JP6049253B2 - Face authentication database management method, face authentication database management apparatus, and face authentication database management program - Google Patents

Description

  According to the present invention, partial image data (hereinafter referred to as “face image data”) including a user's face portion cut out from image data captured by an imaging device such as a camera is used as user identification information ( The present invention relates to a face authentication database management method, a face authentication database management apparatus, and a face authentication database management program that are managed in association with each other.

  Conventionally, ID cards (Identity Cards) such as membership cards and employee ID cards have been issued to users who are subject to authentication, and the user ID recorded on the ID card at the time of each user's entry / exit is read with a reader / writer. A personal authentication system that performs personal authentication by reading is known. Also, a personal authentication technique for performing personal authentication using a user's fingerprint or the like is known in order to deal with fraud such as forgery of the ID card.

  However, when the above personal authentication technology is used, every time a user is authenticated, the user must be stopped and a fingerprint or the like must be collected, which is not preferable from the viewpoint of user convenience. For this reason, recently, face authentication technology that can perform personal authentication without stopping a walking user has attracted attention. For example, in Patent Document 1, the feature amount of the user's face image data is registered as dictionary information in advance, and when the walking user approaches the camera, the user's face portion is captured by the camera. A technique is disclosed in which face image data is acquired, and the user's face authentication is performed by comparing the feature amount of the acquired user's face image data with the contents of dictionary information.

  Here, in order to perform face authentication, it is necessary to construct a database in which user face image data is stored in advance. In Patent Document 2, it is possible to easily perform automatic database construction simply by having the user walk in front of the camera. For example, it is determined whether or not the detected person is the same as the person already registered in the database. If it is determined that the detected person is different, the face image data is registered with a new ID. When it is determined that they are the same person, the face image data is added to the ID of this person. This reduces the burden on the operator when constructing the database.

JP 2009-104599 A JP 2010-238181 A

  However, in the case of automatic construction of a face authentication database in which a plurality of face image data is registered in association with each user, even a registered person is erroneously determined to be an unregistered person, and the same There are cases where a plurality of user IDs are given to a person. In addition, when a plurality of face image data is registered in association with one user, face image data of a plurality of different persons may be associated with one user ID due to erroneous authentication. Therefore, it is necessary to correct the face authentication database. However, when the association between the user ID and the face image data is corrected manually by an operator, it takes a lot of time and labor. If the association between the user ID and the face image data is not corrected correctly, erroneous authentication is induced.

  The present invention has been made to solve the above-described problems of the prior art, and is a face authentication database management capable of automatically correcting the association between a user ID and face image data in the face authentication database. It is an object to provide a method, a face authentication database management apparatus, and a face authentication database management program.

In order to solve the above-described problems and achieve the object, the present invention stores user identification information for uniquely identifying a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A face authentication database management method for managing a face authentication database, wherein a similarity between a plurality of face image data associated with each user identification information and a mutual correlation between face image data associated with different user identification information. A similarity calculation step for calculating a similarity between the predetermined face image data among a plurality of face image data associated with each user identification information based on the similarity calculated by the similarity calculation step only contains the association modifying step of automatically modifying the association to associate the other user identification information, the association modifying step is associated with a first user identification information A first similarity that is a maximum value of the similarity between the isolated face image data of the plurality of face image data and the plurality of face image data associated with the second user identification information, More than or equal to a second similarity that is the maximum value of each similarity between the isolated face image data and the other face image data associated with the first user identification information, and the second similarity is The association is automatically corrected so as to associate the isolated face image data with the second user identification information when the threshold value is equal to or greater than a predetermined recognition threshold value .

  Further, in the present invention according to the above invention, the association correcting step temporarily stores user identification information as a transfer destination of the isolated face image data for each user identification information, and temporarily stores the isolated face image data. Is automatically corrected so as to be associated with the user identification information as the transfer destination.

  Further, according to the present invention, in the above invention, in the association correction step, when there are a plurality of pieces of user identification information to which one isolated face image data is transferred, the isolated face image data is converted to the first similarity degree. The association is corrected so as to be associated with the user identification information that maximizes.

  Also, in the present invention according to the above-described invention, the association correcting step may identify the specified isolated face image data each time the user identification information as the transfer destination of the isolated face image data is specified for each user identification information. The association is automatically corrected so as to be associated with the user identification information as the transfer destination.

The present invention also provides a face authentication database management method for managing a face authentication database that stores user identification information for uniquely identifying a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A similarity calculation step of calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data respectively associated with different user identification information And associating predetermined face image data of a plurality of face image data associated with each user identification information with other user identification information based on the similarity calculated in the similarity calculation step and a association modifying step of automatically modifying the said association modifying step, a plurality of face image data and the second usage associated with the first user identification information The second user identification when the predetermined number or more of the plurality of similarities respectively calculated with the plurality of face image data associated with the identification information is equal to or more than a predetermined authentication threshold. The association is automatically corrected so that a plurality of face image data associated with the information is associated with the first user identification information.

  Further, the present invention is characterized in that, in the above-described invention, a correction start determination step is further included in which the process of the association correction step is started when update registration of a predetermined number of face image data is performed.

  Moreover, the present invention is characterized in that, in the above-mentioned invention, a correction start determination step for starting processing of the association correction step when new user identification information is registered.

  In the invention described above, the invention further includes a correction start determination step of starting the association correction step when a predetermined time is reached.

  In the invention described above, the invention further includes a correction start determination step of starting the processing of the association correction step when an association correction instruction is received from a predetermined input unit.

