WO2012105085A1

WO2012105085A1 - Image authentication device, image authentication method, program, and recording medium

Info

Publication number: WO2012105085A1
Application number: PCT/JP2011/071843
Authority: WO
Inventors: 植木　一也
Original assignee: Ｎｅｃソフト株式会社
Priority date: 2011-01-31
Filing date: 2011-09-26
Publication date: 2012-08-09

Abstract

Provided is an image authentication device with which it is possible to reduce authentication processing load and carry out authentication in a short amount of time. This image authentication device (1) comprises: an affiliation cluster determination means (21) for extracting feature value data from a plurality of logged images, and selecting, from among a plurality of clusters which are formed by grouping similar images in a feature space on the basis of the feature value data, one cluster to which an image to be assessed is affiliated, based on the feature value data which is extracted from the image to be assessed; and a first similarity calculation means (33) for carrying out a similarity calculation between the image to be assessed and a logged image in the cluster which the affiliation cluster determination means (21) has selected.

Description

Image authentication apparatus, image authentication method, program, and recording medium

The present invention relates to an image authentication device, an image authentication method, a program, and a recording medium.

In the field of face authentication technology and the like, an image authentication apparatus that authenticates an image by performing similarity calculation between a plurality of registered images and a determination target image is used (for example, see Patent Document 1).

JP 2008-191743 A

However, in the image authentication apparatus described in Patent Document 1 and the like, since the similarity calculation between all of the plurality of registered images and the determination target image is performed, the amount of authentication processing is large, and authentication takes a long time. It was.

Therefore, an object of the present invention is to provide an image authentication apparatus, an image authentication method, a program, and a recording medium that can reduce the amount of authentication processing and can perform authentication in a short time.

In order to achieve the above object, an image authentication apparatus according to the present invention includes:
Feature value data extracted from a plurality of registered images, and extracted from the determination target image from a plurality of clusters formed by grouping similar images in the feature space based on the feature value data Based on the data, the belonging cluster determining means for selecting one cluster to which the determination target image belongs, and the similarity calculation between the determination target image and the registered image in the cluster selected by the belonging cluster determining means are performed. And a first similarity calculation unit.

The image authentication method of the present invention includes:
Feature value data extracted from a plurality of registered images, and extracted from the determination target image from a plurality of clusters formed by grouping similar images in the feature space based on the feature value data Based on the data, the belonging cluster determination step of selecting one cluster to which the determination target image belongs, and the similarity calculation between the determination target image and the registered image in the cluster selected in the belonging cluster determination step A first similarity calculation step is included.

The program of the present invention causes a computer to execute the image authentication method of the present invention.

The recording medium of the present invention records the program of the present invention.

According to the present invention, the amount of authentication processing can be reduced and authentication can be performed in a short time.

FIG. 1 is a block diagram illustrating an example (first embodiment) of an image authentication apparatus according to the present invention. FIG. 2 is a schematic diagram showing an example of the belonging cluster determination step of the present invention. FIG. 3 is a schematic diagram showing an example of the first similarity calculation step of the present invention. FIG. 4 is a block diagram showing another example (embodiment 2) of the image authentication apparatus of the present invention. FIG. 5 is a schematic diagram showing an example of the cluster range determining step of the present invention. FIG. 6 is a schematic diagram showing another example of the first similarity calculation step of the present invention. FIG. 7 is a block diagram showing still another example (third embodiment) of the image authentication apparatus of the present invention. FIG. 8 is a schematic diagram showing an example of the feature space production process and the cluster production process of the present invention. FIG. 9 is a schematic diagram showing an example of the registration process of the present invention. FIG. 10 is a schematic diagram showing another example of the belonging cluster determination step of the present invention. FIG. 11 is a block diagram showing still another example (embodiment 4) of the image authentication apparatus of the present invention. FIG. 12 is a schematic diagram showing an example of the cluster range production process of the present invention. FIG. 13 is a block diagram showing still another example (embodiment 5) of the image authentication apparatus of the present invention.

