Feature amount conversion learning device, authentication device, feature amount conversion learning method, authentification method and program

Abstract

To provide a feature amount conversion learning device, an authentication device, a feature amount conversion learning method, an authentication method, and a recording medium, which perform feature amount super resolution with higher accuracy.SOLUTION: A feature amount conversion learning device 200 includes: an image patch acquisition unit 201 that acquires a first image; an image reduction unit 202 that reduces the first image to a second image having a resolution lower than that of the first image; an image enlargement unit 103 that enlarges the second image to a third image having the same resolution as that of the first image; a feature amount extraction unit 104 that extracts a first feature amount as a feature amount of the first image and a second feature amount as a feature amount of the third image; a feature amount conversion unit 105 that converts the second feature amount into the third feature amount; and a learning control unit 208 that allows the feature amount conversion unit 105 to learn a feature amount conversion method on the basis of a comparison result between the first feature amount and the third feature amount.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a feature transformation learning device, an authentication device, a feature transformation learning method, an authentication method, and a recording medium. [Background technology]

[0002] A technology for converting an image into an image with a higher resolution than the original image is called super-resolution (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2013-31163 Summary of the Invention [Problem to be solved by the invention]

[0004] There may be cases where the resolution of the image to be authenticated is insufficient, such as when performing iris authentication using a face image. In such cases, it is possible to use super-resolution to convert the image to a higher resolution image and use it for authentication, but since it is necessary to generate a high-resolution image, super-resolution may be computationally expensive.

[0005] An example of an objective of this disclosure is to provide a feature transformation learning device, an authentication device, a feature transformation learning method, an authentication method, and a recording medium that can solve the above-mentioned problems. [Means for solving the problem]

[0006] According to a first aspect of this disclosure, a feature transformation learning device includes an image acquisition means for acquiring a first image, an image reduction means for reducing the first image into a second image having a lower resolution than the first image, an image enlargement means for enlarging the second image into a third image having the same resolution as the first image, a feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the third image, a feature transformation means for converting the second feature into a third feature, and a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature.

[0007] According to a second aspect of this disclosure, an authentication device includes an authentication target image acquisition means for acquiring an authentication target image, an image enlargement means for enlarging the authentication target image, a feature extraction means for extracting features of the enlarged image of the authentication target image in the same manner as the feature extraction method for a comparison image having a higher resolution than the authentication target image, a feature conversion means for converting features of the enlarged image of the authentication target image based on learning results of a feature conversion method using a loss function whose value becomes smaller the smaller the difference between the features after conversion from the features of the learning image and the features of a degraded image obtained by reducing and then enlarging the learning image, and a feature comparison means for comparing the features after conversion by the feature conversion means with the features of the comparison image.

[0008] According to a third aspect of the disclosure, a feature transformation learning method includes obtaining a first image, reducing the first image into a second image having a lower resolution than the first image, enlarging the second image into a third image having the same resolution as the first image, extracting a first feature that is a feature of the first image and a second feature that is a feature of the third image, converting the second feature into a third feature, and learning a feature transformation method that converts the second feature into the third feature based on a comparison result between the first feature and the third feature.

[0009] According to a fourth aspect of this disclosure, an authentication method includes obtaining an authentication target image, enlarging the authentication target image, extracting features of the enlarged image of the authentication target image using the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image, converting features of the enlarged image of the authentication target image based on learning results of a feature conversion method using a loss function whose value becomes smaller the smaller the difference between features after conversion from the features of the learning image and features of a degraded image obtained by reducing and then enlarging the learning image, and comparing the features after conversion from the features of the enlarged image of the authentication target image with the features of the comparison image.

[0010] According to a fifth aspect of this disclosure, the recording medium is a recording medium having recorded thereon a program for causing a computer to execute the following operations: acquiring a first image; reducing the first image into a second image having a lower resolution than the first image; enlarging the second image into a third image having the same resolution as the first image; extracting a first feature that is a feature of the first image and a second feature that is a feature of the third image; converting the second feature into a third feature; and learning a feature conversion method for converting the second feature into the third feature based on a comparison result between the first feature and the third feature.

[0011] According to a sixth aspect of this disclosure, a recording medium is a recording medium that records a program for executing the following operations: acquiring an authentication target image; enlarging the authentication target image; extracting features of the enlarged image of the authentication target image using the same method as the feature extraction method for a comparison image having a higher resolution than the authentication target image; converting features of the enlarged image of the authentication target image based on learning results of a feature conversion method using a loss function whose value becomes smaller the smaller the difference between features after conversion from the features of the learning image and features of a degraded image obtained by reducing and then enlarging the learning image; and comparing the features after conversion from the features of the enlarged image of the authentication target image with the features of the comparison image. [Brief description of the drawings]

[0012] [Figure 1] 1 is a diagram illustrating an example of the configuration of an authentication device according to a first embodiment. [Diagram 2] 5 is a flowchart showing an example of a processing procedure for performing iris authentication by the authentication device according to the first embodiment. [Diagram 3] FIG. 11 is a diagram illustrating an example of the configuration of a feature transformation learning device according to a second embodiment. [Figure 4] FIG. 11 is a diagram illustrating an example of a processing procedure in which the feature transformation learning device according to the second embodiment learns a feature transformation unit. [Diagram 5] FIG. 11 is a diagram illustrating an example of a processing procedure for calculating a loss by the feature transformation learning device according to the second embodiment. [Figure 6] FIG. 13 is a diagram illustrating an example of the configuration of a feature amount extraction unit according to the third embodiment. [Figure 7] FIG. 13 is a diagram illustrating an example of the configuration of an attention processing unit according to the third embodiment. [Figure 8] FIG. 13 is a diagram illustrating an example of the configuration of an attention block 302 according to a first modified example of the attention processing unit according to the third embodiment. [Figure 9] FIG. 13 is a diagram illustrating an example of the configuration of an attention processing unit 303 according to a second modified example of the attention processing unit in the third embodiment. [Figure 10] FIG. 13 is a diagram illustrating an example of the configuration of an attention processing unit 304 according to a second modified example of the attention processing unit according to the third embodiment. [Figure 11] FIG. 13 is a diagram illustrating an example of the configuration of a feature extraction unit according to a modified example of the feature extraction unit in the fourth embodiment. [Figure 12] FIG. 13 is a diagram illustrating an example of a hierarchical structure of a neural network that configures a feature extraction unit according to the fourth embodiment. [Figure 13] FIG. 13 is a diagram illustrating an example of the configuration of an authentication device according to a fifth embodiment. [Figure 14] FIG. 13 is a diagram illustrating an example of the configuration of a feature transformation learning device according to a sixth embodiment. [Figure 15] FIG. 23 is a diagram illustrating an example of the configuration of an authentication device according to the seventh embodiment. [Figure 16]FIG. 23 is a diagram illustrating an example of the configuration of a feature transformation learning device according to the eighth embodiment. [Figure 17] FIG. 13 is a diagram illustrating an example of the configuration of an authentication device according to a ninth embodiment. [Figure 18] FIG. 23 is a diagram illustrating an example of the configuration of a feature transformation learning device according to a tenth embodiment. [Figure 19] FIG. 23 is a diagram illustrating an example of the configuration of an authentication device according to an eleventh embodiment. [Figure 20] FIG. 23 is a diagram illustrating an example of the configuration of a feature transformation learning device according to a twelfth embodiment. [Figure 21] FIG. 23 is a diagram illustrating an example of the configuration of a feature transformation learning device according to a thirteenth embodiment. [Figure 22] FIG. 23 is a diagram illustrating an example of the configuration of an authentication device according to a fourteenth embodiment. [Diagram 23] 23 is a flowchart showing an example of a processing procedure in a feature transformation learning method according to the fifteenth embodiment. [Figure 24] 23 is a flowchart showing an example of a processing procedure in an authentication method according to the sixteenth embodiment. [Diagram 25] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] First Embodiment Fig. 1 is a diagram showing an example of the configuration of an authentication device according to the first embodiment. In the configuration shown in Fig. 1, an authentication device 100 includes an authentication target image acquisition unit 101, a comparison image acquisition unit 102, an image enlargement unit 103, a feature extraction unit 104, a feature conversion unit 105, and a feature comparison unit 106.

[0014] The authentication device 100 extracts a feature amount of an image and performs authentication using the extracted feature amount. In the following, an example will be described in which the authentication device 100 performs iris authentication using a low-resolution image. However, the authentication performed by the authentication device 100 is not limited to iris authentication. The resolution of two images shall be compared with the images being of the same size.

[0015] The authentication target image acquiring unit 101 acquires a photographed image of the eyes of a person to be authenticated, which is an authentication target image. The authentication target image acquiring unit 101 corresponds to an example of an authentication target image acquiring means. The comparison image acquisition unit 102 acquires a comparison image to be compared with the authentication target image. The comparison image may be registered in the authentication device 100 in advance.

[0016] Here, it is assumed that the authentication target image has a lower resolution than the comparison image. For example, the authentication target image may be an image of the eye part of the whole face image of the person to be authenticated. On the other hand, the comparison image may be an image obtained by photographing only the eye part for iris authentication. In this way, if iris authentication can be performed using the image of the eye part of the whole face image, it becomes possible to perform both face authentication and iris authentication to improve authentication accuracy.

[0017] The image enlargement unit 103 enlarges the authentication target image, which is a low-resolution image. The image enlargement unit 103 corresponds to an example of an image enlargement means. The enlarged authentication target image is also called an enlarged image of the authentication target image, or simply an enlarged image. The feature extraction unit 104 extracts the feature of the authentication target image using the same method as for extracting the feature of the comparison image.

[0018] The feature extraction unit 104 extracts the feature of the enlarged authentication target image and the feature of the comparison image. The feature extraction unit 104 applies the same algorithm to the enlarged authentication target image and the comparison image to extract the feature of each. The authentication device 100 may include two identical feature extraction units 104. Alternatively, one feature extraction unit 104 may extract the feature of the enlarged authentication target image and the feature of the comparison image by, for example, time-division processing. In addition, the feature of the comparison image may be registered in the authentication device 100 in advance, such as by the feature extraction unit 104 extracting the feature of the comparison image in advance. The feature extraction unit 104 corresponds to an example of a feature extraction means.

[0019] The feature amount conversion unit 105 converts the feature amount of the enlarged authentication target image. The feature amount conversion unit 105 converts the feature amount so that the difference between the feature amount of the enlarged authentication target image and the feature amount of the high resolution image becomes small. The feature amount conversion unit 105 corresponds to an example of a feature amount conversion means.

[0020] The conversion of features by the feature conversion unit 105 is also referred to as feature super-resolution. The feature after conversion by the feature conversion unit 105 is also referred to as a super-resolution feature. The feature conversion unit 105 learns how to convert features by learning using a loss function that becomes smaller as the difference between the features of the training image and the features of a degraded image obtained by reducing and then enlarging the training image becomes smaller.

[0021] The feature comparison unit 106 compares the feature converted by the feature conversion unit 105 with the feature of the comparison image to determine, for example, whether the person in the authentication target image and the person in the comparison image are the same person. The feature amount comparison unit 106 corresponds to an example of a feature amount comparison means.

[0022] FIG. 2 is a flowchart showing an example of a processing procedure for the authentication device 100 to perform iris authentication. In the process of FIG. 2, the authentication target image acquiring unit 101 acquires an authentication target image (step S111).

[0023] Next, the image enlargement unit 103 enlarges the authentication target image by interpolating pixels of the authentication target image so that the authentication target image, which is a low-resolution image, has the same resolution as the comparison image, which is a high-resolution image (step S112). The algorithm by which the image enlargement unit 103 enlarges the image is not limited to a specific algorithm. For example, the image enlargement unit 103 may enlarge the image using a bicubic method or a bilinear method, but is not limited to these. The image enlarged by the image enlargement unit 103 is also called an enlarged image.

[0024] Next, the feature extraction unit 104 extracts features of the enlarged image (step S113). The feature extraction unit 104 may be, but is not limited to, a feature extractor trained for authentication using high-resolution images. The feature extraction unit 104 may be configured using a trained model based on a deep neural network (DNN) such as VGG or ResNet (Residual Network). The feature extraction unit 104 may output the features in the form of a one-dimensional vector. Alternatively, the feature extraction unit 104 may output the features in the form of a two-dimensional or three-dimensional tensor that maintains image information.

[0025] Next, the feature amount conversion unit 105 converts the feature amount of the enlarged image (step S114). The feature amount conversion unit 105 converts the feature amount of the enlarged image so that it can be compared with the feature amount of the high-resolution image.

[0026] Moreover, the comparison image acquisition unit 102 acquires a comparison image (step S121). Then, the feature extraction unit 104 extracts the feature of the comparison image (step S122). As described above, the feature extraction unit 104 extracts the feature of the comparison image using the same algorithm as that used for extracting the feature of the enlarged image. Steps S111 to S114 and steps S121 to S122 may be executed in parallel. Alternatively, the processes from steps S121 to S122 may be performed in advance, and authentication device 100 may store the feature amounts of the comparison image.

