Information processing systems, information processing methods, and programs

Abstract

An information processing system (100) is provided with an extraction unit (112), a calculation unit (113), and an updating unit (114). The extraction unit (112) uses an extraction model that performs learning with training images including an iris as input, to extract iris information related to the iris. The calculation unit (113) uses resolution of an iris region included in the training images to calculate loss between the iris information and correct answer information. The updating unit (114) uses the calculated loss to update the value of a parameter of the extraction model.

Description

Technical Field

[0001] This disclosure relates to an information processing system , love an information processing method, and program related thereto.

Background Art

[0002] Techniques using a learning model have been variously proposed to extract iris feature amounts (features of the iris) used in iris authentication and the like (see, for example, Patent Documents 1 to 3). For example, Patent Document 2 describes using L2 softmax loss, cosine loss, ArcFace, CosFace, SphereFace, or AdaCos as a loss.

[0003] In addition, Non-Patent Documents 1 to 4 disclose CosFace, ArcFace, MagFace, and T-center, which are examples of loss functions.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Patent Document 3

Non-Patent Documents

[0005]

Non-Patent Document 1

[0006] [Non-Patent Document 2] Jiankang Deng, et al., "ArcFace: Additive Angular Margin Loss for Deep Face Recognition", [online], CVPR 2019, [searched on December 20, 2022], Internet <URL: https: / / openaccess.thecvf.com / content_CVPR_2019 / html / Deng_ArcFace_Additive_Angular_Margin_Loss_for_Deep_Face_Recognition_CVPR_2019_paper.html>

[0007] [Non-Patent Document 3] Qiang Meng, et al., "MagFace: A Universal Representation for Face Recognition and Quality Assessment", [online], 2021 IEEE Conference Publication, [searched on December 20, 2022], Internet <URL: https: / / ieeexplore.ieee.org / document / <9578764>

[0008] [Non-Patent Document 4] Yifeng Chen, et al., "T-Center: A Novel Feature Extraction Approach Towards Large-Scale Iris Recognition", [online], February 12, 2020, IEEE, [searched on December 20, 2022], Internet <URL: https: / / ieeexplore.ieee.org / document / <8995585> Summary of the Invention Problems to be Solved by the Invention

[0009] This disclosure aims to improve the technologies described in the above-mentioned prior art documents. Means for Solving the Problems

[0010] According to one aspect of the present invention, extraction means for extracting iris information related to the iris by using an extraction model that learns with a learning image including the iris as an input; calculation means for calculating a loss between the iris information and correct information by using the resolution of the iris region included in the learning image; update means for updating the value of the parameter of the extraction model by using the calculated loss, are provided 、 The calculation means is <​​​​​​​​​​​​​​​​​Update the values of the parameters of the extraction model using the calculated loss This includes, Calculating the aforementioned loss is Determine the parameter values ​​for the parameters used to calculate the aforementioned loss, This includes calculating the loss between the iris information and the ground truth information using the aforementioned parameter values. An information processing method is provided

[0015] According to one aspect of the present invention On a first computer group composed of one or more computers By using an extraction model that learns with a learning image including an iris as an input, extract iris information regarding the iris Using the resolution of the iris region included in the learning image, calculate the loss between the iris information and the correct answer information Execute updating the values of the parameters of the extraction model using the calculated loss 、 Calculating the aforementioned loss is Determine the parameter values ​​for the parameters used to calculate the aforementioned loss, This includes calculating the loss between the iris information and the correct information using the parameter values ​​mentioned above. Program is provided 。

Brief Description of the Drawings

[0017] [Figure 1] It is a diagram showing an overview of the information processing system according to Embodiment 1 [Figure 2] It is a diagram showing an overview of the learning device according to Embodiment 1 [Figure 3] It is a diagram showing an overview of the collation device according to Embodiment 1 [Figure 4] It is a flowchart showing an overview of the first example of information processing according to Embodiment 1 [Figure 5] It is a flowchart showing an overview of the second example of information processing according to Embodiment 1 [Figure 6] It is a diagram showing a configuration example of the information processing system according to Embodiment 1 [Figure 7] It is a diagram showing a functional configuration example of the learning device according to Embodiment 1 [Figure 8] This figure shows an example of a training image. [Figure 9] This figure shows an example of the functional configuration of the parameter calculation unit. [Figure 10] This figure shows an example of the physical configuration of the learning device 101 according to Embodiment 1. [Figure 11] This flowchart shows an example of the learning process according to Embodiment 1. [Figure 12] This flowchart shows an example of the parameter calculation process according to Embodiment 1. [Modes for carrying out the invention]

[0018] Hereinafter, one embodiment of the present invention will be described with reference to the drawings. In all drawings, similar components are denoted by the same reference numerals, and their descriptions are omitted where appropriate.

