Discrimination model generation device, discrimination device, and their control method and program

Abstract

To improve the accuracy of personal authentication using an iris image even in an environment with outside light.SOLUTION: An identification model generation apparatus that generates an identification model for identifying a person from an image of an eyeball comprises: an acquisition section that acquires the image of the eyeball and data on image capturing status when the image of the eyeball is captured; an inference section that infers a feature amount of the eyeball on the basis of the image of the eyeball and the data on the image capturing status; and a learning section that performs learning in the inference section to approximate the feature amount of the eyeball obtained from the inference section to ground truth data that is a ground truth value of the feature amount of the eyeball corresponding to the image of the eyeball and the data on the image capturing status, and generates the identification model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for identifying an individual using an iris image. [Background technology]

[0002] In recent years, personal authentication technology using iris images has begun to be used for a variety of purposes, such as facility access control and personal authentication when logging in to PCs, smartphones, websites, etc.

[0003] In such a personal authentication device, for example, a specific user who will use the device is identified and registered, and then authentication for logging in to the provided service is performed based on this registered information. Also, the personal authentication device is primarily intended for use indoors with little external light.

[0004] However, in recent years, personal identification technology using iris recognition has begun to be used in devices that are often used outdoors, such as smartphones, and situations in which iris recognition is performed under ambient light are also increasing. When iris recognition is performed under ambient light that contains a large amount of near-infrared components, such as sunlight, near-infrared light emitted by the sun or near-infrared light reflected from objects illuminated by the sun or other sources is reflected over a wide area on the eyeball. As a result, when performing iris recognition under ambient light, the iris code calculated from an eyeball image captured under ambient light may not match the pre-registered iris code, making personal authentication impossible.

[0005] Patent Document 1 discloses a technology that performs machine learning using a large amount of data in which eyeball images are associated with scene images of the viewing direction, eliminating the need for calibration. This learning requires a large data set as training data. For example, gaze recognition is performed by combining eyeball images with facial feature points, camera angles, capture programs (data obtained when gazing at markers displayed on the top, bottom, left, and right sides), etc., as training data. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2020-140630 Summary of the Invention [Problem to be solved by the invention]

[0007] As mentioned above, in Patent Document 1, machine learning is performed to detect the gaze by combining data related to the gaze direction, such as scene images in a direction corresponding to the eyeball image and camera angles, as training data.

[0008] However, when external light is incident on the eyeball, it becomes difficult to recognize the direction of the eyeball, and the accuracy of gaze recognition drops significantly.

[0009] The present invention has been made in view of the above-mentioned problems, and its object is to improve the accuracy of personal authentication using iris images even in an environment with external light. [Means for solving the problem]

[0010] The discrimination model generating device of the present invention is a discrimination model generating device that generates a discrimination model for identifying individuals from an image of an eyeball, and is characterized by comprising: an acquisition means for acquiring an image of the eyeball and data on the shooting conditions when the image of the eyeball was captured; an inference means for inferring features of the eyeball based on the image of the eyeball and the data on the shooting conditions; and a learning means for training the inference means so that the eyeball features obtained from the inference means approach correct data, which is the correct value of the eyeball features corresponding to the image of the eyeball and the data on the shooting conditions, and generating the discrimination model. [Effects of the Invention]

[0011] According to the present invention, it is possible to improve the accuracy of personal authentication using an iris image even in an environment with external light. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram showing the configuration of a learning device according to an embodiment of the present invention. [Figure 2] FIG. 4 is a diagram showing the software configuration of a personal identification model unit. [Figure 3] FIG. 1 is a diagram showing external light incident on the eyeball. [Figure 4] 10 is a flowchart showing an eyeball image learning operation. [Figure 5] FIG. 1 is a diagram illustrating the configuration of an inference device according to an embodiment of the present invention. [Figure 6] FIG. 4 is a diagram showing the software configuration of a personal identification model unit. [Figure 7] 10 is a flowchart showing the process of personal registration inference of an eyeball image. [Figure 8] 10 is a flowchart showing the process of personal authentication inference of an eyeball image. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention claimed. Although multiple features are described in the embodiments, not all of these multiple features are necessarily essential to the invention, and multiple features may be combined arbitrarily. Furthermore, in the accompanying drawings, the same reference numerals are used to designate the same or similar components, and redundant explanations will be omitted.

[0014] In the description of the embodiments of the present invention, external light refers to light incident on the eyeball from sunlight or lighting equipment, and glare refers to a brightness pattern formed by external light in a captured eyeball image.