[0027] After steps S114 and S122, the feature amount comparison unit 106 compares the feature amount of the enlarged image with the feature amount of the comparison image (step S131). For example, the feature amount comparison unit 106 calculates the similarity between the feature amount of the enlarged image and the feature amount of the comparison image, and if the similarity is equal to or greater than a predetermined threshold, determines that the person in the authentication target image and the person in the comparison image are the same person. The method of calculating the similarity used by the feature amount comparison unit 106 can be any of various methods capable of calculating the similarity of vectors or tensors, and is not limited to a specific method. For example, the feature amount comparison unit 106 may calculate the L2 distance or cosine similarity between the feature amount of the enlarged image and the feature amount of the comparison image. After step S131, authentication device 100 ends the process in FIG.

[0028] As described above, the authentication target image acquisition unit 101 acquires the authentication target image. The image enlargement unit 103 enlarges the authentication target image. The feature extraction unit 104 extracts the feature of the enlarged image of the authentication target image using the same method as the feature extraction method for the comparison image having a higher resolution than the authentication target image. The feature conversion unit 105 converts the feature of the enlarged image of the authentication target image based on learning using a loss function that becomes smaller as the difference between the feature of the learning image and the feature of the degraded image obtained by enlarging the learning image after reducing it becomes smaller. The feature comparison unit 106 compares the feature after conversion by the feature conversion unit 105 with the feature of the comparison image.

[0029] According to the authentication device 100, by enlarging a low-resolution image, extracting features, and generating super-resolution features, it is possible to obtain features that can be compared with the features of a comparison image, which is a high-resolution image, as features of the authentication target image, which is a low-resolution image. In this respect, according to the authentication device 100, authentication can be performed even if the input image is a low-resolution image.

[0030] In addition, it is expected that the feature amount of the enlarged authentication target image will approach the feature amount when the authentication target image is photographed at a higher resolution due to the feature amount conversion performed by the feature amount conversion unit 105. In this respect, it is expected that the authentication device 100 can perform authentication with high accuracy. Furthermore, the authentication device 100 can perform iris authentication using a low-resolution image. This allows the authentication device 100 to use a single face image to perform face authentication and iris authentication in combination, without requiring a high-resolution camera, and is expected to improve authentication accuracy.

[0031] <Second embodiment> Fig. 3 is a diagram showing an example of the configuration of a feature transformation learning device according to the second embodiment. In the configuration shown in Fig. 3, the feature transformation learning device 200 includes an image batch acquisition unit 201, an image reduction unit 202, an image enlargement unit 103, a feature extraction unit 104, a feature transformation unit 105, a loss function calculation unit 203, and a learning control unit 208. The loss function calculation unit 203 includes a reconstruction loss calculation unit 204, a feature discrimination unit 205, an adversarial loss calculation unit 206, and a similarity loss calculation unit 207. 3, parts having similar functions to those in FIG. 1 are given the same reference numerals (103, 104, 105).

[0032] The feature transformation learning device 200 trains the feature transformation unit 105. The feature transformation unit 105 after training can be used in the authentication device 100. A model having parameters to be adjusted by learning is referred to as a machine learning model. The parameters of the machine learning model are referred to as machine learning model parameters, and the parameter values ​​of the machine learning model are referred to as machine learning model parameter values. Adjusting the machine learning model parameter values ​​corresponds to learning. In the following, an example will be described in which the machine learning model of the feature transform unit 105 is configured using a neural network. However, the configuration of the machine learning model of the feature transform unit 105 is not limited to a specific one.

[0033] In the following, an example will be described in which the feature transformation learning device 200 uses a loss function for learning. In this case, the feature transformation learning device 200 performs learning so as to reduce the loss function value. However, the feature transformation learning device 200 may perform learning using an evaluation function whose function value increases as the evaluation becomes higher. In this case, the feature transformation learning device 200 performs learning so that the evaluation function value becomes larger.

[0034] The image batch acquisition unit 201 acquires a training data set used for learning by the feature conversion unit 105. The training data acquired by the image batch acquisition unit 201 includes a plurality of combinations of captured images of human eyes and class labels indicating correct classes in classifying the images. The training data set is also simply referred to as training data. Each combination of one image and one class label included in the training data set is also referred to as a labeled image. In the second embodiment, the images acquired by the image batch acquisition unit 201 are also referred to as high-resolution images. The image batch acquisition unit 201 corresponds to an example of an image acquisition means. The high-resolution image acquired by the image batch acquisition unit 201 corresponds to an example of a first image.

[0035] The image reducing section 202 reduces the high-resolution image. The pixels are thinned out by the image reduction performed by the image reducing section 202, resulting in a lower resolution than the high-resolution image. The image reduced by the image reducing section 202 is referred to as a reduced image. The image reducing section 202 may reduce the image by randomly determining the size or reduction ratio of the reduced image using a random number. The image reducing unit 202 is an example of an image reducing means. The reduced image is an example of a second image.

[0036] In the feature transformation learning device 200, the image enlargement unit 103 enlarges the image by the reduction ratio by which the image reduction unit 202 reduced the image. The image enlargement unit 103 interpolates pixels of the reduced image to enlarge the reduced image so that the image after enlarging the reduced image, which is a low-resolution image, has the same resolution as the high-resolution image. The image obtained by the image enlargement unit 103 enlarging the reduced image is also called a degraded image. As described above, the image enlarging unit 103 is an example of an image enlarging means. The degraded image is an example of a third image.

[0037] In the feature transformation learning device 200, the feature extraction unit 104 extracts the feature of the high-resolution image and the feature of the degraded image using the same algorithm. As described above, the feature extraction unit 104 is an example of a feature extraction means. In the feature transformation learning device 200, the feature transformation unit 105 transforms the feature of a degraded image. The feature transformation unit 105 transforms the feature of the degraded image so as to bring it closer to the feature of a high-resolution image. As in the first embodiment, the feature after transformation by the feature transformation unit 105 is also referred to as a super-resolution feature. As described above, the feature transformation unit 105 corresponds to an example of feature transformation means.

[0038] The loss function calculation unit 203 calculates a loss function for learning of the feature transformation unit 105. The loss function calculation unit 203 corresponds to an example of a loss function calculation means. The reconstruction loss calculation unit 204 calculates the reconstruction loss. The reconstruction loss is a loss whose value decreases as the similarity between the feature amount of the high-resolution image and the super-resolution feature amount increases as a vector or a tensor. The reconstruction loss, which the loss function calculation unit 203 calculates, corresponds to an example of an index value of the similarity between the feature amount of the high-resolution image and the super-resolution feature amount. The total loss function that uses the reconstruction loss as one of the variables corresponds to an example of a loss function whose loss decreases as the feature amount of the high-resolution image and the super-resolution feature amount become more similar.

[0039] The feature discrimination unit 205 discriminates between a feature of a high-resolution image and a super-resolution feature. Specifically, the feature discrimination unit 205 receives an input of a feature, and determines whether the input feature is a feature of a high-resolution image or a super-resolution feature. The feature discrimination unit 205 is also configured to be capable of learning. For example, the machine learning model of the feature discrimination unit 205 may be configured using a neural network. The feature determining unit 205 corresponds to an example of a feature determining means.

[0040] The adversarial loss calculation unit 206 calculates a loss for learning of the feature conversion unit 105 and a loss for learning of the feature discrimination unit 205 based on the discrimination result (judgment result) of the feature discrimination unit 205. The loss for learning of the feature conversion unit 105 is also referred to as a loss of the feature conversion unit 105. The loss for learning of the feature discrimination unit 205 is also referred to as a loss of the feature discrimination unit 205. The learning of the feature quantity conversion unit 105 and the learning of the feature discrimination unit 205 are alternately performed in one batch of learning under the control of the learning control unit 208.

[0041] In the learning of the feature conversion unit 105, the adversarial loss calculation unit 206 calculates the adversarial loss. The adversarial loss here is a loss whose value becomes small when the feature discrimination unit 205 erroneously determines that a super-resolution feature is a feature of a high-resolution image. The total loss function using the adversarial loss as one of the variables corresponds to an example of a loss function whose loss becomes small when the feature discrimination unit 205 cannot distinguish between a feature converted by the feature conversion unit 105 and other features. On the other hand, in the learning of the feature discriminator 205, the adversarial loss calculator 206 calculates a loss that becomes small when the feature discriminator 205 correctly discriminates between the super-resolution feature and the feature of the high-resolution image.

[0042] The similarity loss calculation unit 207 calculates the similarity loss. The similarity loss is a loss whose value is small when the class of the classification result in the classification using the super-resolution feature matches the correct class. The total loss function using the similarity loss as one of the variables corresponds to an example of a loss function whose loss is small when the class in the classification based on the super-resolution feature matches the correct class.

[0043] The learning control unit 208 updates the machine learning model parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205 using the loss calculated by the loss function calculation unit 203. The learning control unit 208 corresponds to an example of a learning control means. The learning control unit 208 adjusts the machine learning model parameter values ​​of the feature conversion unit 105 for conversion to features using the loss calculated by the loss function calculation unit 203. In this respect, the learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on the loss function. The update of the machine learning model parameter values ​​of the feature conversion unit 105 corresponds to the learning of the feature conversion unit 105. The update of the machine learning model parameter values ​​of the feature discrimination unit 205 corresponds to the learning of the feature discrimination unit 205.

[0044] FIG. 4 is a diagram showing an example of a processing procedure in which the feature transformation learning device 200 performs learning of the feature transformation unit 105. 4, the image batch acquisition unit 201 acquires a training data set including a plurality of combinations of high-resolution images, which are images of human eyes, and class labels, which indicate correct classes of the high-resolution images (step S211). As described above, each combination of one high-resolution image and one class label included in the training data set is also referred to as a labeled image. Next, the image batch acquisition unit 201 acquires batch data from the training data set (step S212). For example, the image batch acquisition unit 201 randomly selects a number of labeled images of a predetermined batch size from the labeled images included in the training data set. The batch data acquired by the image batch acquisition unit 201 is also referred to as image batch data. The batch size of the image batch data acquired by the image batch acquisition unit 201 is not limited to a specific size. For example, the image batch acquisition unit 201 may acquire a training data set including 128 labeled images, but is not limited thereto.

[0045] Next, the feature transformation learning device 200 starts a loop L11 for processing each labeled image included in the image batch data (step S213). The feature transformation learning device 200 may execute the processing of the loop L11 in parallel or sequentially.

[0046] In the processing of loop L11, the image reducing unit 202 reduces the high-resolution image of the labeled image (step S221). The image reducing unit 202 may randomly determine the size of the reduced image and reduce the image. The image reducing unit 202 reduces the image by thinning out the pixels of the image. Therefore, the resolution of the image decreases due to the reduction of the image by the image reducing unit 202. The reduced image corresponds to a low-resolution image.

[0047] Next, the image enlargement unit 103 enlarges the reduced image (step S222). The image obtained by the image enlargement unit 103 enlarging the reduced image is also called a degraded image. As described above, the image enlargement unit 103 enlarges the reduced image by interpolating pixels of the reduced image so that the degraded image has the same resolution as the high-resolution image.

[0048] Next, the feature extraction unit 104 extracts the feature of the degraded image (step S223). Next, the feature amount transforming unit 105 transforms the feature amount of the degraded image (step S224). As described above, the feature amount transformed by the feature amount transforming unit 105 is also referred to as a super-resolution feature amount.

[0049] Furthermore, the feature extraction unit 104 extracts features of a high-resolution image of the labeled image (step S231). Steps S221 to S224 and step S231 may be executed in parallel.

[0050] After steps S224 and S231, the feature transformation learning device 200 performs a termination process of the loop L11 (step S241). Specifically, the feature transformation learning device 200 waits for the completion of the process of the loop L11 for all labeled images included in the learning batch. When the feature transformation learning device 200 detects that the process of the loop L11 has been completed for all labeled images included in the learning batch, it ends the loop L11.

[0051] After completing loop L11, the feature transformation learning device 200 calculates a loss for learning of the feature transformation unit 105 (step S242). The feature transformation learning device 200 calculates one total loss function value based on the result of processing loop L11 for all labeled images included in the image batch data. 5 is a diagram showing an example of a processing procedure for calculating a loss by the feature transformation learning device 200. The feature transformation learning device 200 performs the processing of FIG. 5 in step S242 of FIG.

[0052] In the process of Fig. 5, the feature discrimination unit 205 outputs a vector for determining (discriminating) whether the input feature is a super-resolution feature or a feature of a high-resolution image (step S251). For example, the feature discrimination unit 205 may calculate the probability that the input feature is a feature of a high-resolution image, and output a vector including the probability as an element. This probability takes a real value between 0 and 1, and the type of the input image can be discriminated, for example, by indicating that the closer to 1 the more likely it is that the feature is a high-resolution image, and by indicating that the closer to 0 the more likely it is that the feature is a super-resolution feature.

[0053] Next, the adversarial loss calculation unit 206 calculates a loss using the vector output by the feature discrimination unit 205 (step S252). The feature conversion unit 105 and the feature discrimination unit 205 have a Generative Adversarial Network (GAN) structure, and the adversarial loss calculation unit 206 calculates the adversarial loss using the vector output by the feature discrimination unit 205. The adversarial loss here is a loss whose value becomes small when the discriminator makes an erroneous judgment.

