside the living body part are observed as shadows. The vein shadow observed in this way is called a vein pattern. The vein pattern is an example of the biological pattern. In order to image such a vein pattern well, it is preferable to use near-infrared light having a predetermined wavelength band of about 600 nm to 1300 nm (preferably about 700 nm to 900 nm).

  For example, when the wavelength of near-infrared light emitted by the LED 102 is less than 600 nm or more than 1300 nm, the proportion of the near-infrared light absorbed by hemoglobin in blood becomes small, and a good vein pattern cannot be obtained. On the contrary, when the wavelength of the near infrared light irradiated by the LED 102 is about 700 nm to 900 nm, the near infrared light is specifically absorbed by both deoxygenated hemoglobin and oxygenated hemoglobin, A good vein pattern is obtained. In addition, the near infrared which can irradiate near-infrared light of such a wavelength band by combining LED102 which can emit light including the said wavelength band, and the band pass filter (not shown) which passes the said wavelength band is combined. A light source can be easily realized.

  The near-infrared light (irradiation light L11) emitted from the LED 102 as described above is scattered into the vein inside the finger FG and enters the image sensor 104 as scattered light L13. Then, the imaging element 104 generates a captured image (A) with the incident scattered light L13. As the imaging element 104, for example, a charge coupled devices (CCD) or a complementary metal oxide semiconductor (CMOS) is used. The captured image (A) output from the image sensor 104 is subjected to predetermined image processing by an image processing unit (not shown) to generate a vein image (B). The predetermined image processing performed here is, for example, edge detection processing, smoothing processing, binarization processing, thinning processing, and the like.

  As described above, the vein authentication sensor irradiates the finger FG with the irradiation light L11 by the LED 102, images the scattered light L13 that has passed through the inside of the finger FG with the imaging element 104, and displays a predetermined image on the captured image (A). By performing image processing, a vein image (B) is generated (S10). The vein image (B) generated in this way is input as biometric information to the discrimination criterion determining device 110 or the biometric authentication device 130 (see FIG. 1). However, biometric information (sample data) acquired from a biological sample is input to the discrimination criterion determination device 110, and biometric information (authentication data; reference data, input data) to be authenticated is input to the biometric authentication device 130. Is done.

(1-2: Functional Configuration of Discrimination Criteria Determination Device 110)
Next, the discrimination criterion determination device 110 will be described. First, the functional configuration of the discrimination criterion determining apparatus 110 according to the present embodiment will be described with reference to FIG. FIG. 3 is an explanatory diagram showing a functional configuration example of the discrimination criterion determining apparatus 110 according to the present embodiment.

  As shown in FIG. 3, the discrimination criterion determination device 110 mainly includes a storage unit 112, a correlation calculation unit 114, a principal component analysis unit 116, a Mahalanobis distance calculation unit 118, and a threshold line determination unit 120. . As described above, a large number of sample data is input from the biometric sensor 100 to the discrimination criterion determination device 110. These input sample data are recorded in the storage unit 112 and used to determine the discrimination reference value.

  The storage unit 112 is an example of a sample data storage unit. The principal component analysis unit 116 is an example of a one-dimensional data generation unit, a coordinate conversion unit, and a first principal component extraction unit. The Mahalanobis distance calculation unit 118 is an example of a Mahalanobis distance calculation unit. The threshold straight line determination unit 120 is an example of a discrimination reference value determination unit.

(Correlation calculator 114)
First, the correlation calculation unit 114 will be described. The correlation calculation unit 114 reads a large number of sample data recorded in the storage unit 112. The sample data recorded in the storage unit 112 is, for example, image data obtained by photographing a biological pattern such as a vein pattern and performing image processing S10. This image data is converted into the coordinate information of the feature points included in the biological pattern and the information of line segments connecting between the feature points (hereinafter, line segment information) in the image processing S10. In other words, the coordinate information and line segment information of the feature points forming each sample data are recorded in the storage unit 112. Therefore, the correlation calculation unit 114 calculates a pixel correlation value and a Hough correlation value, which will be described later, using the coordinate information and line segment information of each sample data.

(About pixel correlation values)
The pixel correlation value here means a correlation coefficient calculated using coordinate information of sample data corresponding to two biological patterns. For example, the correlation coefficient is defined by the following equation (1). However, the coordinates of each pixel are expressed as (m, n), and the pixel information of the two biological samples to be compared is expressed as f ₁ and f ₂ . For example, if there is a feature point of sample data corresponding to the pixel information f ₁ at the coordinates (10, 15), it is expressed as f ₁ (10, 15) = 1. On the other hand, when there is no feature point of sample data corresponding to the pixel information f ₁ at the coordinates (100, 2), it is expressed as f ₁ (100, 2) = 0.

The correlation coefficient S ₁ (f ₁ , f ₂ ) shown in the following equation (1) is a statistical index indicating the similarity between the _two sample data f ₁ and f ₂ . As the correlation coefficient S ₁ is close to 1, two sample data are similar. Conversely, the closer the correlation coefficient S ₁ is 0, two sample data are different. Therefore, if the two sample data f ₁ and f ₂ are completely coincident with each other, the value of the correlation coefficient S ₁ is 1. That is, the correlation coefficient S ₁ of sample data acquired from the same biological sample is a value close to 1. Of course, the correlation coefficient S ₁ calculated using the same sample data is 1. On the other hand, the correlation coefficient S ₁ of the sample data obtained from different biological samples is expected to be sufficiently smaller than 1.

…（１）

... (1)

(About Hough correlation value)
On the other hand, the above-mentioned Hough correlation value is a phase calculated using the following equation (2) for each feature point on the Hough space obtained by Hough transforming each line segment (straight line) of the biological pattern. which means that the relationship between the number of S _2. The Hough transform is, for example, a method in which a straight line L defined in the xy space is expressed by two parameters r and θ, and the straight line L is converted to one point in the r-θ space. For example, when the conversion equation is expressed as r = x · cos θ + y · sin θ, the coordinate point (x ₀ , y ₀ ) on the straight line L is converted into a curve r = x ₀ · cos θ + y ₀ · sin θ on the r-θ space. The Similarly, for k = 1,..., N−1, the coordinate point (x _k , y _k ) on the straight line L is converted to a curve r = x _k · cos θ + y _k · sin θ on the r-θ space.

At this time, since (x _k , y _k ) is on the straight line L, N curves r = x _{k ′} · cos θ + y _{k ′} · sin θ (k ′ = 0,..., N) drawn on the r-θ space. -1) intersects at one point. This point is a feature point on the Hough space (r-θ space) that characterizes the straight line L. For example, the line segment L ₁₂ connecting the feature points (x _k , y _k ) (k = 1, 2) of the biological pattern is represented by r = x _k · cos θ + y _k · sin θ (k = 1, 2) on the Hough space. Characterized by the intersection (r ₁₂ , θ ₁₂ ). Thus, by using the Hough transform, each line segment connecting the feature points of the biological pattern is obtained as the distribution of the feature points in the Hough space.

For example, the coordinates of the Hough space (m ', n') and expressed, the Hough transform values characterizing each segment for two sample data to be compared is expressed as hf _1, hf _2. However, Hough conversion value _{hf 1 (m ', n'} ) is Hough transform values corresponding to the line segment of the sample data is the coordinates _{(m ', n') hf} 1 when present in (m ', n') = 1, and hf ₁ (m ′, n ′) = 0 when there is no such value. Similarly, the Hough transform value _{hf 2 (m ', n'} ) is Hough transform values corresponding to the line segment of the sample data is the coordinates _{(m ', n') hf} 2 (m when present in ', n' ) = 1, and hf ₂ (m ′, n ′) = 0 when it does not exist. With such a representation, the Hough correlation value S ₂ can be expressed as the following equation (2).