[0015] A personal identification system according to one embodiment of the present invention includes a learning device that generates a discrimination model for identifying a user from an image of the user's eyeball, calculates weight coefficients for a convolutional neural network (CNN) using the eyeball image as input, and performs training. The system also includes an inference device that uses the discrimination model trained by the learning device to identify individuals.

[0016] [Learning device configuration] FIG. 1 is a diagram showing the configuration of a learning device (personal identification model generating device) 100 in the personal identification system of this embodiment.

[0017] The learning device 100 includes an imaging unit 101 , a storage unit 102 , an image processing unit 103 , an individual identification model unit 104 , and a CPU 110 .

[0018] The imaging unit 101 is equipped with an eyepiece 10 for forming an image of the eyeball of the user 1 looking into the learning device 100, and the reflected light from the eyeball that passes through the eyepiece 10 is formed on an imaging element 11 in which photoelectric elements such as CMOS type elements are arranged two-dimensionally.

[0019] The memory unit 102 stores a machine learning program and also stores individual eyeball images captured by the imaging unit 101 or individual eyeball images captured by a device outside the personal identification system, such as a digital single-lens reflex camera or a mirrorless camera. The machine learning program is processed by the image processing unit 103, and is a program that calculates and learns weight coefficients for the CNN in the individual identification model unit 104 based on training data stored in the memory unit 102, and further updates the weight coefficients to create a highly accurate inference model. The machine learning program is executed by the CPU 110 to perform processing to create an inference model. The memory unit 102 also holds situational data when the eyeball image was captured, such as the time of capture, the location of capture, the direction the capturing eye was looking, and the season.

[0020] The image processing unit 103 detects brightness in the eyeball region in the eyeball image stored in the storage unit 102, for example, the luminance value of the image, and determines whether or not the region is an external light region using an external light evaluation value that associates brightness in the eyeball image with a position for each brightness. The image processing unit 103 also combines the eyeball image including external light with situation data at the time of photographing the eyeball, which is stored in the storage unit 102, to create training data for learning.

[0021] The individual identification model unit 104 is an inference device configured by CNN in which weighting coefficients are set based on the training data created by the image processing unit 103 and stored in the storage unit 102. The configuration and operation flow of the individual identification model unit 104 will be described in detail later.

[0022] Although the image processing unit 103 and the individual identification model unit 104 are shown as independent blocks in FIG. 1, they are actually realized by the CPU 110 executing a program stored in the memory unit 102.

[0023] [Configuration of personal identification model unit 104] Next, the configuration of the individual identification model unit 104 of the learning device 100 will be described with reference to FIG.

[0024] The training data acquisition unit 201 acquires an eyeball image 203 and shooting condition data 205 when the eyeball image was captured as training data TD from the storage unit 102. When the training data TD is input to the CNN 207, the machine learning program stored in the storage unit 102 performs learning by calculating a weighting coefficient 219 for the CNN 207 so that an output value 217 of the CNN 207 approaches a correct value (correct data) of the output value. The individual identification model unit 104 creates an inference model using the output weighting coefficient 219 of the CNN 207. Note that in FIG. 2, the output value 217 output from the CNN 207 is a feature amount of the iris image, and correct data, which is a correct value of the output value (feature amount) corresponding to the training data TD, is assumed to have been created in advance by a known method.

[0025] As described above, in this embodiment, the eyeball image 203 and data 205 on the situation in which the eyeball image was captured are combined to form training data TD, which is input to the CNN 207. For example, because the amount of external light entering the image differs depending on the height of the sun, data on the season when the image was captured, the time of capture, the location of the image, the direction of the image, and the like are stored in the storage unit 102 as the shooting situation data 205. In addition, the brightness value of each pixel in the image is acquired, and a brightness distribution is calculated for each range of brightness values. This can be combined with data on the location of the image, the direction of the image, or the like to improve the accuracy of the training data.

[0026] In addition, the brightness of the eyeball image is determined as a method for estimating the direction of external light coming from the sun. As shown in Figure 3, when sunlight, an example of external light, enters the eyeball, the eyeball becomes brighter on the side toward which the sun is coming in. Therefore, the brightness value of each pixel in the eyeball image is acquired, and the direction in the image that is bright is determined, thereby estimating the direction of the sun.

[0027] By combining this information with captured eye images containing external light and using them as training data, it becomes possible for the CNN 207 to learn the external light characteristics that differ for each user.

