US20090289760A1  Biometric authentication system, authentication client terminal, and biometric authentication method  Google Patents
Biometric authentication system, authentication client terminal, and biometric authentication method Download PDFInfo
 Publication number
 US20090289760A1 US20090289760A1 US12/379,715 US37971509A US2009289760A1 US 20090289760 A1 US20090289760 A1 US 20090289760A1 US 37971509 A US37971509 A US 37971509A US 2009289760 A1 US2009289760 A1 US 2009289760A1
 Authority
 US
 United States
 Prior art keywords
 enrolled
 claimant
 authentication
 user
 users
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Granted
Links
 238000004364 calculation method Methods 0.000 claims description 67
 230000000875 corresponding Effects 0.000 claims description 31
 238000000605 extraction Methods 0.000 claims description 16
 230000004044 response Effects 0.000 claims description 4
 238000000034 method Methods 0.000 description 43
 238000004891 communication Methods 0.000 description 10
 238000010586 diagram Methods 0.000 description 8
 102100015745 ILF3 Human genes 0.000 description 3
 101700006288 ILF3 Proteins 0.000 description 3
 210000000554 Iris Anatomy 0.000 description 3
 230000005540 biological transmission Effects 0.000 description 3
 238000007796 conventional method Methods 0.000 description 3
 230000000694 effects Effects 0.000 description 3
 238000005516 engineering process Methods 0.000 description 3
 239000000284 extract Substances 0.000 description 2
 230000037071 Jmax Effects 0.000 description 1
 240000007072 Prunus domestica Species 0.000 description 1
 230000003044 adaptive Effects 0.000 description 1
 238000004422 calculation algorithm Methods 0.000 description 1
 230000001413 cellular Effects 0.000 description 1
 238000010348 incorporation Methods 0.000 description 1
 238000007477 logistic regression Methods 0.000 description 1
 238000010801 machine learning Methods 0.000 description 1
 239000000203 mixture Substances 0.000 description 1
 230000004048 modification Effects 0.000 description 1
 238000006011 modification reaction Methods 0.000 description 1
 238000003909 pattern recognition Methods 0.000 description 1
 238000001617 sequential probability ratio test Methods 0.000 description 1
 230000003068 static Effects 0.000 description 1
 238000007619 statistical method Methods 0.000 description 1
 238000010792 warming Methods 0.000 description 1
Classifications

 G—PHYSICS
 G06—COMPUTING; CALCULATING; COUNTING
 G06K—RECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
 G06K9/00—Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
 G06K9/62—Methods or arrangements for recognition using electronic means
 G06K9/6267—Classification techniques
 G06K9/6268—Classification techniques relating to the classification paradigm, e.g. parametric or nonparametric approaches
 G06K9/6277—Classification techniques relating to the classification paradigm, e.g. parametric or nonparametric approaches based on a parametric (probabilistic) model, e.g. based on NeymanPearson lemma, likelihood ratio, Receiver Operating Characteristic [ROC] curve plotting a False Acceptance Rate [FAR] versus a False Reject Rate [FRR]

 G—PHYSICS
 G06—COMPUTING; CALCULATING; COUNTING
 G06K—RECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
 G06K9/00—Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
 G06K9/00885—Biometric patterns not provided for under G06K9/00006, G06K9/00154, G06K9/00335, G06K9/00362, G06K9/00597; Biometric specific functions not specific to the kind of biometric
Abstract
A biometric authentication system, authentication client terminal, and biometric authentication method are provided to reduce an expected value of the number of inputs of biometric data for authentication, while effectively preventing forgery. In a biometric authentication system, prior probabilities of enrolled users u_{n }and nonenrolled user u_{0 }are previously set. 1:N matching is performed between feature data of a claimant v and matching feature data. The matching score is calculated for each enrolled user u_{n}. A ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }is calculated for each enrolled user u_{n }using the calculated matching scores. Posterior probabilities of the enrolled users u_{n }and nonenrolled user u_{0 }are calculated, using the likelihood ratios, and the prior probabilities of both the enrolled users u_{n }and the nonenrolled user u_{0}. Then, determination is made by comparing each posterior probability with a first threshold.
Description
 This application claims priority based on Japanese patent applications, No. 2008119131 filed on Apr. 30, 2008 and No. 2008312411 filed on Dec. 8, 2008, the entire contents of which are incorporated herein by reference.
 The present invention relates to a biometric authentication system using biometric information to verify identities.
 Biometric authentication is known as an authentication technology with advantages that the forgery of information input for authentication (for example, fingerprint) is more difficult than the authentication based on the password, IC card, or other identifying information, and that the information input for authentication is not forgotten.
 In the biometric authentication, biometric data is first acquired from a user, and information called the feature is extracted from the biometric data and is registered in advance. This registered information is called the enrolled template. The user who has been registered in advance is called the enrolled user. Upon authentication of a user, the biometric data is acquired from the user, and the feature extracted from the biometric data is checked against the enrolled template to perform authentication (identity verification). The user to be identified is called the claimant.
 One of biometric authentication technologies that identifies a claimant to find out the identical enrolled user by matching the claimant with each of N enrolled users (hereinafter referred to as “1:N matching”), is called biometric identification. In conventional biometric identification, when an enrolled user identified as the claimant (hereinafter referred to as “identified user”) exists, authentication success is determined with the identified user as the determination result. While authentication failure is determined when no identified user exists. Examples of the biometric authentication system using biometric identification are a time and attendance management system, and a system for credit payment only using biometric authentication instead of using a credit card (hereinafter referred to as “cardless credit payment system”). Biometric identification does not require the claimant to present a card or other means of identification, and has an advantage of high availability.

FIG. 5 shows types of authentication error rate in biometric identification. The authentication error rate in biometric identification can be classified into the following three types.  (1) Error rate that occurs when a claimant previously registered is successfully authenticated as another enrolled user (hereinafter referred to as “Enrollee False Acceptance Rate: EFAR).
(2) Error rate that occurs when a claimant previously registered fails to be authenticated (hereinafter referred to as “Enrollee False Rejection Rate: EFRR”).
(3) Error rate that occurs when a user not previously registered (hereinafter referred to as “nonenrolled user”) is successfully authenticated (hereinafter referred to as “NonEnrollee False Acceptance Rate: NFAR”).  When EFAR or NFAR is high the possibility that the claimant will be successfully authenticated as another person increases, resulting in a decrease in security. Such an authentication error of when the claimant is successfully authenticated as another person, is called erroneous identification. When EFRR is high the possibility that the registered claimant will fail to be authenticated increases, resulting in a decrease in availability.
 There are authentication methods proposed to increase the authentication accuracy in biometric identification. This authentication method asks the claimant to input plural biometric information (for example, biometric information of different types such as fingerprint, iris, voice, or biometric information of the same type collected from different areas of the body such as index fingerprint, middle fingerprint, and fourth fingerprint), and integrates the input biometric data to determine the identity of the claimant (hereinafter referred to as “1:N combined authentication”). For example, in U.S. Pat. No. 7,277,891 B2, the claimant candidates (enrolled users who may possibly be identical to a claimant) are narrowed down by first biometric data (for example, face), and the remaining candidates are further narrowed down by second biometric data (for example, fingerprint). Finally the last remaining candidates are output as the determination result.
 However, in 1:N combined authentication, the claimant needs to input plural types of biometric information, so that the authentication procedure is complicated, and availability decreases. In addition, the time from the input of the first biometric data to the output of the authentication result (hereinafter referred to as “authentication time”) is longer than the case of authentication using only one piece of biometric data, resulting in a further decrease in availability. However, no measure has been taken to solve such problems in U.S. Pat. No. 7,277,891 B2.
 In U.S. Pat. No. 7,277,891 B2, the system may output plural candidates. For example, in an application such as cardless credit payment, it is necessary to narrow down possible claimants to only one user. Thus, one identified user is manually selected after plural candidates are output, resulting in a further decrease in availability.
 An approach to solve the problem is proposed in Hideki Noda, “Sequential Probability Ratio Test for Adaptive Speaker Identification”, IEICE technical report DII Vol. J84DII, No. 1, pp. 211213 (2001). This approach uses the distribution p_{n }(x_{j}) of the features of enrolled users u_{n }(n=1 to N) and the distribution p_{0}(x_{J}) of the feature of all enrolled users, to calculate the likelihood ratio l_{n }for each of the enrolled users, each time the voice feature x_{J }(J=1, 2, and so on) is acquired, by the following equation.

l _{n}=π_{j=1} ^{J} p _{n}(x _{j})/p _{0}(x _{j}) (1)  When the obtained likelihood ratio l_{n }is larger than a threshold A, authentication success is determined with the enrolled user u_{n }at this time as the identified user. When the obtained likelihood ratio l_{n }is smaller than a threshold B, one or more enrolled users u_{n }corresponding to the likelihood ratio l_{n }are excluded from the matching target in the subsequent steps, which is hereinafter referred to as “pruning”. When no identified user is obtained, another feature x_{J }is acquired to repeat determination until the identified user is obtained. As described above, the determination of the claimant is performed by comparing the likelihood ratio with the threshold A each time the feature is acquired, in order to reduce the number of inputs of the biometric data necessary for authentication. Further, the time for 1:N matching is substantially proportional to the number of enrolled users N to be matched. Thus, the time for 1:N matching is reduced by pruning the enrolled user(s) based on the result of comparing each of the likelihood ratios with the threshold B. In this way, the authentication time is further reduced.
 In the technology disclosed in the Noda, the determination is made by comparing the likelihood ratios with the threshold each time the feature is acquired, in order to aim at reducing the number of inputs of the biometric data necessary for authentication. However, the method of biometric identification that calculates likelihood ratios from the feature distribution and compares the obtained likelihood ratios with the threshold, provides no logical guarantee for minimizing the expected value of the number of inputs of the biometric data necessary for determination. In other words, the technology of the Noda has left room for improvement in reduction of the expected value of the number of inputs of the biometric data necessary for authentication.
 Further, in conventional biometric identification systems, there is a problem that a malicious user, who has not been registered, attempts authentication over and over again to be successful in forgery.
 The present invention addresses the above identified problems by providing a biometric authentication system, an authentication client terminal, and a biometric authentication method that can significantly reduce the expected value of the number of inputs of the biometric data necessary for authentication, and effectively prevent forgery.
 In order to solve the above problems, a biometric authentication system according to the present invention includes: a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N); a feature extraction unit for extracting feature data of a claimant v based on biometric data acquired from the claimant v; a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0}; a likelihood ratio calculation unit for calculating a ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the matching feature data held in the database and the extracted feature data of the claimant; a posterior probability calculation unit for calculating posterior probabilities of the enrolled users u_{n }and a posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the prior probabilities of the enrolled users u_{n}, and the prior probability of the nonenrolled user u_{0}; and a determination unit for comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, and when the posterior probability showing the maximum value is one of the posterior probabilities of the enrolled users u_{n}, the determination unit determining that the corresponding enrolled user and the claimant v are identical, while when the posterior probability showing the maximum value is the posterior probability of the nonenrolled user u_{0}, the determination unit determining that the claimant v is a user other than the enrolled users u_{n}.
 Another embodiment of the present invention further includes a 1: N matching unit for performing 1:N matching between the extracted claimant feature data and the feature data held in the database, and calculating a matching score of each of the enrolled users u_{n}. The likelihood ratio calculation unit calculates the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
 In still another embodiment of the present invention, when the maximum value of the posterior probability is equal to or less than the first threshold, and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit requests to reacquire the biometric data of the claimant v, while when the number of acquisitions of the biometric data is equal to or more than the predetermined value, the determination unit determines “authentication failure”. In this case, it is possible to further include an enrolled user pruning unit. When the determination unit requests to reacquire the biometric data, the enrolled user pruning unit compares each of the posterior probabilities calculated by the posterior probability calculation unit with a second threshold, and excludes a enrolled user u_{n }corresponding to the posterior probability smaller than the second threshold, from the target to be matched to the claimant v.
 In still another embodiment of the present invention, the posterior probability calculation unit compares the posterior probability of the nonenrolled user u_{0 }calculated in the current acquisition of the biometric data with the prior probability of the nonenrolled user u_{0 }when the number of acquisitions of the biometric data is one, or with the prior probability of the nonenrolled user u_{0 }calculated in the previous acquisition of the biometric data when the number of acquisitions is two or more. When the posterior probability of the nonenrolled user u_{0 }calculated in the current acquisition of the biometric data is higher, the posterior probability calculation unit does not use the feature data extracted in the current acquisition in the posterior probability calculation in the subsequent acquisitions of the biometric data.
 In still another embodiment of the present invention, the database holds plural different types of matching feature data belonging to plural modalities with respect to each of the enrolled users u_{n}. The embodiment further includes a biometric information input sensor that allows for acquisition of the biometric data only once.
 In still another embodiment of the present invention, the database holds plural different types of matching feature data belonging to the same modality with respect to each of the enrolled users u_{n}. The 1:N matching unit determines “authentication failure” when the best matching score is obtained for the same biometric information several times until input of the biometric data in this time.
 In still another embodiment of the present invention, the database holds plural different types of matching feature data belonging to the same modality with respect to each of the enrolled users u_{n}. When the best matching score is obtained in several times for the same biometric information until input of the biometric data in this time, the 1:N matching unit ensures not to use the matching scores obtained in inputs other than the input with the best matching score achieved in the calculation of likelihood ratios by the likelihood ratio calculation unit.
 In still another embodiment of the present invention, the prior probability setting unit increases the prior probability of the nonenrolled user u_{0 }when the determination unit determines that the claimant v is a user other than the enrolled users u_{n}.
 In still another embodiment of the present invention, the first threshold is larger than 0.5.
 In any of the above described embodiments, when it is determined that the claimant v is a user other than the enrolled users u_{n}, the determination unit can output forgery warning information.
 Still another embodiment of the present invention further includes a loss calculation unit, as well as a prior probability and threshold setting unit. Each time authentication is performed, the loss calculation unit calculates a loss W_{1 }when the claimant v is an enrolled user and is identified as another enrolled user, and a loss W_{0 }when the claimant v is a nonenrolled user and is identified as a certain enrolled user. The prior probability and threshold setting unit sets the prior probabilities of the enrolled users u_{n }and a corresponding threshold, as well as the prior probability of the nonenrolled user u_{0 }and a corresponding threshold, respectively, using the losses W_{1 }and W_{0}. With this setting, EFRR as well as an expected value R, of the loss in the determination of the nth enrolled user are equal to or less than respective desired values.
 Another biometric authentication system according the present invention, includes: a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N); a feature extraction unit for extracting feature data of a claimant v based on biometric data acquired from the claimant v; a 1:N matching unit for performing 1:N matching between the extracted claimant feature data and the matching feature data held in the database, and calculating a matching score of each of the enrolled users u_{n}; a likelihood ratio calculation unit for calculating the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n }by using the calculated matching scores; and a determination unit for comparing each of the calculated likelihood ratios with a first threshold, when a maximum value of the likelihood ratio is larger than the first threshold, the determination unit determining that the enrolled user corresponding to the likelihood ratio showing the maximum value is identical to the claimant v, while when the maximum value of the likelihood ratio is equal to or less than the first threshold, and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit requesting to reacquire the biometric data of the claimant v, and when the number of acquisitions is equal to or more than the predetermined value, the determination unit determining “authentication failure”.
 Further, it is possible to further include an enrolled user pruning unit. When the determination unit requests to reacquire the biometric data, the enrolled user pruning unit compares each of the likelihood ratios calculated by the likelihood calculation unit with a second threshold, and excludes a enrolled user u_{n }corresponding to the likelihood ratio smaller than the second threshold, from the target to be matched to the claimant v.
 In order to solve the above described problem, an authentication client terminal according to the present invention is configured to be able to communicate with an authentication server terminal. The authentication server terminal includes a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N), and a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0}. The authentication client terminal includes: a feature extraction unit for extracting the feature data of a claimant v based on the biometric data acquired from the claimant v; an authentication request unit for transmitting the extracted claimant feature data to the authentication server terminal, calculating a ratio of likelihood v=u_{n }to likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the claimant feature data and the matching feature data held in the database, calculating the posterior probabilities of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the set prior probabilities of the enrolled users u_{n}, and the set prior probability of the nonenrolled user u_{0}, comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, the authentication request unit requesting to transmit the information of the enrolled user u_{n }or nonenrolled user u_{0 }corresponding to the posterior probability showing the maximum value, and acquiring the information transmitted from the authentication server terminal in response to the request; and a determination unit for determining that when the acquired information is of the corresponding enrolled user u_{n}, the enrolled user u_{n }and the claimant v are identical, and that when the acquired information is of the nonenrolled user u_{0}, the claimant v is a user other than the enrolled users u_{n}.
 In the embodiments of the present invention, the authentication request unit performs 1:N matching between the extracted claimant feature data and the matching feature data held in the database, calculates the matching score of each of the enrolled users u_{n}, and calculates the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
 In order to solve the above described problem, a biometric authentication method according to the present invention, includes: a step of extracting feature data of a claimant v based on the biometric data acquired from the claimant v; a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0}; a step of calculating a ratio of likelihood v=u_{n }to likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the matching feature data held in the database with respect to each of the enrolled users u_{n }(n=1 to N), as well as on the extracted claimant feature data; a step of calculating posterior probabilities of the enrolled users u_{n }and a posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the set prior probabilities of the enrolled users u_{n}, and the set prior probability of the nonenrolled user u_{0}; and a step of comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, and when the posterior probability showing the maximum value is one of the posterior probabilities of the enrolled users u_{n}, determining that the corresponding enrolled user and the claimant v are identical, while when the posterior probability showing the maximum value is the posterior probability of the nonenrolled user u_{0}, determining that the claimant v is a user other than the enrolled users u_{n}.
 The embodiments of the present invention, further includes a step of performing 1:N matching between the extracted claimant feature data and the matching feature data held in the database, and calculating a matching score of each of the enrolled users u_{n}. The likelihood ratio calculating step involves calculating the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
 According to the present invention, the determination is made by calculating the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, normalizing the likelihood ratios to the posterior probabilities of the enrolled users u_{n }and to the posterior probability of the nonenrolled user u_{0}, and comparing each of the posterior probabilities with a threshold. This method significantly reduces (logically minimizes) the expected value of the number of inputs of the biometric data necessary for authentication.
 Alternatively, the determination is made by calculating the matching score of each of the enrolled users u_{n}, calculating the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores, and comparing each of the likelihood ratios with a threshold. This method also significantly reduces (logically minimizes) the expected value of the number of inputs of the biometric data necessary for authentication.
 Further, the prior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0 }are both set in advance, whereby not only the posterior probabilities of the enrolled user u_{n }but also the posterior probability of the nonenrolled user u_{0 }is calculated. This method enables to clearly distinguish between the case in which the claimant v is identical to the nonenrolled user u_{0 }(forgery), and the case in which the claimant v is not identified (authentication failure). When the forgery user can be clearly recognized, it is possible to take measures (for example, such as presenting a warning information message, and stopping the operation of the sensor) to effectively prevent such a malicious user forgery.
 These and other benefits are described throughout the present specification. A further understanding of the nature and advantages of the invention may be realized by reference to the remaining portions of the specification and the attached drawings.