…（２）

... (2)

  As described above, the correlation calculation unit 114 calculates pixel correlation values and Hough correlation values for all sample data recorded in the storage unit 112. When the distribution of (pixel correlation value, Hough correlation value) calculated by the correlation calculation unit 114 is two-dimensionally plotted, a distribution diagram illustrated in FIG. 6 is obtained. The correlation calculation unit 114 holds identification information for distinguishing between the pixel correlation value and the Hough correlation value calculated for the same biological sample and the pixel correlation value and the Hough correlation value calculated for different biological samples. Yes. Therefore, in the distribution diagram illustrated in FIG. 6, it is possible to distinguish between a distribution related to the same biological sample (hereinafter referred to as “personal distribution”) and a distribution related to different biological samples (hereinafter referred to as “other person distribution”).

  As a result, it is possible to determine a boundary line (person identification threshold value) that separates the individual distribution from the other person distribution from the distribution diagram of FIG. However, as is apparent from the distribution chart illustrated in FIG. 6, the identity distribution and the identity distribution are slightly overlapped in the vicinity of the boundary line. The result is likely to contain errors.

  Therefore, in the present embodiment, the discrimination reference value is not determined from the distribution of the pixel correlation value-the Hough correlation value, and further converted into a distribution form in which the identity distribution and the other person distribution can be clearly distinguished, and then the discrimination reference value Make a decision. Therefore, the pixel correlation value and the Hough correlation value calculated by the correlation calculation unit 114 are input to the principal component analysis unit 116 (see FIG. 3). Further, identification information for distinguishing the personal distribution from the other person distribution is input from the correlation calculation unit 114 to the principal component analysis unit 116.

(Principal component analysis unit 116)
Next, the principal component analysis unit 116 will be described. Referring to the distribution chart illustrated in FIG. 6, it can be seen that the spread of the personal distribution and the spread of the other person have different directions. Therefore, in the present embodiment, a method is proposed in which the distribution form is converted so that the person distribution and the other person distribution can be more clearly distinguished in consideration of the difference in directionality of both distributions. In particular, in the present embodiment, the distribution of pixel correlation value-Hough correlation value (hereinafter referred to as PH distribution) is rotated in the principal component direction using principal component analysis, and the PH distribution of the sample data is projected onto the first principal component axis. Thus, a method of converting to a one-dimensional distribution is proposed. Hereinafter, this method will be described in detail.

(About principal component analysis)
First, the principal component analysis will be briefly described. In the principal component analysis, as shown in FIG. 8, the direction in which the spread of the point group formed by a large number of points is the largest is the first principal component axis, is orthogonal to the first principal component axis, and This is an analysis method for extracting the direction with the next largest group spread as the second principal component axis. This analysis technique can be used for multidimensional distributions as well. In the case of a point group having a spread in an N-dimensional space, N principal component axes can be extracted by principal component analysis. That is, a new coordinate system defined by the principal component axis is formed by principal component analysis.

Here, a specific calculation method of the principal component axis will be briefly described. For simplicity, a point group Q distributed on a two-dimensional space (xy space) as shown in FIG. 8 is considered. In order to obtain the principal component axis, the eigenvalues λ ₁ and λ ₂ and the eigenvectors E ₁ and E ₂ of the variance-covariance matrix C regarding the point group Q may be obtained. However, eigenvalues λ ₁ and λ ₂ are scalar quantities, and eigenvectors E ₁ and E ₂ are two-dimensional vector quantities. The expected value of the coordinate point (x _k , y _k ) (k = 0,..., N−1) constituting the point group Q is expressed as (μ _x , μ _y ) (where μ _x = Σx _k / N, μ _y = Σy _k / N, Σ: sum regarding k = 0 to N−1), the variance-covariance matrix C is expressed by the following equations (3) and (4).

…（３）

…（４）

... (3)

... (4)

Therefore, the eigenvalues lambda _1, lambda ₂ is obtained by solving the characteristic equation represented by the following formula (5). Here, I represents a unit matrix of 2 rows and 2 columns. Det (...) represents a determinant. Further, eigenvectors E ₁ and E ₂ are obtained from the following relational expressions (6) and (7) and the conditional expression | E ₁ | = | E ₂ | = 1. Since each principal component axis passes through the center of gravity P ₀ of the point group Q, if the eigenvectors E ₁ and E ₂ are expressed as E ₁ = (e ₁₁ , e ₁₂ ), E ₂ = (e ₂₁ , e ₂₂ ), λ _1> lambda case _2, the first principal component axis is expressed by equation (8) below, the second principal component axis is represented by the formula (9) below.

…（５）

…（６）

…（７）

…（８）

…（９）

... (5)

... (6)

... (7)

... (8)

... (9)

  As described above, the principal component axis of the point group Q can be calculated by performing calculations related to the above equations (3) to (9). In the case of this embodiment, a PH distribution (personal distribution, distribution of others) is given as the point group Q. Further, the principal component axis is calculated for each of the principal distribution and the other person distribution using the above calculation method.

(Application to this embodiment)
First, principal component analysis unit 116 calculates the expected value of the personal distribution _{Q 1} inputted from the correlation calculating unit 114 (μ1x, μ1y). Next, the principal component analysis unit 116 calculates the variance-covariance matrix C ₁ of the principal distribution Q ₁ based on the above equations (3) and (4). Next, the principal component analysis unit 116 solves the eigen equation corresponding to the above equation (5) for the variance-covariance matrix C ₁ and calculates eigen values λ ₁₁ and λ ₁₂ . Next, the principal component analysis unit 116 calculates eigenvectors E ₁₁ and E ₁₂ corresponding to the eigenvalues λ ₁₁ and λ ₁₂ based on the above equations (6) and (7), respectively. As a result, the principal component axis is determined based on the above equations (8) and (9).

Similarly, the principal component analysis unit 116 calculates the expected values of others distributions _{Q 2} to which is input from the correlation calculating unit 114 (μ2x, μ2y). Then, principal component analysis unit 116 calculates the covariance matrix _{C 2} of the principal distribution _{Q 2} based on the above equation (3) and (4). Next, the principal component analysis unit 116 solves the eigen equation of the above equation (5) for the variance-covariance matrix C ₂ and calculates eigen values λ ₂₁ and λ ₂₂ . Next, the principal component analysis unit 116 calculates eigenvectors E ₂₁ and E ₂₂ corresponding to the eigenvalues λ ₂₁ and λ ₂₂ based on the above equations (6) and (7), respectively. As a result, the principal component axis is determined based on the above equations (8) and (9).

In this way, the principal component axis is calculated for each of the principal distribution and the other person distribution. When the main component axis of each distribution is calculated, the principal component analysis unit 116, each point of the point group Q _1, Q ₂ is projected to the first principal component axis of each distribution point group Q _1, Q ₂ Reduce dimension to one dimension. Specifically, the coordinate point group Q _1, Q ₂ is converted to principal component direction of the distributions, only the coordinate values corresponding to the first principal component is extracted.