[0028] 2, in this embodiment, the person identification model to be trained is a CNN 207. The CNN 207 includes, for example, a convolution layer 209, a pooling layer 211, a fully connected layer 213, and an output layer 215.

[0029] The convolution layer 209 is an example of an input layer, and is a layer that performs convolution operations on an image. Image convolution corresponds to the process of calculating correlation with a certain filter. The pooling layer 211 is a layer that performs pooling processing. Pooling processing sets the maximum value among the output values ​​of the neurons in the convolution layer 209 that are connected, thereby achieving invariance in response to minute changes in the features that appear in the image.

[0030] The fully connected layer 213 is a layer that connects all neurons between adjacent layers. That is, each neuron included in the fully connected layer 213 is connected to all neurons included in the adjacent layer. Although the fully connected layer 213 is shown as one layer in FIG. 2, it may be composed of multiple layers. The output layer 215 is the layer located on the most output side of the CNN 207. A threshold is set for each neuron, and the output of each neuron is basically determined by whether or not the sum of the products of each input and each weight coefficient exceeds the threshold.

[0031] In the CNN 207 configured as above, when training data TD, which is a combination of an eyeball image 203 containing external light and situation data 205 at the time the eyeball image 203 was captured, is input to the convolutional layer 209 on the most input side, an output value (feature amount of the iris) corresponding to the eyeball image 203 is output from the output layer 215. The machine learning program held in the storage unit 102 updates the weighting coefficients 219 between the neurons so that the output value when the training data TD is input approaches the above-mentioned predetermined correct answer data, and stores the updated value in the storage unit 102. Learning is performed in this manner.

[0032] The individual identification model unit 104 creates an inference model for determining individual identification using information indicating the configuration of the CNN 207 obtained by the learning process, the weighting coefficients between the neurons, and the threshold values ​​of the neurons.

[0033] [Explanation of the learning flowchart] FIG. 4 is a flowchart showing an example of the processing procedure of the individual identification model unit 104. The operation of the individual identification model unit 104 will be described using FIG. 4. Note that the processing procedure related to machine learning by the learning device 100 described below is an example, and each process may be changed as much as possible. Furthermore, steps may be omitted, replaced, or added as appropriate, as necessary. Furthermore, the processing of FIG. 4 is realized by the CPU 110 of the learning device 100 executing a program stored in the storage unit 102.

[0034] In step S401, the teacher data acquisition unit 201 (CPU 110) acquires the teacher data TD for learning stored in the storage unit .

[0035] In step S402, the CPU 110 uses the acquired training data TD as an input to the CNN 207, calculates the weighting coefficient 219 so that the output value 217 of the CNN 207 approaches the correct data, and executes machine learning of the CNN 207.

[0036] As a specific example, a CNN 207 is prepared as a target for the learning process. The configuration of the prepared CNN 207 and the weighting coefficients 219 between each neuron may be acquired using a template or may be provided by operator input. Furthermore, when re-learning is performed, the individual identification model unit 104 (CPU 110) may prepare the CNN 207 based on the weighting coefficients 219 of the CNN that is the target for re-learning.

[0037] In step S403, CPU 110 determines whether the machine learning in step S402 has been performed a predetermined number of times. If it has been performed the predetermined number of times, CPU 110 proceeds to step S404. If it has not been performed the predetermined number of times, step S402 is repeated. The predetermined number of times is adjusted so as not to cause overlearning and so that both the learning accuracy and the prediction accuracy are maximized.

[0038] In step 404, the CPU 110 updates the configuration of the constructed CNN 207 and the weighting coefficients 219 as information indicating the thresholds of each neuron, and stores them in the storage unit 102. This completes the learning process of the CNN 207.

[0039] As described above, in this embodiment, by performing learning using training data of an eyeball image including external light and shooting situation data when the paired eyeball image was captured, it is possible to determine external light characteristics depending on the season, shooting time, shooting location, and shooting direction. Therefore, it is possible to generate an individual identification model without reducing the accuracy of individual identification due to reflection of external light such as sunlight on the eyeball.

[0040] Furthermore, the shooting condition data 205 described in this embodiment may be any data that can determine the direction of incident external light on the eyeball and the intensity of the external light, and does not have to be the type of data described above as long as it is data that can be used to learn the external light characteristics relatively in combination with the eyeball image.