[0054] In the learning of the feature conversion unit 105, the adversarial loss calculation unit 206 calculates the adversarial loss using a loss function that reduces the adversarial loss when the feature discrimination unit 205 erroneously determines that the super-resolution feature is a feature of a high-resolution image. The learning control unit 208 adjusts the machine learning model parameter value of the feature conversion unit 105 so that the adversarial loss calculated by the adversarial loss calculation unit 206 is reduced. In this way, the feature conversion unit 105 generates a super-resolution feature so as to deceive the feature discrimination unit 205. For example, the feature conversion unit 105 may be trained using a cross-entropy loss so that the adversarial loss calculation unit 206 outputs 1 when a super-resolution feature is input.

[0055] On the other hand, in the learning of the feature discrimination unit 205, which is performed alternately with the learning of the feature conversion unit 105, the adversarial loss calculation unit 206 calculates the loss using a loss function that reduces the loss when the feature discrimination unit 205 correctly discriminates between the super-resolution feature and the feature of the high-resolution image. The learning control unit 208 adjusts the machine learning model parameter value of the feature discrimination unit 205 so that the loss calculated by the adversarial loss calculation unit 206 is reduced. Therefore, the feature discrimination unit 205 is adjusted so that the feature discrimination unit 205 can correctly discriminate between the super-resolution feature and the feature of the high-resolution image.

[0056] For example, the feature discrimination unit 205 may output the discrimination result as a real number having a value range from 0 to 1. As described above, the feature discrimination unit 205 may output the probability that the input feature is a feature of a high-resolution image. Then, the adversarial loss calculation unit 206 may calculate the adversarial loss by binary cross entropy so that the adversarial loss calculation unit 206 outputs 1 when a feature of a high-resolution image is input, and outputs 0 when a super-resolution feature is input.

[0057] When "0" of the output of feature discrimination unit 205 represents a super-resolution feature and "1" represents a high-resolution feature, the learning control unit 208 adjusts the machine learning model parameter values ​​of the feature conversion unit 105 so that the output of feature discrimination unit 205 approaches 1 when the input to feature discrimination unit 205 is a super-resolution feature. In this way, the feature conversion unit 105 generates a super-resolution feature so as to deceive the feature discrimination unit 205.

[0058] On the other hand, the learning control unit 208 adjusts the machine learning model parameter values ​​of the feature discrimination unit 205 so that when the input to the feature discrimination unit 205 is a super-resolution feature, the output of the feature discrimination unit 205 approaches 0, and when the input to the feature discrimination unit 205 is a feature of a high-resolution image, the output of the feature discrimination unit 205 approaches 1. In other words, the feature discrimination unit 205 is adjusted so as to increase the possibility of correctly discriminating between the super-resolution feature and the feature of the high-resolution image.

[0059] Furthermore, the reconstruction loss calculation unit 204 calculates a reconstruction loss that decreases as the similarity between the super-resolution feature and the feature of the high-resolution image as a vector or tensor increases (step S261). The learning control unit 208 adjusts the machine learning model parameter values ​​of the feature conversion unit 105 so as to reduce the reconstruction loss. As a result, the learning control unit 208 adjusts the machine learning model parameter values ​​of the feature conversion unit 105 so as to bring the super-resolution feature closer to the feature of the high-resolution image.

[0060] The index value calculated by the reconstruction loss calculation unit 204, which indicates the similarity between the super-resolution feature and the feature of the high-resolution image as a vector or tensor, is not limited to a specific one. For example, the reconstruction loss calculation unit 204 may calculate the reconstruction loss such that the reconstruction loss decreases as the L2 distance between the super-resolution feature and the feature of the high-resolution image decreases. Alternatively, the reconstruction loss calculation unit 204 may calculate the reconstruction loss using the L1 distance instead of the L2 distance.

[0061] In addition, the similarity loss calculation unit 207 calculates the similarity loss using the class label. The similarity loss calculated by the similarity loss calculation unit 207 can be various losses whose value becomes small when the class classification result using the super-resolution feature matches the correct answer, and is not limited to a specific one.

[0062] For example, one linear layer capable of learning may be added after the output layer of the super-resolution features of the feature conversion unit 105, and the output of the linear layer may be one-hot vectors the same number as the number of classes. The learning control unit 208 causes the linear layer to learn so that the linear layer performs class estimation based on the super-resolution features. As a result, the one-hot vector output by the linear layer indicates the class of the estimation result.

[0063] Then, the similarity loss calculation unit 207 may input each element of the one-hot vector output by the linear layer to a softmax function, and calculate the similarity loss by cross-entropy using the class label.

[0064] However, the similarity loss used by the similarity loss calculation unit 207 is not limited to a specific one. For example, the similarity loss calculation unit 207 may use L2 Softmax Loss, Cosine Loss, ArcFace, CosFace, SphereFace, AdaCos, or the like as the similarity loss.

[0065] Alternatively, the similarity loss calculation unit 207 may use triplet loss, center loss, contrastive loss, or the like as the similarity loss without adding the one linear layer.

[0066] Steps S251 to S252, S261, and step S271 may be executed in parallel. After steps S252, S261, and S271, the loss function calculation unit 203 calculates a total loss function value based on the adversarial loss, the reconstruction loss, and the similarity loss (step S261). The loss function calculation unit 203 calculates the loss function value using a loss function in which the smaller the value of the adversarial loss, the smaller the value of the reconstruction loss, and the smaller the value of the similarity loss, the smaller the loss function value. For example, the loss function calculation unit 203 multiplies each loss by a coefficient and takes the sum to calculate the total loss function. The values ​​of these coefficients are not particularly limited.

[0067] However, the loss function calculation unit 203 may calculate the loss function value using only one or two of the adversarial loss, the reconstruction loss, and the similarity loss. Alternatively, the loss function calculation unit 203 may calculate the loss function value by another method without using any of the adversarial loss, the reconstruction loss, and the similarity loss. After step S261, the feature transformation learning device 200 ends the processing in FIG.

[0068] After the process of FIG. 5 is completed, the learning control unit 208 calculates the gradient of the parameters of the neural network of the feature conversion unit 105 using backpropagation (step S243). Then, the learning control unit 208 uses the calculated gradient to update the parameter values ​​of the feature conversion unit 105 (step S244). The update of the parameter values ​​in step S244 corresponds to the learning of the feature conversion unit 105.

[0069] For example, the learning control unit 208 optimizes the parameter values ​​so that the loss function value is minimized. Examples of the optimization method used by the learning control unit 208 include, but are not limited to, Stochastic Gradient Descent (SGD) and Adam.

[0070] Next, the feature transformation learning device 200 calculates a loss for learning of the feature discrimination unit 205 (step S245). As described above, in learning of the feature discrimination unit 205, the adversarial loss calculation unit 206 calculates the loss using a loss function that reduces the loss when the feature discrimination unit 205 correctly discriminates between the super-resolution feature and the feature of the high-resolution image.

[0071] Next, the learning control unit 208 calculates the gradient of the parameters of the neural network of the feature discriminator 205 using the error backpropagation method (step S246). Then, the learning control unit 208 uses the calculated gradient to update the parameter values ​​of the feature determining unit 205 (step S244). The update of the parameter values ​​in step S244 corresponds to the learning of the feature determining unit 205.

[0072] In this way, the learning control unit 208 alternately optimizes the parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205. When optimizing the parameter values ​​of the feature conversion unit 105, the learning control unit 208 fixes the parameter values ​​of the feature discrimination unit 205. When optimizing the parameter values ​​of the feature discrimination unit 205, the learning control unit 208 fixes the parameter values ​​of the feature conversion unit 105.

[0073] After step S244, the learning control unit 208 determines whether or not a learning end condition is satisfied (step S248). The learning end condition here is not limited to a specific condition. For example, the learning control unit 208 may determine that the end condition is satisfied when the number of times the loop from steps S212 to S248 is repeated reaches a predetermined number of times.

[0074] Alternatively, after updating the parameter values ​​in step S247, the feature transformation learning device 200 may have the learning control unit 208 or the loss function calculation unit 203 calculate the matching accuracy using super-resolution features based on evaluation data obtained in the middle of the loop. Then, the learning control unit 208 may determine that the termination condition is met when the matching accuracy using super-resolution features becomes equal to or greater than a predetermined accuracy.

[0075] In step S248, when the learning control unit 208 determines that the end condition is not satisfied (step S248: NO), the process returns to step S212. In this case, the feature transformation learning device 200 continues to train the feature transformation unit 105 and the feature discrimination unit 205. On the other hand, if the learning control unit 208 determines in step S248 that the end condition is met (step S248: YES), the feature transformation learning device 200 ends the processing in FIG.

[0076] As described above, the image batch acquisition unit 201 acquires a high-resolution image. The image reduction unit 202 reduces the high-resolution image to a reduced image having a lower resolution than the high-resolution image. The image enlargement unit 103 enlarges the reduced image to a degraded image having the same resolution as the high-resolution image. The feature extraction unit 104 extracts the feature of the high-resolution image and the feature of the degraded image. The feature conversion unit 105 converts the feature of the degraded image to generate a super-resolution feature. The learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a comparison result between the feature of the high-resolution image and the super-resolution feature. The learning of the feature conversion method here refers to adjusting the machine learning model parameter value that converts the feature.

[0077] In the feature transformation learning device 200, the feature transformation unit 105 can learn to make the feature transformation unit 105 bring the feature of an image obtained by enlarging a reduced image closer to the feature of a high-resolution image. It is expected that the authentication device 100 using the trained feature transformation unit 105 can perform authentication with a relatively high accuracy even when a low-resolution image is input.

[0078] Furthermore, the feature discrimination unit 205 receives an input of a feature of a high-resolution image or a super-resolution feature and determines whether the input feature is a super-resolution feature. The loss function calculation unit 203 causes the feature conversion unit 105 to learn a feature conversion method based on a loss function that reduces loss when the feature discrimination unit 205 cannot distinguish between a feature converted by the feature conversion unit 105 and other features.

[0079] According to the feature transformation learning device 200, it is expected that the feature transformation unit 105 outputs super-resolution features close to the features of the high-resolution image so that the feature discrimination unit 205 fails to discriminate between the features of the high-resolution image and the super-resolution features. As a result, it is expected that the authentication device 100 using the trained feature transformation unit 105 can perform authentication with a relatively high accuracy even when a low-resolution image is input.

[0080] Furthermore, the loss function calculation unit 203 calculates an index value of the similarity between the feature of the high-resolution image and the super-resolution feature. The learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a loss function in which the loss becomes smaller as the feature of the high-resolution image and the super-resolution feature become more similar.

[0081] According to the feature transformation learning device 200, it is expected that the feature transformation unit 105 will output super-resolution features similar to the features of a high-resolution image. As a result, it is expected that the authentication device 100 using the trained feature transformation unit 105 will be able to perform authentication with a relatively high degree of accuracy even when a low-resolution image is input.

[0082] In addition, the learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a loss function that reduces the loss when a class resulting from classification based on super-resolution features matches a correct class.

[0083] As a result, it is expected that the authentication device 100 using the trained feature transformation unit 105 will be able to perform authentication with a relatively high degree of accuracy even when a low-resolution image is input.

[0084] Moreover, the image reducing unit 202 reduces the high-resolution image to a size determined by using a random number to generate a reduced image. According to the feature transformation learning device 200, it is expected that the feature transformation unit 105 can output super-resolution features close to the features of high-resolution images for enlarged images enlarged from images of various resolutions. As a result, it is expected that the authentication device 100 using the trained feature transformation unit 105 can perform authentication with high accuracy for input images of various resolutions.

[0085] <Third embodiment> In the third embodiment, a configuration example of the feature extraction unit 104 will be described with reference to Fig. 1 and Fig. 3. It is preferable that the authentication device 100 in Fig. 1 and the feature transformation learning device 200 in Fig. 3 include the feature extraction unit 104 having the same configuration, but only one of the authentication device 100 and the feature transformation learning device 200 may include the feature extraction unit according to the third embodiment.

[0086] Fig. 6 is a diagram showing an example of the configuration of the feature extraction unit 104 according to the third embodiment. In the configuration shown in Fig. 6, the feature extraction unit 104 includes an attention processing unit 301. In the configuration shown in Fig. 6, the attention processing units 301 are connected in series. The first attention processing unit 301 receives input of image data, and the last attention processing unit 301 outputs features. The number of attention processing units 301 included in the feature extraction unit 104 may be one or more.

[0087] The attention here means emphasizing a part of an image or a feature corresponding thereto. The emphasized feature is more likely to be reflected in the authentication result. For example, in iris authentication, it is considered that authentication is performed based on the features of the iris part of the eye image. Therefore, the attention processing unit 301 may emphasize the feature corresponding to the iris part among the features extracted from the eye image by the feature extraction unit 104. In this way, the attention processing unit 301 emphasizes the feature corresponding to the characteristic part of the image, and it is expected that the authentication accuracy of the authentication device 100 will be improved. The attention processing unit 301 corresponds to an example of an attention processing means. As described above, the feature extraction unit 104 in FIG. 6 corresponds to an example of the feature extraction unit 104 in FIG. 1 and FIG.

[0088] In the authentication device 100 or the feature transformation learning device 200, the feature extraction unit 104 extracts features from each of the high-resolution image and the enlarged low-resolution image. It is preferable to train the feature extraction unit 104 in advance using high-resolution data.