For example, it is assumed that the inclination of the first principal component axis calculated for the principal distribution Q ₁ is θ ₁ . In this case, the coordinate points (x _k , y _k ) (k = 0,..., N−1) included in the principal distribution Q ₁ and the other person distribution Q ₂ have the first principal component axis as the p1x axis and the second principal component. It is converted into a coordinate point (p1x _k , p1y _k ) in a new coordinate system with the axis as the p1y axis. The conversion formula of this coordinate conversion is expressed by the following formula (10). After this coordinate transformation, the principal component analysis unit 116 extracts coordinate points p1x _k after the conversion as a one-dimensional data DQ ₁ related person distribution _{Q 1.}

Similarly, it is assumed that the inclination of the first principal component axis calculated for others distribution Q ₂ is theta _2. In this case, the coordinate points (x _k , y _k ) (k = 0,..., N−1) included in the principal distribution Q ₁ and the other person distribution Q ₂ have the first principal component axis as the p2x axis and the second principal component. It is converted to a coordinate point (p2x _k , p2y _k ) in a new coordinate system with the axis as the p2y axis. The conversion formula of this coordinate conversion is expressed by the following formula (11). After this coordinate transformation, the principal component analysis unit 116 extracts coordinate points p2x _k after the conversion as a one-dimensional data DQ ₂ regarding others distributions _{Q 2.}

…（１０）

…（１１）

(10)

... (11)

As described above, each point included in the PH distribution, the principal component analysis unit 116, and is converted for the principal distribution _{Q 1} one-dimensional data DQ _1, and the one-dimensional data DQ ₂ for others distributions _{Q 2.} The one-dimensional data DQ ₁ and DQ ₂ converted by the principal component analysis unit 116 are input to the Mahalanobis distance calculation unit 118.

(Mahalanobis distance calculator 118)
As described above, the Mahalanobis distance calculation unit 118, each of the principal first principal component axis distribution Q _1, and PH distribution is projected to each of the first principal component axis of others distributions Q ₂ by principal component analysis unit 116 One-dimensional data DQ ₁ and DQ ₂ corresponding to the points are input. Therefore, the Mahalanobis distance calculation unit 118 calculates the Mahalanobis distance between each point and the distribution center for each of the one-dimensional data DQ ₁ and DQ ₂ . Here, the Mahalanobis distance will be briefly described.

  For example, when there are a large number of sample points that are expected to belong to a set, the possibility that a sample point belongs to the set is considered to be higher as the sample point is closer to the center of gravity of all the sample points. From this idea, it is considered that the possibility that each sample point belongs to the set can be determined based on the Euclidean distance from the center of gravity to the sample point. However, it is very difficult to determine the maximum Euclidean distance that should be determined that a sample point belongs to a set. Therefore, when considering the possibility of belonging to a set for each sample point, a Mahalanobis distance, which is a kind of distance taking into account statistics such as the density and spread of the sample points, is used. For example, the Mahalanobis distance is often used when clustering data.

For example, the observation point is x = (x ₁ ,..., X _p ) ^T , the center of the point group Q in the p-dimensional space is μ = (μ ₁ ,..., Μ _p ) ^T , and the variance covariance of the point group Q When the matrix is expressed as C, the p-dimensional Mahalanobis distance D _m (x) related to the observation point x is defined by the following equation (12). Further, when the number of dimensions p = 1, the variance-covariance matrix C becomes variance σ ² , so the one-dimensional Mahalanobis distance D _m (x) is expressed as in the following equation (13). Where σ represents the standard deviation of the point group Q.

…（１２）

…（１３）

(12)

... (13)

In the present embodiment, the Mahalanobis distance is calculated for the one-dimensional data DQ ₁ and DQ ₂ related to the principal distribution Q ₁ and the stranger distribution Q ₂ input from the principal component analysis unit 116. Thus, when the calculation target of the Mahalanobis distance is one-dimensional, the above equation (13) is used. First, the Mahalanobis distance calculation unit 118 calculates standard deviations σ ₁ and σ ₂ for each of the one-dimensional data DQ ₁ and DQ ₂ . Next, the Mahalanobis distance calculation unit 118 calculates the Mahalanobis distances D _m1 and D _m2 corresponding to the one-dimensional data DQ ₁ and DQ ₂ based on the above equation (13). The Mahalanobis distances D _m1 and D _m2 respectively corresponding to the principal distribution Q ₁ and the other person distribution Q ₂ calculated by the Mahalanobis distance calculation unit 118 are input to the threshold straight line determination unit 120.

(Threshold line determination unit 120)
Now, when the Mahalanobis distances D _m1 and D _m2 are plotted as two-dimensional data, a distribution diagram illustrated in FIG. 10 is obtained. As is clear from the distribution diagram of FIG. 10, the distribution QM ₁ Mahalanobis distance corresponding to the principal distribution Q _1, and the distribution QM ₂ Mahalanobis distance corresponding to others distribution Q _2, widely separated. Compared with the PH distribution shown in FIG. 6, the degree of separation of the distributions QM ₁ and QM ₂ in FIG. 10 is very large. The reason for indicating such a degree of separation is that the principal component analysis unit 116 projects the coordinate points on the first principal component axis of each distribution and considers the correlation between the data using the Mahalanobis distance.

The Mahalanobis distance assumes that the data distribution is symmetric with respect to the center of gravity. Therefore, even if the distributions QM ₁ and QM ₂ are calculated by applying the Mahalanobis distance to each point of the PH distribution, the degree of separation of the distributions QM ₁ and QM ₂ as shown in FIG. 10 cannot be obtained. However, in the present embodiment, the principal component analysis unit 116, the direction of the differences existing between the principal distribution Q _1, others distribution Q ₂ is considered. For this reason, the distributions QM ₁ and QM ₂ related to the Mahalanobis distance have an effect due to the consideration of the difference in directionality, and the identity distribution Q ₁ and the other person distribution Q ₂ have been clearly separated.

Therefore, the threshold straight line determination unit 120 determines a threshold straight line (y = m * x) at a position where the identity distribution QM ₁ and the stranger distribution QM ₂ can be clearly separated in the distribution map regarding the Mahalanobis distance. This threshold straight line becomes the discrimination reference value. Further, as the discrimination reference value, for example, the slope m of the threshold line may be determined. As shown in FIG. 10, when the technique of the present embodiment is used, since the principal distribution QM ₁ and the stranger distribution QM ₂ are clearly separated, a certain amount of width is allowed for the slope of the threshold line. Therefore, any method may be used for determining the threshold line as long as the person distribution QM ₁ and the other person distribution QM ₂ can be sufficiently distinguished.

For example, the inclination of the threshold line can be determined as a straight line passing through the midpoint of the line segment connecting the center point of the principal distribution QM _{1 and} the center point of the other person distribution QM ₂ . Also, the point closest to the axis (y-axis) corresponding to the Mahalanobis distance D _m2 among the points included in the principal distribution QM ₁ and the axis corresponding to the Mahalanobis distance D _m1 among the points included in the other-person distribution QM ₂ The threshold straight line can be determined as a straight line passing through the midpoint of the line segment connecting the point closest to (x-axis). In the present embodiment, in addition to these exemplified methods, any method may be adopted as the threshold straight line determination method as long as the person distribution QM ₁ and the other person distribution QM ₂ can be sufficiently distinguished. be able to.