[0041] Next, the inference device of this embodiment will be described. The inference device calculates the feature amount of an individual's eyeball image using the CNN weighting coefficients for each eyeball image obtained from the individual identification model generated as described above, and compares the feature amount with the iris feature amount of registered individuals to determine whether or not the individual is a specific individual.

[0042] [Configuration of inference device] FIG. 5 is a diagram illustrating the configuration of the inference device.

[0043] The inference device 500 performs individual registration inference, which infers the features of an individual's eyeball image and registers the inference results of the features in the inference device 500. Alternatively, the inference device 500 performs individual identification inference, which compares the registered eyeball features with the features of the eyeball image of a specific person input into the inference device 500 and makes an individual identification determination.

[0044] The inference device 500 includes an imaging unit 501, a memory unit 502, an image processing unit 503, a personal identification model unit 504, an authentication unit 505, and a CPU 510. Here, the imaging unit 501, the memory unit 502, and the image processing unit 503 perform the same operations as the imaging unit 101, the memory unit 102, and the image processing unit 103 in the learning device 100, and therefore their explanations will be omitted.

[0045] Although the image processing unit 503, the personal identification model unit 504, and the authentication unit 505 are shown as independent blocks in FIG. 5, in reality, they are realized by the CPU 510 executing a program stored in the memory unit 502.

[0046] The individual identification model unit 504 calculates the features of the eyeball (iris) of a specific person using CNN weighting coefficients calculated from training data, which is a combination of eyeball images including external light and situation data when the eyeball images were captured by the learning device 100. In the case of individual registration inference, as with the learning device 100, an inference model is created by associating the eyeball image and the shooting situation data of the eyeball with features that combine the eyeball features of a specific person and the characteristics of external light, and the inference model is recorded (registered) in the memory unit 502. In addition, in the case of individual identification inference, an image of the eyeball of the person to be identified is first captured by the imaging unit 501, and the eyeball features of the specific person registered in the memory unit 502 by the individual registration inference are compared with the eyeball features inferred from the eyeball image of the person to be identified, and the comparison result is stored in the memory unit 502.

[0047] Based on the identification result (the above-mentioned comparison result) stored in memory unit 502 during personal identification inference, authentication unit 505 determines whether the person to be authenticated is a specific person registered in memory unit 502 and outputs the authentication result. The authentication result is notified to the person to be authenticated by inference device 500. Methods of notification include, for example, display on a display attached to the inference device (not shown), audio notification from a speaker, display on an external display connected to the inference device via a dedicated cable, audio notification from an external speaker, etc.

[0048] [Configuration of the personal identification model unit 504] Next, the configuration of the individual identification model unit 504 in this embodiment will be described with reference to FIG.

[0049] The inference image acquisition unit 601 acquires as input data an eyeball image 603 containing external light obtained by the image processing unit 503 and shooting condition data 605 when the eyeball image was captured. With regard to the shooting condition data 605, for example, because the way external light enters differs depending on the height of the sun, data on the season when the image was captured and the shooting time is stored in the storage unit 502. In addition, by storing data on the shooting location and shooting direction, the direction of external light incident from the sun can also be determined, and therefore this is combined with the captured eyeball image containing external light and used as input data.

[0050] The individual identification model unit 504 performs individual registration inference, which registers eyeball features in the inference device, using the CNN weighting coefficients 219 obtained from the learning device 100 described above, or compares the registered features with the features of the eyeball image of a specific person input to the inference device.

[0051] 6, in this embodiment, the inference model that outputs the feature quantities of the user's eyeball image is a CNN 607. The CNN 607 includes a convolutional layer 609, a pooling layer 611, a fully connected layer 613, and an output layer 615, and is configured in the same manner as the CNN 207 of the individual identification model unit 104 of the learning device 100. Furthermore, since each of the layers 609 to 615 is the same as each of the layers 209 to 215 in the CNN 207 of the individual identification model unit 104 of the learning device 100, a description thereof will be omitted.

[0052] [Explanation of the flowchart for personal registration inference] 7 is a flowchart showing the processing steps of individual registration inference by the individual identification model unit 504. The processing in FIG. 7 is realized by the CPU 510 of the inference device 500 executing a program stored in the storage unit 502.

[0053] In step S701, the CPU 510 acquires from the memory unit 502 the eyeball image 603 of the person whose eyeball features are to be registered, captured by the imaging unit 501, and the shooting condition data 605 when the eyeball image was captured.

[0054] In step S702, the CPU 510 inputs to the CNN 607 the eyeball image 603 acquired in step S701 and the shooting condition data 605 when the eyeball image was shot.