FIG. 1 is a diagram showing an example of the configuration of a biometric authentication system; 
FIG. 2 is a diagram showing a hardware configuration of an authentication client terminal and an authentication server terminal; 
FIG. 3 is a diagram showing the procedure of the authentication process by the biometric authentication system; 
FIG. 4 is a schematic diagram showing the concept of authentication algorithm employed in the biometric authentication system; 
FIG. 5 is a schematic diagram showing the concept of authentication error rate in biometric identification; 
FIG. 6 is a diagram showing an example of the configuration of another biometric authentication system; 
FIG. 7 is a diagram showing an example the configuration of still another biometric authentication system; and 
FIG. 8 is a diagram showing the procedure of the authentication process by still another biometric authentication system.  Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.
 A biometric authentication system 1 of this embodiment is a biometric authentication system that performs biometric identification between a claimant v and N enrolled users u_{n }(n=1 to N), using plural biometric data. The claimant is a user to be identified through an authentication client terminal 100. The enrolled user is a user whose feature data of the biometric data is previously registered for the purpose of matching.

FIG. 1 shows an example of the configuration of the biometric authentication system 1. The biometric authentication system 1 includes the authentication client terminal 100 and an authentication server terminal 110. The authentication client terminal 100 acquires the feature data from the biometric data of the claimant v, and presents the information such as the authentication result to the claimant v. The authentication server 110 performs biometric identification or other processing. The authentication client terminal 100 and the authentication server terminal 110 are coupled through a network 140 so that data can be communicated between the two terminals.  The biometric authentication system 1 can be applied, for example, to a cardless credit payment system. In this case, the authentication client terminal 100 corresponds to an authentication apparatus installed in a member store, the authentication server terminal 110 corresponds to a server installed in a data center, and the network 140 corresponds to the Internet. Further, the biometric authentication system 1 can also be applied to an inhouse time and attendance management system. In this case, the authentication client terminal 100 corresponds to an authentication apparatus installed in a room, the authentication server terminal 110 corresponds to a server installed in a server room, and the network 140 corresponds to a corporate intranet.
 Incidentally, the network 140 may be not only a network using WAN, LAN and the like, but also a wired communication line using USB, IEEE 1394 and the like, a wireless communication line using a cellular phone network, Bluetooth and the like, or a combination of the wired and wireless communication lines.
 The authentication client terminal 100 includes a feature extraction unit 102 and an authentication request unit 104. The feature extraction unit 102 and the authentication request unit 104 may be functionally realized by a central processing unit (CPU) executing a predetermined program (not shown) in the authentication client terminal 100, or may be realized by providing hardware having functions of the relevant processing units described later.
 The biometric authentication system 1 further includes a plurality (M) of biometric information input sensors 101 coupled to the authentication client terminal 100. The biometric information input sensors 101 are devices for acquiring plural biometric data 1 to M from the claimant v.
 Incidentally, the biometric information includes various types of information derived from human, such as fingerprint, voice pattern, iris pattern, handwriting, keystroke (the number of pressed keys) for inputting a predetermined character string through a keyboard, dynamic signature, and a way of walking. The dynamic signature is the biometric information including, for example, not only the information relating to handwriting shape (static signature), but also the information relating to dynamic feature such as speed of pen motion in signature writing.
 Here, the type of biometric information that a single sensor can acquire is called modality. For example, as in the case of fingerprint collection, different types of biometric information may be acquired from different areas of the human body (for example, index finger and middle finger). Thus, it may happen that plural biometric information belonging to the same modality can be acquired by a single sensor 101. In this embodiment, it is assumed that authentication is performed using plural different types of biometric information belonging to plural modalities.
 The feature extraction unit 102 extracts the feature data of a claimant v based on the biometric data acquired from the claimant v. Note that the feature data can be extracted by any conventional method.
 The authentication request unit 104 requests the authentication server terminal 110 to identify the claimant v. Details of the determination method will described later with reference to
FIG. 3 .  Further, the authentication request unit 104 receives the determination result information about the result of the determination from the authentication server terminal 110, and performs processing according to the determination result information. Details of the processing will be described later.
 The authentication server terminal 110 includes a storage unit 117, a prior probability setting unit 111, a 1:N matching unit 112, a likelihood ratio calculation unit 118, a posterior probability calculation unit 113, a determination unit 114, and an enrolled user pruning unit 115. Note that the prior probability setting unit 111, the 1:N matching unit 112, the likelihood ratio calculation unit 118, the posterior probability calculation unit 113, the determination unit 114, and the enrolled user pruning unit 115 may be functionally realized by a central processing unit (CPU) executing a predetermined program (not shown) in the authentication server terminal 110, or may be realized by providing hardware having functions of the relevant processing units. Further, the respective processing units are not necessarily separate from each other, and any of the plural processing units may be put together as a single unit.
 The storage unit 117 stores a enrolled user database 120 for holding the ID of each of the enrolled users u_{n}, as well as the matching feature data (hereinafter referred to as “biometric information enrolled templates”) 122 that are extracted from plural biometric information 1 to M for each of the enrolled users u_{n}. The storage unit 117 also stores determination history information 130 indicating the final determination results (“authentication success”, “forgery”, or “authentication failure”) of D authentication attempts (where D is an integer of one or more). The storage unit 117 is realized by a memory device such as RAM or hard disk drive.
 The prior probability setting unit 111 initializes the prior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0}. In this embodiment, as shown in
FIG. 4 , the concept of nonenrolled user u_{0 }is introduced for the purpose of detecting forgery. Thus, the prior probability is also set to the nonenrolled user u_{0}. The prior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0}, are both probabilities set for the enrolled users u_{n }and the nonenrolled user u_{0}, under the condition that no data (matching scores in this embodiment) is obtained. More specifically, when a certain claimant v performs the identify verification with the matching scores being unknown, the prior probabilities are v=u_{1}, v=u_{2}, and v=u_{n}, as well as v=u_{0}. In this embodiment, the prior probability setting includes the following methods. One involves using a predetermined fixed value (for example, a value included in the source code of a program) as the prior probability. The other involves previously setting a prior probability calculation formula to calculate prior probabilities according to the various data, such as the number of data in the enrolled user database 120 and the content of the determination history information 130. Details of the methods of setting prior probabilities will be described later.  The prior probability setting unit 111 updates the prior probabilities in the course of repeating the authentication process. More specifically, when the determination unit 114, to be described later, determines that the claimant v is a user (forgery user) other than the enrolled users u_{n}, the prior probability setting unit 111 increases the prior probability of the nonenrolled user u_{0}. Accordingly, the prior probability setting unit 111 reduces the prior probability of each of the enrolled users u_{n }so that the sum of the prior probability of the nonenrolled user u_{0 }and the prior probability of each of the enrolled users u_{n }is equal to one. This process may be performed immediately after the determination unit 114 determines “forgery”, or in the next authentication process (for example, upon execution of step S302 described later with reference to
FIG. 3 ). The effect obtained by increasing the prior probability of the nonenrolled user u_{0 }will be described later.  The 1:N matching unit 112 performs 1:N matching between the feature data of the claimant v that is extracted by the feature extraction unit 102 of the authentication client terminal 100, and the biometric information enrolled template 122 held in the enrolled user database 120. Then, the 1:N matching unit 112 obtains the matching score of each of the enrolled users u_{n}. The matching score can be obtained by any conventional method.
 The likelihood ratio calculation unit 118 calculates the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the matching score obtained by the 1:N matching unit 112.
 The posterior probability calculation unit 113 calculates the posterior probabilities of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the likelihood ratio calculated by the likelihood ratio calculation unit 113, and using the prior probabilities of both the enrolled users u_{n }and the nonenrolled user u_{0}, which are set by the prior probability setting unit 111. The posterior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0}, are both conditional probabilities set for the enrolled users u_{n }and the nonenrolled user u_{0}, under conditions that the data (the matching scores in this embodiment) are obtained and taken into account. More specifically, the posterior probabilities of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0 }are calculated, respectively, using the prior probabilities of the enrolled users u_{n}, the prior probability of the nonenrolled user u_{0}, and the likelihood ratios. Details of the posterior probability calculation method will be described later.
 The determination unit 114 compares each of the posterior probabilities calculated by the posterior probability calculation unit 113, with a first threshold (threshold A), and verifies the identity of the claimant v according to the following criteria. That is, when the maximum value of the posterior probability is larger than the predetermined threshold A, and when the posterior probability showing the maximum value is one of the posterior probabilities of the enrolled users u_{n}, the determination unit 114 determines that the corresponding enrolled user and the claimant v are identical (authentication success), and outputs the determination result information indicating authentication success. On the other hand, when the posterior probability showing the maximum value is the posterior probability of the nonenrolled user u_{0}, the determination unit 114 determines that the claimant v is a user (forgery user) other than the enrolled users u_{n}, and outputs the information (forgery warning information) indicating the determination result. Further, when the maximum value of the posterior probability is equal to or less than the threshold A (namely, when it is undeterminable), and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit 114 outputs the information requesting to reacquire the biometric data of the claimant v, as the determination result information indicating that it is necessary to perform the identity verification again. On the other hand, when the number of acquisitions of the biometric data is equal to or more than the predetermined value, the determination unit 114 determines that the claimant v is not identified to any of the enrolled users u_{n }(authentication failure), and outputs the determination result information indicating authentication failure.
 When the determination unit 114 requests to reacquire the biometric data from the claimant v, the enrolled user pruning unit 115 compares each of the posterior probabilities calculated by the posterior probability calculation unit 113, with a second threshold (threshold B), and excludes the enrolled user(s) u_{n }and nonenrolled user u_{0 }corresponding to the posterior probabilities smaller than the threshold B, from the target to be matched to the claimant v. The exclusion of the corresponding enrolled user u_{n }and nonenrolled user u_{0 }from the target to be matched to the user v, is called the pruning of enrolled user u_{n }and nonenrolled user u_{0}. With respect to the pruned enrolled user u_{n }and the pruned nonenrolled user u_{0}, the given values such as the matching score, likelihood ratio, and posterior probability are not calculated in the subsequent authentication process. The target of the pruning may be limited to the enrolled users u_{n}.

