JP2008065651A - Face image authentication method, face image authentication apparatus and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a procedure for improving the determination accuracy of face authentication processing to be generated by changes in face directions, changes of expressions, illumination conditions, or combinations of these conditions. <P>SOLUTION: A face image authentication apparatus 10 is provided with: a registered face image acquisition part 13 for acquiring a registered face image photographed under a photographing condition including an element of a prescribed direction; an authentication face image acquisition part 14 for acquiring a plurality of authentication face images of a target person newly photographed under photographing conditions including elements of a plurality of directions; a similarity degree data calculation part 16 for collating respective similarity degrees of the plurality of authentication face images with the acquired registered face image and relating respective similarity degrees with the elements of the directions; and an authentication part 17 for determining whether the related information data belong to a same-person similarity group or a different person similarity group, by comparing these information data with group discrimination data, and determining whether the plurality of authentication face images and the registered face image are obtained by photographing the same person or not to identify the authentication face images with the registered face image. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a face image authentication method for authenticating a face image and related technology.

  The face image authentication method calculates the degree of similarity between face image data (registered face image) of a person registered in advance in a database or the like and face image data (input image) of a person to be authenticated photographed at the time of authentication, If the similarity is greater than or equal to a preset threshold, the person to be authenticated is determined to be the same as the person of the face image registered in the database (matching is established). If the similarity is less than the threshold, the person to be authenticated Is a technique for determining that the face image person registered in the database is not the same as the person in the face image (matching failure).

  In general, in the face image authentication method, a change in photographing condition (factor 1) including a face direction, lighting conditions, differences in photographing devices, etc., or a facial expression including a facial expression, makeup, presence / absence of an attachment, time passage, etc. Since the captured face images differ greatly depending on the change in the state (factor 2) and the like, the desired similarity cannot be obtained even for face images of the same person, and the matching results often become inconsistent. is there. For this reason, sufficient matching accuracy cannot be obtained unless the use conditions are limited so that the change in the appearance of the face image does not occur.

  However, in the actual usage scene of the face image authentication method, it is often difficult to set the shooting conditions and the face state to be suitable for the collation process, especially when targeting an unspecified number of users. Since it is more difficult to set shooting conditions and the like, it has been difficult to perform stable and accurate collation processing.

  For this problem, in the conventional face image authentication method, in order to ensure higher collation accuracy, a plurality of face images are collected and collated for each, and the maximum value of similarity is adopted, or In calculating the similarity, there are a method of removing a lighting change component from an image, a method using a stable face image feature even with respect to a change in lighting or the like.

Non-Patent Document 1 discloses a method of using a face image feature that is stable against changes in the facial expression of the person to be authenticated using a graph matching method or the like in calculating similarity.
Non-Patent Document 2 discloses a method for preliminarily estimating a change in lighting and a change in the appearance of a face image due to the face orientation when calculating the similarity.
Non-Patent Document 3 discloses a method for calculating the degree of similarity in a stable manner even with respect to a change in how a face image is projected due to a change in illumination conditions by devising a similarity degree calculation algorithm.
In Patent Document 1, for each registrant, a plurality of registrant face image data having different lighting conditions are collated with the certifier face image data obtained from the face image of the person to be authenticated, and the degree of similarity is obtained. An authentication method is disclosed as to whether or not the person to be authenticated is a registrant based on the sum of similarities calculated for each registrant that includes registrant face image data whose similarity is greater than the extraction threshold. .

Masato Kawate and Sei Hosoi “Robust Face Recognition Technology in Real Environments and Its Applications” IIP2000 Information, Intelligence and Precision Instruments Lecture Proceedings The Japan Society of Mechanical Engineers (March 2000) Hitoshi Imaoka and Kei Sato “Face Matching Algorithm Using Discriminant Analysis and Perturbation Image Method” Proceedings of the 4th Information Science and Technology Forum (2005) Fukui Kazuhiro et al., “Face Image Recognition Robust to Environmental Changes Using the Constraint Mutual Subspace Method-Learning Constraint Subspace to Reduce the Effects of Lighting Variation” IEICE Transactions D-II Vol.J82-D- II, No.4, p.613-620,1999 JP 2005-56004 A