As described above, the identity distribution Q ₁ (QM ₁ ) and the stranger distribution Q ₂ (QM ₂ ) are clarified by the correlation calculation unit 114, the principal component analysis unit 116, the Mahalanobis distance calculation unit 118, and the threshold line determination unit 120. A discrimination reference value that can be discriminated in a short time is determined. The discrimination reference value (for example, the slope m of the threshold line) determined by the threshold line determination unit 120 is input to the biometric authentication device 130. The biometric authentication device 130 also receives information indicating the inclinations θ ₁ and θ ₂ of the first principal component axes of the principal distribution Q ₁ and the other person distribution Q ₂ from the principal component analysis unit 116.

  Heretofore, the functional configuration of the discrimination criterion determining apparatus 110 according to the present embodiment has been described in detail for each component. Therefore, the overall processing flow by the discrimination criterion determining apparatus 110 will be described below. Also, in this description, the detailed processing flow in each processing step will be briefly described.

(Regarding the flow of processing by the discrimination criterion determination device 110)
First, with reference to FIGS. 4, 5, 7, and 9, the flow of processing by the discrimination criterion determining apparatus 110 will be briefly described. FIG. 4 is an explanatory diagram showing the overall processing flow by the discrimination criterion determining apparatus 110. FIG. 5 is an explanatory diagram showing the flow of processing for calculating the pixel correlation value and the Hough correlation value. FIG. 7 is an explanatory diagram showing a flow of processing for generating one-dimensional data using principal component analysis. FIG. 9 is an explanatory diagram showing the flow of the Mahalanobis distance calculation process.

(Overall flow)
First, referring to FIG. As shown in FIG. 4, first, sample data is input from the biometric sensor 100 and recorded in the storage unit 112 (S102). Next, the correlation calculation unit 114 calculates a pixel correlation value and a Hough correlation value (PH distribution) (S104). Next, the principal component analysis is performed by the principal component analysis unit 116, and each point of the PH distribution is projected onto one-dimensional data on the first principal component axis for each of the principal distribution Q1 and the other person distribution Q2 (S106). Next, the Mahalanobis distance calculation unit 118 calculates the Mahalanobis distance based on the one-dimensional data converted by the principal component analysis unit 116 (S108). Next, the threshold line determining unit 120 determines a threshold line (discrimination reference value) based on the Mahalanobis distance (S110).

  Hereinafter, the processes of steps S104, S106, and S108 will be described in more detail.

(Flow of PH distribution calculation process)
Reference is now made to FIG. FIG. 5 shows the detailed processing flow of step S104 in FIG. As shown in FIG. 5, first, pixel correlation values are calculated using the pixel values of feature points between vein pattern images of sample data (S122). Next, each line segment connecting the feature points included in the vein pattern of the sample data is subjected to Hough transform (S124). Next, the correlation coefficient of the Hough transform value is calculated as the Hough correlation value (S126). Next, a set of pixel correlation values and Hough correlation values for each sample data is output (S128).

(Principal component analysis and one-dimensional processing flow)
Reference is now made to FIG. FIG. 7 shows the detailed processing flow of step S106 in FIG. As shown in FIG. 7, first, the coordinate values of the distribution centers of the principal distribution and the stranger distribution in the space (PH space) around the pixel correlation value and the Hough correlation value are calculated (S130). Next, a covariance matrix for each distribution is calculated (S132). Next, eigenvalues and eigenvectors of the covariance matrix corresponding to each distribution are calculated (S134). Next, the first principal component axis of each distribution is calculated based on the eigenvalue and eigenvector calculated for each distribution (S136). Next, each distribution is converted into a coordinate system having the first principal component axis as a horizontal axis (S138). Next, the sample data is projected onto each horizontal axis after coordinate conversion, and one-dimensional data for each distribution is generated (S140).

(Mahalanobis distance calculation process flow)
Reference is now made to FIG. FIG. 9 shows the detailed processing flow of step S108 in FIG. As shown in FIG. 9, first, a standard deviation for one-dimensional data of each distribution is calculated (S142). Next, the Mahalanobis distance is calculated using the expected value and standard deviation regarding the one-dimensional data of each distribution (S144). However, the expected value for the one-dimensional data of each distribution has already been calculated as the distribution center of each distribution. That is, since the principal component axis passes through the distribution center and is not changed by coordinate transformation, the coordinate value of each distribution center can be used as the expected value of each one-dimensional data.

  The flow of processing by the discrimination criterion determination device 110 has been described above. The contents of the processing by the discrimination criterion determining apparatus 110 have been described in detail together with the functional configuration. Further, the flow of processing by each component is as described above. As described above, in the present embodiment, the discrimination reference value for discriminating between the principal distribution and the other person distribution is determined based on the Mahalanobis distance in consideration of the directionality of the distribution by principal component analysis. Hereinafter, a method of performing biometric authentication using this discrimination reference value will be described.

(1-3: Functional Configuration of Biometric Authentication Device 130)
Next, the functional configuration of the biometric authentication device 130 according to the present embodiment will be described with reference to FIG. FIG. 11 is an explanatory diagram illustrating a functional configuration example of the biometric authentication device 130 according to the present embodiment. It is assumed that the biometric information of the registered user (hereinafter, registered data) is input to the biometric authentication device 130 as “reference data”. In addition, a discrimination reference value or the like is input from the discrimination reference determination device 110.

  The biometric authentication device 130 is a means for collating biometric information (input data) input using the biometric authentication sensor 100 with registered data and specifying a registered user corresponding to the input data. In particular, the biometric authentication device 130 performs a discrimination process using the discrimination reference value determined by the discrimination criteria determination device 110 when authenticating the user by comparing the input data with the registration data. That is, in the biometric authentication device 130, the pixel correlation value and the Hough correlation value between the input data and the registration data are projected onto the first principal component axis of the individual distribution and the other person distribution calculated by the discrimination criterion determination device 110. The identity between the data is determined using the Mahalanobis distance of the obtained one-dimensional data. Hereinafter, each functional configuration of the biometric authentication device 130 will be described.

  As shown in FIG. 11, the biometric authentication device 130 mainly includes a correlation calculation unit 132, a Mahalanobis distance calculation unit 134, an identity determination unit 136, and a storage unit 138. The storage unit 138 stores registration data input using the biometric sensor 100. In addition, the storage unit 138 stores the discrimination reference value input from the discrimination criterion determination device 110, the inclination for specifying the first principal component axis of the individual / others distribution, the distribution center of each distribution, Information including a standard deviation and the like regarding the one-dimensional data (hereinafter referred to as coordinate conversion parameters) is stored.

  The storage unit 138 is an example of a registered pattern storage unit. The Mahalanobis distance calculation unit 134 is an example of a one-dimensional data generation unit and a Mahalanobis distance calculation unit. The identity determination unit 136 is an example of a similarity determination unit.

(Correlation calculator 132)
First, the correlation calculation unit 132 will be described. As described above, the input data to be authenticated is input to the correlation calculation unit 132 using the biometric sensor 100. When input data is input, the correlation calculation unit 132 reads registration data recorded in the storage unit 138. Then, the correlation calculation unit 132 calculates a pixel correlation value and a Hough correlation value using the coordinate information and line segment information of the input data and the registration data. The pixel correlation value is calculated based on the above equation (1). The Hough correlation value is calculated based on the above equation (2). The correlation calculation unit 132 calculates a pixel correlation value and a Hough correlation value between the input data and the registration data for a part or all of the registration data stored in the storage unit 138. The pixel correlation value and the Hough correlation value calculated by the correlation calculation unit 132 are input to the Mahalanobis distance calculation unit 134.