In addition, the present invention provides a face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A similarity calculation means for calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data respectively associated with different user identification information. And associating predetermined face image data of a plurality of face image data associated with each user identification information with other user identification information based on the similarity calculated by the similarity calculation means and a association modifying means for automatically correcting the said association modifying means is isolated from among the plurality of face image data associated with the first user identification information The first similarity that is the maximum value of the similarity between the image data and the plurality of face image data associated with the second user identification information is the isolated face image data and the first user identification. When it is equal to or higher than a second similarity that is the maximum value of each similarity with other face image data associated with the information, and the second similarity is equal to or higher than a predetermined recognition threshold, The association is automatically corrected so as to associate the isolated face image data with the second user identification information .
In addition, the present invention provides a face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A similarity calculation means for calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data respectively associated with different user identification information. And associating predetermined face image data of a plurality of face image data associated with each user identification information with other user identification information based on the similarity calculated by the similarity calculation means And an association correction unit that automatically corrects the plurality of facial image data associated with the first user identification information and the second usage. The second user identification when the predetermined number or more of the plurality of similarities respectively calculated with the plurality of face image data associated with the identification information is equal to or more than a predetermined authentication threshold. The association is automatically corrected so that a plurality of face image data associated with the information is associated with the first user identification information.

The present invention is also uniquely identifying user identification information a Subscriber, face recognition database management for managing the face authentication database that associates and stores the plurality of face image data of a user captured by the imaging device A face authentication database management program executed by the apparatus, wherein a similarity between a plurality of face image data associated with each user identification information and between face image data associated with different user identification information Based on the similarity calculation procedure for calculating the similarity and the similarity calculated by the similarity calculation procedure, predetermined face image data of a plurality of face image data associated with each user identification information is obtained. to execute the association repair procedure for automatically modifying an association to associate the other user identification information to a computer, the association modifying step, first The first maximum value of the similarity between the isolated face image data of the plurality of face image data associated with the user identification information and the plurality of face image data associated with the second user identification information. Is equal to or higher than a second similarity that is the maximum value of each similarity between the isolated face image data and the other face image data associated with the first user identification information, When the second similarity is equal to or higher than a predetermined recognition threshold value, the association is automatically corrected so as to associate the isolated face image data with the second user identification information .
In addition, the present invention provides a face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of face image data of the user imaged by the imaging device in association with each other. The face authentication database management program executed in the above, the similarity between a plurality of face image data associated with each user identification information and between the face image data associated with different user identification information Based on the similarity calculation procedure for calculating the similarity and the similarity calculated by the similarity calculation procedure, predetermined face image data of a plurality of face image data associated with each user identification information is And an association correction procedure for automatically correcting the association so as to be associated with the user identification information of the user. A predetermined number or more of a plurality of similarities calculated between the plurality of face image data associated with the user identification information and the plurality of face image data associated with the second user identification information, The association is automatically corrected so that a plurality of face image data associated with the second user identification information are respectively associated with the first user identification information when the predetermined authentication threshold value is exceeded. And

  According to the present invention, the similarity between a plurality of face image data associated with each user identification information and the similarity between the face image data respectively associated with different user identification information are calculated and calculated. Since the configuration is such that the association is automatically corrected so as to associate the predetermined face image data among the plurality of face image data associated with each user identification information with other user identification information based on the similarity degree The association between the user identification information and the face image data in the face authentication database can be automatically corrected. As a result, the authentication accuracy is improved, the labor required for workers is reduced, and the amount of data in the face authentication database Can be reduced.

FIG. 1 is a diagram illustrating a configuration of an entrance / exit management device according to the present embodiment. FIG. 2 is a diagram illustrating an example of face image data acquired by the face image detection unit illustrated in FIG. 1. FIG. 3 is an explanatory diagram for explaining the concept of the isolated face image data migration processing by the association correcting unit shown in FIG. FIG. 4 is an explanatory diagram for explaining the concept of the user ID integration processing by the association correcting unit shown in FIG. FIG. 5 is an explanatory diagram for explaining a response when there are a plurality of user IDs to which isolated face image data is transferred. FIG. 6 is a flowchart showing a migration processing procedure of isolated face image data by the association correcting unit shown in FIG. FIG. 7 is a flowchart showing a user ID integration processing procedure by the association correcting unit shown in FIG. FIG. 8 is a flowchart (1) illustrating the procedure of the isolated face image data transfer process by the association correcting unit according to the second embodiment. FIG. 9 is a flowchart (2) illustrating the migration processing procedure of the isolated face image data by the association correcting unit according to the second embodiment.

  Exemplary embodiments of a face authentication database management method, a face authentication database management apparatus, and a face authentication database management program according to the present invention will be described below in detail with reference to the accompanying drawings. In the following first and second embodiments, a case where the face authentication database management method, the face authentication database management device, and the face authentication database management program according to the present invention are applied to an entrance / exit management device will be described.

  First, the features and configuration of the entrance / exit management device 1 according to the first embodiment will be described. FIG. 1 is a diagram illustrating the configuration of the entrance / exit management device 1 according to the first embodiment. The entrance / exit management device 1 shown in FIG. 1 is a device that performs entrance / exit management of the user M using face authentication. In the first embodiment, an ID card is not given to the user M, but the face image data of the user M is acquired and registered in the real environment. When the user M is a new user, a new user ID is assigned, and the user ID and face image data are newly registered in the face authentication database 5. Further, when the user M is a new user, the worker is called and the gate 10 is opened to permit entry into the room under the presence of the worker.

  On the other hand, if the user M is a registered user, the gate 10 is opened and entry is permitted on the condition that the user M is authenticated as a legitimate registrant. Further, new face image data acquired at the time of face authentication of the user M is additionally registered in the face authentication database 5 in association with the user ID.