In the conventional face image authentication method, the above-described improvement has been made against the erroneous authentication caused by the change in how the face image is reflected, but the countermeasure is still insufficient.
Above all, it cannot be said that sufficient measures have been taken against mis-authentication due to changes in the orientation of the face, and if the user is not familiar with the face image authentication method, the person himself / herself is not conscious. There has been a problem that misauthentication may occur even when the face orientation changes.
In addition, changes in the appearance of the face are often caused not only by changes in the orientation of the face, but also by a combination of the various factors described above. When such changes occur in combination, there is a problem that authentication accuracy further decreases.
Further, in the face collation method described in Patent Document 1, each similarity obtained by collating the similarity of each of the plurality of authentication face images with the registered face image is associated with the direction element, and based on the associated information. It was not done from the viewpoint of authenticating.

  It is an object of the present invention to provide a technique for improving determination accuracy in face authentication processing that occurs when a change in face orientation, a change in facial expression, illumination conditions, or a combination of these changes occur.

  The face image authentication method provided by the present invention is a method performed by a computer having a memory and a control means, wherein the control means is a plurality of faces photographed under photographing conditions including elements in a plurality of directions for one person. Features of face images generated by collecting images for a plurality of persons and associating similarity obtained by matching face images of the same person and face images of different persons with each of the elements in the plurality of directions. The process is a process for deriving group discrimination data for discriminating between the same-person similarity group data and the different-person similarity group data included in the group data representing the set of the images, and the shooting was performed under the shooting conditions including elements in a predetermined direction. A registered face image and a plurality of authentication face images of a person to be authenticated that are newly photographed under photographing conditions including elements in a plurality of directions are acquired, stored in the memory, and stored. It is obtained by collating the similarity of each of the plurality of authentication face images with the recorded face image, and associates each similarity with a direction element, and the associated information is the same person similarity group or different person similarity. The process of determining whether the plurality of authentication face images and the registered face images are images obtained by photographing the same person by comparing which group belongs to the group discrimination data. For this reason, it is possible to correctly determine the identity between the authentication face image and the registered face image by verifying the degree of correlation between the image data of the authentication face image and the registered face image in a multiple manner.

  In one embodiment, the control means checks the registered face image or the authentication face image against a set of face images and elements of the direction of the face image, and compares the features of the face image to thereby compare the registration image. The elements of the predetermined direction of the face image or the elements of the plurality of directions of the plurality of authentication face images are estimated, and the attribution of the information including the similarity associated with the elements of the estimated direction is determined. For this reason, it is possible to correctly determine the direction condition of the authentication face image or the registered face image.

  In one embodiment, the direction element is a photographing direction of a face image or a light irradiation direction at the time of photographing a face image. For this reason, it is a face direction included in either a face image photographed by a plurality of photographing means having different photographing directions or a face image photographed by the photographing means in a situation where light irradiation direction at the time of photographing is different. Can also be judged correctly.

  The face image authentication apparatus provided by the present invention collects a plurality of face images taken for a person under shooting conditions including elements in a plurality of directions, and the face images of the same person and different persons are collected. Same-person similarity group data and different-person similarity included in group data representing a set of facial image features generated by associating similarity obtained by matching face images with each of the elements in the plurality of directions Acquire a registered face image photographed under photographing conditions including a database in which group discrimination data for discriminating the degree group data are registered, database management means for managing registration information of the database, and elements in a predetermined direction. Registered face image acquisition means, authentication face image acquisition means for acquiring a plurality of authentication face images of a person to be authenticated newly photographed under shooting conditions including elements in a plurality of directions, and Obtained by comparing the similarity of each of the plurality of authentication face images with respect to the registered face image, and a similarity calculating means for associating each similarity with a direction element, and these related by the similarity calculating means Whether the information belongs to the same person similarity group or the different person similarity group is compared with the group discrimination data, so that the plurality of authentication face images and the registered face images photograph the same person. Determination means for determining whether or not the image is an image, and determining the identity between the authentication face image and the registered face image.