The image authentication apparatus of the present invention further includes a cluster range determination unit that determines a cluster range centered on a cluster selected by the cluster determination unit based on a predetermined criterion as a target of similarity calculation, It is preferable that one similarity calculation unit is a similarity calculation unit that performs similarity calculation between the determination target image and the registered image within the determined cluster range. Similarly, in the image authentication method of the present invention, a cluster range determination step for determining a cluster range centered on the cluster selected in the belonging cluster determination step as a target for similarity calculation based on a predetermined criterion. Preferably, the first similarity calculation step is a similarity calculation step of calculating a similarity between the determination target image and the registered image within the determined cluster range.

In the image authentication apparatus of the present invention, a feature space creating means for creating a feature space using a plurality of feature space creation images, and feature quantities are extracted by mapping the feature space creation image to the feature space In the feature space, based on the feature amount, cluster creation means for forming each cluster, and feature amount data is extracted from a plurality of registered images, and on the basis of the feature amount data, a plurality of feature spaces in the feature space are extracted. It is preferable that registration means for registering the registration image in any one of the clusters is included. Similarly, in the image authentication method of the present invention, a feature space creation step of creating a feature space using a plurality of feature space creation images, and a feature space mapping to the feature space A feature is extracted, and in the feature space, a cluster creation step for forming each cluster based on the feature value, feature value data is extracted from a plurality of registered images, and the feature space is based on the feature value data. And a registration step of registering the registration image in any one of the plurality of clusters.

In the image authentication apparatus of the present invention, it is preferable that the image authentication apparatus further includes a cluster range creating means for creating a cluster range, which is a target of similarity calculation, for each cluster based on a predetermined criterion. Similarly, the image authentication method of the present invention may further include a cluster range creation step of creating a cluster range to be subjected to similarity calculation for each cluster based on a predetermined criterion in cluster units. preferable.

In the image authentication apparatus of the present invention, further, second similarity calculation means for performing similarity calculation different from the similarity calculation performed by the first similarity calculation means, and the first similarity calculation It is preferable that the calculation result by the means and the calculation result by the second similarity calculation means are integrated to determine the similarity. Similarly, in the image authentication method of the present invention, the second similarity calculation step for performing similarity calculation different from the similarity calculation performed in the first similarity calculation step, It is preferable to include an integrated determination step of determining a similarity by integrating the calculation result of the similarity calculation step and the calculation result of the second similarity calculation step.

In the image authentication apparatus according to the present invention, it is preferable that the feature space creating means includes a neural network (Neural Network). Similarly, in the image authentication method of the present invention, it is preferable that the feature space creation step is performed by a neural network.

In the image authentication device and the image authentication method of the present invention, the image may be a face image, for example. Examples of the image include a determination target image, a registered image, and a feature space creation image.

Next, an image authentication apparatus, an image authentication method, a program, and a recording medium according to the present invention will be described with examples. However, the present invention is not limited to the following examples. In addition, in the following FIGS. 1 to 13, the same reference numerals are given to the same portions.

(Embodiment 1)
FIG. 1 is a block diagram of an image authentication apparatus according to this embodiment. The image authentication apparatus 1 according to the present embodiment includes a belonging cluster determination unit 21 and a first similarity calculation unit 33 as main components. The affiliation cluster determination unit 21 is connected to the first similarity calculation unit 33. The first similarity calculation unit 33 is connected to the determination unit 34. The determination unit 34 is an arbitrary component and may or may not be included. Each unit of the affiliation cluster determination unit 21, the first similarity calculation unit 33, and the determination unit 34 may be configured using, for example, dedicated hardware (for example, a central processing unit (CPU)). It can also be realized on a computer by software processing. In the image authentication apparatus of the present embodiment, feature spaces and clusters created in advance are used. The feature space can be produced by, for example, a feature space production process of the third embodiment described later, but is not limited to this, and may be produced by any method. The cluster is formed by extracting feature amount data from a plurality of registered images and grouping similar images into groups in the feature space based on the feature amount data. Specifically, the cluster can be produced by, for example, a cluster production process of Embodiment 3 to be described later, but is not limited to this, and may be produced by any method.

Hereinafter, taking the case where the image is a face image as an example, the image authentication apparatus and the image authentication method of the present embodiment will be described in more detail.