  As described above, the entrance / exit management device 1 does not perform initial registration using an ID card or the like, and presupposes that a new user ID is automatically given each time a new user M arrives. Therefore, when the degree of similarity between the face image data registered in the face authentication database 5 and the face image data captured by the camera 3 is low even though the user M is the same user M, the same person is different. There is a problem that the user ID is given. In addition, when the degree of similarity is high even though they are different persons, there is also a problem that face image data of different persons is registered in one user ID.

  Therefore, in the entrance / exit management device 1 according to the first embodiment, the face image data of another person isolated from the face image of the original user within one user ID (hereinafter referred to as “isolated face image data”). If it is determined that the isolated face image data is transferred to the user ID of the user who should be, the association is corrected. However, even if it is determined that there is a high possibility that the isolated face image data is included, if the user ID of the other person is not registered, a new user ID at that time Migration processing is not performed on the condition of grant. This is because a new user ID is given when the other person subsequently enters the room.

  Further, even when there is no isolated face image data in the user ID, a plurality of user IDs may be given to the same user. In such a case, the association is corrected so that the face image data in the plurality of user IDs are integrated into one user ID. This is because when a plurality of user IDs are given to the same user, there is a problem that the amount of data in the face image database 5 increases and a problem that authentication accuracy is lowered.

  Specifically, in the entrance / exit management device 1 according to the first embodiment, is there any isolated face image data, that is, face image data of another person among a plurality of face image data included in the same user ID? If it is determined that the isolated face image data is included, the process of specifying the user ID to which the isolated face image data is to be transferred is repeated. On the other hand, if it is determined that the isolated face image data is not included, it is determined whether or not a plurality of user IDs should be integrated. Then, the process of correcting the association is performed so that the isolated face image data is finally transferred to the transfer destination user ID.

  Here, when determining whether or not the isolated face image data is present in each user ID, the similarity Sw between the face image data located in the same user ID, and other The similarity Sb between the face image data located in the user ID and the authentication threshold value T, which is a reference when determining that they are the same person, are used. The authentication threshold T is assigned a predetermined value empirically obtained for all users. If the relationship “Sb ≧ Sw ≧ T” is established, it is determined that isolated face image data that should originally belong to another user ID exists in a certain user ID.

  On the other hand, when determining whether or not to integrate a plurality of user IDs, the similarity Sb between each face image data belonging to one user ID and the face image data belonging to the other user ID is obtained. When these similarities Sb are all higher than a predetermined value (for example, authentication threshold value T), the face image data in both user IDs are integrated. At the time of such integration, the face image data located in one user ID is transferred to the other user ID, and the user ID whose face image data no longer exists is discarded.

  As shown in FIG. 1, the entrance / exit management device 1 according to the first embodiment includes a camera 3, a control unit 4, a face authentication database 5, a gate 10, a storage unit 6, and an input / output unit 8. Have. In addition, the control unit 4 includes a face image detection unit 12, a registration processing unit 13, a face authentication processing unit 14, an association correction unit 15, a correction start determination unit 16, and an alarm output unit 17. Although FIG. 1 illustrates the configuration of the entrance side of the entrance / exit management device 1, the face authentication database 5 can be commonly used on the exit side.

  The camera 3 is an imaging device including a CCD (Charge Coupled Device) element that images the user M and outputs image data. The camera 3 is set in advance to an angle of view that can capture the face portion of the user M on the way to the gate 10. In the first embodiment, the image data captured by the camera 3 is a black and white image with 256 gradations (the pixel value of the black pixel is “0” and the pixel value of the white pixel is “255” for convenience of explanation). )), But color images can also be targeted.

  The control unit 4 is a computer that controls the entrance / exit management device 1 as a whole, and is connected to the camera 3, the face authentication database 5, the input / output unit 8, and the gate 10. The control unit 4 performs a process of automatically registering and automatically updating and registering the face image data of the user M in the face authentication database 5 in association with the user ID. Specifically, when the face image data of the user M is captured, the similarity between the face image data of the user M and each face image data in all user IDs in the face authentication database 5 is calculated. The similarity is compared with an authentication threshold value. As a result, if each similarity does not exceed the authentication threshold, it is determined that the user is a new user, a new user ID is issued, and a face associated with the user ID is issued. Image data is automatically registered in the face authentication database 5.

  Further, the control unit 4 performs a process of correcting the association of each face image data in the face authentication database 5. This is because it is necessary to correct the association when a plurality of user IDs are registered even though they are the same user, or when isolated face image data exists in one user ID. is there.

  Further, the control unit 4 performs the face authentication process of the user M using the face image data registered in the face authentication database 5 and manages the entrance and exit of the user M based on the authentication result. Specifically, the control unit 4 opens the gate 10 when the user M can be authenticated, and performs control to maintain the closed state of the gate 10 until the arrival of the worker when the user M cannot be authenticated. . In addition, since the well-known technique can be used for the opening / closing control of the gate 10, the detailed description is abbreviate | omitted here.

  The face authentication database 5 is a database that stores at least a user ID and one or a plurality of face image data in association with each other. In this face authentication database 5, the face image data of the user M is registered at the first use. The control unit 4 serves as a database management device (DBMS) for the face authentication database 5.

  The storage unit 6 is a storage device including a hard disk device, a nonvolatile memory, and the like, and stores an authentication threshold value T used when the face authentication processing unit 14 performs face authentication processing. The input / output unit 8 is realized by various input devices and output devices. In particular, the input / output unit 8 inputs an instruction to start association correction processing by the association correction unit 15. The input / output unit 8 can input and set various types of information with reference to the output information.