FIG. 2 shows a hardware configuration of the authentication client terminal 100 and the authentication server terminal 110 in this embodiment. As shown in the figure, the authentication client terminal 100 and the authentication server terminal 110 can be configured using a typical information processor including a CPU 200, a memory (such as RAM) 201, an HDD 202, an input device (such as a keyboard and a mouse) 203, an output device (such as a display and a speaker) 204, and a communication device (such as a LAN adaptor or an infrared adaptor) 205. 
FIG. 3 shows the procedure of the authentication process by the biometric authentication system 1 of this embodiment, and the data flow associated with the authentication process.  First, the authentication server terminal 110 initializes the number of inputs J of biometric data from the claimant v (J=0) (step S301). This process is performed, for example, by the determination unit 114.
 Then, the prior probability setting unit 111 of the authentication server terminal 110 initializes the prior probability P (v=u_{n}) of each of the enrolled users u_{n}, as well as the prior probability P (v=u_{0}) of the nonenrolled user u_{0 }(step S302). The method for setting prior probabilities will be described below.
 For example, the prior probability P (v=u_{n}) of each of the enrolled users u_{n }(n=1 to N) and the prior probability P (v=u_{0}) of the nonenrolled user u_{0 }can be set to the same value as follows.

P(v=u _{n})=1/(N+1) 
P(v=u _{0})=1/(N+1)  It is also possible that the prior probability P (v=u_{n}) of each of the enrolled users u_{n }(n=1 to N) is given as

P(v=u _{n})=1/(N+1)  and that the prior probability P (v=u_{0}) of the nonenrolled user u_{0 }is given as

P(v=u _{0})=0  The prior probability of the nonenrolled user u_{0 }to 0, under the assumption that the claimant v is necessarily one of the N enrolled users u_{n}. This setting ensures that the claimant v is not identified as the nonenrolled user u_{0}.
 For example, an entrance management apparatus is installed outside the entrance door of a room to control the unlocking and locking of the door by reading the information of an IC card, and an entrance/exit management system is established based on biometric identification to limit persons entering the room. In this case, when only the enrolled users u_{n }can unlock the entrance door and hardly anyone would impersonate to unlock the entrance door, the prior probability of the nonenrolled user u_{0 }is set to 0 as described above. This ensures that the claimant v will not be identified as the nonenrolled user u_{0 }by mistake.
 Further, the prior probability setting unit 111 of the authentication server terminal 110 increases the prior probability of the nonenrolled user u_{0}, based on the determination history information 130 stored in the storage unit 117. Accordingly, the prior probability setting unit 111 reduces the prior probability of each of the enrolled users u_{n }so that the sum of the prior probability of the nonenrolled user u_{0 }and the prior probability of each of the enrolled users u_{n }are equal to one. More specifically, in step S302, the prior probability setting unit 111 refers to the determination history information 130 stored in the storage unit 117. When the determination result from the determination unit 114 in the last authentication process is “forgery”, the number of determinations of “forgery” in D attempts of authentication is given by D_{0 }and is obtained using the final determination results (“authentication successful”, “forgery”, or “authentication failure”) of the D authentication attempts. Then, the prior probability P (v=u_{n}) of each of the enrolled users u_{n }and the prior probability P (v=u_{0}) of the nonenrolled user u_{0}, are given as follows.

P(v=u _{n})=(1−D _{0} /D)/(N+1) 
P(v=u _{0})=(N×D _{0} /D+1)/(N+1)  As described above, because the prior probability setting unit 111 updates the prior probability values, even if a malicious claimant v, who has not been registered, repeats authentication attempt until forgery is achieved, the prior probability P (v=u_{0}) of the nonenrolled user u_{0 }increases each time the claimant v is identified as a user “forgery”. As a result, the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }also increases as described later, making it more difficult to achieve authentication. This leads to the effect that the security of the system is further increased.
 In an application with a high probability of forgery attacks by malicious users, the prior probability of the nonenrolled user u_{0 }can be set to a higher value to achieve high security.
 As described above, the biometric authentication system of this embodiment flexibly controls the prior probability of the nonenrolled user u_{0 }by taking into account the possibility of forgery attacks. Thus, it is possible to achieve appropriate operation both in terms of availability and security according to the situation to which the system is applied.
 In response to an input of biometric data from the claimant v (for example, by putting the index finger on the biometric information 1 input sensor 101 to present the fingerprint), the biometric information input sensor 101 of the authentication client terminal 100 acquires the biometric data of the claimant v (step S303). Here, the biometric information input sensor 101 may permit the acquisition of biometric data only once. In this embodiment, there is a risk that a user, whose biometric data is similar to another person's biometric data, repeatedly inputs the similar biometric data to gradually increase the posterior probability of the specific person, and is finally authenticated as the specific person, resulting in a decrease in security. However, as described above, by ensuring that the once input modality is not input twice, it is possible to prevent forgery of the user repeatedly inputting the same biometric data. In this way, security deterioration can be prevented.
 The feature extraction unit 102 of the authentication client terminal 100 extracts the feature data of the claimant v, based on the biometric data acquired from the claimant v through the biometric information input sensor 101 (step S304).
 The authentication request unit 104 of the authentication client terminal 100 transmits the claimant feature data extracted by the feature extraction unit 102, to the authentication server terminal 110 through the communication device 205. At the same time, the authentication request unit 104 requests the authentication server terminal 110 to identify the claimant v (step S305).
 Upon receiving the claimant feature data from the authentication client terminal 100, the authentication server terminal 110 increments the number of inputs J of the biometric data from the claimant v to J=J+1 (step S306). This process is performed, for example, by the determination unit 114.
 Next, the 1:N matching unit 112 of the authentication server terminal 110 performs 1:N matching between the claimant feature data extracted by the feature extraction unit 102 of the authentication client terminal 100, and the enrolled template 122 held in the enrolled user database 120 with respect to each of the N enrolled users u_{n}. Then, the 1:N matching unit 112 calculates the matching score s_{Jn }of each of the enrolled users u_{n }(step S307). An example of the method for calculating matching scores is the method of measuring the distance in the feature space. However, any other calculation methods may be used. Incidentally, with respect to the pruned enrolled user u_{n}, the matching process is not performed and the matching score is not calculated. Thus, the process is speeded up.
 Then, the likelihood ratio calculation unit 118 of the authentication server terminal 110 calculates the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }(the likelihood ratio) with respect to each of the enrolled users u_{n}, using the matching scores calculated by the 1:N matching unit 112 (step S308). Before calculating likelihood ratios it is necessary to calculate the matching scores of all the enrolled users u_{n}. However, the likelihood ratio of the pruned enrolled user u_{n }is not calculated and its matching score is not necessary. In this way, the process of calculating the likelihood ratio of the pruned enrolled user u_{n }can be eliminated to speed up the authentication process.
 The posterior probability calculation unit 113 of the authentication server terminal 110 calculates the posterior probability P (v=u_{n}S) of each of the enrolled users u_{n }and the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0}, using the likelihood ratios calculated by the likelihood ratio calculation unit 118, and using the prior probabilities of both the enrolled users u_{n }and the nonenrolled user u_{0}, which are set by the prior probability setting unit 111 (step S309). Here, S=[s_{jn}j=1 to J, n=1 to N].
 At this time, when the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0}, which is calculated in the current acquisition of the biometric data, is higher than the posterior probability in the previous (J−1)th acquisition of the biometric data (or higher than the prior probability when J=1), N matching scores (s_{J1}, s_{J2}, . . . , s_{JN}) calculated from the feature data extracted in the current acquisition, may not be used in the posterior probability calculation in the (J+1)th and subsequent acquisitions. In this embodiment, when a enrolled user attempts authentication under bad conditions involving the posture of inputting the biometric data (such as the finger position and the face direction), the lighting environment, and the like, the posterior probability of the enrolled user decreases each time the enrolled user inputs the biometric data. This even makes it difficult to correctly authenticate the enrolled user, even resulting in a decrease in availability. When the conditions such as the posture and the lighting environment are bad, the input biometric data is not similar to any of the biometric data of the enrolled users. Thus, the posterior probability of the nonenrolled user increases. As described above, when the posterior probability of the nonenrolled user in the current acquisition is higher than the posterior probability in the previous acquisition, the matching score calculated from the feature data extracted in the acquisition at this time, is not used in the subsequent calculation of posterior probabilities. In this way, it is possible to prevent a decrease in availability when the biometric data is input under bad conditions involving the posture of inputting the biometric data, the lighting environment, and the like. Hereinafter, a description will be given of the method for calculating the likelihood ratio, the posterior probability P (v=u_{n}S) of each of the enrolled users u_{n}, as well as the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0}.
 The posterior probability P (v=u_{n}S) of each of the enrolled users u_{n}, and the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }can be transformed into the following equations, respectively, using the Bayesian theory.

$\begin{array}{cc}P\ue8a0\left(v={u}_{n}S\right)=\frac{P\ue8a0\left(v=u\right)\ue89eP\ue8a0\left(Sv={u}_{n}\right)}{P\ue8a0\left(S\right)}\ue89e\text{}\ue89e\phantom{\rule{7.5em}{7.5ex}}=\frac{P\ue8a0\left(v={u}_{n}\right)\ue89eP\ue8a0\left(Sv={u}_{n}\right)}{\sum _{x=0}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89eP\ue8a0\left(Sv={u}_{x}\right)}& \left(2\right)\\ P\ue8a0\left(v={u}_{0}S\right)=\frac{P\ue8a0\left(v={u}_{0}\right)\ue89eP\ue8a0\left(Sv={u}_{0}\right)}{P\ue8a0\left(S\right)}\ue89e\text{}\ue89e\phantom{\rule{7.5em}{7.5ex}}=\frac{P\ue8a0\left(v={u}_{0}\right)\ue89eP\ue8a0\left(Sv={u}_{0}\right)}{\sum _{x=0}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89eP\ue8a0\left(Sv={u}_{x}\right)}& \left(3\right)\end{array}$  Here, the distribution of the first person to first person matching score s (firstperson distribution) is given by f(s), and the distribution of the first person to third person matching score s (thirdperson distribution) is given by g(s). Each of f(s) and g(s) can be calculated from the likelihood ratio assuming a normal distribution, Bayesian estimation, logistic regression or other statistical methods, as described in C. M. Bishop, “Pattern Recognition and Machine Learning”, SpringerVerlag; New Edition (2006) (the disclosure of this document is incorporated herein by reference in its entirety). In this embodiment, f(s) and g(s) may be calculated for each enrolled user, or may be shared by all the enrolled users.
 Using these, P (Sv=u_{n}) and P (Sv=u_{0}) can be expressed by the following equations, respectively.

$\begin{array}{cc}P\ue8a0\left(Sv={u}_{n}\right)=\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eP\ue8a0\left({S}_{\mathrm{ji}}v={u}_{n}\right)\ue89e\text{}\ue89e\phantom{\rule{7.2em}{7.2ex}}=\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jn}}\right)\ue89e\prod _{i=1,i\ne n}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)& \left(4\right)\\ P\ue8a0\left(Sv={u}_{0}\right)=\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eP\ue8a0\left({S}_{\mathrm{ji}}v={u}_{i}\right)\ue89e\text{}\ue89e\phantom{\rule{7.2em}{7.2ex}}=\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)& \left(5\right)\end{array}$  Thus, by substituting the above equations into equations (2) and (3), the posterior probability P (v=u_{n}S) of each of the enrolled users u_{n }and the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }can be expressed by the following equations (posterior probability calculation functions), respectively.

$\begin{array}{cc}\begin{array}{c}P\ue89e\left(v={u}_{n}S\right)=\ue89e\frac{P\ue8a0\left(v={u}_{n}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jn}}\right)\ue89e\prod _{i=1,i\ne n}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)}{\begin{array}{c}\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jx}}\right)\ue89e\prod _{i=1,i\ne x}^{N}\ue89eg\ue89e\left({S}_{\mathrm{ji}}\right)+\\ P\ue8a0\left(v={u}_{0}\right)\ue89e\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)\end{array}}\\ =\ue89e\frac{P\ue8a0\left(v={u}_{n}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jn}}\right)/g\ue8a0\left({S}_{\mathrm{ji}}\right)}{\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jx}\ue89e\phantom{\rule{0.3em}{0.3ex}}}\right)/g\ue8a0\left({S}_{\mathrm{jx}}\right)+P\ue8a0\left(v={u}_{0}\right)}\end{array}& \left(6\right)\\ \begin{array}{c}P\ue89e\left(v={u}_{0}S\right)=\ue89e\frac{P\ue8a0\left(v={u}_{0}\right)\ue89e\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)}{\begin{array}{c}\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jx}}\right)\ue89e\prod _{i=1,i\ne x}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)+\\ P\ue89e\left(v={u}_{0}\right)\ue89e\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89eg\ue8a0\left({S}_{\mathrm{ji}}\right)\end{array}}\\ =\ue89e\frac{P\ue8a0\left(v={u}_{0}\right)}{\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jx}}\right)/g\ue8a0\left({S}_{\mathrm{jx}}\right)+P\ue8a0\left(v={u}_{0}\right)}\end{array}& \left(7\right)\end{array}$  Here, the following equation included in the equations is the equation (likelihood ratio function) to calculate the likelihood ratio (the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n}), under the condition that the matching scores s_{1n }to s_{Jn }corresponding to the enrolled users u_{n }are obtained.