[0089] 6 shows an example in which the feature extraction unit 104 is configured using a neural network, and each of the attention processing units 301 corresponds to a layer of the neural network. Attention processing may be performed in each layer of the neural network. Alternatively, attention processing may be performed only in some layers of the neural network. Alternatively, attention processing may be performed as pre-processing or post-processing of feature extraction.

[0090] Fig. 7 is a diagram showing a configuration example of the attention processing unit 301. In the configuration shown in Fig. 7, the attention processing unit 301 includes two-dimensional convolution calculation units 311a and 311b, a function calculation unit 312, and a multiplication unit 313. The two-dimensional convolution calculation units 311a and 311b are also collectively referred to as the two-dimensional convolution calculation unit 311.

[0091] The two-dimensional convolution calculation unit 311 performs a convolution calculation on data in the form of a two-dimensional matrix. The two-dimensional convolution calculation unit 311a performs a convolution calculation for extracting a feature amount. The two-dimensional convolution calculation unit 311b performs a convolution calculation as filtering for attention to the feature amount extracted by the two-dimensional convolution calculation unit 311a.

[0092] The function calculation unit 312 applies a predetermined function to the output of the two-dimensional convolution calculation unit 311b. For example, the function calculation unit 312 replaces the value of each element of the data in the form of a two-dimensional matrix indicating the calculation result of the two-dimensional convolution calculation unit 311b with a value to which a Softmax function is applied. As a result, the function calculation unit 312 calculates a weighting coefficient for emphasizing the part detected by the two-dimensional convolution calculation unit 311b from the output data of the two-dimensional convolution calculation unit 311a. The application of the function by the function calculation unit 312 also corresponds to the calculation of the activation function of a neural network.

[0093] The multiplication unit 313 multiplies the output of the two-dimensional convolution operation unit 311a by the output of the function calculation unit 312 for each element of the data in the form of a two-dimensional matrix. In this way, the multiplication unit 313 weights the feature amount extracted by the two-dimensional convolution operation unit 311a with the weighting coefficient calculated by the function calculation unit 312. The attention processing unit 301 may learn a method for detecting a target region of attention. Specifically, the attention processing unit 301 may learn a filter of the two-dimensional convolution calculation unit 311b so as to reduce the loss calculated by the learning control unit 208.

[0094] 8 is a diagram showing an example of the configuration of an attention block 302 according to a first modified example of the attention processing unit 301. The feature extraction unit 104 shown in FIG. 6 may be equipped with the attention block 302 instead of the attention processing unit 301. In the configuration shown in FIG. 8, the attention block 302 includes a channel separation unit 321, the attention processing unit 301, and a channel combination unit 322.

[0095] 8, multiple attention processing units 301 are arranged in parallel. Each of the attention processing units 301 receives data input from the channel separation unit 321 and outputs data to the channel combination unit 322. However, the number of attention processing units 301 included in the attention block 302 may be one or more.

[0096] The channel separation unit 321 acquires input data to the attention block 302 as input data to the channel separation unit 321, and outputs the input data or a part of the data to each of the attention processing units 301. The process performed by the channel separation unit 321 corresponds to a process of separating channels of a neural network into a plurality of blocks.

[0097] The method in which the channel separation unit 321 generates data for each attention processing unit 301 from the input data is not limited to a specific method. For example, the channel separation unit 321 may output the input data to each attention processing unit 301 as it is. Alternatively, the channel separation unit 321 may divide the input data based on the position in the image, such as dividing the input data into data corresponding to the left, center, and right of the original image, and output each data to the attention processing unit 301. In addition, the channel separation unit 321 may divide the input data based on the position in the channel, such as dividing the input data into a plurality of channels from the original number of channels, and output each data to the attention processing unit 301.

[0098] Each of the attention processing units 301 extracts features from input data to the attention processing unit 301 itself, and emphasizes and outputs a part of the extracted features. The attention processing unit 301 in FIG. 7 can be used as the attention processing unit 301 in the attention block 302.

[0099] The channel combining unit 322 combines the data output by each of the attention processing units 301 into one data. The method of combining the data by the channel combining unit 322 can be various methods corresponding to the method by which the channel separating unit 321 generates data for each attention processing unit 301 from one data.

[0100] For example, when the channel separation unit 321 outputs the input data directly to each of the attention processing units 301, the channel combination unit 322 may combine the data in the form of a two-dimensional matrix output from each of the attention processing units 301 into one data by averaging or summing each element.

[0101] In addition, when the channel separation unit 321 divides the input data based on its position in the image, the channel combination unit 322 may combine the data from each of the attention processing units 301 into one data by arranging and combining the data according to its position in the image.

[0102] Here, in an image used for iris authentication, many features for identifying an individual are present, particularly in the iris and the area around the eyes. In this way, it is considered that features for identifying an individual are present in multiple parts of the image to be authenticated. Therefore, in the attention block 302, the channel of the neural network is separated into multiple blocks, attention processing is performed for each block, and the output of each block is combined. In this way, the attention block 302 can extract features by focusing on each part that has many features for identifying an individual, such as the iris, eyelids, and eyebrows.

[0103] The attention block 302 may emphasize some channels of the feature vectors instead of image regions. 9 is a diagram showing an example of the configuration of the attention processing unit 303 according to a second modified example of the attention processing unit 301. The attention block 302 shown in FIG.

[0104] In the configuration shown in FIG. 9, the attention processing unit 303 includes an image reducing unit 331, a linear operation unit 332, a function calculation unit 312, and a multiplication unit 313. Comparing the attention processing unit 303 with the attention processing unit 301, the two-dimensional convolution calculation unit 311b of the attention processing unit 301 is replaced with an image reduction unit 331 and a linear calculation unit 332 in the attention processing unit 303. That is, in the attention processing unit 303, the two-dimensional convolution calculation unit 311 is configured to be separated into an image reduction unit 331 and a linear calculation unit 332. In other respects, the attention processing unit 303 is similar to the attention processing unit 301.

[0105] When distinguishing between the two image reducing units 331, the image reducing unit 331 corresponding to the two-dimensional convolution calculation unit 311a is referred to as image reducing unit 331a, and the image reducing unit 331 corresponding to the two-dimensional convolution calculation unit 311b is referred to as image reducing unit 331b. When distinguishing between the two linear calculation units 332, the linear calculation unit 332 corresponding to the two-dimensional convolution calculation unit 311a is referred to as linear calculation unit 332a, and the linear calculation unit 332 corresponding to the two-dimensional convolution calculation unit 311b is referred to as linear calculation unit 332b.

[0106] The image reducing unit 331 reduces an input image to a size of 1x1. That is, the image reducing unit 331 reduces a tensor having a size of channel C, height H, and width W to a tensor having a size of channel C, height 1, and width 1. The method by which the image reducing unit 331 reduces an image is not limited to a specific method. For example, the image reducing unit 331 may reduce an image using an averaging method, but is not limited thereto.

[0107] The linear operation unit 332 performs linear operation on the channel. The input tensor and output tensor of the linear operation unit 332 each have a size of (C, 1, 1). The function calculation unit 312 applies a softmax function to the output of the linear calculation unit 332b to calculate weights for emphasizing the channels.

[0108] 10 is a diagram illustrating an example of the configuration of an attention processing unit 304 according to a third modified example of the attention processing unit 301. The feature extraction unit 104 illustrated in FIG. 10, the attention processing unit 304 includes a two-dimensional convolution calculation unit 311a, a multiplication unit 313, and an attention map acquisition unit 341. The two-dimensional convolution calculation unit 311a is also referred to as a two-dimensional convolution calculation unit 311.

[0109] The process performed by the two-dimensional convolution calculation unit 311a is similar to that of the two-dimensional convolution calculation unit 311a of the attention processing unit 301. The process performed by the multiplication unit 313 is similar to that of the multiplication unit 313 of the attention processing unit 301. The attention map acquisition unit 341 acquires an attention map. The attention map is data indicating a part of the input data to the attention processing unit 304 that should be emphasized by the attention processing unit 304. When the attention processing unit 304 uses the attention map, the attention map is, so to speak, correct answer data for attention, and is included in the training data.

[0110] In the attention processing unit 304, the multiplication unit 313 weights the output from the two-dimensional convolution operation unit 311a using an attention map. In addition to the feature extraction method in the two-dimensional convolution operation unit 311a, the attention processing unit 304 may also learn a method of highlighting the feature by attention. For example, the attention processing unit 304 may multiply the attention map input from the training data by a temperature coefficient S and then input the result to a softmax function, and the multiplication unit 313 may multiply the result of the convolution operation unit 311 by the obtained function value. In this case, the temperature coefficient S may be a parameter adjusted by learning.

[0111] As described above, the attention processing unit 301 performs weighting to emphasize a feature corresponding to a specific portion of an image or channel that is the subject of feature extraction. It is expected that the authentication device 100 will thereby be able to perform authentication with a relatively high degree of accuracy by using feature amounts in which parts containing many features for identifying an individual are emphasized.

[0112] <Fourth embodiment> Fig. 11 is a diagram showing a configuration example of a feature extraction unit 114 according to a modified example of the feature extraction unit 104. Both or either one of the authentication device 100 in Fig. 1 and the feature transformation learning device 200 in Fig. 3 may be provided with the feature extraction unit 114 instead of the feature extraction unit 104.

[0113] 11, the feature extraction unit 114 includes a first feature extraction unit 114a and a second feature extraction unit 114b. The output of the first feature extraction unit 114a is input to the feature conversion unit 105, and the output of the feature conversion unit 105 is input to the second feature extraction unit 114b. The first feature extraction unit 114a and the second feature extraction unit 114b are configured as partial networks obtained by dividing the neural network configuring the feature extraction unit 104 into two between layers.

[0114] FIG. 12 is a diagram showing an example of a hierarchical structure of a neural network that constitutes the feature extraction unit 104. As shown in FIG. In the example of Fig. 12, the feature extraction unit 104 is configured using a neural network having multiple layers. The layers of the neural network configuring the feature extraction unit 104 are grouped into a group of one or more layers from the front side and a group of one or more layers from the back side. The front side here refers to the input layer side, and the back side refers to the output layer side.

[0115] The first feature extraction unit 114a is configured using a neural network that corresponds to one or more layers from the front of these two groups, and the second feature extraction unit 114b is configured using a neural network that corresponds to one or more layers from the back of these two groups.

[0116] The output of the first feature amount extraction unit 114a corresponds to the intermediate feature amount in the feature amount extraction unit 104. The output of the first feature amount extraction unit 114a when a low-resolution image is input as image data to the first feature amount extraction unit 114a is also referred to as a low-resolution intermediate feature amount. 11, the feature conversion unit 105 receives low-resolution intermediate features and converts the input low-resolution intermediate features. The output of the feature conversion unit 105 in this case is also referred to as super-resolution intermediate features. The second feature extraction unit 114b receives the super-resolution intermediate feature and further extracts a feature. The output of the second feature extraction unit 114b corresponds to the super-resolution feature output by the feature conversion unit 105 in the configurations of Figs. 1 and 3.

[0117] In this way, the feature conversion unit 105 may perform feature super-resolution using features in intermediate layers, not limited to features output by the neural network constituting the feature extraction unit 104. For example, VGG16 is composed of 13 convolutional layers and three fully connected layers. Of these layers, the feature conversion unit 105 may perform feature super-resolution on intermediate features output by the second layer from the front of the convolutional layer, and the super-resolution intermediate features output by the feature conversion unit 105 may be input to the third layer from the front of the convolutional layer. Alternatively, the feature conversion unit 105 may perform feature super-resolution on intermediate features output by the fourth layer from the front of the convolutional layer, and the super-resolution intermediate features output by the feature conversion unit 105 may be input to the fifth layer from the front of the convolutional layer.

[0118] As described above, the feature transform unit 105 transforms the intermediate features generated by the feature extraction unit. This allows the authentication device 100 to obtain an effect between image super-resolution, which super-resolutions an input image to the feature extractor, and feature super-resolution, which super-resolutions an output feature. In other words, by not limiting the portion to be super-resolved to the input or output to the feature converter 105, feature super-resolution can be performed with intermediate features that can maximize authentication performance at a desired magnification ratio, in accordance with the desired magnification ratio. As a result, feature super-resolution can be performed with even higher accuracy.

[0119] <Fifth embodiment> Fig. 13 is a diagram showing an example of the configuration of an authentication device according to the fifth embodiment. In the configuration shown in Fig. 13, an authentication device 411 includes an authentication target image acquisition unit 101, a comparison image acquisition unit 102, a feature extraction unit 104, a feature conversion unit 105, and a feature comparison unit 106.

[0120] 13, parts having similar functions to those in FIG. 1 are given the same reference numerals (101, 12, 104, 105, 106) and will not be described in detail here. The authentication device 100 in Fig. 1 and the authentication device 411 in Fig. 13 use different images for authentication. The authentication device 100 acquires a low-resolution image as an authentication target image, and acquires a high-resolution image as a comparison image. In contrast, the authentication device 411 acquires an iris image when wearing glasses as an authentication target image, and acquires an iris image with the naked eye with the same resolution as the authentication target image as a comparison image. Note that the resolution of the authentication image and the comparison image need only be about the same, and do not need to be exactly the same.