  The face image authentication program provided by the present invention collects a plurality of face images photographed under photographing conditions including elements in a plurality of directions with respect to one person, face images of the same person, and Different from the same person similarity group data included in the group data representing the set of facial image features generated by associating the similarity obtained by collating face images of different persons with each of the elements in the plurality of directions. Registered face image acquisition means for acquiring a registered face image photographed under a photographing condition including an element in a predetermined direction from a computer that can access a database in which group discrimination data for identifying person similarity group data is registered; Authentication face image acquisition means for acquiring a plurality of authentication face images of a person to be authenticated newly photographed under shooting conditions including elements in a plurality of directions, Obtained by collating the similarities of each of the number of authentication face images, the similarity calculating means for associating each similarity with a direction element, and the information related by the similarity calculating means is the same person Whether the plurality of authentication face images and the registered face images are images obtained by photographing the same person by comparing with the group discrimination data whether it belongs to the similarity group or the different person similarity group. It functions as a determination means for determination.

  According to the present invention, whether a plurality of authentication face images and the registered face images are images of the same person using group discrimination data for discriminating between the same person similarity group data and different person similarity group data. If the face direction of the registered face image and the authentication face image is different, the facial expression or illumination direction of the registered face image and the authentication face image is different, or the face direction of the registered face image and the authentication face image Even when facial expressions and lighting directions are all different, highly accurate face image authentication can be performed. Further, since the face direction of the registered face image or the authentication face image can be determined, it is possible to perform highly accurate face image authentication even when the face direction of the registered face image or the authentication face image is not constant. Further, even when the proof irradiation direction is different when the registered face image and the authentication face image are photographed, it is possible to perform highly accurate face image authentication as in the case where the face directions are different.

Embodiments of a face image authentication device to which the present invention is applied will be described below. Here, the person to be authenticated means a person who receives face image authentication by the face image authentication apparatus of the present invention, and the face image of the person to be authenticated photographed by photographing means described later is called an authentication face image. The registered face image refers to a person who has registered his / her face image in the registered image holding means in advance to receive authentication. That is, if the person to be authenticated holds a correct registered face image, the identity of the authenticated face image and the registered face image is authenticated by the face image device to which the present invention is applied.
Further, in this embodiment, a face direction with respect to the camera (to be described later) (hereinafter referred to as “face direction”) is used as a direction element at the time of photographing a face image, and a predetermined registered face image in which a registrant's face image is registered in advance is stored. As an example, an example in which an IC card held by a registrant is used will be given.

[Constitution]
FIG. 1 is a diagram illustrating a configuration example of the face image authentication apparatus according to the present embodiment. The face image authentication device 10 includes a database 11 in which same-person similarity group data, different-person similarity group data, and face direction set data, which will be described later, are registered, a database management unit 12 that manages registration information of the database 11, registration A registered face image acquisition unit 13 that acquires a face image, an authentication face image acquisition unit 14 that acquires an authentication face image, a face direction determination unit 15 that determines the face direction of the registered face image and the authentication face image, a registered face image and an authentication face The similarity data calculation unit 16 that calculates the similarity of the image and associates the calculated similarity with the face direction of the face image, and whether or not the plurality of authentication face images and the registered face images are images of the same person. An authentication unit 17 for determination is provided.
The face image authentication device 10 also has an internal memory (not shown) for storing a registered face image, an authentication face image, a similarity, and a face direction. Each part 11-17 in the face image authentication apparatus 10 can access this internal memory as needed.

The same person similarity group data and different person similarity group data registered in the database 11 are obtained by collecting a plurality of face images photographed under photographing conditions including elements in a plurality of directions for one person. Group data representing a set of facial image features generated by associating similarity obtained by collating face images of the same person and face images of different persons with each of the elements in the plurality of directions. Similarity group data obtained for face images of the same person and face images of different persons included. Details of this will be described later.
The face direction set data is data obtained by associating face images of a large number of unspecified persons with the face directions when the face images are taken.
Here, set data (hereinafter referred to as face direction set data) in which the face direction and the face direction used by the face direction determination unit 15 for determining the face direction of the registered face image and the authentication face image are registered in the database 11. The face direction set data may be registered in a separate and independent database.