$\begin{array}{cc}\prod _{j=1}^{J}\ue89ef\ue8a0\left({S}_{\mathrm{jn}}\right)/g\ue8a0\left({S}_{\mathrm{jn}}\right)& \left(8\right)\end{array}$  The likelihood ratio calculation unit 118 calculates likelihood ratios using the likelihood ratio function described above.
 Then, the posterior probability calculation unit 113 calculates the posterior probability of each of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the likelihood ratios calculated by the likelihood ratio calculation unit 118 using the above posterior probability calculation functions. Note that since the posterior probabilities of the pruned enrolled user u_{n }and nonenrolled user u_{0 }are not calculated, the authentication process can be speeded up. Further, before calculating likelihood ratios it is necessary to calculate the matching scores of all the enrolled users u_{n}. However, the matching score of the pruned enrolled user u_{n }is not calculated as described above. At this time, the likelihood ratio of the pruned enrolled user u_{n }may be set to 0, or may be the same as the likelihood ratio calculated in the likelihood ratio calculation process (step S308) just before pruning. By eliminating the process of calculating the matching score, the likelihood ratio, and the posterior probability with respect to the pruned enrolled user u_{n }(and the pruned nonenrolled user u_{0 }when the nonenrolled user u_{0 }is also the target to be pruned), the authentication process can be speeded up.
 Another approach is described in Vladimir P. Dragalin, et al., “Multihypothesis Sequential Probability Ratio Tests, PART I: Asymptotic Optimality”, IEEE Trans., Information Theory, Volume 45, Issue 7, pp. 24482461 (1999) (the disclosure of this document is incorporated herein by reference in its entirety). This shows that in testing to determine which is true of L (≧2) hypotheses, under the assumption that the observational data is i.i.d. (independently following the same distribution) and that the decision error probabilities are sufficiently small, the likelihood ratio of each hypothesis is normalized to the probability value of achieving the specific hypothesis. The determination is made by comparing each of the probability values with a threshold A, thereby enabling to minimize the average number of observations necessary until the determination is completed.
 In this embodiment, the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }is calculated with respect to each of the enrolled users u_{n}. The calculated likelihood ratios are normalized to the posterior probabilities of the enrolled user u_{n}, and to the posterior probability of the nonenrolled user u_{0}. Then, the determination is made by comparing each of the posterior probabilities with the threshold A. This method minimizes the expected value of the number of inputs of the biometric data necessary for authentication.
 As shown in
FIG. 4 (the figure showing the first biometric data input), the determination unit 114 of the authentication server terminal 110 compares each of the posterior probabilities calculated by the posterior probability calculation unit 113 in step S309, with the threshold A. Then, the determination unit 114 verifies the identity of the claimant v according to the above described criteria (the description on the determination unit 114 inFIG. 1 ) (step S310).  In this embodiment, the threshold A used in the determination process is set to a value larger than 0.5. This setting of the threshold A ensures that the number of posterior probabilities exceeding the threshold A is not more than one. Because the number of posterior probability exceeding the threshold A is not more than one, a difference naturally occurs between the maximum value of the posterior probability and the next larger value of the posterior probability. Thus, it is possible to reduce the occurrence of misidentification (and increase security). This effect is increased by further increasing the value of the threshold A.
 As described above, the biometric authentication system 1 of this embodiment calculates the likelihood ratios and the posterior probabilities, using the firstperson matching score distribution and the thirdperson matching score distribution. The firstperson matching score distribution f(s) and the thirdperson matching score distribution g(s) are estimated using the matching scores of all the enrolled users u_{n}, instead of using the matching score of a single enrolled user u_{n}. For this reason, the distributions can be reliably estimated even if the amount of the matching score data previously collected from each of the enrolled users u_{n }is small. Thus, the likelihood ratio and the posterior probability can be reliably estimated, and EFAR, EFRR, NFAR can be reduced. As a result, the security of the system is increased, and its availability is further increased.
 Next, the determination unit 114 of the authentication server terminal 110 separates into the following processes according to the determination result in step S310.
 That is, when it is determined that the claimant v is identical to one of the enrolled users u_{n }(authentication success) in step S310, the determination unit 114 transmits the determination result information indicating authentication success, to the authentication client terminal 100 through the communication device 205 (
FIG. 2 ) (step S312). Further, when the biometric authentication system 1 is applied to the cardless credit payment system, the payment process and other related processes are performed. When the biometric authentication system 1 is applied to the time and attendance system, the door unlocking process, the automatic clockin process, and other related processes are performed.  In step S310, when it is determined that the claimant v is the nonenrolled user u_{0 }(forgery user), the determination unit 114 transmits the forgery warming information to the authentication client terminal 100 through the communication device 205 (
FIG. 2 ) (step S313).  If indeterminable in step S310, the determination unit 114 determines whether the number of acquisitions J of the biometric data from the claimant v is less than the predetermined value (J<J_{max}) (step S314).
 When the result of the determination in step S314 is “Yes”, in other words, when the number of acquisitions J of the biometric data does not reach the predetermined value yet, the enrolled user pruning unit 115 compares each of the posterior probabilities calculated by the posterior probability calculation unit 113, with the threshold B. Then, the enrolled user pruning unit 115 prunes the enrolled user u_{n }and nonenrolled user u_{0 }corresponding to the posterior probabilities smaller than the threshold B (step S315).
 For example, the authentication server terminal 110 is configured that the ID of the enrolled user u_{n }pruned away in step S315 is held in the memory. With this configuration, the respective processing units of the authentication server terminal 110 can recognize the pruned enrolled user u_{n}. When the pruning of enrolled users u_{n }is performed, the matching targets are narrowed down, for example, as in the case of
FIG. 4 in which although all the enrolled users u_{n }and nonenrolled user no are the matching targets in the first biometric data input, only the enrolled users u_{n1}, u_{nN }and the nonenrolled user u_{0}, are the matching targets in the second biometric data input.  With the biometric authentication system 1 of this embodiment, the pruning is performed based on comparing the threshold B to the normalized posterior probability of each of the enrolled users u_{n}, and to the normalized posterior probability of the nonenrolled user u_{0}. Thus, as described in the Vladimir P. Dragalin, et al., it is possible to minimize the expected value of the number of inputs J of the biometric data necessary for pruning the corresponding enrolled users u_{n }and nonenrolled user no. This is the same meaning as maximizing the expected value of the number of enrolled users u_{n }and nonenrolled user u_{0 }to be pruned away when the input number J is a certain value. Thus, the time for 1:N matching can be reduced, and availability can be further increased.
 After the pruning process (step S315), the authentication server terminal 110 (more specifically, for example, the determination unit 114) outputs the information requesting to reacquire the biometric data of the claimant v, as the determination result information indicating that it is necessary to perform the identity verification again, to the authentication client terminal 100 through the communication device 205 (
FIG. 2 ) (step S316).  Then, the authentication server terminal 110 waits until another determination request is issued from the authentication client terminal 100, in other words, until the claimant feature is transmitted from the authentication client terminal 100 (step S317). Upon receiving the claimant feature data from the authentication client terminal 100, the authentication server terminal 110 performs the processes in step S306 and the subsequent steps.
 On the other hand, when the determination result in step S314 is “No”, in other words, when the number of acquisitions J of the biometric data reaches the predetermined value, the authentication server terminal 110 outputs the determination result information indicating authentication failure, and transmits the information to the authentication client terminal 100 through the communication device (
FIG. 2 ) (step S318).  The authentication request unit 104 of the authentication client terminal 100 performs, for example, the following processes according to the determination result information transmitted from the authentication server terminal 110 (step S319).
 When receiving the determination result information indicating authentication success, the authentication request unit 104 presents (for example, displays or voices) a message saying “Authentication is successful”, to the claimant v through the output device 204.
 When receiving the forgery warning information, the authentication request unit 104 provides punishment for the forgery user, such as stopping the operation of the biometric information input sensor 101 to reject the authentication attempt, or alarming through the output device 204.
 When receiving the determination result information requesting to reacquire the biometric data, the authentication request unit 104 presents (for example, displays or voices) a message to ask the claimant v to input the biometric data, through the output device 204, such as saying “Put the index fingerprint on the sensor”.
 When receiving the determination result information indicating authentication failure, the authentication request unit 104 presents (for example, displays or voices) a message to ask the claimant v to reattempt to perform the identity verification, through the output device 204, such as saying “Authentication is failed. Start again from the beginning”.
 In the authentication server terminal 110, the determination unit 114 obtains the final determination result (“authentication success”, “forgery”, or “authentication failure”), and records the final determination result to the determination history information 130 stored in the storage unit 117 (step S320).
 In the biometric authentication system 1 of this embodiment, the prior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0 }are set in advance. Whereby, not only the posterior probabilities of the enrolled users u_{n }but also the posterior probability of the nonenrolled user u_{0 }is calculated using the prior probabilities and the likelihood ratios. When the claimant v is identified as the nonenrolled user u_{0}, the system determines “forgery”. When the claimant v is not identified (not identified as the enrolled user u_{n }or the nonenrolled user u_{0}), the system determines “authentication failure”. The two cases “forgery” and “authentication failure” are clearly distinguished, so that it is possible to provide punishments such as the processes in step S319, only when the system determines that the claimant v not previously registered is attempting forgery. This can effectively prevent malicious user forgery, and can increase the security of the system.
 In this embodiment, authentication is performed using plural different types of biometric information belonging to the same modality. The second embodiment will be described below focusing on the difference from the first embodiment.
 In step S307 of
FIG. 3 , the 1:N matching unit 112 of the authentication server terminal 110 performs 1:N matching between the claimant feature data extracted by the feature extraction unit 102 of the authentication client terminal 100, and M enrolled templates 122 held in the enrolled user database 120 with respect to each of N enrolled users u_{n}. Then, the 1:N matching unit 112 calculates the matching score s_{jnm }of the mth enrolled template with respect to each of the enrolled users u_{n}. In the following description, the distance in the feature space is used as the method for calculating matching scores. In this case, the smaller the matching score the better. In other words, the smaller the matching score, the more two biometric information sources are likely to be identical. At this time, authentication failure is determined when the smallest matching score is obtained several times with respect to the same biometric data of the same user until this (Jth) input. For example, assuming that the number of inputs is two, authentication failure is determined when the biometric data indicating the smallest matching score in a certain input is identical to the biometric data indicating the smallest matching score in the other input.  There is a risk that a user, whose biometric data is similar to another person's biometric data, repeatedly inputs the similar biometric data to gradually increase the posterior probability of the specific person, and is finally authenticated as the specific person, resulting in a decrease in security. This embodiment uses only one modality, and it is difficult to take measures to ensure that the once input modality will not be input twice. However, as described above, authentication failure is determined when the smallest matching score is obtained several times with respect to the same biometric data, to prevent forgery of a user repeatedly inputting the same biometric data. In this way, it is possible to prevent a decrease in security.
 However, in the above described case, authentication failure is typically determined when the claimant inputs the same biometric data multiple times. This results in a decrease in availability. In order to avoid this disadvantage, when a smallest matching score is obtained multiple times (K times) with respect to the same biometric data until this (Jth) input, ((K−1)×N×M) matching scores obtained in inputs other than the input achieving the most smallest matching score, are not used in the likelihood ratio calculation in step S308. This prevents forgery of a user repeatedly inputting the same biometric data to gradually increase the posterior probability. At the same time, it is possible to avoid the problem of authentication failure typically occurring when the claimant inputs the same biometric data multiple times. As a result, it is possible to prevent a decrease in security and availability.
 In steps S308 and S309, the likelihood ratio, the posterior probability P (v=u_{n}S) of each of the enrolled users u_{n}, as well as the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }are respectively calculated as follows. Here, S=[s_{jnm}j=1 to J, n=1 to N, m=1 to M]
 The posterior probability P (v=u_{n}S) of each of the enrolled users u_{n }and the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }can be transformed into the following equations, respectively, using the Bayesian theory.

$\begin{array}{cc}\begin{array}{c}P\ue8a0\left(v={u}_{n}S\right)=\ue89e\frac{P\ue8a0\left(v={u}_{n}\right)\ue89eP\ue8a0\left(Sv={u}_{n}\right)}{P\ue8a0\left(S\right)}\\ =\ue89e\frac{P\ue8a0\left(v={u}_{n}\right)\ue89eP\ue8a0\left(Sv={u}_{n}\right)}{\sum _{x=0}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89eP\ue8a0\left(Sv={u}_{x}\right)}\end{array}& \left(9\right)\\ \begin{array}{c}P\ue8a0\left(v={u}_{0}S\right)=\ue89e\frac{P\ue8a0\left(v={u}_{0}\right)\ue89eP\ue8a0\left(Sv={u}_{0}\right)}{P\ue8a0\left(S\right)}\\ =\ue89e\frac{P\ue8a0\left(v={u}_{0}\right)\ue89eP\ue8a0\left(Sv={u}_{0}\right)}{\sum _{x=0}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89eP\ue8a0\left(Sv={u}_{x}\right)}\end{array}& \left(10\right)\end{array}$  Here, assuming that the probability that the claimant will input the mth biometric data (m=1, 2, . . . M) is all equal to 1/M, P (v=u_{n}S) and P (v=u_{0}S) are given by the following equations, respectively.

$\begin{array}{cc}\begin{array}{c}P\ue8a0\left(Sv={u}_{n}\right)=\ue89e\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89e\prod _{k=1}^{M}\ue89eP\ue8a0\left({S}_{\mathrm{jik}}v={u}_{n}\right)\\ =\ue89e\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89e\prod _{k=1}^{M}\\ \ue89e\left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89eP\ue8a0\left(\begin{array}{c}{S}_{\mathrm{jik}}v={u}_{n},\\ \begin{array}{c}\begin{array}{c}\mathrm{Input}\ue89e\phantom{\rule{0.8em}{0.8ex}}\ue89em\ue89e\text{}\ue89e\mathrm{th}\\ \mathrm{biometric}\ue89e\phantom{\rule{1.1em}{1.1ex}}\ue89e\mathrm{information}\end{array}\\ \mathrm{for}\ue89e\phantom{\rule{1.1em}{1.1ex}}\ue89eJ\ue89e\text{}\ue89e\mathrm{th}\ue89e\phantom{\rule{1.1em}{1.1ex}}\ue89e\mathrm{input}\end{array}\ue89e\phantom{\rule{0.3em}{0.3ex}}\end{array}\right)\right\}\\ =\ue89e\prod _{j=1}^{J}\ue89e\left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89ef\ue8a0\left({S}_{\mathrm{jnm}}\right)/g\ue8a0\left({S}_{\mathrm{jnm}}\right)\right\}\ue89e\prod _{i=1}^{N}\ue89e\prod _{k=1}^{M}\ue89eg\ue8a0\left({S}_{\mathrm{jik}}\right)\end{array}& \left(11\right)\\ P\ue8a0\left(Sv={u}_{0}\right)=\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89e\prod _{k=1}^{M}\ue89eP\ue8a0\left({S}_{\mathrm{jik}}v={u}_{0}\right)\ue89e\text{}\ue89e\phantom{\rule{7.5em}{7.5ex}}=\prod _{j=1}^{J}\ue89e\prod _{i=1}^{N}\ue89e\prod _{k=1}^{M}\ue89eg\ue8a0\left({S}_{\mathrm{jik}}\right)& \left(12\right)\end{array}$  Thus, by substituting equations (11) and (12) into equations (9) and (10), the posterior probability P (v=u_{n}S) of each of the enrolled users u_{n }and the posterior probability P (v=u_{0}S) of the nonenrolled user u_{0 }are given by the following equations (posterior probability functions), respectively.