[0121] Moreover, the authentication device 411 does not include the image enlargement unit 103 included in the authentication device 100. In the authentication device 411, the authentication target image acquisition unit 101 outputs the acquired authentication target image to the feature amount extraction unit 104. In other respects, authentication device 411 is similar to authentication device 100 .

[0122] In the authentication device 411 of the fifth embodiment, the feature extraction unit 104 extracts the feature of the authentication target image by the same method as the feature extraction method for the comparison image, which is an image of the iris of the naked eye. The feature extraction unit 104 may extract the feature of the comparison image at the time of authentication. Alternatively, the authentication device 100 may store the feature of the comparison image in advance. In the authentication device 411 of the fifth embodiment, the feature conversion unit 105 receives an input of the feature of the image of the iris when the glasses are worn, and converts the feature so that the difference from the feature of the image of the iris when the glasses are worn is small. The machine learning parameter value of the feature conversion unit 105 is adjusted by learning using a loss function whose value becomes smaller as the difference between the feature after conversion from the feature of the image of the iris when the glasses are worn and the feature of the image of the iris when the naked eye is small. In this case, the image of the iris when the glasses are worn corresponds to an example of the first learning image, and the image of the iris when the naked eye is corresponds to an example of the second learning image. A feature conversion learning device 412 according to the sixth embodiment performs this learning.

[0123] As described above, the authentication target image acquisition unit 101 acquires an image of the iris when wearing glasses as the authentication target image. The feature amount extraction unit 104 extracts the feature amount of the authentication target image by the same method as the feature amount extraction method for the comparison image, which is an image of the iris with the naked eye. The feature amount conversion unit 105 converts the feature amount of the authentication target image based on the learning result of the feature amount conversion method using a loss function that has a smaller value as the difference between the feature amount after conversion from the feature amount of the first learning image, which is an image of the iris with glasses, and the feature amount of the second learning image, which is an image of the iris with the naked eye, becomes smaller. The feature amount comparison unit 106 compares the feature amount after conversion by the feature amount conversion unit 105 with the feature amount of the comparison image.

[0124] It is expected that the authentication device 411 according to the fifth embodiment can bring the feature amount of an iris image when wearing glasses closer to the feature amount of an iris image when wearing naked eyes. In this respect, it is expected that the authentication device 411 according to the fifth embodiment can perform iris authentication with a relatively high accuracy even when the person to be authenticated is wearing glasses.

[0125] Sixth embodiment 14 is a diagram showing an example of the configuration of a feature transformation learning device according to the sixth embodiment. A feature transformation learning device 412 of the sixth embodiment performs learning of the feature transformation unit 105 of the authentication device 411 of the fifth embodiment.

[0126] 14, the feature transformation learning device 412 includes an image batch acquisition unit 413, a feature extraction unit 104, a feature transformation unit 105, a loss function calculation unit 203, and a learning control unit 208. The loss function calculation unit 203 includes a reconstruction loss calculation unit 204, a feature discrimination unit 205, an adversarial loss calculation unit 206, and a similarity loss calculation unit 207. 14, parts having similar functions to those in FIG. 3 are given the same reference numerals (104, 105, 203-208), and detailed description thereof will be omitted here.

[0127] The images contained in the training data of the feature transformation learning device 200 in FIG. 3 and the feature transformation learning device 412 in FIG. 14 are different. The image batch acquisition unit 201 of the feature transformation learning device 200 acquires training data including a plurality of combinations of high-resolution images and class labels indicating correct classes in classifying the images.

[0128] In response to this, the image batch acquisition unit 413 of the feature transformation learning device 412 acquires training data that includes multiple combinations of iris images taken with glasses and with the naked eye, and class labels indicating correct classes in classifying these images. Images of the same person and with the same resolution are used as the iris images taken with glasses and with the naked eye. The image batch acquisition unit 413 is an example of an image acquisition unit. The image of the naked eye iris included in the training data is an example of the first image, and the image of the iris when wearing glasses is an example of the second image.

[0129] Furthermore, the feature transformation learning device 412 does not include the image reducing unit 202 and the image enlarging unit 103 that the feature transformation learning device 200 includes. The image batch acquisition unit 201 outputs the image of the iris when the eyeglasses are worn and the image of the iris when the naked eye is worn to the feature amount extraction unit 104. The feature amount extraction unit 104 extracts the feature amount of the image of the iris when the eyeglasses are worn and the feature amount of the image of the iris when the naked eye is worn. The feature amount of the image of the iris without glasses corresponds to an example of the first feature amount, and the feature amount of the image of the iris with glasses corresponds to an example of the second feature amount.

[0130] The feature amount conversion unit 105 receives an input of the feature amount of the image of the iris when wearing glasses, and converts the feature amount so that the difference from the feature amount of the image of the iris when wearing the naked eye becomes small. The feature amount after conversion from the feature amount of the image of the iris when wearing glasses corresponds to an example of the third feature amount. The loss function calculation unit 203 calculates a loss based on the feature amount converted by the feature amount conversion unit 105 and the feature amount of the naked eye iris image.

[0131] The learning control unit 208 updates the machine learning model parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205 based on the loss calculated by the loss function calculation unit 203. The learning control unit 208 updates the machine learning model parameter values ​​so as to reduce the difference between the feature after conversion by the feature conversion unit 105 and the feature of the naked eye iris image.

[0132] As described above, the image batch acquisition unit 413 acquires a first image which is an image of an iris without glasses and a second image which is an image of an iris when wearing glasses. The feature extraction unit 104 extracts a first feature which is a feature of the first image and a second feature which is a feature of the second image. The feature conversion unit 105 converts the second feature into a third feature. The learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a comparison result between the first feature and the third feature.

[0133] According to the feature conversion learning device 412 of the sixth embodiment, the feature conversion unit 105 can be trained to convert the feature of an iris image when wearing eyeglasses into a feature of an iris image with the naked eye. By having the authentication device 411 perform authentication using this feature conversion unit 105, it is expected that iris authentication can be performed with a relatively high degree of accuracy even when the person to be authenticated is wearing eyeglasses.

[0134] Seventh embodiment Fig. 15 is a diagram showing an example of the configuration of an authentication device according to the seventh embodiment. In the configuration shown in Fig. 15, an authentication device 411 includes an authentication target image acquisition unit 101, a comparison image acquisition unit 102, a feature extraction unit 104, a feature conversion unit 105, and a feature comparison unit 106.

[0135] The configuration of the authentication device 411 according to the seventh embodiment is similar to the configuration of the authentication device 411 according to the fifth embodiment. Among the units in Fig. 15, those having similar functions to those in Fig. 13 are given the same reference numerals (101, 102, 104, 105, 106, 411) and will not be described in detail here.

[0136] In the seventh embodiment, the image used by the authentication device 411 for authentication is different from that in the fifth embodiment. The authentication device 411 in the fifth embodiment acquires an iris image when wearing glasses as an authentication target image, and acquires an iris image without glasses with the same resolution as the authentication target image as a comparison image. In contrast, the authentication device 411 in the seventh embodiment acquires an iris image photographed from an oblique angle as an authentication target image, and acquires an iris image photographed from the front with the same resolution as the authentication target image as a comparison image.

[0137] In the authentication device 411 of the seventh embodiment, the feature extraction unit 104 extracts the feature of the authentication target image by the same method as that for the comparison image, which is an iris image captured from the front. The feature extraction unit 104 may extract the feature of the comparison image at the time of authentication. Alternatively, the authentication device 100 may store the feature of the comparison image in advance. In the authentication device 411 of the seventh embodiment, the feature conversion unit 105 receives an input of the feature of the image of the iris photographed obliquely, and converts the feature so that the difference from the feature of the image of the iris photographed from the front becomes small. The machine learning parameter value of the feature conversion unit 105 is adjusted by learning using a loss function whose value becomes smaller as the difference between the feature after conversion from the feature of the image of the iris photographed obliquely and the feature of the image of the iris photographed from the front becomes smaller. In this case, the image of the iris photographed obliquely corresponds to an example of the first learning image, and the image of the iris photographed from the front corresponds to an example of the second learning image. A feature conversion learning device 412 according to the eighth embodiment performs this learning.

[0138] As described above, the authentication target image acquisition unit 101 acquires an image of an iris photographed obliquely as an authentication target image. The feature amount extraction unit 104 extracts the feature amount of the authentication target image by the same method as the feature amount extraction method for the comparison image, which is an image of an iris photographed from the front. The feature amount conversion unit 105 converts the feature amount of the authentication target image based on the learning result of the feature amount conversion method using a loss function that has a smaller value as the difference between the feature amount after conversion from the feature amount of the first learning image, which is an image of the iris photographed from the oblique direction, and the feature amount of the second learning image, which is an image of the iris photographed from the front, becomes smaller. The feature amount comparison unit 106 compares the feature amount after conversion by the feature amount conversion unit 105 with the feature amount of the comparison image.

[0139] It is expected that the authentication device 411 according to the seventh embodiment can bring the image feature amount of the iris photographed obliquely closer to the image feature amount of the iris photographed from the front. In this respect, it is expected that the authentication device 411 according to the seventh embodiment can perform iris authentication with a relatively high accuracy even when an image of the iris of the person to be authenticated photographed obliquely is used as the authentication target image.

[0140] Eighth embodiment 16 is a diagram showing an example of the configuration of a feature transformation learning device according to the eighth embodiment. A feature transformation learning device 412 of the eighth embodiment performs learning of the feature transformation unit 105 of the authentication device 411 of the seventh embodiment.

[0141] 16 , the feature transformation learning device 412 includes an image batch acquisition unit 413, a feature extraction unit 104, a feature transformation unit 105, a loss function calculation unit 203, and a learning control unit 208. The loss function calculation unit 203 includes a reconstruction loss calculation unit 204, a feature discrimination unit 205, an adversarial loss calculation unit 206, and a similarity loss calculation unit 207.

[0142] The configuration of the feature transformation learning device 412 according to the eighth embodiment is similar to the configuration of the feature transformation learning device 412 according to the sixth embodiment. Among the units in Fig. 16, the units having the same functions as those in Fig. 14 are given the same reference numerals (104, 105, 203-208, 312, 412, 413), and detailed description thereof will be omitted here.

[0143] In the eighth embodiment, images included in training data acquired by the feature transformation learning device 412 are different from those in the sixth embodiment. In the sixth embodiment, an image batch acquisition unit 413 of a feature transformation learning device 412 acquires training data including multiple combinations of iris images with eyeglasses and naked-eye iris images, and class labels indicating correct classes in classifying these images.

[0144] In contrast, in the eighth embodiment, an image batch acquisition unit 413 of a feature transformation learning device 412 acquires training data including a plurality of combinations of an iris image photographed from the front and an iris image photographed from an oblique angle, and a class label indicating a correct class in classifying the images. Images of the same person and with the same resolution are used as the iris image photographed from the front and the iris image photographed from an oblique angle. The image batch acquisition unit 413 is an example of an image acquisition unit. The image of the iris captured from the front included in the training data is an example of the first image, and the image of the iris captured from an oblique angle is an example of the second image.

[0145] In the eighth embodiment, the image batch acquisition unit 201 outputs an image of the iris photographed obliquely and an image of the iris photographed from the front to the feature amount extraction unit 104. The feature amount extraction unit 104 extracts feature amounts of the image of the iris photographed obliquely and the image of the iris photographed from the front. The feature amount of an image of an iris photographed from the front corresponds to an example of the first feature amount, and the feature amount of an image of an iris photographed from an oblique angle corresponds to an example of the second feature amount.

[0146] The feature amount conversion unit 105 receives the feature amount of the image of the iris photographed obliquely, and converts the feature amount so as to reduce the difference with the feature amount of the image of the iris photographed from the front. The feature amount after conversion from the feature amount of the image of the iris photographed obliquely corresponds to an example of a third feature amount. The loss function calculation unit 203 calculates a loss based on the feature amount after conversion by the feature amount conversion unit 105 and the feature amount of the image of the iris photographed from the front.

[0147] The learning control unit 208 updates the machine learning model parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205 based on the loss calculated by the loss function calculation unit 203. The learning control unit 208 updates the machine learning model parameter values ​​so as to reduce the difference between the feature values ​​after conversion by the feature conversion unit 105 and the feature values ​​of an iris image captured from the front.

[0148] As described above, the image batch acquisition unit 413 acquires a first image which is an image of the iris photographed from the front and a second image which is an image of the iris photographed from an oblique angle. The feature amount extraction unit 104 extracts a first feature amount which is a feature amount of the first image and a second feature amount which is a feature amount of the second image. The feature amount conversion unit 105 converts the second feature amount into a third feature amount. The learning control unit 208 causes the feature amount conversion unit 105 to learn a feature amount conversion method based on a comparison result between the first feature amount and the third feature amount.

[0149] According to the feature conversion learning device 412 of the eighth embodiment, the feature conversion unit 105 can be trained to convert the feature of an iris image photographed obliquely so as to approximate the feature of an iris image photographed from the front. By having the authentication device 411 perform authentication using this feature conversion unit 105, it is expected that iris authentication can be performed with relatively high accuracy even when an iris image photographed obliquely is used as the authentication target image.