  The face image detection unit 12 is a processing unit that cuts out rectangular face image data including a human face portion present in black and white grayscale image data (hereinafter simply referred to as “image data”) captured by the camera 3. The face image data cut out by the face image detection unit 12 is also a black and white image with 256 gradations. FIG. 2 is a diagram illustrating an example of face image data detected by the face image detection unit 12. As shown in the figure, here, the user M is included in the image data 20 captured by the camera 3, and the face image data 21 of the user M is cut out. Here, when the face image data of the user M is cut out, a template (mask pattern) indicating the outline of a general face of a person is prepared, and this template is shifted to the image data 20 and matched to thereby obtain image data. It is possible to detect the face image data 21 existing within. In addition to such template matching, various known face detection algorithms (image processing techniques) can be used.

  The registration processing unit 13 is a processing unit that registers the face image data of the user M in the face authentication database 5 in association with the user ID. As already described, in the first embodiment, when the user M enters the room for the first time, a new user ID is issued, and the face image data of the user M is registered in the face authentication database 5 under the actual environment. It will be.

  The registration processing unit 13 additionally registers new face image data in the face authentication database 5 in association with the user ID when the user M enters or leaves the room for the second time or later. When the detected face image data of the user M is authenticated, if the face image data in the face authentication database 5 has not reached a predetermined number that can be registered, the face image data is additionally registered as it is. In addition, when face image data reaches a predetermined number and there is face image data whose frequency of use for authentication (hereinafter referred to as “authentication use degree”) is less than a threshold value, the oldest of them exists The face image data is deleted, and the authenticated face image data is updated and registered. On the other hand, if the face image data reaches a predetermined number and there is no authentication usage less than a predetermined threshold, all face image data is frequently used. The registered face image data is deleted without registration, and the currently registered face image data is maintained.

  The face authentication processing unit 14 is a processing unit that authenticates whether the user M is a registered user registered in the face authentication database 5. Specifically, the similarity between the face image data of the user M detected by the face image detection unit 12 and the face image data registered in the face authentication database 5 is calculated, and the calculated similarity and the authentication threshold are calculated. Based on the value T, it is authenticated whether or not the user M is a registered user registered in the face authentication database 5. This is because the user M registered in the face authentication database 5 needs to control the opening of the gate 10. Here, as an example of the degree of similarity, a cross-correlation coefficient between face image data can be used. When both face image data includes the face portion of the same person, the degree of similarity increases, and both face images The similarity is low when the face portion of a different person is included in the data. As the similarity, a known similarity calculation technique other than the cross-correlation coefficient can be used.

  When there is only one face image data registered in the face authentication database 5, the user M enters the face authentication database 5 when the similarity between the face image data is equal to or greater than a predetermined threshold. If there are a plurality of face image data registered in the face authentication database 5 as determined to have been registered, the face image data of the user M detected by the face image detection unit 12 and registered in the face authentication database 5 If, for example, the top three similarities are extracted from the similarities of the plurality of face image data obtained, and the average value of the extracted top three similarities is equal to or greater than a predetermined threshold, the user M It can be determined that the face authentication database 5 has already been registered. For this determination process, an existing well-known technique can be used.

  The association correction unit 15 is a processing unit that corrects the association between each face image data registered in the face image database 5 and the user ID. If isolated face image data exists within a certain user ID, the isolated face image data is corrected so as to be transferred to an appropriate user ID. Further, when there are a plurality of user IDs for the same user, the association is corrected so that each face image data is aggregated into one user ID.

  Specifically, the similarity Sw between face image data existing in each user ID is calculated in advance, and the similarity having the maximum value in the user ID is set as the maximum similarity SWmax in ID. For example, when face image data A1, A2 and A3 exist within a certain user ID, the similarity Sw1 of the face image data A1 and A2, the similarity Sw2 of the face image data A2 and A3, the face image data A1 and The similarity Sw3 of A3 is obtained, and the maximum value is set as Swmax.

  In addition, the similarity Sb between certain face image data in a certain user ID and each face image data in another user ID is calculated, and the similarity Sb that is the maximum value is set as the maximum similarity Sbmax between IDs. . For example, when face image data A1 in a certain user ID and face image data B1, B2, and B3 in another user ID exist, the similarity Sb1 between the face image data A1 and B1, the face image data A1 And the similarity Sb2 of B2 and the similarity Sb3 of the face image data A1 and B3 are obtained, respectively, and the maximum value is defined as Sbmax.

  Then, when the condition “maximum similarity between IDs Sbmax ≧ maximum similarity within ID Swmax ≧ authentication threshold T” is satisfied, it is considered that the face image data A1 is isolated face image data, and the user The face image data A1 in the ID is corrected to be associated with another user ID. Such a series of processing is repeated for all user IDs. Thereby, the isolated face image data located in the user ID can be transferred to the original user ID.

  However, when there is no isolated face image data in the user ID, but there are a plurality of user IDs of the same user, the problem can be solved even if the above-described series of processing is performed. Can not. For this reason, the similarity Sb between each face image data in the user ID and each face image data in the other user ID is calculated, and all the similarities or a predetermined number of similarities or more are the authentication threshold value T. If it exceeds, a process of integrating both user IDs is performed.

  The correction start determination unit 16 is a processing unit that determines the start time of the association correction process by the association correction unit 15. For example, the correction start determination unit 16 determines to start the association correction process when update registration of a predetermined number of face image data is performed. The correction start determination unit 16 determines to start the association correction process when a new user ID is registered. The correction start determination unit 16 determines to start the association correction process when a predetermined time (for example, every day at midnight) is reached. Further, the correction start determination unit 16 determines to start the association correction process when receiving the association correction instruction from the input / output unit 8.