[0055] In step S703, the CPU 510 executes inference calculation processing by the CNN 607 using the eyeball image 603 input in step S702 and the shooting condition data 605 when the eyeball image was shot.

[0056] In step S704, CPU 510 applies CNN weighting coefficients 219 obtained from individual identification model unit 104 of learning device 100 to CNN 607, and calculates features corresponding to the eyeball image of the person for whom individual authentication data is to be registered, using CNN 607. The result of individual registration inference is obtained as eyeball information, which is information that associates the eyeball features obtained from CNN 607 with shooting situation data 605 of the eyeball image, for example.

[0057] In step S705, an inference model is created that summarizes the eyeball information obtained in step S704.

[0058] In step S706, the inference model is stored in the memory unit 502 and personal registration is performed.

[0059] As described above, when performing personal registration inference, the personal identification model unit 504 infers eyeball information from the CNN weighting coefficients 219 obtained in advance from the eyeball image including external light, the eyeball image 603 including external light of the person for whom personal registration is to be performed, and the shooting situation data 605 when the eyeball image was captured, and stores the inference model of the person to be registered in the storage unit 502. This completes the processing of personal registration inference.

[0060] [Explanation of the flowchart for individual identification inference] Fig. 8 is a flowchart showing the processing steps of the individual identification inference of the individual identification model unit 504. Note that the software configuration for the individual identification inference is the same as the software configuration for the individual registration inference, and therefore will be described with reference to Figs. 5 and 6. Note that the processing of Fig. 8 is realized by the CPU 510 of the inference device 500 executing a program stored in the memory unit 502.

[0061] In step S801, CPU 510 acquires eyeball image 603, including external light, of a person for whom personal authentication is to be performed, captured by imaging unit 501, and image capturing condition data 605 when the eyeball image was captured.

[0062] In step S802, the CPU 510 inputs to the CNN 607 the eyeball image 603 containing external light of the person to be authenticated, which was stored in step S801, and the shooting condition data 605 when the eyeball image was shot.

[0063] In step S803, the CPU 510 executes calculation processing by the CNN 607 using the eyeball image 603 input in step S802 and the shooting condition data 605 when the eyeball image was shot.

[0064] In step S804, CPU 510 applies CNN weighting coefficients 219 obtained from individual identification model unit 104 of learning device 100 to CNN 607, and calculates features corresponding to the eyeball image of the person to be identified using CNN 607. The result of individual identification inference is obtained as eyeball information, which is information that associates the eyeball features obtained from CNN 607 with shooting situation data 605 of the eyeball image, for example.

[0065] In step S805, the CPU 510 inputs the inference model obtained in step S705, which is the result of the personal registration inference, from the memory unit 502 to the authentication unit 505. The CPU 510 also inputs the eyeball information, which is the inference result obtained in step 804, to the authentication unit 505. The authentication unit 505 performs personal identification processing by comparing the inference model obtained in step S705 with the eyeball information obtained in step 804.

[0066] In step S806, authentication unit 505 (CPU 510) determines whether the person undergoing personal authentication is registered in inference device 500 based on the personal identification result obtained in step S805, and outputs the authentication result.

[0067] In step S807, the authentication result obtained in step S806 is notified to the person performing personal authentication by inference device 500. Methods of notification include, for example, display on a display attached to the inference device (not shown), audio notification from a speaker, display on an external display connected to the inference device via a dedicated cable, audio notification from an external speaker, etc.

[0068] As described above, when performing personal identification inference, the personal identification model unit 504 compares the inference model registered in the personal registration inference with the eyeball information obtained during identification, and if the identification result matches the inference model to a certain extent, it authenticates the person as a specific person and notifies the person performing the authentication. This completes the personal authentication inference process.

[0069] The shooting condition data 605 described in this embodiment may be any data that can determine the direction of incident external light on the eyeball and the intensity of the external light, and does not have to be the type of data described above as long as it is data that can be used to learn the external light characteristics relatively in combination with the eyeball image.

[0070] (Other embodiments) The present invention can also be realized by supplying a program that realizes one or more of the functions of the above-described embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program.The present invention can also be realized by a circuit (e.g., ASIC) that realizes one or more of the functions.