[0150] <Ninth embodiment> Fig. 17 is a diagram showing an example of the configuration of an authentication device according to the 9th embodiment. In the configuration shown in Fig. 17, an authentication device 411 includes an authentication target image acquisition unit 101, a comparison image acquisition unit 102, a feature extraction unit 104, a feature conversion unit 105, and a feature comparison unit 106.

[0151] The configuration of the authentication device 411 according to the 9th embodiment is similar to the configuration of the authentication device 411 according to the 5th embodiment. Among the units in Fig. 17, those having similar functions to those in Fig. 13 are given the same reference numerals (101, 102, 104, 105, 106, 411) and will not be described in detail here.

[0152] In the ninth embodiment, the image used by the authentication device 411 for authentication is different from that in the fifth embodiment. The authentication device 411 of the fifth embodiment acquires an image of the iris when wearing glasses as an authentication target image, and acquires an image of the iris without glasses as a comparison image with the same resolution as the authentication target image. In contrast, the authentication device 411 of the ninth embodiment acquires an image of the iris photographed with visible light as an authentication target image, and acquires an image of the iris photographed with light other than visible light, such as near-infrared light, with the same resolution as the authentication target image as a comparison image. Note that the wavelength of the image for authentication is not limited to visible light, and may be an image photographed with other wavelengths, such as ultraviolet light or near-infrared light. In addition, the number of channels representing the color of the input image is not particularly limited.

[0153] In the authentication device 411 of the ninth embodiment, the feature extraction unit 104 extracts the feature of the authentication target image by the same method as the feature extraction method for the comparison image, which is an iris image captured with light other than visible light. The feature extraction unit 104 may extract the feature of the comparison image at the time of authentication. Alternatively, the authentication device 100 may store the feature of the comparison image in advance. In the authentication device 411 of the ninth embodiment, the feature conversion unit 105 receives an input of the feature of an iris image captured with visible light, and converts the feature so that the difference between the feature of an iris image captured with light other than visible light is reduced. The machine learning parameter value of the feature conversion unit 105 is adjusted by learning using a loss function whose value decreases as the difference between the feature after conversion from the feature of the iris image captured with visible light and the feature of the iris image captured with light other than visible light decreases. In this case, the iris image captured with visible light corresponds to an example of the first learning image, and the iris image captured with light other than visible light corresponds to an example of the second learning image. A feature conversion learning device 412 according to the tenth embodiment performs this learning.

[0154] As described above, the authentication target image acquisition unit 101 acquires an iris image photographed with visible light as the authentication target image. The feature amount extraction unit 104 extracts the feature amount of the authentication target image by the same method as the feature amount extraction method for the comparison image, which is an iris image photographed with light other than visible light. The feature amount conversion unit 105 converts the feature amount of the authentication target image based on the learning result of the feature amount conversion method using a loss function that has a smaller value as the difference between the feature amount after conversion from the feature amount of the first learning image, which is an iris image photographed with visible light, and the feature amount of the second learning image, which is an iris image photographed with light other than visible light, becomes smaller. The feature amount comparison unit 106 compares the feature amount after conversion by the feature amount conversion unit 105 with the feature amount of the comparison image.

[0155] It is expected that the authentication device 411 according to the ninth embodiment can bring the image feature of the iris photographed with visible light closer to the image feature of the iris photographed with light other than visible light. In this respect, it is expected that the authentication device 411 according to the seventh embodiment can perform iris authentication with a relatively high accuracy using an image photographed with visible light, even if the comparison image is photographed with light other than visible light.

[0156] Tenth embodiment 18 is a diagram showing an example of the configuration of a feature transformation learning device according to the tenth embodiment. A feature transformation learning device 412 according to the tenth embodiment performs learning of the feature transformation unit 105 of an authentication device 411 according to the ninth embodiment. In the tenth embodiment, the wavelength of the image for authentication is not limited to visible light, and may be an image captured at other wavelengths such as ultraviolet light or near infrared light. Furthermore, the number of channels representing the color of the input image is not particularly limited.

[0157] 18, the feature transformation learning device 412 includes an image batch acquisition unit 413, a feature extraction unit 104, a feature transformation unit 105, a loss function calculation unit 203, and a learning control unit 208. The loss function calculation unit 203 includes a reconstruction loss calculation unit 204, a feature discrimination unit 205, an adversarial loss calculation unit 206, and a similarity loss calculation unit 207.

[0158] The configuration of the feature transformation learning device 412 according to the tenth embodiment is similar to the configuration of the feature transformation learning device 412 according to the sixth embodiment. Among the units in Fig. 18, those having similar functions to those in Fig. 14 are given the same reference numerals (104, 105, 203-208, 312, 412, 413) and will not be described in detail here.

[0159] In the tenth embodiment, images included in the training data acquired by the feature transformation learning device 412 are different from those in the sixth embodiment. In the sixth embodiment, an image batch acquisition unit 413 of a feature transformation learning device 412 acquires training data including multiple combinations of iris images with eyeglasses and naked-eye iris images, and class labels indicating correct classes in classifying these images.

[0160] In contrast, in the tenth embodiment, an image batch acquisition unit 413 of a feature transformation learning device 412 acquires training data including a plurality of combinations of an iris image captured with light other than visible light, an iris image captured with visible light, and a class label indicating a correct class in classifying these images. Images of the same person and with the same resolution are used as the iris image captured with light other than visible light and the iris image captured with visible light. The image batch acquisition unit 413 is an example of an image acquisition unit. The image of the iris captured with light other than visible light included in the training data is an example of the first image, and the image of the iris captured with visible light is an example of the second image.

[0161] In the tenth embodiment, the image batch acquisition unit 201 outputs an iris image captured with visible light and an iris image captured with light other than visible light to the feature amount extraction unit 104. The feature amount extraction unit 104 extracts feature amounts of the iris image captured with visible light and feature amounts of the iris image captured with light other than visible light. The feature amount of an iris image captured using light other than visible light is an example of the first feature amount, and the feature amount of an iris image captured using visible light is an example of the second feature amount.

[0162] The feature conversion unit 105 receives an input of the feature of the image of the iris captured with visible light, and converts the feature so as to reduce the difference with the feature of the image of the iris captured with light other than visible light. The feature after conversion from the feature of the image of the iris captured with visible light corresponds to an example of a third feature. The loss function calculation unit 203 calculates a loss based on the feature after conversion by the feature conversion unit 105 and the feature of the image of the iris captured with light other than visible light.

[0163] The learning control unit 208 updates the machine learning model parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205 based on the loss calculated by the loss function calculation unit 203. The learning control unit 208 updates the machine learning model parameter values ​​so as to reduce the difference between the feature after conversion by the feature conversion unit 105 and the feature of an iris image captured with light other than visible light.

[0164] As described above, the image batch acquisition unit 413 acquires a first image which is an image of an iris captured with light other than visible light, and a second image which is an image of an iris captured with visible light. The feature extraction unit 104 extracts a first feature which is a feature of the first image, and a second feature which is a feature of the second image. The feature conversion unit 105 converts the second feature into a third feature. The learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a comparison result between the first feature and the third feature.

[0165] According to the feature conversion learning device 412 of the tenth embodiment, the feature conversion unit 105 can be trained to convert the feature of an iris image captured with visible light into a feature of an iris image captured with light other than visible light. By having the authentication device 411 perform authentication using this feature conversion unit 105, it is expected that iris authentication can be performed with a relatively high degree of accuracy using an image captured with visible light even when a comparison image is captured with light other than visible light.

[0166] <Eleventh embodiment> Fig. 19 is a diagram showing an example of the configuration of an authentication device according to the 11th embodiment. In the configuration shown in Fig. 19, an authentication device 411 includes an authentication target image acquisition unit 101, a comparison image acquisition unit 102, a feature extraction unit 104, a feature conversion unit 105, and a feature comparison unit 106.

[0167] The configuration of the authentication device 411 according to the 11th embodiment is similar to the configuration of the authentication device 411 according to the fifth embodiment. Among the units in Fig. 19, those having similar functions to those in Fig. 13 are given the same reference numerals (101, 102, 104, 105, 106, 411), and detailed description thereof will be omitted here.

[0168] In the eleventh embodiment, the image used by the authentication device 411 for authentication is different from that in the fifth embodiment. The authentication device 411 in the fifth embodiment acquires an iris image when wearing glasses as an authentication target image, and acquires an iris image without glasses with the same resolution as the authentication target image as a comparison image. In contrast, the authentication device 411 in the eleventh embodiment acquires an iris image with a blurred image as an authentication target image, and acquires an iris image with a focused image with the same resolution as the authentication target image as a comparison image. When something is in focus, it is also called the in-focus position. When something is out of focus, it is also called the out-of-focus position.

[0169] In the authentication device 411 of the eleventh embodiment, the feature extraction unit 104 extracts the feature of the authentication target image by the same method as that for the comparison image, which is an iris image in focus. The feature extraction unit 104 may extract the feature of the comparison image at the time of authentication. Alternatively, the authentication device 100 may store the feature of the comparison image in advance. In the authentication device 411 of the ninth embodiment, the feature conversion unit 105 receives an input of a feature of an iris image that is out of focus, and converts the feature so that the difference between the feature of an iris image that is in focus is reduced. The machine learning parameter value of the feature conversion unit 105 is adjusted by learning using a loss function whose value decreases as the difference between the feature after conversion from the feature of the iris image that is out of focus and the feature of the iris image that is in focus is reduced. In this case, the iris image that is out of focus corresponds to an example of the first learning image, and the iris image that is in focus corresponds to an example of the second learning image. A feature conversion learning device 412 according to the twelfth embodiment performs this learning.

[0170] As described above, the authentication target image acquisition unit 101 acquires an image of an iris that is out of focus as the authentication target image. The feature amount extraction unit 104 extracts the feature amount of the authentication target image by the same method as the feature amount extraction method for the comparison image that is an image of an iris that is in focus. The feature amount conversion unit 105 converts the feature amount of the authentication target image based on the learning result of the feature amount conversion method using a loss function that has a smaller value as the difference between the feature amount after conversion from the feature amount of the first learning image that is an image of an iris that is out of focus and the feature amount of the second learning image that is an image of an iris that is in focus is smaller. The feature amount comparison unit 106 compares the feature amount after conversion by the feature amount conversion unit 105 with the feature amount of the comparison image.

[0171] It is expected that the authentication device 411 according to the eleventh embodiment can bring the feature amount of an out-of-focus iris image closer to the feature amount of an in-focus iris image. In this respect, it is expected that the authentication device 411 according to the eleventh embodiment can perform iris authentication with a relatively high degree of accuracy even when an out-of-focus iris image is used as an authentication target image.

[0172] <Twelfth embodiment> 20 is a diagram showing an example of the configuration of a feature transformation learning device according to the 12th embodiment. A feature transformation learning device 412 of the 12th embodiment performs learning of the feature transformation unit 105 of the authentication device 411 of the 11th embodiment.

[0173] 20, the feature transformation learning device 412 includes an image batch acquisition unit 413, a feature extraction unit 104, a feature transformation unit 105, a loss function calculation unit 203, and a learning control unit 208. The loss function calculation unit 203 includes a reconstruction loss calculation unit 204, a feature discrimination unit 205, an adversarial loss calculation unit 206, and a similarity loss calculation unit 207.

[0174] The configuration of the feature transformation learning device 412 according to the 12th embodiment is similar to the configuration of the feature transformation learning device 412 according to the 6th embodiment. Among the units in Fig. 20, those having similar functions to those in Fig. 14 are given the same reference numerals (104, 105, 203-208, 312, 412, 413), and detailed description thereof will be omitted here.

[0175] In the twelfth embodiment, images included in the training data acquired by the feature transformation learning device 412 are different from those in the sixth embodiment. In the sixth embodiment, an image batch acquisition unit 413 of a feature transformation learning device 412 acquires training data including multiple combinations of iris images with eyeglasses and naked-eye iris images, and class labels indicating correct classes in classifying these images.

[0176] In contrast, in the twelfth embodiment, the image batch acquisition unit 413 of the feature transformation learning device 412 acquires training data including a plurality of combinations of an out-of-focus iris image and an in-focus iris image, and a class label indicating a correct class in the classification of the images. Images of the same person and with the same resolution are used as the out-of-focus iris image and the in-focus iris image. The image batch acquisition unit 413 is an example of an image acquisition unit. An in-focus iris image in the image included in the training data is an example of a first image, and an out-of-focus iris image in the image is an example of a second image.

[0177] In the twelfth embodiment, the image batch acquisition unit 201 outputs an iris image in an out-of-focus image and an iris image in an in-focus image to the feature amount extraction unit 104. The feature amount extraction unit 104 extracts a feature amount of the iris image in an out-of-focus image and a feature amount of the iris image in an in-focus image. The feature amount of an iris image when the image is in focus corresponds to an example of the first feature amount, and the feature amount of an iris image when the image is out of focus corresponds to an example of the second feature amount.

[0178] The feature amount conversion unit 105 receives an input of the feature amount of the out-of-focus iris image and converts the feature amount so as to reduce the difference from the feature amount of the in-focus iris image. The feature amount after conversion from the feature amount of the out-of-focus iris image corresponds to an example of a third feature amount. The loss function calculation unit 203 calculates a loss based on the feature amount after conversion by the feature amount conversion unit 105 and the feature amount of the in-focus iris image.