The database management unit 12 controls reading / writing of information registered in the three data stored in the database 11.
The registered face image acquisition unit 13 is connected to the card reader 3 and acquires a registered face image registered in the IC card 2 read by the card reader 3.
The authentication face image acquisition unit 14 is connected to nine cameras 4a to 4i that are imaging means capable of capturing a face image from a plurality of directions for one person to be authenticated. Although only two cameras 4a and 4i are shown in FIG. 1 to simplify the drawing, in practice, authentication face images are taken from a total of nine directions in front and in two stages, up, down, left, and right. It is arranged.
Here, since the authentication target person's face image only needs to be acquired from at least two directions, the camera for capturing the authentication target person's face image can be authenticated from two or more directions. What is necessary is just to be comprised so that a person's face can be image | photographed.
FIG. 2 illustrates processing steps in which the registered face image acquisition unit 13 and the authentication face image acquisition unit 14 acquire the registered face images and the authentication face images a to i, respectively.

The face direction determination unit 15 extracts the registered face image and the authentication face image pattern by a known pattern recognition method such as a subspace method, and the extracted pattern is included in the face direction set data registered in the database 11. Compared with the characteristics (pattern). Further, by performing this comparison process statistically on the set of face images included in the face direction set data, as shown in FIG. 2, the registered face images a to i and the face directions D1 to D9 of the authentication face image and Estimate D0.
For example, face images taken from a total of nine directions of front and top / bottom / left / right for a sufficient number of persons are registered in the face direction set data in advance as a sample, and the face direction of a certain authentication face image indicates which face in the sample The face directions D1 to D9 of the authentication face images a to i can be estimated by comparing features using a pattern recognition technique such as a subspace method to determine whether the face image is close to the face direction.
Note that the face direction D0 of the registered face image may be the same as the direction of any one of the authentication face images a to i, or may be a different direction.

The face direction determination unit 15 also uses the subspace method for the registered face image representing one face photograph printed on the IC card 2 as in the face direction determination of the authentication face images a to i. The face direction D0 is estimated by statistical processing.
The face direction D0 of the registered face image may be registered in advance in the IC card 2. In this case, the face direction estimated by the face direction determination unit 15 is the authentication face image taken by the cameras 4a to 4i. Therefore, the face direction registered in the IC card 2 may be used as the face direction D0 of the registered face image required in the following processing.
Also, here, a mode has been described in which the face direction of the registered face image and the authentication face image is determined by performing face image pattern recognition using the subspace method. Other methods may be used as long as they are simple techniques.

As shown in FIG. 2, the similarity data calculation unit 16 calculates the similarity Ra to Ri between the registered face images a to i and the authentication face images whose face directions have been determined by the direction determination unit 15, and the calculated similarities Ra to Ri are associated with face directions D1 to D9 and D0 of the face image. That is, as shown in FIG. 2, nine similarities Ra are obtained by matching nine authentication face images a to i (face directions D1 to D9) against one registered face image (face direction D0). -Ri are calculated, and the combinations (directions D0-D1 to D0-D9) of the registered face images and the authentication face images a-i from which the respective similarities Ra-Ri are obtained are referred to as similarities Ra-Ri. Associate.
For example, as shown in FIG. 3, the respective similarities are obtained as Ra = 0.52, Rb = 0.38,..., Ri = 0.22.
The nine similarities Ra to Ri and the combinations of face directions (directions D0 to D1 to D0 to D9) obtained in this way constitute a nine-dimensional vector.

Here, a person identity determination and a similarity calculation method in the face image authentication technique using the similarity of face images will be described.
In general, in the face image authentication technology, for example, the similarity between a plurality of face images of the same person or different persons registered in a predetermined database is derived in advance, and the face image registered in the database and the authentication The similarity with the face image of the person to be authenticated photographed at times is calculated. If the similarity is a certain value or more, it is determined that the person of the registered face image and the person to be authenticated are the same person. On the other hand, if the similarity is less than a certain value, it is determined that the person who registered the face image and the person to be authenticated are not the same person.