Affiliation Cluster Determination Process An example of the belonging cluster determination process is shown in the schematic diagram of FIG. Based on the feature amount data extracted from the determination target image, the affiliation cluster determination unit 21 selects a cluster to which the determination target image belongs from a plurality of clusters in the feature space (in this example, the face feature space). Select one. In FIG. 2, a cluster storing a registered image A of a middle-aged woman and a registered image B of a young woman is selected as a cluster to which the determination target image A11 of a middle-aged woman belongs.

First Similarity Calculation Step Next, the first similarity calculation means 33 performs similarity calculation between the registered image in the cluster selected in the cluster determination step and the determination target image. . For the similarity calculation, a conventionally known method such as a method using a distance (for example, Euclidean distance) in the face feature space can be applied. An example of this process is shown in the schematic diagram of FIG. In FIG. 3, the similarity calculation between the determination target image A11 and the registered images A and B among the registered images included in the cluster is shown as a representative. As a result of the similarity calculation, the determination unit 34 determines that the registered image A has the highest similarity with the determination target image A11, and the determination target image A11 belongs to the same person as the registration image A. It is authenticated as (face image). As described above, by narrowing the range for performing the similarity calculation to the registered images in the cluster, the image authentication apparatus of the present embodiment can reduce the authentication processing amount and can perform authentication in a short time.

(Embodiment 2)
The block diagram of FIG. 4 shows another configuration of the image authentication apparatus of the present invention. As shown in FIG. 4, the image authentication device 200 has the same configuration as the image authentication device 1 shown in FIG. 1 except that it further includes a cluster range determination unit 32. The cluster range determination unit 32 is located between the belonging cluster determination unit 21 and the first similarity calculation unit 33 and is connected to them.

Cluster Range Determination Step An example of the cluster range determination step is shown in the schematic diagram of FIG. In this example, the shape of the eyebrows when the determination target image A12 is acquired is different from the determination target image A11 in the first embodiment. For this reason, in the affiliation cluster determination step, as the cluster to which the determination target image A12 belongs, not the cluster in which the registered images A and B are stored, but the nearest cluster is selected. In the cluster range determining step, the cluster range determining means 32 determines a cluster range centered on the cluster selected in the belonging cluster determining step as a target for similarity calculation based on a predetermined criterion. Specifically, for example, the cluster range is determined by adding close clusters to the similarity calculation target until the number of data that is the target of similarity calculation reaches the number set by the user. In FIG. 5, a cluster included in a range surrounded by a broken line is determined as the cluster range.

First Similarity Calculation Step Next, the first similarity calculation means 33 performs similarity calculation between the registered image and the determination target image within the determined cluster range. For example, a conventionally known method can be applied to the similarity calculation. Another example of this process is shown in the schematic diagram of FIG. In FIG. 6, the similarity calculation between the determination target image A12 and the registered images A and B among the registered images included in the cluster range is shown as a representative. As a result of the similarity calculation, the determination unit 34 determines that the registered image A has the highest similarity with the determination target image A12, and the determination target image A12 is of the same person as the registration image A ( Face image). As described above, by narrowing the range in which the similarity calculation is performed to the registered images within the cluster range, the image authentication apparatus according to the present embodiment can reduce the authentication processing amount and perform the authentication in a short time. .