$\begin{array}{cc}P\ue8a0\left(v={u}_{n}S\right)=\frac{P\ue8a0\left(v={u}_{n}\right)\ue89e\prod _{j=1}^{J}\ue89e\left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89ef\ue8a0\left({S}_{\mathrm{jnm}}\right)/g\ue8a0\left({S}_{\mathrm{jnm}}\right)\right\}}{\begin{array}{c}\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\\ \left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89ef\ue8a0\left({S}_{\mathrm{jxm}}\right)/g\ue8a0\left({S}_{\mathrm{jxm}}\right)\right\}+P\ue8a0\left(v={u}_{0}\right)\end{array}}& \left(13\right)\\ P\ue8a0\left(v={u}_{0}S\right)=\frac{P\ue8a0\left(v={u}_{0}\right)}{\begin{array}{c}\sum _{x=1}^{N}\ue89eP\ue8a0\left(v={u}_{x}\right)\ue89e\prod _{j=1}^{J}\\ \left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89ef\ue8a0\left({S}_{\mathrm{jxm}}\right)/g\ue8a0\left({S}_{\mathrm{jxm}}\right)\right\}+P\ue8a0\left(v={u}_{0}\right)\end{array}}& \left(14\right)\end{array}$  Here, the following equation included in the above posterior probability calculation functions, is the equation (likelihood ratio function) to calculate the likelihood ratio (the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n}), under the condition that the matching scores s_{1nm }to s_{Jnm }corresponding to the mth biometric data of the enrolled users u_{n}, are obtained.

$\begin{array}{cc}\prod _{j=1}^{J}\ue89e\left\{\frac{1}{M}\ue89e\sum _{m=1}^{M}\ue89ef\ue8a0\left({S}_{\mathrm{jnm}}\right)/g\ue8a0\left({S}_{\mathrm{jnm}}\right)\right\}& \left(15\right)\end{array}$  The likelihood ratio calculation unit 118 calculates likelihood ratios using the likelihood ratio function in step S308. Then, the posterior probability calculation unit 113 calculates the posterior probability of each of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the likelihood ratios calculated by the likelihood ratio calculation unit 118 using the posterior probability calculation functions.
 A biometric authentication system 2 of this embodiment is a cardless credit payment system. The third embodiment will be described below focusing on the difference from the first embodiment.

FIG. 7 shows an example of the configuration of the biometric authentication system 2.  A loss calculation unit 102 a calculates a loss W_{1 }occurring when v is a enrolled user and is identified as another enrolled user, and a loss W_{0 }occurring when v is a nonenrolled user and is identified as a certain enrolled user, based on the price Q of a product the claimant v purchased.
 A prior probability initial setting unit 111 a initializes the posterior probabilities of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}.
 A prior probability and threshold setting unit 111 b sets the prior probability of each of the enrolled users u_{n }and a corresponding threshold, as well as the prior probability of the nonenrolled user u_{0 }and a corresponding threshold, using the losses W_{1 }and W_{0}. So this ensures that both EFRR and the expected value R_{n }of the loss when v is identified as the nth enrolled user, are equal to or less than the desired values EFRR′ and R_{n}′, respectively. In this embodiment, the threshold is set for each enrolled user and nonenrolled user.

FIG. 8 shows the procedure of the authentication process by the biometric authentication system 2 of this embodiment, and the data flow associated with the authentication process.  In step S302, the prior probability initial setting unit 111 a of the authentication server terminal 110 initializes the prior probability P (v=u_{n}) of each of the enrolled users u_{n}, as well as the prior probability P (v=u_{0}) of the nonenrolled user u_{0}. Here, the initial values of the prior probabilities P (v=u_{n}) and P (v=u_{0}) are given by π_{n}, π_{0}, respectively.
 In step S302 a, the loss calculation unit 102 a of the authentication client terminal 100 calculates the loss W_{1 }occurring when v is a enrolled user and is identified as another enrolled user, and the loss W_{0 }occurring when v is a nonenrolled user and is identified as a certain enrolled user, based on the price Q of a product the claimant v purchased. The price Q is input, for example, by a store clerk through a keyboard. The losses W_{1 }and W_{0 }are, for example, assumed to be proportional to the price Q, and are calculated using the following equations, where α, β are invariables.

W_{1}=αQ (16) 
W_{0}=βQ (17)  In step S305, the authentication request unit 104 of the authentication client terminal 100 transmits the claimant feature data extracted by the feature extraction unit 102, as well as the losses W_{1 }and W_{0 }calculated by the loss calculation unit 102 a, to the authentication server terminal 110 through the communication device 205. At the same time, the authentication request unit 104 requests the authentication server terminal 110 to identify the claimant v.
 In step S306 a, the prior probability and threshold setting unit 111 b of the authentication server terminal 110 sets the prior probability P (v=u_{n}) of each of N enrolled users u_{n }and a threshold A_{n}, as well as the prior probability P (v=u_{0}) of the nonenrolled user u_{0 }and a threshold A_{0}, using the losses W_{1 }and W_{0 }received from the authentication client terminal 100. So this ensures that both EFRR and the expected value R_{n }of the loss when v is identified as the nth enrolled user, are equal to or less than the desired values EFRR′ and R_{n}′, respectively. The prior probabilities and the thresholds are given by the following equations.

$\begin{array}{cc}P\ue8a0\left(v={u}_{n}\right)=\frac{{\pi}_{n}\ue89e{W}_{1}}{\sum _{k=1}^{N}\ue89e{\pi}_{k}\ue89e{W}_{1}+{\pi}_{0}\ue89e{W}_{0}}& \left(18\right)\\ {A}_{n}=\frac{{W}_{1}\ue89e\mathrm{exp}\ue8a0\left[{a}_{n}\right]}{1+{W}_{1}\ue89e\mathrm{exp}\ue8a0\left[{a}_{n}\right]}& \left(19\right)\\ P\ue8a0\left(v={u}_{0}\right)=\frac{{\pi}_{0}\ue89e{W}_{0}}{\sum _{k=1}^{N}\ue89e{\pi}_{x}\ue89e{W}_{1}+{\pi}_{0}\ue89e{W}_{0}}& \left(20\right)\\ {A}_{0}=\frac{{W}_{0}\ue89e\mathrm{exp}\ue8a0\left[{a}_{0}\right]}{{W}_{1}+{W}_{0}\ue89e\mathrm{exp}\ue8a0\left[{a}_{0}\right]}& \left(21\right)\end{array}$  Incidentally, a_{n }and a_{0 }are variables expressed by the following equations.

$\begin{array}{cc}{a}_{n}=\mathrm{log}\ue8a0\left[\frac{{\pi}_{n}}{{R}_{n}^{\prime}}\right]& \left(22\right)\\ {a}_{0}=\mathrm{log}\ue8a0\left[\frac{{\pi}_{0}}{\left(1{\pi}_{0}\right)\ue89e{\mathrm{EFRR}}^{\prime}}\right]& \left(23\right)\end{array}$  When the prior probabilities and the thresholds are set as described above, EFRR and R_{n }are equal to or less than the desired values EFRR′ and R_{n}′, respectively. This can be proved as follows.
 According to the Vladimir P. Dragalin, et al., the expected value R_{i }of the loss when the claimant v is identified as the ith enrolled user, can be expressed as follows, using the probability a_{ji }that v=u_{j }will be identified as v=u_{i }and using the losses W (j, i) at this time.

$\begin{array}{cc}{R}_{i}=\sum _{j=0,j\ne i}^{N}\ue89e{\pi}_{j}\ue89eW\ue8a0\left(j,i\right)\ue89e{\alpha}_{\mathrm{ji}}& \left(24\right)\end{array}$  Here, let the probability be denoted by a_{ji }and let the losses be denoted by W (j, i). Then, using the following equation, the fact that EFRR and R_{n }are equal to or less than the desired values EFRR′ and R_{n}′, can be derived referring to the Vladimir P. Dragalin, et al.