There are various methods for calculating the degree of similarity between face images, and one example is the eigenface method.
In this method, the face image is regarded as a feature vector having the number of dimensions of the image size (number of pixels), and principal component analysis (KL expansion: also called Karhunen-Loeve expansion) is performed on the feature vectors of many face images collected in advance. )I do. Among the eigenvectors obtained by this analysis, a plurality of eigenvectors having a high contribution ratio to the principal component formation are selected, and a projection matrix composed of the selected eigenvectors is generated.

Here, using the above projection matrix, the projection of feature vectors of a plurality of face images registered in the database is obtained, and the average vector of the projected plurality of vectors is also registered in the database.
Also, using this projection matrix, a vector obtained from the face image photographed at the time of authentication is projected, and a square error between the projected vector and the average vector registered in the database is obtained. The square error obtained in this way is set as the distance between the registered face image and the face image taken at the time of authentication. The similarity between face images can be obtained from this distance. The distance calculated here is a relationship in which the magnitude of the absolute value is opposite to the similarity between the face images, and takes a small value when they are the same person, and takes a large value when they are not the same person. That is, the similarity between face images takes a large value when they are the same person, and takes a small value when they are not the same person.
With the above-described method, the similarity data calculation unit 16 calculates the similarity Ra to Ri between the registered face image ai to which the face direction is determined by the direction determination unit 15 and the authentication face image.

The authentication unit 17 uses, for example, a group discrimination method realized by a statistical method such as a support vector machine method or a nuclear nonlinear discriminant analysis method to determine whether a plurality of authentication face images and registered face images are images of the same person. Determine if. This determination uses group determination data for determining whether the similarity between face images belongs to the same person similarity group data or different person similarity group data in the n-dimensional feature space.
The authentication unit 17 is connected to the terminal 5, and the terminal 5 receives, for example, a determination result that a plurality of authentication face images and registered face images are images of the same person. The terminal can unlock the gate (not shown). The terminal 5 can be appropriately changed in configuration according to the use of the face image authentication device.

Here, how to obtain the group discrimination data will be described. This group discrimination data is previously obtained by a group discrimination plane calculation unit (not shown) in the face image authentication apparatus according to the present embodiment or a group discrimination plane calculation apparatus (not shown) connected to the face image authentication apparatus. It is calculated and registered in the database 11. Here, a case where data representing a group discrimination plane is calculated by the group discrimination plane calculation unit in the face image authentication apparatus will be described.
First, the group discrimination plane calculation unit collects a plurality of face images taken in a plurality of face directions for one person. This is a set of learning face images. All face images included in this learning face image set have face directions specified, and all face directions that can occur as the face directions of the authentication face image and the registered face image described above. Must be included in the learning face image set.

Next, as shown in FIG. 4 (a), a similarity set obtained by collating face image data of the same person and a similarity set obtained by collating face image data of different persons. Is calculated. Each similarity included in these sets is associated with a face direction at the time of photographing. For example, if the similarity is calculated by collating face image data with face directions D1 and D2, the similarity may be associated with the face direction D1-D2.
The similarity set needs to be configured to include combinations of face directions included in all combinations of the authentication face image and the registered face image to be collated.

  When such similarity calculation and face direction association processing are performed for all learning face images, as shown in FIG. 4B, in the n-dimensional feature space, face images of the same person and different persons are displayed. For the face images, the similarity data associated with the face direction is obtained. The n-dimensional number n is determined by the number of face directions of all face images included in the set of learning face images.

In FIG. 4 (b), the group data represented by the plot of ○ represents the similarity obtained by the face images of the same person, and the group data represented by the x plot in the figure is different people. This represents the degree of similarity obtained between the face images.
Next, the group discriminant plane calculation unit regards all of these similar groups as n-dimensional feature vectors, and uses a pattern recognition technique realized by the support vector machine method, and all the feature vectors are the faces of the same person. Group discrimination data representing a group discrimination plane for discriminating the similarity between images and the similarity between face images of different persons is calculated.
That is, the group discrimination data includes the similarity between face images existing in the n-dimensional feature space shown in FIG. 4C belonging to the similarity group of face images of the same person, or face images of different persons. This is data representing a surface for determining whether or not it belongs to the similarity group.
The group discrimination data is obtained as described above.