[0071] The invention is not limited to the above-described embodiments, and various changes and modifications can be made without departing from the spirit and scope of the invention. Accordingly, the following claims are appended to apprise the public of the scope of the invention. [Explanation of symbols]

[0072] 100: Learning device, 101: Imaging unit, 102: Storage unit, 103: Image processing unit, 104: Personal identification model unit, 110: CPU, 207: CNN, 219: Weighting coefficient

Claims

1. An identification model generation device that generates an identification model for identifying an individual from an image of an eyeball, an acquisition means for acquiring an image of the eyeball and data on a photographing situation when the image of the eyeball was photographed; an inference means for inferring a feature amount of the eyeball based on the image of the eyeball and data on the photographing situation; a learning means for learning the inference means so that the eyeball feature amount obtained from the inference means approaches correct data that is a correct value of the eyeball feature amount corresponding to the image of the eyeball and data on the shooting situation, and generating the discrimination model; 1. A discrimination model generating device comprising:

2. The discrimination model generating device according to claim 1 , wherein the eyeball feature amount is a feature amount of an iris of the eyeball.

3. 3. The discrimination model generating device according to claim 1, wherein the data on the photographing conditions includes any one of a photographing time, a photographing location, a direction in which the eyes are looking, and a season.

4. 4. The discrimination model generating device according to claim 1, wherein the inference means uses a convolutional neural network.

5. 5. The discrimination model generating device according to claim 4, wherein the learning means calculates weighting coefficients of a convolutional neural network based on the eyeball image and the data on the shooting conditions so that the eyeball feature amounts obtained from the inference means approach correct data that are correct values ​​of the eyeball feature amounts corresponding to the eyeball image and the data on the shooting conditions.

6. 6. The discriminative model generating device according to claim 1, wherein the acquisition means includes an imaging means for imaging the eyeball.

7. 7. The discriminative model generating device according to claim 1, further comprising: a detection unit for detecting brightness of the image of the eyeball.

8. 8. The discrimination model generating device according to claim 7, wherein the detecting means outputs an evaluation value that associates brightness in the image of the eyeball with a position for each brightness as the data of the photographing situation.

9. An identification device that identifies an individual from an image of an eyeball, imaging means for imaging the eyeball; an acquisition means for acquiring an image of the eyeball captured by the imaging means and data on a photographing situation when the image of the eyeball was captured; an inference means for inferring a feature amount of the eyeball based on the image of the eyeball and data on the photographing situation; a comparison means for comparing the eye feature amount of a specific person registered in advance with the eye feature amount inferred by the inference means; An identification device comprising:

10. The identification device according to claim 9 , wherein the feature amount of the eyeball is a feature amount of the iris of the eyeball.

11. 11. The identification device according to claim 9, wherein the data on the photographing conditions includes any one of the photographing time, the photographing location, the direction in which the eyeball is looking, and the season.

12. 12. The identification device according to claim 9, wherein the inference means uses a convolutional neural network.

13. 13. The identification device according to claim 9, further comprising a detection unit for detecting brightness of the image of the eyeball.

14. 14. The identification device according to claim 13, wherein the detection means outputs an evaluation value that associates brightness in the image of the eyeball with a position for each brightness as the data on the photographing situation.

15. A control method for a discrimination model generation device that generates a discrimination model for identifying an individual from an image of an eyeball, comprising: an acquisition step of acquiring, by the discriminative model generation device, an image of the eyeball and data on a photographing situation when the image of the eyeball was photographed; an inference step of inferring features of the eyeball based on the image of the eyeball and data on the shooting situation by the discriminative model generation device; a learning process of performing learning of the inference process by the discriminative model generation device so that the eyeball feature amount obtained in the inference process approaches correct data that is a correct value of the eyeball feature amount corresponding to the eyeball image and data on the shooting situation, and generating the discriminative model; 1. A control method for a discrimination model generation device, comprising:

16. A method for controlling an identification device that identifies an individual from an image of an eyeball, comprising: an imaging step of imaging an eyeball by the identification device; an acquisition step of acquiring, by the identification device, the image of the eyeball captured in the imaging step and data on the imaging situation when the image of the eyeball was captured; an inference step of inferring, by the identification device, feature amounts of the eyeball based on the image of the eyeball and data on the photographing situation; a comparison step of comparing, by the identification device, the eye feature amount of a specific person registered in advance with the eye feature amount inferred in the inference step; A method for controlling an identification device, comprising:

17. A program for causing a computer to function as each of the means of the discrimination model generating device according to any one of claims 1 to 7.

18. A program for causing a computer to function as each of the means of the identification device according to any one of claims 9 to 14.

Images