$\begin{array}{cc}{\alpha}_{\mathrm{ji}}=\{\begin{array}{c}\frac{\mathrm{EFAR}}{N1}\ue89e\left(j\ne i,j\ne 0,i\ne 0\right)\\ \mathrm{EFRR}\ue8a0\left(j\ne i,i=0\right)\ue89e\phantom{\rule{4.2em}{4.2ex}}\\ \frac{\mathrm{EFAR}}{N}\ue89e\left(j\ne i,j=0\right)\ue89e\phantom{\rule{3.6em}{3.6ex}}\end{array}& \left(25\right)\\ W\ue8a0\left(j,i\right)=\{\begin{array}{c}{W}_{1}\ue8a0\left(j\ne i,j\ne 0,i\ne 0\right)\\ 1\ue89e\left(j\ne i,j=0\right)\ue89e\phantom{\rule{4.4em}{4.4ex}}\\ {W}_{2}\ue8a0\left(j\ne i,j=0\right)\ue89e\phantom{\rule{3.1em}{3.1ex}}\end{array}& \left(26\right)\end{array}$  As described above, each time authentication is performed, the losses W_{1 }and W_{0 }are calculated, and the prior probabilities and the thresholds are set based on the calculated losses W_{1 }and W_{0}. Thus, it is possible to ensure that EFRR as well as the expected value R_{n }of the loss when the claimant v is identified as the nth enrolled user, are equal to or less than the desired values EFRR′ and R_{n}′, respectively. As a result, the desired availability and security can be reliably achieved.
 While the embodiments of the present invention have been described, it should be understood that the following variant can also be used.
 In the above described embodiments, the threshold A used in the determination process is set to a value larger than 0.5. However, the threshold A may be set to 0.5 or less. In the case of the threshold A set to 0.5 or less, if plural posterior probabilities exceeding the threshold A exist, the determination unit 114 determines based on the posterior probability showing the maximum value of the posterior probabilities exceeding the threshold A.
 Further, in the biometric authentication systems of the above described embodiments, the authentication client terminal 100 and the authentication server terminal 110 are coupled through the network 140. However, the configurations of the terminals 100 and 110 may be put together to constitute a single unit of the biometric authentication system. In such a case, the authentication request unit 104, which is provided in the authentication client terminal 100 in the above embodiments, can be omitted. The authentication client terminal 100 and the authentication server terminal 110 are not limited to the configurations shown in
FIGS. 1 and 7 . For example, the processing units 102, 111 to 115, and 118, which are realized by the CPU executing a program, may be provided either to the authentication client terminal 100 or to the authentication server terminal 110. In addition, all or some of the processing units may be put together as a single unit, or may be broken into separate components.  In the above described embodiments, the likelihood ratio calculation unit 118 calculates the likelihood ratio of each of the enrolled users u_{n}, using the matching scores obtained by 1:N matching unit 112 in step S307 of
FIG. 3 (step S308). However, the likelihood ratio may be calculated for each of the enrolled users u_{n}, based on the biometric information enrolled templates 122 held in the enrolled user database 120, as well as on the claimant feature data extracted by the feature extracting unit 102 of the authentication client terminal 100. More specifically, each time the claimant feature data x_{J }(J=1, 2, and so on) is obtained, the likelihood l_{n }of each of the enrolled users u_{n }is calculated by the following equation, using the distribution p_{n}(x_{J}) of the feature of the individual enrolled users u_{n }(n=1 to N) and using the distribution p_{0}(x_{J}) of the feature of all the enrolled users. 
$\begin{array}{cc}{l}_{n}=\prod _{j=1}^{J}\ue89e{P}_{n}\ue8a0\left({x}_{j}\right)/{P}_{0}\ue8a0\left({x}_{j}\right)& \left(27\right)\end{array}$  It is possible to calculate the posterior probability of each of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, by replacing the calculated likelihood ratios with the likelihood ratios used in the posterior probability calculation functions described in the above embodiments.
 In the above described embodiments, the authentication server terminal 110 makes a determination of the claimant v. However, this determination may be made by the authentication client terminal 100. For example, as shown in
FIG. 6 , the authentication client terminal 100 is provided with a determination unit 106, while the authentication server terminal 110 is provided with a transmission information decision unit 116 in place of the determination unit 114 (FIG. 1 ). The other configurations of the authentication client terminal 100 and the authentication server terminal 110, and the configurations of the biometric information sensor 101 and the network 140, are basically the same as the respective configurations in the above embodiments, and are denoted by the same reference numerals. The authentication process procedure in the present configuration will be descried below.  First, the authentication request unit 104 of the authentication client terminal 100 transmits the claimant feature data extracted by the feature extraction unit 102, to the authentication server terminal 110. At the same time, the authentication request unit 104 calculates the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the claimant feature data and the matching feature data held in the enrolled user database 120. Then, the authentication request unit 104 calculates the posterior probability of each of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, and using the prior probabilities of both the enrolled users u_{n }and the nonenrolled user u_{0}, which are set by the prior probability setting unit 111 of the authentication server terminal 110. Then the authentication request unit 104 compares each of the calculated posterior probabilities, with the threshold A. When the maximum value of the posterior probability is larger than the threshold A, the authentication request unit 104 requests the authentication server terminal 110 to transmit the information of the enrolled user u_{n }or nonenrolled user u_{0 }corresponding to the posterior probability showing the maximum value.
 In the authentication server terminal 110, in response to the request from the authentication client terminal 100, the prior probability setting unit 111, the 1:N matching unit 112, the likelihood ratio calculation unit 118, and the posterior probability calculation unit 113 perform the same processes as in the above embodiments, to calculate posterior probabilities. Then, the transmission information decision unit 116 compares each of the posterior probabilities calculated by the posterior probability calculation unit 113, with the threshold A. When the maximum value of the posterior probability is larger than the threshold A, the transmission information decision unit 116 decides the information of the enrolled user u_{n }or nonenrolled user u_{0 }corresponding to the posterior probability showing the maximum value, and transmits the information to the authentication client terminal 100.
 The authentication request unit 104 of the authentication client terminal 100 receives the information transmitted from the authentication server terminal 110, and asks the determination unit 106 to make a determination of the claimant v.
 When the information received by the authentication request unit 104 is the information of the enrolled user u_{n}, the determination unit 106 of the authentication client terminal 100 determines that the enrolled user u_{n }is identical to the claimant v. While when the information received by the authentication request unit 104 is the information of the nonenrolled user u_{0}, the determination unit 106 determines that the claimant v is a user other than the enrolled users u_{n}.
 Then, the authentication request unit 104 preferably performs a process similar to the above embodiment (step S319 of
FIG. 3 ), based on the determination result by the determination unit 106.  As another embodiment, the determination unit 114 of the authentication server 110 or the determination unit 106 of the authentication client terminal 100, may be configured to perform authentication by comparing the likelihood ratio (the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n}) calculated by the likelihood ratio calculation unit 118 with respect to each of the enrolled users u_{n }in step S308 of
FIG. 3 , with a threshold C, instead of using the posterior probabilities. More specifically, when the maximum value of the likelihood ratio is larger than the threshold C, the determination unit 114 or 106 determines that the claimant v is identical to the enrolled user corresponding to the likelihood ratio showing the maximum value (authentication success). While when the maximum value of the likelihood ratio is equal to or less than the threshold C, and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit 114 or 106 requests to reacquire the biometric data of the claimant v. Further, when the number of acquisitions of the biometric data from the claimant v is equal to or more than the predetermined value, the determination unit 114 or 106 determines “authentication failure”.  In the above case, the concept of prior and posterior probabilities is not used, so that steps S302 and S309 in the flowchart of
FIG. 3 are not performed. Also, the concept of nonenrolled user u_{0 }is not used, so that step S313 is not performed either.  In the above case, for example, when the determination unit 114 or 106 requests to reacquire the biometric data, the enrolled user pruning unit 115 performs a pruning process by comparing each of the likelihood ratios calculated by the likelihood ratio calculation unit 118, with a threshold D, and excluding a enrolled user u_{n }corresponding to the likelihood ratio smaller than the threshold D, from the target to be matched with the claimant v. The timing of the pruning may be between step S314 and step S315 in
FIG. 3 (when the determination unit 114 or 106 requests to reacquire the biometric data), or may be after step S315.  In the foregoing description, the determination is made by calculating the likelihood ratio of each of the enrolled users u_{n }based on the matching score of each of the enrolled users u_{n}, and by comparing each of the calculated likelihood ratios with the threshold C, in order to minimize the expected value of the number of inputs of the biometric data necessary for authentication. Further, in the conventional method for calculating likelihood ratios using the feature distribution, generally the dimensionality of the feature is large (for example, 2048 to 4096 dimensions for iris pattern). Very large numbers of study samples are necessary to estimate the feature distribution, and there is a problem of difficulty in practical application. On the other hand, when the likelihood ratio is calculated using the matching score whose dimensionality is typically one, it is possible to reliably estimate the firstperson matching score distribution and the thirdperson matching score distribution.
 According to the above described embodiments, the determination is made by calculating the ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, normalizing the likelihood ratios to the posterior probabilities of the enrolled users u_{n }and to the posterior probability of the nonenrolled user u_{0}, and comparing each of the posterior probabilities with the threshold. This method significantly reduces (logically minimizes) the expected value of the number of inputs of the biometric data necessary for authentication.
 Alternatively, the determination is made by calculating the matching score of each of the enrolled users u_{n}, calculating the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores, and comparing each of the likelihood ratios with the threshold. This method also significantly reduces (logically minimizes) the expected value of the number of inputs of the biometric data necessary for authentication.
 Further, the prior probabilities of the enrolled users u_{n }and the prior probability of the nonenrolled user u_{0 }are both set in advance, whereby not only the posterior probabilities of the enrolled users u_{n }but also the posterior probability of the nonenrolled user u_{0 }is calculated. This method enables to clearly distinguish between the case in which the claimant v is identical to the nonenrolled user u_{0 }(forgery), and the case in which the claimant v is not identified (authentication failure). When the forgery user can be clearly recognized, it is possible to take measures (for example, such as presenting a warning information message, and stopping the operation of the sensor) to effectively prevent such a malicious user forgery.
 The present invention can be applied to any biometric authentication system to verify identities based on biometric identification. For example, the present invention can be applied to a time and attendance management system, a personal authentication system for credit card payment, and an entrance/exit management system.
 The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. It will, however, be evident that various modifications and changes may be made thereto without departing from the spirit and scope of the invention as set forth in the claims.
Claims (18)
1. A biometric authentication system comprising:
a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N);
a feature extraction unit for extracting feature data of a claimant v based on biometric data acquired from the claimant v;
a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0};
a likelihood ratio calculation unit for calculating a ratio of the likelihood V=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the enrolled feature data held in the database and the extracted claimant feature data of the claimant;
a posterior probability calculation unit for calculating posterior probabilities of the enrolled users u_{n }and a posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the prior probabilities of the enrolled users u_{n}, and the prior probability of the nonenrolled user u_{0}; and
a determination unit for comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, and when the posterior probability showing the maximum value is one of the posterior probabilities of the enrolled users u_{n}, the determination unit determining that the corresponding enrolled user and the claimant v are identical, while when the posterior probability showing the maximum value is the posterior probability of the nonenrolled user u_{0}, the determination unit determining that the claimant v is a nonenrolled user u_{0}.
2. The biometric authentication system according to claim 1 , further comprising a 1:N matching unit for performing 1:N matching between the extracted claimant feature data of the claimant and the enrolled feature data held in the database, and calculating a matching score of each of the enrolled users u_{n},
wherein the likelihood ratio calculation unit calculates the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
3. The biometric authentication system according to claim 1 ,
wherein, when the maximum value of the posterior probability is equal to or less than the first threshold, and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit requests to reacquire the biometric data of the claimant v, while when the number of acquisitions is equal to or more than the predetermined value, the determination unit determines “authentication failure”.
4. The biometric authentication system according to claim 3 , further comprising an enrolled user pruning unit,
wherein, when the determination unit requests to reacquire the biometric data from the claimant v, the enrolled user pruning unit compares each of the posterior probabilities calculated by the posterior calculation unit with a second threshold, and excludes an enrolled user u_{n }corresponding to the posterior probability smaller than the second threshold, from the target to be matched to the claimant v.
5. The biometric authentication system according to claim 3 ,
wherein the posterior probability calculation unit compares the posterior probability of the nonenrolled user u_{0 }calculated in the current acquisition of the biometric data from the claimant v with the prior probability of the nonenrolled user u_{0 }when the number of acquisitions is one, or with the posterior probability of the nonenrolled user u_{0 }calculated in the previous acquisition of the biometric data when the number of acquisitions is two or more, and
wherein, when the posterior probability of the nonenrolled user u_{0 }calculated in the current acquisition is higher, the posterior probability calculation unit does not use the feature data extracted in the current acquisition in the calculation of posterior probabilities in the subsequent acquisitions of the biometric data from the claimant v.
6. The biometric authentication system according to claim 3 ,
wherein the database holds a plurality of different types of feature data belonging to a plurality of modalities with respect to each of the users u_{n}, and
wherein the biometric authentication system further comprises a biometric information input sensor that allows for acquisition of the biometric data only once with respect to each of modalities.
7. The biometric authentication system according to claim 3 ,
wherein the database holds a plurality of different types of feature data belonging to the same modality with respect to each of the enrolled users u_{n}, and
wherein the 1:N matching unit determines “authentication failure” when the best matching score is obtained several times for the same biometric information until acquisition of the biometric data in this time.
8. The biometric authentication system according to claim 3 ,
wherein the database holds a plurality of different types of feature data belonging to the same modality with respect to each of the enrolled users u_{n}, and
wherein, when the best matching score is obtained several times for the same biometric information until acquisition of the biometric information in this time, the 1:N matching unit does not use among the matching score above the matching scores calculated in acquisitions other than the acquisition with the best matching score for the calculation of likelihood ratios by the likelihood ratio calculation unit.
9. The biometric authentication system according to claim 1 ,
wherein, when the determination unit determines that the claimant v is a nonenrolled users u_{0}, the prior probability setting unit increases the prior probabilities of the nonenrolled users u_{0}.
10. The biometric authentication system according to claim 1 ,
wherein the first threshold is larger than 0.5.
11. The biometric authentication system according to claim 1 ,
wherein the determination unit outputs forgery warning information about the forgery when it is determined that the claimant v is the nonenrolled users u_{0}.
12. The biometric authentication system according to claim 1 , further comprising:
a loss calculation unit; and
a prior probability and threshold setting unit,
wherein, each time authentication is performed, the loss calculation unit calculates a loss W_{1 }when the claimant v is an enrolled user and is identified as another enrolled user, and a loss W_{0 }when the claimant v is a nonenrolled user and is identified as a certain enrolled user, and
wherein the prior probability and threshold setting unit sets the prior probabilities of the enrolled users u_{n }and a corresponding threshold, as well as the prior probability of the nonenrolled user u_{0 }and a corresponding threshold, using the losses W_{1 }and W_{0}, respectively, so that EFRR as well as an expected value R_{n }of the loss when the claimant v is identified as the nth enrolled user are equal to or less than respective desired values.
13. A biometric authentication system comprising:
a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N);
a feature extraction unit for extracting feature data of a claimant v based on biometric data acquired from the claimant v;
a 1:N matching unit for performing 1:N matching between the extracted feature data of the claimant and the feature data held in the database, and calculating a matching score of each of the enrolled users u_{n};
a likelihood ratio calculation unit for calculating a ratio of the likelihood v=u_{n }to the likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, using the calculated matching scores; and
a determination unit for comparing each of the likelihood ratios with a first threshold, when a maximum value of the likelihood ratio is larger than the first threshold, the determination unit determining that the enrolled user corresponding to the likelihood ratio showing the maximum value is identical to the claimant v, while when the maximum value of the likelihood ratio is equal to or less than the first threshold, and when the number of acquisitions of the biometric data from the claimant v is less than a predetermined value, the determination unit requesting to reacquire the biometric data of the claimant v, and when the number of acquisitions is equal to or more than the predetermined value, the determination unit determining “authentication failure”.
14. The biometric authentication system according to claim 13 , further comprising an enrolled user pruning unit,
wherein, when the determination unit requests to reacquire the biometric data, the enrolled user pruning unit compares each of the likelihood ratios calculated by the likelihood ratio unit with a second threshold, and excludes an enrolled user u_{n }corresponding to the likelihood ratio smaller than the second threshold, from the target to be matched to the claimant v.
15. An authentication client terminal capable of communicating with an authentication server terminal, the authentication server terminal including a database holding enrolled feature data with respect to each of enrolled users u_{n }(n=1 to N), and a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0},
the authentication client terminal comprising:
a feature extraction unit for extracting the feature data of a claimant v based on the biometric data acquired from the claimant v;
an authentication request unit for transmitting the extracted feature data of the claimant to the authentication server terminal, calculating a ratio of likelihood v=u_{n }to likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the feature data of the claimant and the matching feature data held in the database, calculating the posterior probabilities of the enrolled users u_{n }and the posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the prior probabilities of the enrolled users u_{n}, and the prior probability of the nonenrolled user u_{0}, comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, the authentication request unit requesting to transmit the information of the enrolled user u_{n }or nonenrolled user u_{0 }corresponding to the posterior probability showing the maximum value, and acquiring the information transmitted from the authentication server terminal in response to the request; and
a determination unit for determining that when the acquired information is of the enrolled users u_{n}, the enrolled user u_{n }and the claimant v are identical, and that when the acquired information is of the nonenrolled user u_{0}, the claimant v is a user other than the enrolled users u_{n}.
16. The authentication client terminal according to claim 15 ,
wherein the authentication request unit requests to perform 1:N matching between the extracted feature data of the claimant and the feature data held in the database, calculates the matching score of each of the enrolled users u_{n}, and calculates the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
17. A biometric authentication method comprising:
a step of extracting feature data of a claimant v based on the biometric data acquired from the claimant v;
a prior probability setting unit for setting prior probabilities of the enrolled users u_{n }and a prior probability of a nonenrolled user u_{0};
a step of calculating a ratio of likelihood v=u_{n }to likelihood v≠u_{n }with respect to each of the enrolled users u_{n}, based on the feature data held in the database with respect to each of the enrolled users u_{n }(n=1 to N), as well as on the extracted claimant feature data;
a step of calculating posterior probabilities of the enrolled users u_{n }and a posterior probability of the nonenrolled user u_{0}, using the calculated likelihood ratios, the prior probabilities of the enrolled users u_{n}, and the prior probability of the nonenrolled user u_{0}; and
a step of comparing each of the calculated posterior probabilities with a first threshold, when a maximum value of the posterior probability is larger than the first threshold, and when the posterior probability showing the maximum value is one of the posterior probabilities of the enrolled users u_{n}, determining that the corresponding enrolled user and the claimant v are identical, while when the posterior probability showing the maximum value is the posterior probability of the nonenrolled user u_{0}, determining that the claimant v is a nonenrolled users u_{0}.
18. The biometric authentication method according to claim 17 , further comprising a matching step of performing 1:N matching between the extracted feature data of the claimant and the feature data held in the database, and calculating a matching score of each of the enrolled users u_{n},
wherein the likelihood calculation step involves calculating the likelihood ratio of each of the enrolled users u_{n }by using the calculated matching scores.
Priority Applications (4)
Application Number  Priority Date  Filing Date  Title 

JP2008119131  20080430  
JP2008119131  20080430  
JP2008312411  20081208  
JP2008312411A JP5121681B2 (en)  20080430  20081208  Biometric authentication system, authentication client terminal, and biometric authentication method 
Publications (2)
Publication Number  Publication Date 