(Mahalanobis distance calculator 134)
Next, the Mahalanobis distance calculation unit 134 will be described. As described above, the correlation value indicating the pixel correlation and the Hough correlation between the input data and the registered data is input to the Mahalanobis distance calculation unit 134 from the correlation calculation unit 132. When these correlation values are input, the Mahalanobis distance calculation unit 134 projects the input correlation values onto the first principal component axis of the principal distribution and the other person distribution used when calculating the discrimination reference value, and generates one-dimensional data. Convert.

  In order to perform this conversion processing, first, the Mahalanobis distance calculation unit 134 reads information on the inclination of the first principal component axis recorded in the storage unit 138, and laterally converts the first principal component axis of the personal distribution and the other person distribution. Convert the coordinate system of the correlation value to the coordinate system used as the axis. Next, the Mahalanobis distance calculation unit 134 extracts the first principal component axis component from the correlation value after the coordinate conversion to obtain one-dimensional data. When the one-dimensional data is calculated in this way, the Mahalanobis distance calculating unit 134 calculates the one-dimensional data from the distribution center based on the information indicating the distribution center of the individual distribution and the other person distribution recorded in the storage unit 138 and the standard deviation of each distribution. Calculate the Mahalanobis distance to the data. The Mahalanobis distance calculated by the Mahalanobis distance calculation unit 134 is input to the identity determination unit 136.

(Principal determination unit 136)
Next, the person determination unit 136 will be described. As described above, the identity determination unit 136 calculates the correlation between the input data and the registered data from the Mahalanobis distance calculation unit 134 on each first principal component axis from each distribution center of the individual distribution and the other person distribution. Mahalanobis distance is entered. When the Mahalanobis distance is input, the identity determination unit 136 determines whether the calculation result regarding the correlation between the input data and the registered data belongs to the identity distribution based on the discrimination reference value recorded in the storage unit 138 and the input Mahalanobis distance. It is determined whether it belongs to another person distribution.

When the discrimination reference value m illustrated in FIG. 10 (threshold line y = m * x) is given, the Mahalanobis distance (x ₀ , y ₀₎ calculated based on the correlation value between the input data and the registered data. ) Is y ₀ <m * x ₀ , the identity determination unit 136 determines that it belongs to the other person distribution. Conversely, if the Mahalanobis distance (x ₀ , y ₀ ) calculated based on the correlation value between the input data and the registered data is y ₀ > m * x ₀ , the identity determination unit 136 belongs to the identity distribution. Is determined. When it is determined that it belongs to the personal distribution, the personal determination unit 136 determines that the input data and the registration data are the same. On the other hand, if it is determined that the person belongs to the other person distribution, the person determination unit 136 determines that the input data and the registration data are not identical.

  The functional configuration of the biometric authentication device 130 according to the present embodiment has been described above. As described above, the biometric authentication device 130 does not determine the identity between the input data and the registered data on the pixel correlation value-Hough correlation value space, but is projected onto the first principal component axis of the person distribution and the other person distribution. The identity is determined based on the correlation value. In particular, the identity between the two data to be compared is determined based on the Mahalanobis distance between the correlation value projected on the first principal component axis of the principal distribution and the other person distribution and the distribution center of each distribution. As a result, the determination accuracy can be improved.

(About the flow of processing by the biometric authentication device 130)
Here, the flow of processing by the biometric authentication device 130 will be briefly described with reference to FIG. FIG. 12 is an explanatory diagram showing the flow of processing by the biometric authentication device 130.

  As shown in FIG. 12, the biometric authentication device 130 first calculates a pixel correlation value and a Hough correlation value between authentication data (input data) input from the biometric authentication sensor 100 and registration data (S150). . Next, the biometric authentication device 130 performs coordinate conversion of both correlation values in the principal component directions of the principal distribution and the stranger distribution, and calculates one-dimensional data by extracting the components of the first principal component axes (S152). Next, the biometric authentication device 130 calculates the Mahalanobis distance between the calculated one-dimensional data and each distribution center of the person distribution and the other person distribution (S154). Next, the biometric authentication device 130 performs threshold determination based on the determination reference value using the calculated Mahalanobis distance, and determines the identity between the input data and the registered data (S156). Then, the biometric authentication device 130 outputs the determination result as a result of biometric authentication (S158).

  In accordance with the flow described above, the biometric authentication device 130 executes an authentication process.

(1-4: Overall Process Flow by Biometric Authentication System 10)
Next, an overall processing flow by the biometric authentication system 10 will be described with reference to FIGS. 13 and 14, and an overall configuration of the biometric authentication processing according to the present embodiment will be summarized. FIG. 13 is an explanatory diagram for explaining the overall processing flow by the biometric authentication system 10. FIG. 14 is an explanatory diagram conceptually showing the contents of each process executed in the biometric authentication system 10.

  As shown in FIG. 13, first, a principal distribution and a stranger distribution composed of two-dimensional correlation data are acquired using a large amount of sample data (S162). Next, the first principal component axis is determined for each of the principal distribution and the others distribution (CS1 in FIG. 14). Furthermore, each correlation data is coordinate-transformed in the principal component direction (CS2 in FIG. 14), and projected onto each first principal component axis (CS3 in FIG. 14), thereby generating one-dimensional data corresponding to each distribution (CS3 in FIG. 14). S164). Next, the Mahalanobis distance between the one-dimensional data generated on each first principal component axis and each distribution center is calculated (CS4 in FIG. 14), and each Mahalanobis distance of each distribution is calculated in a coordinate space with an orthogonal axis. A threshold straight line for distinguishing the distribution is determined (S166).

  Next, two-dimensional correlation data indicating the correlation between the authentication data and the registration data is acquired (S168). Next, two-dimensional correlation data is subjected to coordinate transformation in the principal component directions of the above-described principal distribution and others distribution, and each first principal component is extracted to generate one-dimensional data (S170). Next, the Mahalanobis distance between the distribution center of each distribution and the one-dimensional data is calculated (S172; CS5 in FIG. 14). Next, identity determination is performed based on the Mahalanobis distance and the threshold line for each distribution (S174). The discrimination reference value is calculated as described above, and identity determination is performed using the calculated discrimination reference value.

[2: Effects obtained by applying the biometric authentication system 10]
Next, effects obtained by applying the biometric authentication system 10 according to the present embodiment will be described with reference to FIGS. 6 and 15 to 18. The experimental data shown here is sample data obtained by photographing each finger a plurality of times using 100 fingers as biological samples. Therefore, the experimental data includes 100 pieces of two-dimensional correlation data of the individual distribution indicating the correlation of the same living body and 18800 pieces of two-dimensional correlation data of the other person distribution indicating the correlation of different living bodies.

(Distribution characteristics of original data)
First, referring to FIG. As already mentioned, for example, pixel correlation values and Hough correlation values are used to characterize the correlation between sample data. Then, the person distribution and the other person distribution are plotted in a space having the pixel correlation value and the Hough correlation value as coordinate axes, respectively, and the determination criterion for the person or the other person is determined based on the characteristics of each distribution. As shown in FIG. 6, in the pixel correlation value-Hough correlation value distribution (PH distribution), the distribution center of the principal distribution and the distribution center of the stranger distribution are separated from each other. Can do.