(Embodiment 3)
FIG. 7 is a block diagram showing still another configuration of the image authentication apparatus of the present invention. As shown in FIG. 7, the image authentication apparatus 300 further includes a feature space creating unit 10, a cluster creating unit 11, and a mapping unit 20 to the feature space (hereinafter referred to as “mapping unit 20”). 1 is the same configuration as the image authentication apparatus 1 shown in FIG. In the image authentication apparatus of this embodiment, the mapping unit 20 and the assigned cluster determination unit 21 constitute a registration unit. However, the registration means is merely an example. In the present invention, the registration means may have any configuration as long as it can perform the registration process described later. In the image authentication apparatus according to the present embodiment, the mapping unit 20, the belonging cluster determination unit 21, the first similarity calculation unit 33, and the determination unit 34 constitute an authentication unit. In the authentication unit, the mapping unit 20 and the determination unit 34 are optional components and may or may not be included. Although FIG. 7 shows a case where the registration unit and the authentication unit share the mapping unit 20 and the belonging cluster determination unit 21, the registration unit and the authentication unit include, for example, the mapping unit 20 and the belonging cluster determination unit. The means 21 may be provided separately. In this case, the image authentication apparatus 300 includes, for example, two mapping units 20 and belonging cluster determination units 21, respectively. The feature space creating means 10 is connected to the feature space creating image database (DB) 40 and the cluster creating means 11 provided outside the image authentication apparatus 300. The feature space creation image DB 40 stores a plurality of feature space creation images acquired in advance. The mapping unit 20 is connected to a registered image DB 50 and an assigned cluster determination unit 21 provided outside the image authentication apparatus 300. The registered image DB 50 stores a plurality of registered images acquired in advance. The feature space creation image DB 40 and the registered image DB 50 may be incorporated in the image authentication apparatus 300. Each part of the feature space creating unit 10, the cluster creating unit 11, the mapping unit 20, the belonging cluster determining unit 21, the first similarity calculating unit 33, and the determining unit 34 includes dedicated hardware (for example, a central processing unit (CPU)). Etc.) and can be realized on a computer by software processing.

Feature space creation step The feature space creation means 10 creates a feature space using a plurality of feature space creation images stored in the feature space creation image DB 40.

Cluster production step The cluster production means 11 extracts the feature quantity by mapping the feature space production image onto the feature space, and forms each cluster in the feature space based on the feature quantity.

FIG. 8 is a schematic diagram illustrating an example of the feature space manufacturing process and the cluster manufacturing process. First, prior to these steps, a feature space creation image DB 40 is prepared in advance. As the feature space creation image DB 40, for example, a public DB published on the Internet (in this example, a public DB storing a plurality of person's face images) may be used. The feature space creation image (in this example, a plurality of human face images to be determined) may be created in advance, or a plurality of feature space creation images may be created at the site where image authentication is performed. You may produce by acquiring beforehand. The number of persons included in the feature space is not particularly limited and is, for example, in the range of thousands to millions. In FIG. 8, feature space creation images (A1 and A2) for middle-aged women, feature space creation images for young women (B1 and B2), and feature space creation images for middle-aged men (C1 and C2). , And an image (D1 and D2) for producing a feature space of an elderly man. For example, when the feature space creation image DB 40 is created in advance in a laboratory or the like, the feature space of the same person is taken into consideration as shown in FIG. It is preferable to obtain a plurality of production images by changing the shooting conditions. In FIG. 8, although the example which acquired the image for feature space production in the bright environment (A1, B1, C1, and D1) and the dark environment (A2, B2, C2, and D2) was shown, in addition to this, for example, Conditions such as the color of illumination, the intensity of illumination, and the shooting angle (face orientation) of a person may be changed.

Next, the feature space creation means 10 creates a feature space (in this example, a face feature space) using a plurality of feature space creation images. Specifically, for example, the facial feature space is created by collecting images of the same person at a short distance and separating images of different persons at a long distance. For the creation of the face feature space, a conventionally known method can be applied. For example, a technique such as a neural network or a linear discriminant analysis (LDA) can be applied.

Next, the cluster creation means 11 extracts the feature quantity by mapping the feature space creation image onto the feature space, and forms each cluster based on the feature quantity in the face feature space. The feature amount is extracted by performing feature amount conversion of a plurality of feature space creation images. The number of clusters formed in the face feature space is not particularly limited. For example, when an image for creating feature spaces of 1 million people is used, for example, the range is 50 to 100. An image for creating a feature space of 10,000 to 20,000 persons is stored. However, the number of the clusters is merely an example, and may be appropriately set according to the number of feature space creation images. For the formation of each cluster, a conventionally known method can be applied. For example, a technique such as K-means method, clustering based on binary space division, or clustering by Gaussian mixture model (GMM) can be applied. The K-means method is a technique of grouping feature space creation images into a predetermined number (K) of clusters and repeating the grouping so that the correlation with the cluster center is high. The clustering based on the binary space division is a technique for recursively dividing a set of feature space creation images according to similarity. The clustering by GMM is a technique for modeling the appearance probability distribution of the feature space creation image in the form of a normal mixture distribution. Among these, the K-means method is preferable because it is easy to adjust the number of images for creating the feature space of each cluster.