US20090289760A1 true US20090289760A1 (en)  20091126 
US8340361B2 US8340361B2 (en)  20121225 
Family
ID=40911035
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

US12/379,715 Active 20311021 US8340361B2 (en)  20080430  20090227  Biometric authentication system, authentication client terminal, and biometric authentication method 
Country Status (4)
Country  Link 

US (1)  US8340361B2 (en) 
EP (1)  EP2113867A3 (en) 
JP (1)  JP5121681B2 (en) 
CN (1)  CN101571920B (en) 
Cited By (19)
Publication number  Priority date  Publication date  Assignee  Title 

US20070172114A1 (en) *  20060120  20070726  The Johns Hopkins University  Fusing Multimodal Biometrics with Quality Estimates via a Bayesian Belief Network 
US20100214062A1 (en) *  20090220  20100826  Fujitsu Limited  Verification apparatus and authentication apparatus 
US7835548B1 (en)  20100301  20101116  Daon Holding Limited  Method and system for conducting identity matching 
EP2348458A1 (en)  20100126  20110727  Hitachi, Ltd.  Biometric authentication system 
US20120208529A1 (en) *  20090924  20120816  Wave Guard Technologies Ltd.  System and method of online radiation management and control of nonionizing radiation sources 
US20130207779A1 (en) *  20100930  20130815  Fujitsu Limited  Device and method for biometric authentication 
US20140007210A1 (en) *  20111212  20140102  Hitachi, Ltd.  High security biometric authentication system 
US20140314283A1 (en) *  20130416  20141023  Imageware Systems, Inc.  Multimodal biometric database searching methods 
US20140313007A1 (en) *  20130416  20141023  Imageware Systems, Inc.  Conditional and situational biometric authentication and enrollment 
US8989520B2 (en)  20100301  20150324  Daon Holdings Limited  Method and system for conducting identification matching 
US20150222630A1 (en) *  20130521  20150806  Tencent Technology (Shenzhen) Company Limited  Third application login method and corresponding apparatus and terminal 
US20160188863A1 (en) *  20140205  20160630  International Business Machines Corporation  Biometric authentication 
US20160239650A1 (en) *  20150215  20160818  Alibaba Group Holding Limited  System and method for user identity verification, and client and server by use thereof 
US20170142101A1 (en) *  20151116  20170518  International Business Machines Corporation  Secure Biometrics Matching with Split Phase ClientServer Matching Protocol 
CN108540457A (en) *  20180320  20180914  深圳市文鼎创数据科技有限公司  A kind of safety equipment and its biological identification control method and device 
US10332378B2 (en) *  20171011  20190625  Lenovo (Singapore) Pte. Ltd.  Determining user risk 
US10873578B1 (en)  20191209  20201222  Evan Chase Rose  Biometric authentication, decentralized learning framework, and adaptive security protocols in distributed terminal network 
US10902705B1 (en)  20191209  20210126  Evan Chase Rose  Biometric authentication, decentralized learning framework, and adaptive security protocols in distributed terminal network 
US11113665B1 (en)  20200312  20210907  Evan Chase Rose  Distributed terminals network management, systems, interfaces and workflows 
Families Citing this family (12)
Publication number  Priority date  Publication date  Assignee  Title 

US20120150450A1 (en) *  20090909  20120614  Nec Corporation  Biometric authentication system, method, and program 
CN102486832A (en) *  20101206  20120606  中国人民解放军理工大学  User authentication method based on keystroke mode identification 
JP2013137590A (en) *  20111228  20130711  Fujitsu Frontech Ltd  Authentication device, authentication program and authentication method 
JP6164284B2 (en) *  20130226  20170719  日本電気株式会社  Authentication apparatus, authentication method, and computer program 
EP2990999A1 (en) *  20140829  20160302  Accenture Global Services Limited  A machinelearning system to optimise the performance of a biometric system 
JP6238867B2 (en) *  20140930  20171129  株式会社日立製作所  Sequential biometric cryptographic system and sequential biometric cryptographic processing method 
CN105160225B (en) *  20150818  20180727  宇龙计算机通信科技(深圳)有限公司  Electronic device method and device 
CN107305625A (en) *  20160420  20171031  厦门中控智慧信息技术有限公司  A kind of person recognition method based on multimode biometric information 
CN106375336A (en) *  20160928  20170201  北京金山安全软件有限公司  Login information storage method and device, and login information providing method and device 
WO2018060010A1 (en) *  20160929  20180405  British Telecommunications Public Limited Company  Collection of sensor data from sensor devices 
KR101995443B1 (en) *  20170726  20190702  네이버 주식회사  Method for verifying speaker and system for recognizing speech 
KR102098237B1 (en) *  20190626  20200407  네이버 주식회사  Method for verifying speaker and system for recognizing speech 
Citations (55)
Publication number  Priority date  Publication date  Assignee  Title 

US5850470A (en) *  19950830  19981215  Siemens Corporate Research, Inc.  Neural network for locating and recognizing a deformable object 
US20020028021A1 (en) *  19990311  20020307  Jonathan T. Foote  Methods and apparatuses for video segmentation, classification, and retrieval using image class statistical models 
US20020131641A1 (en) *  20010124  20020919  Jiebo Luo  System and method for determining image similarity 
US20020159627A1 (en) *  20010228  20021031  Henry Schneiderman  Object finder for photographic images 
US20020172303A1 (en) *  19991227  20021121  Wataru Matsumoto  Communication device and communication method 
US6624739B1 (en) *  19980928  20030923  Anatoli Stobbe  Access control system 
US20040054572A1 (en) *  20000727  20040318  Alison Oldale  Collaborative filtering 
US20040179719A1 (en) *  20030312  20040916  Eastman Kodak Company  Method and system for face detection in digital images 
US6844506B2 (en) *  20000814  20050118  MettlerToledo Gmbh  Measuring apparatus with user profile 
US20050066180A1 (en) *  20030924  20050324  Sanyo Electric Co., Ltd.  Authentication apparatus and authentication method 
US20050063567A1 (en) *  20030924  20050324  Sanyo Electric Co., Ltd.  Authentication apparatus and authentication method 
US20050163367A1 (en) *  20000504  20050728  Microsoft Corporation  System and method for progressive stereo matching of digital images 
US20050190973A1 (en) *  20040227  20050901  International Business Machines Corporation  System and method for recognizing word patterns in a very large vocabulary based on a virtual keyboard layout 
US6940545B1 (en) *  20000228  20050906  Eastman Kodak Company  Face detecting camera and method 
US20050207626A1 (en) *  20040322  20050922  Lg Electronics Inc.  User identification and verification system and method for a mobile terminal 
US20060093208A1 (en) *  20041029  20060504  Fayin Li  Open set recognition using transduction 
US20060167784A1 (en) *  20040910  20060727  Hoffberg Steven M  Game theoretic prioritization scheme for mobile ad hoc networks permitting hierarchal deference 
US20060222210A1 (en) *  20050331  20061005  Hitachi, Ltd.  System, method and computer program product for determining whether to accept a subject for enrollment 
US20060224898A1 (en) *  20030502  20061005  Ahmed Ahmed E  System and method for determining a computer user profile from a motionbased input device 
US20060251339A1 (en) *  20050509  20061109  Gokturk Salih B  System and method for enabling the use of captured images through recognition 
US7161468B2 (en) *  20030121  20070109  Samsung Electronicss Co., Ltd.  User authentication method and apparatus 
US7167574B2 (en) *  20020314  20070123  Seiko Epson Corporation  Method and apparatus for contentbased image copy detection 
US20070071286A1 (en) *  20050916  20070329  Lee Yong J  Multiple biometric identification system and method 
US7212233B2 (en) *  20000614  20070501  Minolta Co., Ltd.  Image extracting apparatus and image extracting method 
US20070127825A1 (en) *  20051206  20070607  Yuri Ivanov  Weighted ensemble boosting method for classifier combination and feature selection 
US7242810B2 (en) *  20040513  20070710  Proximex Corporation  Multimodal highdimensional data fusion for classification and identification 
US20070189611A1 (en) *  20060214  20070816  Microsoft Corporation  Bayesian Competitive Model Integrated With a Generative Classifier for Unspecific Person Verification 
US20070219801A1 (en) *  20060314  20070920  Prabha Sundaram  System, method and computer program product for updating a biometric model based on changes in a biometric feature of a user 
US7277891B2 (en) *  20021011  20071002  Digimarc Corporation  Systems and methods for recognition of individuals using multiple biometric searches 
US20070258626A1 (en) *  20060427  20071108  Bruce Reiner  Apparatus and method for utilizing biometrics in medical applications 
US20070286463A1 (en) *  20060609  20071213  Sony Ericsson Mobile Communications Ab  Media identification 
US20080080775A1 (en) *  20060929  20080403  Cornell Center For Technology Enterprise & Commercialization  Methods and systems for reconstruction of objects 
US7362884B2 (en) *  20050317  20080422  Imageware Systems, Inc.  Multimodal biometric analysis 
US20080118124A1 (en) *  20061018  20080522  Anant Madabhushi  Systems and methods for classification of biological datasets 
US20080172725A1 (en) *  20070116  20080717  Yoshihiro Fujii  System, apparatus, and program for biometric authentication 
US20080195541A1 (en) *  20000926  20080814  Integrated Technological Systems, Inc.  Integrated technology money transfer system 
US20080193048A1 (en) *  20070213  20080814  Microsoft Corporation  Picture Collage Systems And Methods 
US20080222706A1 (en) *  20070306  20080911  Martin Renaud  Globally aware authentication system 
US20080260212A1 (en) *  20070112  20081023  Moskal Michael D  System for indicating deceit and verity 
US20090025081A1 (en) *  20070720  20090122  Thomas Quigley  Method and system for configuring local and remote resources to accomplish rendering of multimedia content on dissimilar format devices based on user biometric data 
US20090023398A1 (en) *  20040820  20090122  Koninklijke Philips Electronics N.V.  Wireless Terminal, Wireless Module and Method of Manufacturing such a Terminal 
US20090141947A1 (en) *  20071129  20090604  Volodymyr Kyyko  Method and system of person identification by facial image 
US7565548B2 (en) *  20041118  20090721  Biogy, Inc.  Biometric print quality assurance 
US7583822B2 (en) *  20030220  20090901  Sagem Sa  Method for identifying persons and system for carrying out said method 
US7596246B2 (en) *  20041116  20090929  Imageware Systems Inc  Multimodal biometric platform 
US7742641B2 (en) *  20041206  20100622  Honda Motor Co., Ltd.  Confidence weighted classifier combination for multimodal identification 
US7783082B2 (en) *  20030630  20100824  Honda Motor Co., Ltd.  System and method for face recognition 
US20100284575A1 (en) *  20080129  20101111  Kabushiki Kaisha Dds  Hybrid biometric authentication device, hybrid biometric authentication method, and computerreadable medium storing hybrid biometric authentication program 
US7873583B2 (en) *  20070119  20110118  Microsoft Corporation  Combining resilient classifiers 
US7912252B2 (en) *  20090206  20110322  Robert Bosch Gmbh  Timeofflight sensorassisted iris capture system and method 
US7945101B2 (en) *  20070726  20110517  Palo Alto Research Center Incorporated  Innovative OCR systems and methods that combine a template based generative model with a discriminative model 
US7961955B1 (en) *  20080128  20110614  Thomas Cecil Minter  Adaptive bayes feature extraction 
US7974475B1 (en) *  20090820  20110705  Thomas Cecil Minter  Adaptive bayes image correlation 
US7983490B1 (en) *  20071220  20110719  Thomas Cecil Minter  Adaptive Bayes pattern recognition 
US8014570B2 (en) *  20041116  20110906  Activcard, Inc.  Method for improving false acceptance rate discriminating for biometric authentication systems 
Family Cites Families (8)
Publication number  Priority date  Publication date  Assignee  Title 

US5572597A (en) *  19940329  19961105  Loral Corporation  Fingerprint classification system 
US7492928B2 (en) *  20030225  20090217  Activcard Ireland Limited  Method and apparatus for biometric verification with data packet transmission prioritization 
JP2005010826A (en) *  20030616  20050113  Fujitsu Ltd  Authentication terminal device, biometrics information authentication system and biometrics information acquisition system 
JP4348202B2 (en) *  20040121  20091021  日本放送協会  Face image recognition apparatus and face image recognition program 
JP4546168B2 (en) *  20040628  20100915  富士通株式会社  Biometric authentication system registration method, biometric authentication system and program thereof 
JP4221385B2 (en) *  20050221  20090212  日立オムロンターミナルソリューションズ株式会社  Biometric authentication device, terminal device and automatic transaction device 
US7747044B2 (en) *  20060120  20100629  The Johns Hopkins University  Fusing multimodal biometrics with quality estimates via a bayesian belief network 
WO2007146437A2 (en) *  20060614  20071221  Agent Science Technologies, Inc.  User authentication system 

2008
 20081208 JP JP2008312411A patent/JP5121681B2/en active Active

2009
 20090225 EP EP09002648.5A patent/EP2113867A3/en not_active Withdrawn
 20090226 CN CN2009101185305A patent/CN101571920B/en active Active
 20090227 US US12/379,715 patent/US8340361B2/en active Active
Patent Citations (58)
Publication number  Priority date  Publication date  Assignee  Title 