  However, both the person distribution and the other person distribution have a large spread, and it cannot be said that the boundary between the person distribution and the other person distribution is clear. In FIG. 6, the temporarily determined threshold for determining the person is drawn, but data belonging to the person distribution is included in the area of the other person distribution. That is, a determination error occurs for this data. Thus, even if the distribution centers are separated, when there is a spread from the distribution center, it is not clear how far the distribution center can be said to be the range of the set corresponding to the distribution center. Therefore, if a threshold straight line is determined in the PH distribution based on the Euclidean distance from each distribution center, the above determination error occurs.

  As described above, when the threshold determination criterion is provided by directly using the PH distribution, a very high determination accuracy cannot be expected. Therefore, the first possible countermeasure is to use the Mahalanobis distance considering the spread of each distribution.

(Principal / other comparison of Mahalanobis distance to original data)
Therefore, with reference to FIG. 15, the distribution obtained when the Mahalanobis distance is directly calculated from the PH distribution will be considered. FIG. 15A is a plot of the Mahalanobis distance from the distribution center of the other person distribution to all the data points included in the PH distribution. Similarly, FIG. 15B is a plot of Mahalanobis distances from the distribution center of the principal distribution to all data points included in the PH distribution. The horizontal axis is an index indicating what number data point is, and the vertical axis indicates the Mahalanobis distance.

  If the Mahalanobis distance clearly distinguishes the person distribution from the other person distribution, the data point of the person distribution and the data point of the other person distribution are clearly divided into two regions with a certain Mahalanobis distance as a boundary. It should be. That is, if it is possible to clearly distinguish the personal distribution from the other person distribution by the Mahalanobis distance, the region including the data point of the personal distribution and the region including the data point of the other person distribution are shown in the distribution diagram of FIG. It will be clearly separated by a horizontal line located at a certain height.

  With the above consideration in mind, reference is made to the distribution diagram of FIG. In FIG. 15A, the area A1 including the data points of the personal distribution and the area A2 including the data points of the other person distribution cannot be clearly separated by a horizontal line having a certain height. Also in FIG. 15B, the data point of the principal distribution and the data point of the other person distribution are included in the region B1 corresponding to the same Mahalanobis distance. From these results, it can be seen that even if the Mahalanobis distance is simply applied to the PH distribution, the person distribution and the other person distribution cannot be clearly distinguished.

  It is considered that the Mahalanobis distance characteristic is related to the background from which such a result was obtained. As already mentioned, the Mahalanobis distance is a distance that takes into account the spread of the distribution. However, the spread of the distribution considered in the Mahalanobis distance assumes a symmetric distribution. Therefore, if the distribution shape is sufficiently asymmetric, a determination error will occur if an attempt is made to determine the distribution attribute based on the Mahalanobis distance. Further, as shown in FIG. 6, in the PH distribution of biometric information used in biometric authentication, directionality is observed in each of the identity distribution and the other person distribution. Therefore, the threshold used for authentication cannot be determined with high accuracy simply by applying the Mahalanobis distance to the PH distribution.

(Data plot after coordinate transformation by principal component analysis)
In view of this, a method using principal component analysis is devised in order to consider the directionality of the principal distribution and the other person distribution. FIG. 16 shows a PH distribution (A) coordinate-converted in the principal component direction of the other person distribution and a PH distribution (B) coordinate-transformed in the principal component direction of the other person distribution. As shown in FIG. 16, when coordinate conversion is performed by principal component analysis, the data points of the principal distribution and the data points of the other person distribution can be clearly distinguished along the first principal component axis.

  For example, referring to FIG. 16A, the stranger distribution is all included in the region A1 among the regions A1 and A2 clearly distinguished in the first principal component direction. Similarly, the identity distribution is all included in the region A2 among the regions A1 and A2 clearly distinguished in the first principal component direction. Further, referring to FIG. 16B, the stranger distribution is all included in the region B1 among the regions B1 and B2 clearly distinguished in the first principal component direction. Similarly, the identity distribution is all included in the region B2 among the regions B1 and B2 clearly distinguished in the first principal component direction.

  As can be understood from the results of FIG. 16, both distributions can be clearly distinguished in the first principal component direction by considering the directionality of the principal distribution and the other person distribution. However, when only the principal component analysis is used, the spread of the distribution is not considered. Therefore, in order to consider the spread of the distribution, a method of applying the Mahalanobis distance can be considered. However, in order to apply the Mahalanobis distance, the shape of the distribution needs to be symmetric.

  Therefore, in the present embodiment, as described above, a method has been proposed in which PH distribution data points are projected onto the first principal component axis to generate one-dimensional data, and the Mahalanobis distance is applied to the one-dimensional data. . If one-dimensional data is used, spatial asymmetry (difference between the spread in the first principal component axis direction and the spread in the second principal component axis direction) is eliminated, so that the Mahalanobis distance can be correctly applied.

(Principal / other comparison of Mahalanobis distance for one-dimensional data)
As described above, FIG. 17 shows the result of calculating the Mahalanobis distance from the center of distribution to each data point after converting each data point of the PH distribution into one-dimensional data.

  FIG. 17A plots the Mahalanobis distance from the center of the other person's distribution to the one-dimensional data corresponding to each data point of the PH distribution. Similarly, FIG. 17B plots the Mahalanobis distance from the center of the principal distribution to the one-dimensional data corresponding to each data point of the PH distribution. Referring to FIG. 17A, it can be seen that the identity distribution and the stranger distribution are clearly distinguished into regions A1 and A2 divided by a certain Mahalanobis distance. Similarly, referring to FIG. 17B, it can be seen that the principal distribution and the stranger distribution are clearly distinguished into regions B1 and B2 divided by a certain Mahalanobis distance.

  Therefore, when a two-dimensional plot is made with the Mahalanobis distance measured from the distribution center of the other person's distribution on the vertical axis and the Mahalanobis distance measured from the distribution center of the principal distribution on the horizontal axis, a distribution chart as shown in FIG. 18 is obtained. As shown in FIG. 18, the Mahalanobis distance distribution calculated after projecting the data points in the direction of the first principal component axis of each distribution can clearly distinguish the personal distribution from the other person distribution. Compared with the PH distribution shown in FIG. 6, it is understood that the distance between the principal distribution and the other person distribution in the distribution chart of FIG. 18 is very large. The magnitude of this distance indicates the level of authentication accuracy in biometric authentication.

  In this way, it is possible to greatly improve the authentication accuracy in biometric authentication by highlighting the statistical properties between the person distribution and the other person distribution and clarifying the difference between the person and the other person. In the present embodiment, the statistical properties are emphasized in consideration of the difference in directionality appearing between the principal distribution and the distribution of others and the spread of each distribution, so that each distribution can be clearly distinguished as described above. A method was proposed. By using this method, the effect of significantly improving the authentication accuracy in biometric authentication can be obtained.

[3: Hardware Configuration of Discrimination Criteria Determination Device 110 and Biometric Authentication Device 130]
The functions of the constituent elements of the discrimination criterion determining device 110 and the biometric authentication device 130 can be realized by using, for example, the hardware configuration of the information processing apparatus illustrated in FIG. For example, the function of each component is realized by controlling the information processing apparatus shown in FIG. 19 using a computer program. In addition, the form of the information processing apparatus shown here is arbitrary, For example, personal information terminals, such as a personal computer, a mobile telephone, PHS, PDA, a game machine, or various information household appliances are contained in this. However, the above PHS is an abbreviation of Personal Handy-phone System. The PDA is an abbreviation for Personal Digital Assistant.