[Registration process]
Next, the registration unit extracts feature amount data from a plurality of registered images, and registers the registered image in any one of a plurality of clusters in the feature space based on the feature amount data.

An example of this process is shown in the schematic diagram of FIG. First, prior to this process, a registered image DB 50 is prepared in advance. The registered image DB 50 may be created by, for example, acquiring a plurality of registered images (in this example, face images of a plurality of persons to be determined) in advance in a laboratory or the like, or performing image authentication. A plurality of registered images may be obtained in advance. Further, as the registered image DB 50, a part or all of the feature space creation image DB 40 may be used. In FIG. 9, a registered image A of a middle-aged woman, a registered image B of a young woman, a registered image C of a middle-aged man, and a registered image D of an elderly man are shown as representatives.

Next, the feature amount data of each registered image is extracted by performing the feature amount conversion of the plurality of registered images in the same manner as in the feature space creation image.

Next, the registered image is registered in any one of a plurality of clusters in the face feature space by the mapping unit 20 and the assigned cluster determination unit 21 based on the feature amount data. Specifically, for example, a distance between the feature amount data (vector) and the center of each cluster when the registered image is mapped to the face feature space is obtained, and the cluster having the shortest distance is determined as the registered image. Is determined to belong to the cluster.

[Certification process]
Next, the authentication unit authenticates the registered image and the determination target image. As described above, this process includes the belonging cluster determination process and the first similarity calculation process.

Affiliation cluster determination process FIG. 10 is a schematic diagram showing another example of the belonging cluster determination process. First, the mapping unit 20 and the assigned cluster determination unit 21 extract feature amount data from the determination target image in the same manner as in the registered image, and based on the feature amount data, a plurality of pieces in the face feature space are extracted. One cluster to which the determination target image belongs is selected from the clusters. In FIG. 10, the cluster (see FIG. 9) in which the registered image A of the middle-aged woman is stored in the registration process is selected as the cluster to which the determination target image A11 of the middle-aged woman acquired in a bright environment belongs. Has been.

First Similarity Calculation Step Next, the first similarity calculation step is performed in the same manner as in the first embodiment. However, in the present embodiment, the registered image B is not included in the cluster selected by the affiliation cluster determination unit 21 and is therefore not subjected to the similarity calculation. Also in the image authentication apparatus of the present embodiment, by narrowing the range for performing the similarity calculation to the registered images in the cluster, the amount of authentication processing can be reduced and authentication can be performed in a short time.

(Embodiment 4)
The block diagram of FIG. 11 shows still another configuration of the image authentication apparatus of the present invention. As shown in FIG. 11, the image authentication apparatus 400 has the same configuration as the image authentication apparatus 300 shown in FIG. 7 except that the image authentication apparatus 400 further includes a cluster range creation unit 12 and a cluster range determination unit 32. The cluster range preparation means 12 is connected to the cluster preparation means 11. The image authentication method according to the present embodiment is performed in the same manner as in the above-described third embodiment except for the cluster range creation step and the first similarity calculation step described below.

Cluster range creation step The cluster range creation means 12 creates a cluster range for each cluster based on a predetermined criterion in units of clusters. Specifically, for example, the cluster range is created by adding close clusters to the similarity calculation object until the number of data that is the object of similarity calculation reaches the number set by the user. FIG. 12 shows an example of the cluster range production process. In FIG. 12, in the above-described third embodiment, clusters included in a range surrounded by a broken line are subjected to similarity calculation with respect to a cluster storing a middle-aged female feature space creation image (A1 and A2). Is determined as the cluster range.

First similarity calculation step The first similarity calculation step in the present embodiment is performed in the same manner as in the second embodiment. Also in the image authentication apparatus of the present embodiment, by narrowing the range for performing the similarity calculation to the registered images within the cluster range, the authentication processing amount can be reduced and authentication can be performed in a short time.