The authentication unit 17 determines whether the similarity between the registered face image represented by the 9-dimensional vector calculated by the similarity data calculation unit 16 and the authentication face image belongs to a similarity group of face images of the same person. Whether the images belong to a similarity group of face images of different persons is determined based on the group determination data.
If the authentication unit 17 determines that the similarity between the registered face image and the authentication face image belongs to a similarity group of face images of the same person, the authentication unit 17 determines that the registered face image and the authentication face image are the same. On the other hand, if it is determined that the registered face image belongs to the similarity group of face images of persons having different degrees of similarity, the authentication unit 17 determines that the registered face image and the authentication face image are not the same. To do.

  Here, the case of performing group discrimination using the support vector machine method has been described. However, the method of performing group discrimination is not limited to this, and other methods such as a nuclear nonlinear discriminant analysis method can also be used.

FIG. 6 is an explanatory diagram of the authentication processing procedure according to the present embodiment.
First, it is determined whether there is an authentication request from a person to be authenticated (step S1). Specifically, when the registered face image acquisition unit 13 detects that the IC card 2 has been passed through the IC card reader 3 by the person to be authenticated, it is determined that an authentication request has been made.
Next, the registered image acquisition unit 13 acquires a registered face image registered in the IC card 2 through the IC card reader 3 (step S2).

The authentication face image acquisition unit 14 acquires authentication face images a to i photographed by the cameras 4a to 4i (step S3).
Next, the face direction determination unit 15 determines the face directions of the registered face image and the authentication face images a to i (step S4). If the face direction is determined for all registered face images and authentication face images, the processing procedure proceeds to step S5.
In step S4, when the face direction of the determined authentication face image or registered face image does not match the face direction of the sample face image included in the face direction set data, the face direction of the authentication face image or registered face image Is determined to be impossible, and the processing procedure returns to START.

The similar data calculation unit 16 collates each of the registered face image and the authentication face images a to i, and calculates the similarity Ra to Ri (step S5).
The authentication unit 17 reads group discrimination data from the database 11 (step S6).
Further, the authentication unit 17 has a similarity between the registered face image represented by a 9-dimensional vector and the authentication face image belonging to a similarity group of face images of the same person, or similarity between face images of different persons. Whether it belongs to the degree group is determined based on the group determination data (step S7).
If the authentication unit 17 determines that the similarity between the registered face image and the authentication face image belongs to the similarity group of face images of the same person, the authentication face image and the registered face image photograph the same person. It is determined that the image has been processed, and the determination result is output to the predetermined terminal 5 (step S8).
FIG. 5 is a diagram illustrating the relationship between the group discrimination plane in the n-dimensional feature space and the calculated similarity. In this figure, the similarity between the registered face image and the authentication face image is represented by a black circle plot. Since the black circle plot is on the same person similarity group side as the surface represented by the group discrimination data, the authentication unit 17 determines that the registered face image and the authentication face image are the same.
When step S8 ends, the processing procedure according to the present embodiment ends (END).

  Note that if it is determined in step S7 that the registered face image belongs to a similarity group of face images of persons with different degrees of similarity between the authentication face image and the authentication face image, the authentication unit 17 includes a plurality of authentication face images and registered face images. Are not images of the same person, and the processing procedure returns to STRAT.

  As described above, according to the present embodiment, since the identity of the registered face image and the authentication face image is integrally determined using the group determination surface in the n-dimensional feature space, the face directions of the registered face image and the authentication face image are determined. If the registered face image and the authentication face image have different facial expressions or illumination irradiation directions, or if the registered face image and the authentication face image all have different face directions, facial expressions, and illumination directions, high accuracy It is possible to perform face image authentication. Further, since the face direction of the registered face image or the authentication face image can be determined, it is possible to perform highly accurate face image authentication even when the face direction of the registered face image or the authentication face image is not constant. Further, even when the proof irradiation direction is different when the registered face image and the authentication face image are photographed, it is possible to perform highly accurate face image authentication as in the case where the face directions are different.

In the above description, the case where the face direction is used as the direction condition has been described. However, the illumination direction when photographing a face image may be used.
The determination result of the person to be authenticated may be additionally registered as learning data in the same person similarity group data, different person similarity group data, and face direction set data.