  As shown in FIG. 19, this information processing apparatus mainly includes a CPU 902, a ROM 904, a RAM 906, a host bus 908, a bridge 910, an external bus 912, an interface 914, an input unit 916, and an output unit. 918, a storage unit 920, a drive 922, a connection port 924, and a communication unit 926. However, the CPU is an abbreviation for Central Processing Unit. The ROM is an abbreviation for Read Only Memory. The RAM is an abbreviation for Random Access Memory.

  The CPU 902 functions as, for example, an arithmetic processing unit or a control unit, and controls the overall operation of each component or a part thereof based on various programs recorded in the ROM 904, the RAM 906, the storage unit 920, or the removable recording medium 928. . The ROM 904 is a means for storing a program read by the CPU 902, data used for calculation, and the like. In the RAM 906, for example, a program read by the CPU 902, various parameters that change as appropriate when the program is executed, and the like are temporarily or permanently stored.

  These components are connected to each other via, for example, a host bus 908 capable of high-speed data transmission. On the other hand, the host bus 908 is connected to an external bus 912 having a relatively low data transmission speed via a bridge 910, for example. As the input unit 916, for example, a mouse, a keyboard, a touch panel, a button, a switch, a lever, or the like is used. Further, as the input unit 916, a remote controller (hereinafter referred to as a remote controller) capable of transmitting a control signal using infrared rays or other radio waves may be used.

  As the output unit 918, for example, a display device such as a CRT, LCD, PDP, or ELD, an audio output device such as a speaker or a headphone, a printer, a mobile phone, or a facsimile, etc. Or it is an apparatus which can notify audibly. However, the above CRT is an abbreviation for Cathode Ray Tube. The LCD is an abbreviation for Liquid Crystal Display. The PDP is an abbreviation for Plasma Display Panel. Furthermore, the ELD is an abbreviation for Electro-Luminescence Display.

  The storage unit 920 is a device for storing various data. As the storage unit 920, for example, a magnetic storage device such as a hard disk drive (HDD), a semiconductor storage device, an optical storage device, a magneto-optical storage device, or the like is used. However, the HDD is an abbreviation for Hard Disk Drive.

  The drive 922 is a device that reads information recorded on a removable recording medium 928 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, or writes information to the removable recording medium 928. The removable recording medium 928 is, for example, a DVD medium, a Blu-ray medium, an HD DVD medium, a compact flash (registered trademark) (CF), a memory stick, or an SD memory card. Of course, the removable recording medium 928 may be, for example, an IC card on which a non-contact type IC chip is mounted, an electronic device, or the like. However, the above CF is an abbreviation for CompactFlash. The SD is an abbreviation for Secure Digital memory card. The IC is an abbreviation for Integrated Circuit.

  The connection port 924 is a port for connecting an external connection device 930 such as a USB port, an IEEE 1394 port, a SCSI, an RS-232C port, or an optical audio terminal. The external connection device 930 is, for example, a printer, a portable music player, a digital camera, a digital video camera, or an IC recorder. However, the above USB is an abbreviation for Universal Serial Bus. The SCSI is an abbreviation for Small Computer System Interface.

  The communication unit 926 is a communication device for connecting to the network 932. For example, a wired or wireless LAN, Bluetooth (registered trademark), or a WUSB communication card, an optical communication router, an ADSL router, or various types It is a modem for communication. The network 932 connected to the communication unit 926 is configured by a wired or wireless network, such as the Internet, home LAN, infrared communication, visible light communication, broadcast, or satellite communication. However, the above LAN is an abbreviation for Local Area Network. The WUSB is an abbreviation for Wireless USB. The above ADSL is an abbreviation for Asymmetric Digital Subscriber Line.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to the example which concerns. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

  For example, in the description of the above-described embodiment, a method has been proposed in which biometric information is converted into pixel correlation values and Hough correlation values, and identity authentication is performed with high accuracy using the distribution of the two-dimensional correlation data. However, the application range of the technology according to the present embodiment is not limited to this. For example, biometric information can be converted into a distribution of N-dimensional (N ≧ 3) feature amount data, and the technique of this embodiment can be applied to the distribution. In this case, the N-dimensional feature amount data is projected onto the first principal component axis and converted into one-dimensional data. Then, the Mahalanobis distance is calculated for the one-dimensional data, and the discrimination reference value is calculated based on the distribution of the calculated Mahalanobis distance. Further, the projection to the first principal component axis and the Mahalanobis distance are calculated for the input data and the registered data, and the personal authentication is performed based on the discrimination reference value. Thus, the technique of the present embodiment can be similarly applied to a three-dimensional or higher data distribution.

10 Biometric authentication system 100 Biometric authentication sensor 102 LED
DESCRIPTION OF SYMBOLS 104 Image pick-up element 110 Discrimination reference | standard determination apparatus 112 Memory | storage part 114 Correlation calculation part 116 Principal component analysis part 118 Mahalanobis distance calculation part 120 Threshold straight line determination part 130 Biometric authentication apparatus 132 Correlation calculation part 134 Mahalanobis distance calculation part 136 Personal identification part 138 Memory | storage part FG Finger L11 Irradiation light L13 Scattered light V Vein

Claims (11)