(Embodiment 5)
The block diagram of FIG. 13 shows still another configuration of the image authentication apparatus of the present invention. As shown in FIG. 13, the image authentication apparatus 500 includes an image authentication apparatus 400 shown in FIG. 11 except that the authentication means includes a second similarity calculation means 35 and the determination means 34 is an integrated determination means 36. It is the same composition as. The second similarity calculation means 35 is connected to the cluster range determination means 32. The integrated determination unit 36 is connected to the first similarity calculation unit 33 and the second similarity calculation unit 35.

Second similarity calculation step In the second similarity calculation step, a cluster centered on the cluster selected in the belonging cluster determination step is the same as in the first similarity calculation step of the fourth embodiment. The similarity calculation between the registered image and the determination target image within the range is performed. The order of execution of the second similarity calculation step and the first similarity calculation step is not particularly limited, and both steps may be performed simultaneously, or one of the steps may be performed first. . Examples of the second similarity calculation unit 35 include a conventionally known subspace method similarity calculation unit and an inter-data distance calculation unit. The inter-data distance calculation means is means for calculating a distance between the determination target image mapped to the face feature space and each data included in the cluster range.

Subspace Method Similarity Calculation Unit First, with reference to FIG. 5, an example in which the second similarity calculation unit 35 is a subspace method similarity calculation unit in the image authentication apparatus of the present embodiment will be described. . In this example, the similarity between the determination target image A12 and the registered image B is 0.75, and the similarity between the determination target image A12 and the registered image A is 0. If it is calculated as 74, it is determined that the registered image B, rather than the registered image A, has the highest similarity with the determination target image A12. Even in such a case, the integration determination unit 36 integrates the calculation result of the first similarity calculation unit 33 and the calculation result of the second similarity calculation unit (subspace method similarity calculation unit) 35 to determine. Thus, the determination target image A12 can be authenticated as the same person (face image) as the registered image A. Thus, according to this example, by using two different similarity calculation means, erroneous authentication can be prevented and the authentication accuracy can be further improved.

Next, the data distance calculation means, with reference to FIG. 5, the image recognition apparatus of this embodiment, the second similarity calculation unit 35, an example of a case where the data distance calculation means. In this example, the distance between the determination target image A12 mapped in the face feature space and each data (each registered image) included in the cluster range is calculated by the inter-data distance calculation means. . As a result, it is calculated that the distance between the determination target image A12 and the registered image B is 0.43, and the distance between the determination target image A12 and the registered image A is 0.17. Assume that A is determined to have the highest similarity with the determination target image A12. In this case, the registered image A is similar to the determination target image A12 in both the calculation result by the first similarity calculation unit 33 and the calculation result of the second similarity calculation unit (inter-data distance calculation unit) 35. Since it is determined that the degree is the highest, the integrated determination unit 36 authenticates that the determination target image A12 belongs to the same person as the registered image A. Also in this example, the authentication accuracy can be further improved by using two different similarity calculation means.

In the image authentication apparatus of the present embodiment, the example in which the second similarity calculation unit 35 and the integrated determination unit 36 are applied to the image authentication apparatus of the above-described fourth embodiment has been described. However, the image of the above-described first to third embodiments is illustrated. The second similarity calculation unit 35 and the integrated determination unit 36 may be applied to the authentication device.

In Embodiments 1 to 5, a face image is used as the image, but an image of the entire head may be used instead of the face image. Thereby, the authentication accuracy can be further improved.

In Embodiments 1 to 5, a face image is used as the image, but a fingerprint image, a vein image, or the like may be used instead of the face image.

[Embodiment 6]
The program of this embodiment is a program that can execute the above-described image authentication method on a computer. The program of this embodiment may be recorded on a recording medium, for example. The recording medium is not particularly limited, and examples thereof include a random access memory (RAM), a read-only memory (ROM), a hard disk (HD), an optical disk, and a floppy (registered trademark) disk (FD).

As mentioned above, although this invention was demonstrated with reference to embodiment, this invention is not limited to the said embodiment. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

This application claims priority based on Japanese Patent Application No. 2011-17663 filed on January 31, 2011, the entire disclosure of which is incorporated herein.