  The alarm output unit 17 is a processing unit that outputs an alarm when the face authentication processing unit 14 determines that the user has not been registered in the face authentication database 5. As the alarm output, in addition to the actual voice output, it may be possible to notify a predetermined worker's mobile terminal of “there is a new user”. Note that the closed state of the gate 10 is maintained while the alarm is output.

  Next, the concept of the isolated face image data migration process by the association correcting unit 15 shown in FIG. 1 will be described. FIG. 3 is an explanatory diagram for explaining the concept of the isolated face image data migration processing by the association correcting unit 15 shown in FIG. Hereinafter, the user ID of the face image data transfer destination is referred to as a transfer destination ID, and the user ID of the face image data transfer source is referred to as a transfer source ID.

  As shown in the figure, three face image data D11, D12 and D13 are registered in the transfer source ID (user ID; 0002), and in the transfer destination ID (user ID; 0001). Three face image data D1, D2, and D3 are registered. First, the similarity Sw1 of the face image data D11 and D12 in the migration source ID, the similarity Sw2 of the face image data D12 and D13, and the similarity Sw3 of the face image data D11 and D13 are obtained, and the maximum value is obtained as the maximum similarity in the ID. Degree Swmax. Here, a situation is shown in which the similarity Sw1 between the face image data D11 and D12 is the maximum similarity SWmax in the ID.

  Thereafter, the similarity between the face image data D11 in the migration source ID and the face image data D1, D2, and D3 in the migration destination ID is calculated. Specifically, the similarity Sb1 of the face image data D11 and D1, the similarity Sb2 of the face image data D11 and D2, and the similarity Sb3 of the face image data D11 and D3 are calculated, and the maximum value is calculated as the maximum similarity between IDs. Let Sbmax. Here, a situation is shown in which the similarity Sb3 between the face image data D11 and D3 is the ID maximum similarity Sbmax.

  Then, it is determined whether or not the relationship “maximum similarity between IDs Sbmax ≧ maximum similarity within ID Swmax ≧ authentication threshold value T” is satisfied. As described above, the face image data D11 is associated with the integration destination ID, and the association with the integration source ID is canceled. In the face image database 5, the association between the user ID and the face image data is held in a link relationship. Therefore, when the association is corrected, the link may be changed.

  Next, the concept of user ID integration processing by the association correcting unit 15 shown in FIG. 1 will be described. FIG. 4 is an explanatory diagram for explaining the concept of user ID integration processing by the association correcting unit 15 shown in FIG.

  As shown in the figure, three face image data D11, D12 and D13 are registered in the transfer source ID (user ID; 0002), and in the transfer destination ID (user ID; 0001). Three face image data D1, D2, and D3 are registered. Here, it is assumed that no isolated face image data exists in the migration source ID and the migration destination ID.

  Here, the similarity Sb between the face image data D11, D12 and D13 in the migration source ID and the face image data D1, D2 and D3 in the migration destination ID is calculated. Here, a total of nine similarities Sb are obtained.

  Thereafter, it is checked whether or not all the similarities Sb are equal to or higher than the authentication threshold T. If all the similarities Sb are equal to or higher than the authentication threshold T, the face image data D11, D12 in the migration source ID and The association is corrected so that D13 is migrated to the migration destination ID, and the migration source ID is deleted. Here, for convenience of explanation, it is assumed that all the similarities Sb are equal to or greater than the authentication threshold value. However, the present invention is not limited to this, and a predetermined number of all similarities ( For example, when 2/3) is equal to or greater than the authentication threshold, the user IDs can be integrated.

  Next, an isolated face image data transfer processing procedure by the association correcting unit 15 shown in FIG. 1 will be described. FIG. 5 is a flowchart showing the migration processing procedure of isolated face image data by the association correcting unit 15 shown in FIG. As shown in the figure, the association correction unit 15 first determines whether or not an instruction to start association correction related to the correction of isolated face image data has been received from the correction start determination unit 16 (step S101). If the association correction start instruction has not been received (step S101, No), this determination process is repeated. On the other hand, when an instruction to start association correction is received (step S101, Yes), the processing flag for all user IDs and the processing flag for all face image data are set to unprocessed (step S102).

  Thereafter, after obtaining the authentication threshold value T from the storage unit 6 (step S103), the maximum ID similarity Swmax of each user ID is calculated (step S104). Specifically, the degree of similarity Sw between the face image data in the user ID is obtained, and the process of setting the maximum value of the obtained similarity degree Sw as the maximum degree of similarity SWmax in the ID is repeated for each user ID. Become.

  Thereafter, one user ID whose processing flag is unprocessed is selected (step S105), unprocessed face image data within the selected user ID is selected (step S106), and the maximum similarity between IDs is selected. The degree Sbmax is calculated (step S107). Specifically, the similarity Sb between the selected face image data and each face image data in another user ID is calculated, and the similarity Sb having the maximum value is set as the maximum ID similarity Sbmax. Since this process is repeated for each other user ID, the inter-ID maximum similarity Sbmax is obtained for each other user ID.

  Thereafter, it is determined whether or not there exists a user ID that satisfies the relationship of “maximum similarity between IDs Sbmax ≧ maximum similarity within ID Swmax ≧ authentication threshold T” (step S108), and the corresponding user ID. Is present (step S108, Yes), the processing flag of the face image data is set to “transfer target” and the corresponding user ID is temporarily stored as the transfer destination ID (step S109). When there are a plurality of corresponding user IDs, the user ID having the maximum inter-ID maximum similarity Sbmax is selected as the migration destination ID. By this processing, the migration destination ID of the isolated face image data can be specified. On the other hand, if there is no user ID that satisfies the above relational expression (No in step S108), “processed” is set in the processing flag of the face image data (step S110).