The figure which shows the structural example of the face image authentication apparatus by this embodiment. The figure which shows the functional block of the face image authentication apparatus by this embodiment. The figure which shows the relationship between the similarity of a face image, and a face direction. The figure which shows the calculation procedure of a group discrimination | determination surface, and the group discrimination | determination surface in n-dimensional feature space. The figure which shows the relationship between the group discrimination | determination surface in n-dimensional feature space, and the calculated similarity. The figure which shows the process sequence by the face image authentication apparatus by this embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Face image authentication apparatus, 11 ... Database, 12 ... Database management part, 13 ... Registered face image acquisition part, 14 ... Authentication face image acquisition part, 15 ... Face direction determination part, 16 ... Similarity data calculation part, 17 ... Authentication department.

Claims (5)

A method performed by a computer having a memory and control means, wherein the control means comprises:
Similarities obtained by collating face images of the same person and face images of different persons, by collecting multiple face images taken for a person with shooting conditions including elements in multiple directions. Group discrimination for discriminating between the same person similarity group data and the different person similarity group data included in the group data representing the set of facial image features generated by associating each of the plurality of direction elements with each other Processing to derive data;
A registered face image photographed under photographing conditions including elements in a predetermined direction, and a plurality of authentication face images of a person to be authenticated newly photographed under photographing conditions including elements in a plurality of directions, respectively Obtained by collating the similarity of each of the plurality of authentication face images with respect to the registered face image stored in the memory, associating each similarity with a direction element, and the associated information Whether the plurality of authentication face images and the registered face images are images of the same person by comparing with the group discrimination data whether it belongs to the same person similarity group or the different person similarity group. Execute the process of determining,
Face image authentication method.   The control means compares the registered face image or the authentication face image against a set of face images and elements of the direction of the face image, and compares the characteristics of the face image to thereby compare the predetermined face image with the predetermined face image. 2. The face image according to claim 1, wherein an element of a plurality of directions or an element of a plurality of directions of the plurality of authentication face images is estimated, and the attribution of information including a similarity associated with the estimated direction element is determined. Authentication method.   The face image authentication method according to claim 1, wherein the element of the direction is a shooting direction of the face image or a light irradiation direction at the time of shooting the face image. Similarities obtained by collating face images of the same person and face images of different persons, by collecting multiple face images taken for a person with shooting conditions including elements in multiple directions. Group discrimination for discriminating between the same person similarity group data and the different person similarity group data included in the group data representing the set of facial image features generated by associating each of the plurality of direction elements with each other A database with registered data,
Database management means for managing registration information of the database;
Registered face image acquisition means for acquiring a registered face image shot under shooting conditions including an element in a predetermined direction;
Authentication face image acquisition means for acquiring a plurality of authentication face images of a person to be authenticated newly photographed under shooting conditions including elements in a plurality of directions;
Similarity calculation means obtained by collating similarities of each of the plurality of authentication face images with respect to the registered face image, and associating each similarity with a direction element;
By comparing with the group discrimination data whether the information related by the similarity calculation unit belongs to the same person similarity group or the different person similarity group, the plurality of authentication face images And a determination unit that determines whether the registered face image is an image of the same person, and determines the identity of the authentication face image and the registered face image. Similarities obtained by collating face images of the same person and face images of different persons, by collecting multiple face images taken for a person with shooting conditions including elements in multiple directions. Group discrimination for discriminating between the same person similarity group data and the different person similarity group data included in the group data representing the set of facial image features generated by associating each of the plurality of direction elements with each other A computer that can access the database where the data is registered.
Registered face image acquisition means for acquiring a registered face image shot under shooting conditions including elements in a predetermined direction;
Authentication face image acquisition means for acquiring a plurality of authentication face images of a person to be authenticated newly photographed under a shooting condition including elements in a plurality of directions;
A similarity calculation unit that associates each similarity with a direction element obtained by comparing the similarities of each of the plurality of authentication face images with the acquired registered face image; and
By comparing with the group discrimination data whether the information related by the similarity calculation unit belongs to the same person similarity group or the different person similarity group, the plurality of authentication face images And a computer program for functioning as determination means for determining whether the registered face image is an image obtained by photographing the same person.

Images