Principal component analysis is performed on multi-dimensional correlation data regarding the correlation of biological patterns acquired from a plurality of biological samples, the first principal component axis of the correlation data distribution regarding the same biological sample, and the first of the correlation data distribution regarding different biological samples. A principal component analysis unit for determining a principal component axis;
A one-dimensional data generating unit that projects the multi-dimensional correlation data onto a first principal component axis of each distribution determined by the principal component analyzing unit to generate one-dimensional data relating to each distribution;
A Mahalanobis distance calculator for calculating a Mahalanobis distance from the distribution center of each distribution to each point of the one-dimensional data on the first principal component axis of each distribution;
Discrimination for determining a discrimination reference value related to the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the Mahalanobis distance of each distribution calculated by the Mahalanobis distance calculation unit with respect to the plurality of biological samples. A reference value determination unit;
A biometric authentication device. The first correlation value related to the feature point arrangement of each biological pattern and the second correlation value related to the slope of the line segment connecting the feature points of each biological pattern are recorded for the biological patterns acquired from the plurality of biological samples. A sample data storage unit;
The biometric authentication device according to claim 1, wherein the principal component analysis unit performs principal component analysis on the set of the first correlation value and the second correlation value recorded in the sample data storage unit as the multi-dimensional correlation data. . The one-dimensional data generation unit
A coordinate transformation unit that performs coordinate transformation on a coordinate system having the first principal component axis as one coordinate axis for the multi-dimensional correlation data;
A first principal component extraction unit that extracts a component of the first principal component axis from the multi-dimensional correlation data subjected to coordinate transformation by the coordinate transformation unit to generate one-dimensional data;
The biometric authentication device according to claim 2, comprising:   The biometric authentication device according to claim 1, wherein the biometric pattern is one or a combination of a vein pattern, a fingerprint pattern, and an iris pattern. A biometric sensor that acquires a biometric pattern from a predetermined biometric part;
A registered pattern storage unit in which registered biometric patterns are recorded;
The first principal component axis of the correlation data distribution related to the same biological sample and the correlation data related to different biological samples determined by principal component analysis of the first dimension data of multiple dimensions related to the correlation of biological patterns acquired from a plurality of biological samples. For the first principal component axis of the distribution
One-dimensional data for generating one-dimensional data relating to each distribution by projecting a plurality of second-dimensional correlation data relating to the correlation between the biometric pattern acquired by the biometric authentication sensor and the biometric pattern recorded in the registered pattern storage unit A generator,
A Mahalanobis distance calculator that calculates a Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the one-dimensional data generated by the one-dimensional data generator;
A similarity determination unit that determines similarity between a biometric pattern acquired by the biometric sensor and a biometric pattern recorded in the registered pattern storage unit based on the Mahalanobis distance calculated by the Mahalanobis distance calculation unit; ,
A biometric authentication device. Principal component analysis is performed on first-dimensional correlation data of a plurality of dimensions related to the correlation of biological patterns acquired from a plurality of biological samples, and the first principal component axis of the correlation data distribution regarding the same biological sample and the correlation data distribution regarding different biological samples A principal component analysis unit for determining a first principal component axis;
The first one-dimensional data generation for generating the first one-dimensional data relating to each distribution by projecting the plurality of first-dimensional correlation data onto the first principal component axis of each distribution determined by the principal component analysis unit And
A first Mahalanobis distance calculator that calculates a first Mahalanobis distance from the distribution center of each distribution to each point of the first one-dimensional data on the first principal component axis of each distribution;
Discrimination regarding the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the first Mahalanobis distance of each distribution calculated by the first Mahalanobis distance calculation unit with respect to the plurality of biological samples. A discriminant reference value determining unit for determining a reference value;
A biometric authentication device;
A biometric sensor that acquires a biometric pattern from a predetermined biometric part;
A registered pattern storage unit in which registered biometric patterns are recorded;
For the first principal component axis of the correlation data distribution for the same biological sample and the first principal component axis of the correlation data distribution for the different biological samples,
Projecting multi-dimensional second correlation data relating to the correlation between the biometric pattern acquired by the biometric authentication sensor and the biometric pattern recorded in the registered pattern storage unit to generate second one-dimensional data relating to each distribution. A second one-dimensional data generation unit;
A second Mahalanobis distance calculator for calculating a second Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the second one-dimensional data;
The biometric pattern acquired by the biometric sensor and the registered pattern storage unit based on the second Mahalanobis distance calculated by the second Mahalanobis distance calculation unit and the discrimination reference value determined by the discrimination reference value determination unit A similarity determination unit that determines the similarity between the biological patterns recorded in
A biometric authentication device;
Including a biometric authentication system. Principal component analysis is performed on multi-dimensional correlation data regarding the correlation of biological patterns acquired from a plurality of biological samples, the first principal component axis of the correlation data distribution regarding the same biological sample, and the first of the correlation data distribution regarding different biological samples. A principal component analysis step for determining a principal component axis;
Projecting the multi-dimensional correlation data onto a first principal component axis of each distribution determined in the principal component analysis step to generate one-dimensional data relating to each distribution;
A Mahalanobis distance calculating step of calculating a Mahalanobis distance from the distribution center of each distribution to each point of the one-dimensional data on the first principal component axis of each distribution;
Discrimination for determining a discrimination reference value related to the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the Mahalanobis distance of each distribution calculated in the Mahalanobis distance calculation step with respect to the plurality of biological samples. A reference value determining step;
Discrimination criteria determination method including The first principal component axis of the correlation data distribution related to the same biological sample and the correlation data related to different biological samples determined by principal component analysis of the first dimension data of multiple dimensions related to the correlation of biological patterns acquired from a plurality of biological samples. For the first principal component axis of the distribution
One-dimensional data relating to the respective distributions by projecting multi-dimensional second correlation data relating to a correlation between a biometric pattern acquired using a biometric authentication sensor for acquiring a biometric pattern from a predetermined biometric part and a registered biometric pattern A one-dimensional data generation step for generating data;
A Mahalanobis distance calculating step of calculating a Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the one-dimensional data generated in the one-dimensional data generating step;
A similarity determination step for determining a similarity between a biometric pattern acquired using the biometric sensor based on the Mahalanobis distance calculated in the Mahalanobis distance calculation step and the registered biometric pattern;
A biometric authentication method. Principal component analysis is performed on first-dimensional correlation data of a plurality of dimensions related to the correlation of biological patterns acquired from a plurality of biological samples, and the first principal component axis of the correlation data distribution regarding the same biological sample and the correlation data distribution regarding different biological samples A principal component analysis step for determining a first principal component axis;
Projecting the first-dimensional data of the plurality of dimensions onto the first principal component axis of each distribution determined in the principal component analysis step to generate first one-dimensional data relating to each distribution Steps,
A first Mahalanobis distance calculating step of calculating a first Mahalanobis distance from the distribution center of each distribution to each point of the first one-dimensional data on the first principal component axis of each distribution;
Discrimination regarding the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the first Mahalanobis distance of each distribution calculated in the first Mahalanobis distance calculating step with respect to the plurality of biological samples. A discrimination reference value determination step for determining a reference value;
For the first principal component axis of the correlation data distribution for the same biological sample and the first principal component axis of the correlation data distribution for the different biological samples,
A second multi-dimensional correlation data relating to a correlation between a biometric pattern acquired using a biometric authentication sensor for acquiring a biometric pattern from a predetermined biometric part and a registered biometric pattern is projected to obtain a second for each distribution. A second one-dimensional data generation step for generating the one-dimensional data of
A second Mahalanobis distance calculating step of calculating a second Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the second one-dimensional data;
The biometric pattern acquired using the biometric sensor based on the second Mahalanobis distance calculated in the second Mahalanobis distance calculation step and the discrimination reference value determined in the discrimination reference value determination step and the registered A similarity determination step for determining similarity between the biometric patterns of
A biometric authentication method. Principal component analysis is performed on multi-dimensional correlation data regarding the correlation of biological patterns acquired from a plurality of biological samples, the first principal component axis of the correlation data distribution regarding the same biological sample, and the first of the correlation data distribution regarding different biological samples. A principal component analysis function for determining a principal component axis;
A one-dimensional data generation function for projecting the multi-dimensional correlation data onto a first principal component axis of each distribution determined by the principal component analysis function to generate one-dimensional data relating to each distribution;
A Mahalanobis distance calculation function for calculating a Mahalanobis distance from the distribution center of each distribution to each point of the one-dimensional data on the first principal component axis of each distribution;
Discrimination for determining a discrimination reference value related to the Mahalanobis distance for discriminating between the same biological sample and the different biological sample based on the Mahalanobis distance of each distribution calculated by the Mahalanobis distance calculation function with respect to the plurality of biological samples. A reference value determination function;
A program to make a computer realize. The first principal component axis of the correlation data distribution related to the same biological sample and the correlation data related to different biological samples determined by principal component analysis of the first dimension data of multiple dimensions related to the correlation of biological patterns acquired from a plurality of biological samples. For the first principal component axis of the distribution
One-dimensional data relating to the respective distributions by projecting multi-dimensional second correlation data relating to a correlation between a biometric pattern acquired using a biometric authentication sensor for acquiring a biometric pattern from a predetermined biometric part and a registered biometric pattern A one-dimensional data generation function for generating data;
A Mahalanobis distance calculation function for calculating a Mahalanobis distance from the distribution center of each distribution on the first principal component axis of each distribution to the one-dimensional data generated by the one-dimensional data generation function;
A similarity determination function for determining similarity between a biometric pattern acquired using the biometric sensor and the registered biometric pattern based on the Mahalanobis distance calculated by the Mahalanobis distance calculation function;
A program to make a computer realize.

Images