  If there is face image data with the processing flag set to “unprocessed” (step S111, Yes), the process proceeds to step S106 and the same processing is repeated, and the processing flag is set to “unprocessed”. If the set face image data does not exist (step S111, No), “processed” is set to the processing flag of the user ID (step S112).

  After that, when there is a user ID whose processing flag is “unprocessed” (Yes in step S113), the process proceeds to step S105 and the same processing is repeated, and the user ID whose processing flag is “unprocessed”. Is not present (No at step S113), the process of migrating each face image data whose processing flag is set to “move target” is performed (step S114). By performing the above-described series of processing, it is possible to transfer the isolated face image data included in each user ID into the user ID that should be originally.

  By the way, as described above, when the isolated face image data is collectively transferred, a problem occurs when it is determined that the isolated face image data should be transferred to a plurality of user IDs. As shown in FIG. 6, when it is determined that the face image data D11 should be associated and corrected for two or more migration destination IDs (ID = 0001,0003), the maximum between IDs is obtained. It can be associated with the migration destination ID having the maximum similarity Sbmax. As described above, when the association correction processing of each face image data is performed at a time, it is possible to perform the association correction processing in which each user ID is matched equally as compared with the sequential time series processing.

  Next, a user ID integration processing procedure by the association correcting unit 15 shown in FIG. 1 will be described. FIG. 7 is a flowchart showing a user ID integration processing procedure by the association correcting unit 15 shown in FIG. Here, for convenience of explanation, it is assumed that a start instruction relating to integration processing different from the isolated face image data migration processing is received.

  As shown in the figure, the association correction unit 15 first determines whether or not an instruction to start association correction related to user ID integration has been received from the correction start determination unit 16 (step S201). If no association correction start instruction has been received (step S201, No), this determination process is repeated. On the other hand, when an instruction to start association correction is received (step S201, Yes), the processing flags of all user IDs are set to unprocessed (step S202).

  Thereafter, after acquiring the authentication threshold value T from the storage unit 6 (step S203), the user ID whose processing flag is set to “unprocessed” (this is referred to as “user IDa”) is selected ( In step S204), another user ID (this is referred to as "user IDb") that is a candidate for integration of this user IDa is selected (step S205), and each face image data in the user IDa and the user are selected. The similarity Sb with each face image data in IDb is calculated (step S206).

  Then, it is determined whether or not all the similarities Sb are greater than or equal to the authentication threshold T (step S207). If all the similarities Sb are greater than or equal to the authentication threshold T (step S207, Yes). Each face image data in the user IDb is associated to be transferred to the user IDa (step S208), the user IDb is deleted (step S209), and the process flag of the user IDa is set to “processed” (step S209). Step S210).

  After step S210, or when any one of the similarities Sb is less than the authentication threshold value T (step S207, No), an unselected user ID remains as a user IDb that is an integration source candidate. It is determined whether or not (step S210). If an unselected user ID remains as a user IDb that is an integration source candidate (Yes in step S210), the process proceeds to step S205 to select the user IDb. On the other hand, if all user IDs other than the user IDa have been selected as the integration source candidate user IDb, the processing flag of the user IDa is set to “processed” (step S211), and step S212. Migrate to

  Thereafter, when there is an unprocessed user ID (step S212, Yes), the process proceeds to step S204 and the same processing is repeated, and when there is no unprocessed user ID (step S212). , No), the process ends.

  By performing the above-described series of processing, face image data in these user IDs can be integrated when a plurality of user IDs are assigned to the same user. Although the description is omitted here for convenience of explanation, when integrating the face image data in the user ID, if the total number of face image data exceeds a predetermined number (for example, 10), authentication is performed. You can delete them in ascending order of usage.

  As described above, in the first embodiment, the association correction process for transferring the isolated face image data located in the user ID to the original user ID, and a plurality of user IDs are given to the same user. The user ID is integrated in the face authentication database, the association between the user ID and the face image data in the face authentication database is automatically corrected, thereby improving the authentication accuracy and in the face authentication database 5. It is possible to compress the amount of data. In particular, since it is not necessary for the worker to update the face authentication database 5, the human labor can be reduced.

  By the way, in the first embodiment, the case where all the isolated face image data is extracted and then the process of migrating each isolated face image data to another user ID is performed at the same time is described. Without being limited thereto, it is possible to move to another user ID each time isolated face image data is extracted. Therefore, in the second embodiment, a case will be described in which a transition is made to another user ID each time isolated face image data is extracted. Since the configuration diagram of the second embodiment is the same as that of the first embodiment, here, the procedure of the isolated face image data transfer process by the association correcting unit 15 will be described.

  FIG. 8 and FIG. 9 are flowcharts illustrating the isolated face image data transfer processing procedure by the association correcting unit 15 according to the second embodiment. As shown in FIG. 8, the association correction unit 15 first determines whether or not an instruction to start association correction related to correction of isolated face image data has been received from the correction start determination unit 16 (step S <b> 301). If an association correction start instruction has not been received (step S301, No), this determination process is repeated. If an association correction start instruction has been received (step S301, Yes), the user with all user IDs The processing flag for ID and the processing flag for all face image data are all set to “unprocessed” (step S302).

  Thereafter, the authentication threshold value T is acquired from the storage unit 6 (step S303). Further, one user ID whose processing flag is not processed (this is referred to as “user IDa”) is selected (step S304). Furthermore, one user ID (this is referred to as “user IDb”) as a migration source candidate is selected from user IDs other than the selected user IDa (step S305). Furthermore, one face image data with an unprocessed processing flag in the selected user IDb is selected (step S306). After that, an association correction process is performed on the selected face image data to perform a specific association correction (step S307). Such association correction processing will be described with reference to FIG.