US5850470A (en) *  19950830  19981215  Siemens Corporate Research, Inc.  Neural network for locating and recognizing a deformable object 
US6624739B1 (en) *  19980928  20030923  Anatoli Stobbe  Access control system 
US20020028021A1 (en) *  19990311  20020307  Jonathan T. Foote  Methods and apparatuses for video segmentation, classification, and retrieval using image class statistical models 
US20020172303A1 (en) *  19991227  20021121  Wataru Matsumoto  Communication device and communication method 
US6940545B1 (en) *  20000228  20050906  Eastman Kodak Company  Face detecting camera and method 
US20050163367A1 (en) *  20000504  20050728  Microsoft Corporation  System and method for progressive stereo matching of digital images 
US7212233B2 (en) *  20000614  20070501  Minolta Co., Ltd.  Image extracting apparatus and image extracting method 
US20040054572A1 (en) *  20000727  20040318  Alison Oldale  Collaborative filtering 
US6844506B2 (en) *  20000814  20050118  MettlerToledo Gmbh  Measuring apparatus with user profile 
US20080195541A1 (en) *  20000926  20080814  Integrated Technological Systems, Inc.  Integrated technology money transfer system 
US20020131641A1 (en) *  20010124  20020919  Jiebo Luo  System and method for determining image similarity 
US6829384B2 (en) *  20010228  20041207  Carnegie Mellon University  Object finder for photographic images 
US20020159627A1 (en) *  20010228  20021031  Henry Schneiderman  Object finder for photographic images 
US7167574B2 (en) *  20020314  20070123  Seiko Epson Corporation  Method and apparatus for contentbased image copy detection 
US7277891B2 (en) *  20021011  20071002  Digimarc Corporation  Systems and methods for recognition of individuals using multiple biometric searches 
US7161468B2 (en) *  20030121  20070109  Samsung Electronicss Co., Ltd.  User authentication method and apparatus 
US7583822B2 (en) *  20030220  20090901  Sagem Sa  Method for identifying persons and system for carrying out said method 
US7508961B2 (en) *  20030312  20090324  Eastman Kodak Company  Method and system for face detection in digital images 
US20040179719A1 (en) *  20030312  20040916  Eastman Kodak Company  Method and system for face detection in digital images 
US20060224898A1 (en) *  20030502  20061005  Ahmed Ahmed E  System and method for determining a computer user profile from a motionbased input device 
US7783082B2 (en) *  20030630  20100824  Honda Motor Co., Ltd.  System and method for face recognition 
US20050063567A1 (en) *  20030924  20050324  Sanyo Electric Co., Ltd.  Authentication apparatus and authentication method 
US20050066180A1 (en) *  20030924  20050324  Sanyo Electric Co., Ltd.  Authentication apparatus and authentication method 
US20050190973A1 (en) *  20040227  20050901  International Business Machines Corporation  System and method for recognizing word patterns in a very large vocabulary based on a virtual keyboard layout 
US20050207626A1 (en) *  20040322  20050922  Lg Electronics Inc.  User identification and verification system and method for a mobile terminal 
US7242810B2 (en) *  20040513  20070710  Proximex Corporation  Multimodal highdimensional data fusion for classification and identification 
US20090023398A1 (en) *  20040820  20090122  Koninklijke Philips Electronics N.V.  Wireless Terminal, Wireless Module and Method of Manufacturing such a Terminal 
US20060167784A1 (en) *  20040910  20060727  Hoffberg Steven M  Game theoretic prioritization scheme for mobile ad hoc networks permitting hierarchal deference 
US20060093208A1 (en) *  20041029  20060504  Fayin Li  Open set recognition using transduction 
US7606396B2 (en) *  20041116  20091020  Imageware Systems, Inc.  Multimodal biometric platform 
US7596246B2 (en) *  20041116  20090929  Imageware Systems Inc  Multimodal biometric platform 
US8014570B2 (en) *  20041116  20110906  Activcard, Inc.  Method for improving false acceptance rate discriminating for biometric authentication systems 
US7565548B2 (en) *  20041118  20090721  Biogy, Inc.  Biometric print quality assurance 
US7742641B2 (en) *  20041206  20100622  Honda Motor Co., Ltd.  Confidence weighted classifier combination for multimodal identification 
US7362884B2 (en) *  20050317  20080422  Imageware Systems, Inc.  Multimodal biometric analysis 
US20060222210A1 (en) *  20050331  20061005  Hitachi, Ltd.  System, method and computer program product for determining whether to accept a subject for enrollment 
US20060251339A1 (en) *  20050509  20061109  Gokturk Salih B  System and method for enabling the use of captured images through recognition 
US20070071286A1 (en) *  20050916  20070329  Lee Yong J  Multiple biometric identification system and method 
US20070127825A1 (en) *  20051206  20070607  Yuri Ivanov  Weighted ensemble boosting method for classifier combination and feature selection 
US20070189611A1 (en) *  20060214  20070816  Microsoft Corporation  Bayesian Competitive Model Integrated With a Generative Classifier for Unspecific Person Verification 
US20070219801A1 (en) *  20060314  20070920  Prabha Sundaram  System, method and computer program product for updating a biometric model based on changes in a biometric feature of a user 
US20070258626A1 (en) *  20060427  20071108  Bruce Reiner  Apparatus and method for utilizing biometrics in medical applications 
US20070286463A1 (en) *  20060609  20071213  Sony Ericsson Mobile Communications Ab  Media identification 
US20080080775A1 (en) *  20060929  20080403  Cornell Center For Technology Enterprise & Commercialization  Methods and systems for reconstruction of objects 
US20080118124A1 (en) *  20061018  20080522  Anant Madabhushi  Systems and methods for classification of biological datasets 
US20080260212A1 (en) *  20070112  20081023  Moskal Michael D  System for indicating deceit and verity 
US20080172725A1 (en) *  20070116  20080717  Yoshihiro Fujii  System, apparatus, and program for biometric authentication 
US7873583B2 (en) *  20070119  20110118  Microsoft Corporation  Combining resilient classifiers 
US20080193048A1 (en) *  20070213  20080814  Microsoft Corporation  Picture Collage Systems And Methods 
US20080222706A1 (en) *  20070306  20080911  Martin Renaud  Globally aware authentication system 
US20090025081A1 (en) *  20070720  20090122  Thomas Quigley  Method and system for configuring local and remote resources to accomplish rendering of multimedia content on dissimilar format devices based on user biometric data 
US7945101B2 (en) *  20070726  20110517  Palo Alto Research Center Incorporated  Innovative OCR systems and methods that combine a template based generative model with a discriminative model 
US20090141947A1 (en) *  20071129  20090604  Volodymyr Kyyko  Method and system of person identification by facial image 
US7983490B1 (en) *  20071220  20110719  Thomas Cecil Minter  Adaptive Bayes pattern recognition 
US7961955B1 (en) *  20080128  20110614  Thomas Cecil Minter  Adaptive bayes feature extraction 
US20100284575A1 (en) *  20080129  20101111  Kabushiki Kaisha Dds  Hybrid biometric authentication device, hybrid biometric authentication method, and computerreadable medium storing hybrid biometric authentication program 
US7912252B2 (en) *  20090206  20110322  Robert Bosch Gmbh  Timeofflight sensorassisted iris capture system and method 
US7974475B1 (en) *  20090820  20110705  Thomas Cecil Minter  Adaptive bayes image correlation 
NonPatent Citations (2)
Title 

Bazen et al., Likelihood Ratio Based Biometric Verification, January 2004, Vol. 14, No. 1, pp. 8694 * 
R. McDonough, A. Whalen, Detection of Signals in Noise, 1995, AT&T Bell Laboratories and Academic Press, Inc., 2nd Edition, pp. 155173 * 
Cited By (37)
Publication number  Priority date  Publication date  Assignee  Title 

US7747044B2 (en) *  20060120  20100629  The Johns Hopkins University  Fusing multimodal biometrics with quality estimates via a bayesian belief network 
US20070172114A1 (en) *  20060120  20070726  The Johns Hopkins University  Fusing Multimodal Biometrics with Quality Estimates via a Bayesian Belief Network 
US20100214062A1 (en) *  20090220  20100826  Fujitsu Limited  Verification apparatus and authentication apparatus 
US20120208529A1 (en) *  20090924  20120816  Wave Guard Technologies Ltd.  System and method of online radiation management and control of nonionizing radiation sources 
US8948741B2 (en) *  20090924  20150203  Wave Guard Technologies Ltd.  System and method of online radiation management and control of nonionizing radiation sources 
US20110182480A1 (en) *  20100126  20110728  Hitachi, Ltd.  Biometric authentication system 
EP2348458A1 (en)  20100126  20110727  Hitachi, Ltd.  Biometric authentication system 
US8437511B2 (en)  20100126  20130507  Hitachi, Ltd.  Biometric authentication system 
US20110211734A1 (en) *  20100301  20110901  Richard Jay Langley  Method and system for conducting identity matching 
US7835548B1 (en)  20100301  20101116  Daon Holding Limited  Method and system for conducting identity matching 
US8989520B2 (en)  20100301  20150324  Daon Holdings Limited  Method and system for conducting identification matching 
US20130207779A1 (en) *  20100930  20130815  Fujitsu Limited  Device and method for biometric authentication 
US9792420B2 (en) *  20100930  20171017  Fujitsu Limited  Device and method for biometric authentication 
US20140007210A1 (en) *  20111212  20140102  Hitachi, Ltd.  High security biometric authentication system 
US20140314283A1 (en) *  20130416  20141023  Imageware Systems, Inc.  Multimodal biometric database searching methods 
US10580243B2 (en) *  20130416  20200303  Imageware Systems, Inc.  Conditional and situational biometric authentication and enrollment 
US9286528B2 (en) *  20130416  20160315  Imageware Systems, Inc.  Multimodal biometric database searching methods 
US10777030B2 (en)  20130416  20200915  Imageware Systems, Inc.  Conditional and situational biometric authentication and enrollment 
US20140313007A1 (en) *  20130416  20141023  Imageware Systems, Inc.  Conditional and situational biometric authentication and enrollment 
US20150222630A1 (en) *  20130521  20150806  Tencent Technology (Shenzhen) Company Limited  Third application login method and corresponding apparatus and terminal 
US20160188863A1 (en) *  20140205  20160630  International Business Machines Corporation  Biometric authentication 
US10509895B2 (en) *  20140205  20191217  International Business Machines Corporation  Biometric authentication 
US9405893B2 (en) *  20140205  20160802  International Business Machines Corporation  Biometric authentication 
US10528710B2 (en) *  20150215  20200107  Alibaba Group Holding Limited  System and method for user identity verification, and client and server by use thereof 
US20160239650A1 (en) *  20150215  20160818  Alibaba Group Holding Limited  System and method for user identity verification, and client and server by use thereof 
US20170142101A1 (en) *  20151116  20170518  International Business Machines Corporation  Secure Biometrics Matching with Split Phase ClientServer Matching Protocol 
US10778672B2 (en) *  20151116  20200915  International Business Machines Corporation  Secure biometrics matching with split phase clientserver matching protocol 
US10332378B2 (en) *  20171011  20190625  Lenovo (Singapore) Pte. Ltd.  Determining user risk 
CN108540457A (en) *  20180320  20180914  深圳市文鼎创数据科技有限公司  A kind of safety equipment and its biological identification control method and device 
US10902705B1 (en)  20191209  20210126  Evan Chase Rose  Biometric authentication, decentralized learning framework, and adaptive security protocols in distributed terminal network 
US10873578B1 (en)  20191209  20201222  Evan Chase Rose  Biometric authentication, decentralized learning framework, and adaptive security protocols in distributed terminal network 
US10904259B1 (en)  20191209  20210126  Evan Chase Rose  Graphical user interface and console management system for distributed terminal network 
US10911463B1 (en)  20191209  20210202  Evan Chase Rose  Graphical user interface and console management system for distributed terminal network 
US10931677B1 (en)  20191209  20210223  Evan Chase Rose  Graphical user interface and console management system for distributed terminal network 
US11019055B1 (en)  20191209  20210525  Evan Chase Rose  Biometric authentication, decentralized learning framework, and adaptive security protocols in distributed terminal network 
US11108771B2 (en) *  20191209  20210831  Evan Chase Rose  Facial recognition, image analysis, and decentralized learning framework using adaptive security protocols in distributed terminal network 
US11113665B1 (en)  20200312  20210907  Evan Chase Rose  Distributed terminals network management, systems, interfaces and workflows 
Also Published As
Publication number  Publication date 

EP2113867A2 (en)  20091104 
US8340361B2 (en)  20121225 
JP2009289253A (en)  20091210 
CN101571920A (en)  20091104 
CN101571920B (en)  20120627 
JP5121681B2 (en)  20130116 
EP2113867A3 (en)  20150311 
Similar Documents
Publication  Publication Date  Title 

US8340361B2 (en)  Biometric authentication system, authentication client terminal, and biometric authentication method  
US10438054B2 (en)  Biometric identification and verification  
US10580243B2 (en)  Conditional and situational biometric authentication and enrollment  
US8433922B2 (en)  System and method of biometric authentication using multiple kinds of templates  
US6591224B1 (en)  Biometric score normalizer  
US8437511B2 (en)  Biometric authentication system  
Ko  Multimodal biometric identification for large user population using fingerprint, face and iris recognition  
US20060222210A1 (en)  System, method and computer program product for determining whether to accept a subject for enrollment  
JP2006006753A (en)  Registration method of biometrics system, biometrics system, and program thereof  
JP2013122679A (en)  Biometric authentication system with high safety  
WO2007075982A1 (en)  Method and apparatus for multimodel hybrid comparison system  
JP5710748B2 (en)  Biometric authentication system  
US20120150450A1 (en)  Biometric authentication system, method, and program  
KR100734837B1 (en)  System for multimodel biometric identification and method thereof  
US20100045787A1 (en)  Authenticating apparatus, authenticating system, and authenticating method  
WO2013084617A1 (en)  Biometric authentication system  
Gamassi et al.  Accuracy and performance of biometric systems  
JP2010108074A (en)  Biometrics authentication system and method  
Shawkat et al.  The new hand geometry system and automatic identification  
Spanakis et al.  Developing a ContextDependent Tuning Framework of Multichannel Biometrics that Combine AudioVisual Characteristics for Secure Access of an eHealth Platform  
Kraus  PRIOR—/PRIOR A PRIOR—/\PRIOR J PRIOR PRIORS  
Gutkowski  Algorithm for retrieval and verification of personal identity using bimodal biometrics  
Harvey  Biometric Quality and its Impact on Template Ageing in a Longitudinal Fingerprint Study  
Araújo  A probabilistic analysis of the biometrics menagerie existence: case study in fingerprint data  
Chmielewski et al.  Biometric Techniques: The Fundamentals of Evaluation 
Legal Events
Date  Code  Title  Description 

AS  Assignment 
Owner name: HITACHI, LTD., JAPAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:MURAKAMI, TAKAO;TAKAHASHI, KENTA;REEL/FRAME:023066/0415 Effective date: 20090304 

STCF  Information on status: patent grant 
Free format text: PATENTED CASE 

FPAY  Fee payment 
Year of fee payment: 4 

MAFP  Maintenance fee payment 
Free format text: PAYMENT OF MAINTENANCE FEE, 8TH YEAR, LARGE ENTITY (ORIGINAL EVENT CODE: M1552); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY Year of fee payment: 8 