[0179] The learning control unit 208 updates the machine learning model parameter values ​​of the feature conversion unit 105 and the feature discrimination unit 205 based on the loss calculated by the loss function calculation unit 203. The learning control unit 208 updates the machine learning model parameter values ​​so as to reduce the difference between the feature values ​​after conversion by the feature conversion unit 105 and the feature values ​​of an iris image in focus.

[0180] As described above, the image batch acquisition unit 413 acquires a first image which is an image of an iris in focus and a second image which is an image of an iris out of focus. The feature extraction unit 104 extracts a first feature which is a feature of the first image and a second feature which is a feature of the second image. The feature conversion unit 105 converts the second feature into a third feature. The learning control unit 208 causes the feature conversion unit 105 to learn a feature conversion method based on a comparison result between the first feature and the third feature.

[0181] According to the feature transformation learning device 412 of the twelfth embodiment, the feature transformation unit 105 can be trained to transform the feature of an out-of-focus iris image into a feature of an in-focus iris image. By using this feature transformation unit 105 to perform authentication, the authentication device 411 is expected to be able to perform iris authentication with a relatively high degree of accuracy even when an out-of-focus image is used as the authentication target image.

[0182] <Thirteenth embodiment> Fig. 21 is a diagram showing an example of the configuration of a feature transformation learning device according to the 13th embodiment. In the configuration shown in Fig. 21, a feature transformation learning device 610 includes an image acquisition unit 611, an image reduction unit 612, an image enlargement unit 613, a feature extraction unit 614, a feature transformation unit 615, and a learning control unit 616.

[0183] In this configuration, the image acquisition unit 611 acquires a first image. The image reduction unit 612 reduces the first image to a second image having a lower resolution than the first image. The image enlargement unit 613 enlarges the second image to a third image having the same resolution as the first image. The feature extraction unit 614 extracts a first feature which is a feature of the first image and a second feature which is a feature of the third image. The feature conversion unit 615 converts the second feature into a third feature. The learning control unit 616 causes the feature conversion unit 615 to learn a feature conversion method based on a comparison result between the first feature and the third feature.

[0184] The image acquisition unit 611 corresponds to an example of an image acquisition means. The image reduction unit 612 corresponds to an example of an image reduction means. The image enlargement unit 613 corresponds to an example of an image enlargement means. The feature extraction unit 614 corresponds to an example of a feature extraction means. The feature conversion unit 615 corresponds to an example of a feature conversion means. The learning control unit 616 corresponds to an example of a learning control means.

[0185] In the feature conversion learning device 610, the feature conversion unit 615 can make the feature of an image obtained by enlarging a reduced image closer to the feature of a high-resolution image by learning the feature conversion unit 615. It is expected that an authentication device using the trained feature conversion unit 615 can perform authentication with relatively high accuracy even when a low-resolution image is input.

[0186] <Fourteenth embodiment> Fig. 22 is a diagram showing an example of the configuration of an authentication device according to the 14th embodiment. In the configuration shown in Fig. 22, an authentication device 620 includes an authentication target image acquisition unit 621, an image enlargement unit 622, a feature extraction unit 623, a feature conversion unit 624, and a feature comparison unit 625.

[0187] In this configuration, the authentication target image acquisition unit 621 acquires the authentication target image. The image enlargement unit 622 enlarges the authentication target image. The feature extraction unit 623 extracts the feature of the enlarged image of the authentication target image using the same method as the feature extraction method for the comparison image having a higher resolution than the authentication target image. The feature conversion unit 624 converts the feature of the enlarged image of the authentication target image based on the learning result of the feature conversion method using a loss function whose value becomes smaller as the difference between the feature after conversion from the feature of the learning image and the feature of the degraded image obtained by reducing and then enlarging the learning image becomes smaller. The feature comparison unit 625 compares the feature after conversion by the feature conversion unit 624 with the feature of the comparison image.

[0188] The authentication target image acquiring unit 621 corresponds to an example of an authentication target image acquiring means. The image enlargement unit 622 corresponds to an example of an image enlargement means. The feature extraction unit 623 corresponds to an example of a feature extraction means. The feature conversion unit 624 corresponds to an example of a feature conversion means. The feature comparison unit 625 corresponds to an example of a feature comparison means.

[0189] According to the authentication device 620, by enlarging a low-resolution image to extract features and converting the extracted features, it is possible to obtain features that can be compared with the features of a comparison image, which is a high-resolution image, as features of the authentication target image, which is a low-resolution image. In this respect, according to the authentication device 620, authentication can be performed even if the input image is a low-resolution image.

[0190] In addition, it is expected that the feature amount of the enlarged authentication target image will approach the feature amount when the authentication target image is photographed at a higher resolution due to the feature amount conversion performed by the feature amount conversion unit 624. In this respect, it is expected that the authentication device 620 can perform authentication with high accuracy. Furthermore, the authentication device 620 can perform iris authentication using a low-resolution image. This allows the authentication device 620 to use a single face image to perform face authentication and iris authentication in combination, without requiring a high-resolution camera, and is expected to improve authentication accuracy.

[0191] <Fifteenth embodiment> Fig. 23 is a flowchart showing an example of a processing procedure in the feature transformation learning method according to the fifteenth embodiment. The feature transformation learning method shown in Fig. 23 includes acquiring an image (step S611), reducing the image (step S612), enlarging the image (step S613), extracting features (step S614), converting the features (step S615), and controlling learning (step S616).

[0192] In acquiring an image (step S611), a first image is acquired. In reducing the image (step S612), the first image is reduced to a second image having a lower resolution than the first image. In enlarging the image (step S613), the second image is enlarged to a third image having the same resolution as the first image. In extracting features (step S614), a first feature which is a feature of the first image and a second feature which is a feature of the third image are extracted. In converting features (step S615), the second feature is converted to a third feature. In controlling learning (step S616), learning is performed on a feature conversion method for converting the second feature to the third feature based on a comparison result between the first feature and the third feature.

[0193] According to the feature transformation learning method shown in Fig. 23, by learning the feature transformation method, the feature of an image obtained by enlarging a reduced image can be made to approach the feature of a high-resolution image by feature transformation. In authentication using the trained feature transformation method, it is expected that authentication can be performed with relatively high accuracy even when a low-resolution image is input.

[0194] <16th embodiment> Fig. 24 is a flowchart showing an example of a processing procedure in an authentication method according to the 16th embodiment. The authentication method shown in Fig. 24 includes obtaining an authentication target image (step S621), enlarging the image (step S622), extracting a feature amount (step S623), converting the feature amount (step S624), and comparing the feature amounts (step S625).

[0195] In obtaining an authentication target image (step S621), an authentication target image is obtained. In enlarging an image (step S622), the authentication target image is enlarged. In extracting features (step S623), features of an enlarged image of the authentication target image are extracted using the same method as the feature extraction method for a comparison image having a higher resolution than the authentication target image. In converting features (step S624), features of an enlarged image of the authentication target image are converted based on the learning result of a feature conversion method using a loss function whose value decreases as the difference between features after conversion from features of the learning image and features of a degraded image obtained by enlarging the learning image after reduction is decreased. In comparing features (step S625), features after conversion from features of the enlarged image of the authentication target image are compared with features of the comparison image.

[0196] According to the authentication method shown in Fig. 24, by enlarging a low-resolution image, extracting features, and converting the extracted features, it is possible to obtain features that can be compared with the features of a comparison image, which is a high-resolution image, as features of the authentication target image, which is a low-resolution image. In this respect, according to the authentication method shown in Fig. 24, authentication can be performed even if the input image is a low-resolution image.

[0197] In addition, it is expected that the feature amount of the enlarged authentication target image will be closer to the feature amount when the authentication target image is photographed at a higher resolution by the feature amount conversion. In this respect, it is expected that the authentication method shown in FIG. 24 can perform authentication with high accuracy. 24, iris authentication can be performed using a low-resolution image. This allows the authentication device 100 to use a single face image to perform face authentication and iris authentication in combination, without requiring a high-resolution camera, and is expected to improve authentication accuracy.

[0198] <Seventeenth embodiment> The configuration of the authentication device 620 according to the seventeenth embodiment is similar to the configuration of the authentication device 620 according to the fourteenth embodiment. The authentication device 620 according to the seventeenth embodiment further includes a feature discrimination unit that receives an input of a feature of a high-resolution image or a super-resolution feature and determines whether the input feature is a super-resolution feature. In this configuration, the feature discrimination unit causes the feature conversion unit 615 to learn a feature conversion method based on a loss function that reduces loss when the feature conversion unit 615 cannot distinguish between a feature converted by the feature conversion unit 615 and other features.

[0199] <Eighteenth embodiment> The configuration of the authentication device 620 according to the eighteenth embodiment is similar to the configuration of the authentication device 620 according to the fourteenth or seventeenth embodiment. In an authentication device 620 according to the eighteenth embodiment, a learning control unit 616 causes a feature conversion unit 615 to learn a feature conversion method based on a loss function in which the loss decreases as the features of a high-resolution image and super-resolution features become more similar.

[0200] <Nineteenth embodiment> The configuration of the authentication device 620 according to the nineteenth embodiment is similar to the configuration of the authentication device 620 according to the fourteenth, seventeenth or eighteenth embodiment. In an authentication device 620 according to the 19th embodiment, a learning control unit 616 causes a feature conversion unit 615 to learn a feature conversion method based on a loss function that reduces the loss when a class resulting from classification based on super-resolution features matches a correct class.

[0201] <Twentieth embodiment> The configuration of the authentication device 620 according to the twentieth embodiment is similar to the configuration of the authentication device 620 according to the fourteenth, seventeenth, eighteenth or nineteenth embodiment. In an authentication device 620 according to the twentieth embodiment, an image reducing section 612 reduces a high-resolution image to a size determined by using a random number to generate a reduced image.

[0202] FIG. 25 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. In the configuration shown in FIG. 25, a computer 700 includes a CPU (Central Processing Unit) 710, a main memory device 720, an auxiliary memory device 730, and an interface 740.

[0203] Any one or more of the authentication device 100, the feature transformation learning device 200, the authentication device 411, the feature transformation learning device 412, the feature transformation learning device 610, and the authentication device 620, or a part thereof, may be implemented in the computer 700. In this case, the operation of each of the above-mentioned processing units is stored in the auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, expands it in the main storage device 720, and executes the above-mentioned processing according to the program. The CPU 710 also secures a storage area corresponding to each of the above-mentioned storage units in the main storage device 720 according to the program. Communication between each device and other devices is executed by the interface 740 having a communication function and performing communication according to the control of the CPU 710.

[0204] When the authentication device 100 is implemented in a computer 700, the operations of the authentication target image acquisition unit 101, the comparison image acquisition unit 102, the image enlargement unit 103, the feature extraction unit 104, the feature conversion unit 105, and the feature comparison unit 106 are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing according to the program.

[0205] Furthermore, the CPU 710 secures a storage area for processing of the authentication device 100 in the main storage device 720 according to a program. Communication between the authentication device 100 and other devices is performed by an interface 740 having a communication function and performing communication according to the control of the CPU 710. Interaction between the authentication device 100 and a user is performed by the interface 740 having a display device and an input device, displaying various images according to the control of the CPU 710, and accepting user operations.

[0206] When the feature transformation learning device 200 is implemented in a computer 700, the image batch acquisition unit 201, the image reduction unit 202, the image enlargement unit 103, the feature extraction unit 104, the feature transformation unit 105, the loss function calculation unit 203, the learning control unit 208, and the operations of each of these units are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing according to the program.

[0207] Furthermore, the CPU 710 allocates a storage area in the main storage device 720 for processing of the feature transformation learning device 200 in accordance with the program. Communication between the feature transformation learning device 200 and other devices is performed by the interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the feature transformation learning device 200 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0208] When the authentication device 411 is implemented in a computer 700, the operations of the authentication target image acquisition unit 101, the comparison image acquisition unit 102, the feature extraction unit 104, the feature conversion unit 105, and the feature comparison unit 106 are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing according to the program.

[0209] Furthermore, the CPU 710, in accordance with the program, secures a storage area in the main storage device 720 for processing by the authentication device 411. Communication between the authentication device 411 and other devices is performed by an interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the authentication device 411 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0210] When the feature transformation learning device 412 is implemented in a computer 700, the image batch acquisition unit 413, the feature extraction unit 104, the feature transformation unit 105, the loss function calculation unit 203, the learning control unit 208, and the operations of each of these units are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing according to the program.

[0211] Furthermore, the CPU 710, in accordance with the program, secures a storage area in the main storage device 720 for processing of the feature transformation learning device 412. Communication between the feature transformation learning device 412 and other devices is performed by an interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the feature transformation learning device 412 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0212] When the feature conversion learning device 610 is implemented in a computer 700, the operations of the image acquisition unit 611, image reduction unit 612, image enlargement unit 613, feature extraction unit 614, feature conversion unit 615, and learning control unit 616 are stored in the form of programs in an auxiliary storage device 730. The CPU 710 reads the programs from the auxiliary storage device 730, loads them into the main storage device 720, and executes the above-mentioned processes according to the programs.