  Thereafter, the processing flag of the face image data selected in step S306 is set to “processed” (step S308). Further, it is determined whether or not face image data whose processing flag is not processed exists in the selected user IDb (step S309). When face image data with an unprocessed processing flag exists (step S309, Yes), the process proceeds to step S306, and an association correction process is performed on other unprocessed face image data in the selected user IDb. .

  On the other hand, if there is no face image data whose processing flag is unprocessed (No in step S309), all the processing flags of all face image data are set to “unprocessed” (step S310). Thereafter, it is determined whether or not an unselected user ID remains as the migration source candidate user IDb (step S311). If an unselected user ID remains as the migration source candidate user IDb (step S311, Yes), the process proceeds to step S305, and another user IDb is selected and the above-described processing is repeated.

  On the other hand, if all user IDs other than the user IDa have been selected as the migration source candidate user IDb (step S311, No), the processing flag of the selected user IDa is set to “processed”. Setting is performed (step S312). Thereafter, it is further determined whether or not there is a user IDa whose processing flag is not processed (step S313). When there is a user IDa whose processing flag is not processed (step S313, Yes), the process proceeds to step S304, and another user IDa whose processing flag is not processed is selected and the above-described processing is repeated. On the other hand, if there is no user IDa whose processing flag is not yet processed (step S313, No), this processing ends. By following such a processing procedure, it is possible to perform association correction processing for all user IDs for all face image data.

  Here, the association correction processing procedure in step S307 will be described. As shown in FIG. 9, first, a similarity Sw between the selected face image data and the other face image data in the selected user IDb is obtained, and a maximum ID similarity Swmax is calculated (step S401). ). Thereafter, the similarity Sb between the selected face image data and all the face image data in the selected user IDa is obtained, and the inter-ID maximum similarity Sbmax is calculated (step S402).

  Thereafter, it is determined whether or not the relational expression “maximum similarity between IDs Sbmax ≧ maximum similarity within ID Swmax ≧ authentication threshold T” is satisfied (step S403). When this relational expression is established (step S403, Yes), the selected face image data is associated as the face image data of the selected user IDa and a process of removing it from the user IDb is performed (step S404). The process proceeds to step S308. On the other hand, if the relational expression “maximum similarity between IDs Sbmax ≧ maximum similarity in ID Swmax ≧ authentication threshold T” is not satisfied (No in step S403), the process proceeds to step S308 as it is.

  In the above-described first and second embodiments, even when there is one face image data in the transfer destination user ID, “maximum similarity between IDs Sbmax ≧ maximum similarity in ID Swmax ≧ authentication threshold T” However, the present invention is not limited to this. Specifically, the association can be corrected when the following conditions are satisfied with respect to the specified number of face image data to be integrated.

  As a condition for such association correction, “(1) Similarity to at least one face image data of the transfer destination user ID in which the specified number (for example, 30% of the registration upper limit) of face image data is registered. The degree of similarity is greater than or equal to the authentication threshold T, and the similarity to the face image data of the user ID of the transfer destination is greater than or equal to the similarity of all face image data of the user ID of the transfer source. Is mentioned.

  Also, “(2) face images greater than or equal to the designated number (for example, 10% of the registration upper limit) of the transfer destination user IDs in which the specified number (for example, 30% of the registration upper limit) of face image data is registered. Similarity with the face image data of the transfer destination user ID is similar to all face image data of the transfer source user ID, and the similarity with the data is equal to or greater than the authentication threshold T. "(3)" or (3) the specified number of transfer destination user IDs in which the specified number (for example, 30% of the registration upper limit) of face image data is registered (for example, 10% of the registration upper limit). ) The similarity with the above face image data is equal to or greater than the authentication threshold T, and the specified number (for example, equal to or greater than the authentication threshold T) of the face image data of the transfer destination user ID. 30% of the total number of face image data) May be set as a condition and that "it is the similarity or more with all of the face image data of the source of the user ID.

  In the first and second embodiments described above, the maximum value among the similarities Sw in the migration source user ID is used. However, the present invention is not limited to this, and the user ID is within the user ID. The average value of the similarities Sw can also be used. Further, it is preferable that the similarity Sw when the face image data in the migration source user ID is one is replaced with the authentication threshold value T for processing.

  Moreover, in the said Example 1 and 2, although the case where this invention was applied to the entrance / exit management system was shown, this invention is not limited to this, It can apply to various systems, such as an attendance management system. . Thereby, face authentication can be introduced into entrance / exit management and attendance management. Since the face image data for face authentication is automatically registered, the user can unlock the electric lock and stamp the time card without having to hold the ID card.

  The face authentication database management method, face authentication database management apparatus, and face authentication database management program according to the present invention are suitable for an entrance / exit monitoring system, an attendance management system, and the like. It is suitable for suppressing and obtaining stable authentication accuracy.