DESCRIPTION OF SYMBOLS 1,200,300,400,500 Image authentication apparatus 10 Feature space preparation means 11 Cluster preparation means 12 Cluster range preparation means 20 Mapping means 21 to feature space Affiliation cluster determination means 32 Cluster range determination means 33 First similarity calculation Means 34 Determination means 35 Second similarity calculation means 36 Integrated determination means 40 Feature space creation image DB
50 registered image DB

Claims

Feature value data extracted from a plurality of registered images, and extracted from the determination target image from a plurality of clusters formed by grouping similar images in the feature space based on the feature value data Based on the data, the belonging cluster determining means for selecting one cluster to which the determination target image belongs, and the similarity calculation between the determination target image and the registered image in the cluster selected by the belonging cluster determining means are performed. An image authentication apparatus comprising: a first similarity calculation unit.
Furthermore, based on a predetermined criterion, including cluster range determination means for determining a cluster range centered on the cluster selected by the belonging cluster determination means as a target of similarity calculation,
2. The similarity calculation unit according to claim 1, wherein the first similarity calculation unit is a similarity calculation unit that calculates a similarity between the determination target image and the registered image within the determined cluster range. Image authentication device.
Furthermore, a feature space creating means for creating a feature space using a plurality of feature space creating images,
A cluster creation means for extracting a feature quantity by mapping the feature space creation image to the feature space, and forming each cluster in the feature space based on the feature quantity;
And a registration unit that extracts feature data from a plurality of registered images and registers the registered image in any one of a plurality of clusters in the feature space based on the feature data. The image authentication apparatus according to claim 1 or 2.
4. The method according to claim 1, further comprising cluster range creation means for creating a cluster range, which is a target of similarity calculation for each cluster, in units of clusters based on a predetermined criterion. The image authentication device according to claim 1.
And second similarity calculation means for performing similarity calculation different from the similarity calculation performed by the first similarity calculation means,
The integrated determination means for determining the similarity by integrating the calculation result by the first similarity calculation means and the calculation result by the second similarity calculation means. The image authentication device according to any one of the above.
The image authentication apparatus according to any one of claims 3 to 5, wherein the feature space creation means includes a neural network (Neural Network).
The image authentication apparatus according to claim 1, wherein the image is a face image.
Feature value data extracted from a plurality of registered images, and extracted from the determination target image from a plurality of clusters formed by grouping similar images in the feature space based on the feature value data Based on the data, the belonging cluster determination step of selecting one cluster to which the determination target image belongs, and the similarity calculation between the determination target image and the registered image in the cluster selected in the belonging cluster determination step An image authentication method comprising a first similarity calculation step to be performed.
Furthermore, based on a predetermined criterion, including a cluster range determination step for determining a cluster range centered on the cluster selected in the cluster determination step as a target of similarity calculation,
9. The similarity calculation step according to claim 8, wherein the first similarity calculation step is a similarity calculation step of calculating a similarity between the determination target image and the registered image within the determined cluster range. Image authentication method.
Furthermore, a feature space creation process for creating a feature space using a plurality of feature space creation images;
A cluster creation step of extracting a feature amount by mapping the feature space creation image onto the feature space, and forming each cluster in the feature space based on the feature amount; and
And a registration step of extracting feature amount data from a plurality of registered images and registering the registered image in any one of a plurality of clusters in the feature space based on the feature amount data. 10. The image authentication method according to the range 8 or 9.
The cluster range creation step of creating a cluster range to be subjected to similarity calculation for each cluster in cluster units based on a predetermined criterion. The image authentication method according to claim 1.
Furthermore, a second similarity calculation step for performing similarity calculation different from the similarity calculation performed in the first similarity calculation step;
9. The integrated determination step of determining the similarity by integrating the calculation result of the first similarity calculation step and the calculation result of the second similarity calculation step. The image authentication method according to claim 11.
The image authentication method according to any one of claims 10 to 12, wherein the feature space creation step is performed by a neural network (Neural Network).
The image authentication method according to claim 8, wherein the image is a face image.
A program causing a computer to execute the image authentication method according to any one of claims 8 to 14.
A recording medium in which the program according to claim 15 is recorded.