[0213] Furthermore, the CPU 710 allocates a storage area for processing of the feature transformation learning device 610 in the main storage device 720 according to the program. Communication between the feature transformation learning device 610 and other devices is performed by an interface 740 having a communication function and performing communication under the control of the CPU 710. Interaction between the feature transformation learning device 610 and a user is performed by the interface 740 having a display device and an input device, displaying various images under the control of the CPU 710, and accepting user operations.

[0214] When the authentication device 620 is implemented in the computer 700, the operations of the authentication target image acquisition unit 621, the image enlargement unit 622, the feature extraction unit 623, the feature conversion unit 624, and the feature comparison unit 625 are stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing according to the program.

[0215] Furthermore, the CPU 710 secures a storage area for processing of the authentication device 620 in the main storage device 720 according to the program. Communication between the authentication device 620 and other devices is performed by the interface 740 having a communication function and performing communication according to the control of the CPU 710. Interaction between the authentication device 620 and a user is performed by the interface 740 having a display device and an input device, displaying various images according to the control of the CPU 710, and accepting user operations.

[0216] Note that a program for executing all or part of the processing performed by authentication device 100, feature transformation learning device 200, authentication device 411, feature transformation learning device 412, feature transformation learning device 610, and authentication device 620 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing of each part. Note that the term "computer system" here includes hardware such as an OS (Operating System) and peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs (Read Only Memory), and CD-ROMs (Compact Disc Read Only Memory), as well as storage devices such as hard disks built into computer systems. The above-mentioned program may be for realizing part of the above-mentioned functions, or may be capable of realizing the above-mentioned functions in combination with a program already recorded in the computer system.

[0217] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and designs that do not deviate from the gist of the present invention are also included.

[0218] Some or all of the above embodiments may be described as follows, but are not limited to the following:

[0219] (Appendix 1) An image acquisition means for acquiring a first image; an image reducing means for reducing the first image to a second image having a lower resolution than the first image; an image enlarging means for enlarging the second image into a third image having the same resolution as the first image; a feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the third image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature; A feature transformation learning device comprising: (Appendix 2) a feature determining unit that receives an input of the first feature amount or the third feature amount and determines whether the input feature amount is a feature amount converted by a feature amount converting unit, the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function that reduces a loss when the feature discrimination means cannot distinguish between the feature transformed by the feature transformation means and other features; 2. A feature transformation learning device according to claim 1. (Appendix 3) the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function in which a loss becomes smaller as the first feature and the third feature become more similar; 3. The feature transformation learning device according to claim 1 or 2. (Appendix 4) the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function that reduces a loss when a class resulting from the class classification based on the third feature matches a correct class; 4. A feature transformation learning device according to any one of appendix 1 to 3. (Appendix 5) The image reducing means reduces the first image to a size determined by using a random number to generate the second image. 5. A feature transformation learning device according to any one of appendix 1 to 4. (Appendix 6) The feature extraction means Attention processing means for weighting the feature quantity corresponding to a specific part of the image from which the feature quantity is extracted to emphasize the feature quantity 6. The feature transformation learning device according to claim 1, further comprising: (Appendix 7) the feature conversion means converts the intermediate feature generated by the feature extraction means; 7. A feature transformation learning device according to any one of appendix 1 to 6. (Appendix 8) An authentication object image acquiring means for acquiring an authentication object image; an image enlarging means for enlarging the authentication target image; a feature extraction means for extracting a feature of an enlarged image of the authentication target image by the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; a feature conversion means for converting a feature of an enlarged image of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between a feature converted from the feature of a learning image and a feature of a degraded image obtained by enlarging the learning image after being reduced is smaller; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising: (Appendix 9) An image acquisition means for acquiring a first image which is an image of an iris without glasses and a second image which is an image of the iris when wearing glasses; A feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the second image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature; A feature transformation learning device comprising: (Appendix 10) an authentication target image acquisition means for acquiring an iris image when wearing glasses as an authentication target image; a feature extraction means for extracting features of the authentication target image in the same manner as a feature extraction method for a comparison image which is an image of an iris of a naked eye; a feature conversion means for converting the feature of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between the feature converted from the feature of a first learning image, which is an image of the iris when wearing glasses, and the feature of a second learning image, which is an image of the iris with the naked eye, decreases; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising: (Appendix 11) an image acquisition means for acquiring a first image, which is an image of an iris photographed from the front, and a second image, which is an image of the iris photographed from an oblique angle; A feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the second image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature; A feature transformation learning device comprising: (Appendix 12) an authentication target image acquiring means for acquiring an image of an iris photographed from an oblique angle as an authentication target image; a feature extraction means for extracting features of the authentication target image in the same manner as a feature extraction method for a comparison image, which is an image of an iris photographed from the front; a feature conversion means for converting the feature of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between the feature converted from the feature of a first learning image, which is an image of the iris photographed from an oblique angle, and the feature of a second learning image, which is an image of the iris photographed from the front, decreases; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising: (Appendix 13) an image acquisition means for acquiring a first image, which is an image of an iris photographed with light other than visible light, and a second image, which is an image of the iris photographed with visible light; A feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the second image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature; A feature transformation learning device comprising: (Appendix 14) an authentication target image acquisition means for acquiring an image of an iris captured with visible light as an authentication target image; a feature extraction means for extracting features of the authentication target image by the same method as a feature extraction method for a comparison image, which is an image of an iris photographed with light other than visible light; a feature conversion means for converting the feature of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between the feature after conversion from the feature of a first learning image, which is an image of an iris photographed with visible light, and the feature of a second learning image, which is an image of an iris photographed with light other than visible light, decreases; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising: (Appendix 15) an image acquisition means for acquiring a first image which is an image of the iris captured in a focused state and a second image which is an image of the iris captured in a defocused state; A feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the second image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature and the third feature; A feature transformation learning device comprising: (Appendix 16) an authentication target image acquisition means for acquiring an iris image captured in a defocused state as an authentication target image; a feature extraction means for extracting features of the authentication target image in the same manner as a feature extraction method for a comparison image, which is an iris image captured in a focused state; a feature conversion means for converting the feature of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between feature converted from feature of a first learning image, which is an image of the iris captured in a defocused state, and feature of a second learning image, which is an image of the iris captured in a focused state, decreases; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising: (Appendix 17) Obtaining a first image; Reducing the first image to a second image having a lower resolution than the first image; enlarging the second image into a third image having the same resolution as the first image; Extracting a first feature amount that is a feature amount of the first image and a second feature amount that is a feature amount of the third image; converting the second feature amount into a third feature amount; learning a feature transformation method for transforming the second feature into a third feature based on a comparison result between the first feature and the third feature; A feature transformation learning method including: (Appendix 18) Obtaining an image to be authenticated; Enlarging the authentication target image; extracting features of an enlarged image of the authentication target image using the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; converting the feature amount of the enlarged image of the authentication target image based on a learning result of a feature amount conversion method using a loss function whose value decreases as the difference between the feature amount converted from the feature amount of the learning image and the feature amount of a degraded image obtained by reducing and then enlarging the learning image decreases; comparing a feature amount obtained by converting the feature amount of the enlarged image of the authentication target image with a feature amount of the comparison image; Authentication methods, including: (Appendix 19) On the computer, Obtaining a first image; Reducing the first image to a second image having a lower resolution than the first image; enlarging the second image into a third image having the same resolution as the first image; Extracting a first feature amount that is a feature amount of the first image and a second feature amount that is a feature amount of the third image; converting the second feature amount into a third feature amount; learning a feature transformation method for transforming the second feature into a third feature based on a comparison result between the first feature and the third feature; A recording medium for recording a program for executing the program. (Appendix 20) On the computer, Obtaining an image to be authenticated; Enlarging the authentication target image; extracting features of an enlarged image of the authentication target image using the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; converting the feature amount of the enlarged image of the authentication target image based on a learning result of a feature amount conversion method using a loss function whose value decreases as the difference between the feature amount converted from the feature amount of the learning image and the feature amount of a degraded image obtained by reducing and then enlarging the learning image decreases; comparing a feature amount obtained by converting the feature amount of the enlarged image of the authentication target image with a feature amount of the comparison image; A recording medium for recording a program for executing the program. [Industrial Applicability]

[0220] The present disclosure may be applied to a feature transformation learning device, an authentication device, a feature transformation learning method, an authentication method, and a recording medium. [Explanation of symbols]

[0221] 100, 411, 620 Authentication device 101 Authentication target image acquisition unit 102 Comparison image acquisition section 103, 613, 622 Image enlargement section 104, 114, 614, 623 Feature extraction section 105, 615, 624 Feature conversion unit 106, 625 Feature Comparison Section 114a first feature extraction unit 114b second feature extraction unit 200, 412, 610 Feature transformation learning device 201, 413 Image batch acquisition section 202, 612 Image reduction section 203 Loss function calculation part 204 Reconstruction loss calculation section 205 Feature discrimination section 206 Enemy Loss Calculation Department 207 Similarity loss calculation part 208, 616 Learning control unit 301, 303, 304 Attention processing section 302 Attention Block 311, 311a, 311b 2D convolution calculation unit 312 Function Calculation Section 313 Multiplication section 321 Channel Separation Unit 322 Channel Junction 341 Attention Map Acquisition Unit 611 Image Acquisition Unit

Claims

1. An image acquisition means for acquiring a first image; an image reducing means for reducing the first image to a second image having a lower resolution than the first image; an image enlarging means for enlarging the second image into a third image having the same resolution as the first image; a feature extraction means for extracting a first feature which is a feature of the first image and a second feature which is a feature of the third image; a feature conversion means for converting the second feature into a third feature; a learning control means for causing the feature transformation means to learn a feature transformation method based on a comparison result between the first feature amount and the third feature amount; A feature transformation learning device comprising:

2. a feature determining unit that receives an input of the first feature amount or the third feature amount and determines whether the input feature amount is a feature amount converted by a feature amount converting unit, the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function that reduces a loss when the feature discrimination means cannot distinguish between the feature transformed by the feature transformation means and other features; The feature transformation learning device according to claim 1 .

3. the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function in which a loss becomes smaller as the first feature and the third feature become more similar; 3. The feature transformation learning device according to claim 1 or 2.

4. the learning control means causes the feature transformation means to learn a feature transformation method based on a loss function that reduces a loss when a class resulting from the class classification based on the third feature matches a correct class; The feature transformation learning device according to claim 1 .

5. The image reducing means reduces the first image to a size determined by using a random number to generate the second image. The feature transformation learning device according to claim 1 .

6. The feature extraction means Attention processing means for weighting the feature quantity corresponding to a specific part of the image from which the feature quantity is extracted to emphasize the feature quantity The feature transformation learning device according to claim 1 , further comprising:

7. the feature conversion means converts the intermediate feature generated by the feature extraction means; The feature transformation learning device according to claim 1 .

8. An authentication object image acquiring means for acquiring an authentication object image; an image enlarging means for enlarging the authentication target image; a feature extraction means for extracting a feature of an enlarged image of the authentication target image by the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; a feature conversion means for converting a feature of an enlarged image of the authentication target image based on a learning result of a feature conversion method using a loss function whose value decreases as the difference between a feature converted from the feature of a learning image and a feature of a degraded image obtained by enlarging the learning image after being reduced is smaller; a feature comparison means for comparing the feature converted by the feature conversion means with the feature of the comparison image; An authentication device comprising:

9. acquiring a first image; reducing the first image to a second image having a lower resolution than the first image; enlarging the second image into a third image having the same resolution as the first image; extracting a first feature amount that is a feature amount of the first image and a second feature amount that is a feature amount of the third image; converting the second feature amount into a third feature amount; learning a feature transformation method for transforming the second feature into a third feature based on a comparison result between the first feature and the third feature; A feature transformation learning method including:

10. Obtaining an image to be authenticated; Enlarging the authentication target image; extracting features of an enlarged image of the authentication target image using the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; converting the feature amount of the enlarged image of the authentication target image based on a learning result of a feature amount conversion method using a loss function whose value becomes smaller as the difference between the feature amount converted from the feature amount of the learning image and the feature amount of a degraded image obtained by reducing and then enlarging the learning image becomes smaller; comparing a feature amount obtained by converting a feature amount of an enlarged image of the authentication target image with a feature amount of the comparison image; Authentication methods, including:

11. On the computer, acquiring a first image; reducing the first image to a second image having a lower resolution than the first image; enlarging the second image into a third image having the same resolution as the first image; extracting a first feature amount that is a feature amount of the first image and a second feature amount that is a feature amount of the third image; converting the second feature amount into a third feature amount; learning a feature transformation method for transforming the second feature into a third feature based on a comparison result between the first feature and the third feature; A recording medium for recording a program for executing the program.

12. On the computer, Obtaining an image to be authenticated; Enlarging the authentication target image; extracting features of an enlarged image of the authentication target image using the same method as a feature extraction method for a comparison image having a higher resolution than the authentication target image; converting the feature amount of the enlarged image of the authentication target image based on a learning result of a feature amount conversion method using a loss function whose value becomes smaller as the difference between the feature amount converted from the feature amount of the learning image and the feature amount of a degraded image obtained by reducing and then enlarging the learning image becomes smaller; comparing a feature amount obtained by converting a feature amount of an enlarged image of the authentication target image with a feature amount of the comparison image; A recording medium for recording a program for executing the program.

Images