DESCRIPTION OF SYMBOLS 1 Entrance / exit management apparatus 3 Camera 4 Control part 5 Face authentication database 6 Storage part 8 Input / output part 10 Gate 12 Face image detection part 13 Registration process part 14 Face authentication process part 15 Association correction part 16 Correction start determination part 17 Alarm output part 20 Image data 21 Face image data M User

Claims (13)

A face authentication database management method for managing a face authentication database for associating and storing user identification information for uniquely identifying a user and a plurality of face image data of a user imaged by an imaging device,
A similarity calculation step of calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data respectively associated with different user identification information;
Based on the similarity calculated in the similarity calculation step, the association is automatically performed to associate predetermined face image data among a plurality of face image data associated with each user identification information with other user identification information. only contains the association modification step of modifying,
The association correcting step includes
The maximum value of the similarity between the isolated face image data of the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information. A certain first similarity is equal to or greater than a second similarity that is the maximum value of each similarity between the isolated face image data and the other face image data associated with the first user identification information. And face authentication , wherein when the second similarity is equal to or higher than a predetermined recognition threshold, the association is automatically corrected so as to associate the isolated face image data with the second user identification information. Database management method. The association correcting step includes
For each user identification information, the user identification information as the transfer destination of the isolated face image data is temporarily stored, and the association is automatically performed so that the temporarily stored isolated face image data is associated with the user identification information as the transfer destination. The face authentication database management method according to claim 1 , wherein correction is performed. The association correcting step includes
When there are a plurality of pieces of user identification information to which one isolated face image data is to be transferred, the association is corrected so as to associate the isolated face image data with the user identification information having the maximum first similarity. The face authentication database management method according to claim 2 . The association correcting step includes
Each time the user identification information that becomes the migration destination of the isolated face image data is specified for each user identification information, the association is automatically corrected so that the identified isolated face image data is associated with the user identification information that becomes the migration destination. The face authentication database management method according to claim 1 . A face authentication database management method for managing a face authentication database for associating and storing user identification information for uniquely identifying a user and a plurality of face image data of a user imaged by an imaging device,
A similarity calculation step of calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data respectively associated with different user identification information;
Based on the similarity calculated in the similarity calculation step, the association is automatically performed to associate predetermined face image data among a plurality of face image data associated with each user identification information with other user identification information. The association correction process to be corrected and
Including
The association correcting step includes
A predetermined number of the plurality of similarities respectively calculated between the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information When the above is equal to or greater than a predetermined authentication threshold value, the association is automatically corrected so that a plurality of face image data associated with the second user identification information is associated with the first user identification information, respectively. face authentication database management how to, characterized in that. 6. The method according to claim 1 , further comprising a correction start determination step of starting processing of the association correction step when update registration of a predetermined number of face image data is performed. Face authentication database management method. The face authentication according to any one of claims 1 to 5 , further comprising a correction start determination step of starting processing of the association correction step when new user identification information is registered. Database management method. The face authentication database management method according to any one of claims 1 to 5 , further comprising a correction start determination step of starting processing of the association correction step when a predetermined time is reached. 6. The method according to claim 1 , further comprising a correction start determination step of starting processing of the association correction step when receiving an association correction instruction from a predetermined input unit. Face authentication database management method. A face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of user face image data captured by the imaging device in association with each other,
Similarity calculating means for calculating the similarity between a plurality of face image data associated with each user identification information and the similarity between the face image data associated with different user identification information;
Based on the similarity calculated by the similarity calculation means, the association is automatically performed to associate predetermined face image data among the plurality of face image data associated with each user identification information with other user identification information. An association correcting means for correcting , and
The association correcting means includes
The maximum value of the similarity between the isolated face image data of the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information. A certain first similarity is equal to or greater than a second similarity that is the maximum value of each similarity between the isolated face image data and the other face image data associated with the first user identification information. Yes, if the second similarity is greater than or equal to a predetermined recognition threshold, the association is automatically corrected to associate the isolated face image data with the second user identification information
A face authentication database management device.   A face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of user face image data captured by the imaging device in association with each other,
  Similarity calculating means for calculating the similarity between a plurality of face image data associated with each user identification information and the similarity between the face image data associated with different user identification information;
  Based on the similarity calculated by the similarity calculation means, the association is automatically performed to associate predetermined face image data among the plurality of face image data associated with each user identification information with other user identification information. Association correction means to correct
  With
  The association correcting means includes
  A predetermined number of the plurality of similarities respectively calculated between the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information When the above is equal to or greater than a predetermined authentication threshold value, the association is automatically corrected so that a plurality of face image data associated with the second user identification information is associated with the first user identification information, respectively.
  A face authentication database management device. Face authentication executed by a face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A database management program,
A similarity calculation procedure for calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data associated with different user identification information;
Based on the similarity calculated by the similarity calculation procedure, association is automatically performed so that predetermined face image data among a plurality of face image data associated with each user identification information is associated with other user identification information. Let the computer execute the association correction procedure to correct ,
The association correction procedure includes:
The maximum value of the similarity between the isolated face image data of the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information. A certain first similarity is equal to or greater than a second similarity that is the maximum value of each similarity between the isolated face image data and the other face image data associated with the first user identification information. Yes, if the second similarity is greater than or equal to a predetermined recognition threshold, the association is automatically corrected to associate the isolated face image data with the second user identification information
A face authentication database management program characterized by that.   Face authentication executed by a face authentication database management device that manages a face authentication database that stores user identification information that uniquely identifies a user and a plurality of face image data of the user imaged by the imaging device in association with each other. A database management program,
  A similarity calculation procedure for calculating a similarity between a plurality of face image data associated with each user identification information and a similarity between the face image data associated with different user identification information;
  Based on the similarity calculated by the similarity calculation procedure, association is automatically performed so that predetermined face image data among a plurality of face image data associated with each user identification information is associated with other user identification information. Association correction procedure to correct
  To the computer,
  The association correction procedure includes:
  A predetermined number of the plurality of similarities respectively calculated between the plurality of face image data associated with the first user identification information and the plurality of face image data associated with the second user identification information When the above is equal to or greater than a predetermined authentication threshold value, the association is automatically corrected so that a plurality of face image data associated with the second user identification information is associated with the first user identification information, respectively.
  A face authentication database management program characterized by that.

Images