[0019] <Embodiment 1> (overview) Figure 1 is a diagram showing an overview of the information processing system 100 according to Embodiment 1. The information processing system 100 comprises an extraction unit 112, a calculation unit 113, and an update unit 114.

[0020] The extraction unit 112 extracts iris information related to the iris by using an extraction model that is trained on training images including the iris as input.

[0021] The calculation unit 113 calculates the loss between iris information and ground truth information using the resolution of the iris region included in the training image.

[0022] The update unit 114 updates the parameter values ​​of the extraction model using the calculated loss.

[0023] According to this information processing system 100, even if the resolution of the iris region in a target image including the iris is low, it becomes possible to obtain accurate information from the image of the iris region.

[0024] Figure 2 is a diagram showing an overview of the learning device 101 according to Embodiment 1. The learning device 101 comprises an extraction unit 112, a calculation unit 113, and an update unit 114.

[0025] The extraction unit 112 extracts iris information related to the iris by using an extraction model that is trained on training images including the iris as input.

[0026] The calculation unit 113 calculates the loss between iris information and ground truth information using the resolution of the iris region included in the training image.

[0027] The update unit 114 updates the parameter values ​​of the extraction model using the calculated loss.

[0028] This learning device 101 makes it possible to obtain accurate information from the iris region even when the resolution of the iris region in a target image including the iris is low.

[0029] Figure 3 shows an overview of the matching device 103 according to Embodiment 1. The matching device 103 comprises a matching extraction unit 131 and a matching unit 132.

[0030] The matching extraction unit 131 takes a target image including the target iris as input and extracts iris information about the target iris using an extraction model learned using the learning device 101 described above.

[0031] The matching unit 132 compares the target iris information with pre-registered data.

[0032] According to this matching device 103, even if the resolution of the iris region in the target image including the target iris is low, it becomes possible to obtain accurate information from the iris region.

[0033] Figure 4 is a flowchart illustrating the overview of the first example of information processing according to Embodiment 1.

[0034] The extraction unit 112 extracts iris information related to the iris by using an extraction model that learns from training images including the iris as input (step S102).

[0035] The calculation unit 113 calculates the loss between iris information and ground truth information using the resolution of the iris region included in the training image (step S103).

[0036] The update unit 114 updates the parameter values ​​of the extraction model using the calculated loss (step S104).

[0037] This information processing method makes it possible to obtain accurate information from the iris region even when the resolution of the iris region in the target image, including the iris, is low.

[0038] Figure 5 is a flowchart showing an overview of a second example of information processing according to Embodiment 1.

[0039] The matching extraction unit 131 takes a target image including the target iris as input and uses the extraction model learned by the learning device 101 by executing the information processing method described above to extract iris information about the target iris (step S201).

[0040] The matching unit 132 compares the target iris information with pre-registered registration data (step S202).

[0041] This information processing method makes it possible to obtain accurate information from the iris region even when the resolution of the iris region in the target image, including the iris, is low.

[0042] The following describes a detailed example of the information processing system 100 according to Embodiment 1.

[0043] (detail) Generally, when obtaining an image of a subject that includes the iris, such as through photography using a camera, the resolution of the iris region included in the image may be low. The resolution of the iris region is the size of the iris region included in the image. In other words, even if the resolution (number of pixels) of the image is constant, the resolution of its iris region can vary.

[0044] When training a learning model to obtain iris information (information about the iris) from such target images, a general loss function, as described in prior art literature, is often used. However, in learning models trained using such loss functions in a general manner, the accuracy of the iris information obtained from the target image may vary depending on the resolution of the iris region. In particular, the accuracy of iris information obtained from target images with low iris resolution is often lower than the accuracy of iris information obtained from target images with high iris resolution.

[0045] One example of the object of the present invention is, in view of the above-mentioned problems, to provide an information processing system, learning device, matching device, information processing method, recording medium, etc., that can obtain accurate information from an image of the iris region even when the resolution of the iris region in a target image including the iris is low.

[0046] (Example of the configuration of the information processing system 100 according to Embodiment 1) Figure 6 shows an example configuration of the information processing system 100 according to Embodiment 1. The information processing system 100 is a system for matching objects using target images and for training an extraction model. The extraction model is a machine learning model that takes target images as input and outputs information used for matching and other purposes.

[0047] The information processing system 100 includes a learning device 101, a shooting device 102, and a matching device 103.

[0048] The learning device 101, the imaging device 102, and the matching device 103 are connected to each other via a network NT, which is configured as wired, wireless, or a combination thereof, and they send and receive information from each other via the network NT.

[0049] The target image is an image that includes the iris of the subject. The subject is, for example, a person. However, the subject is not limited to a person; it may also be an animal such as a dog or a snake.

[0050] (Example of functional configuration of the learning device 101 according to Embodiment 1) Figure 7 shows an example of the functional configuration of the learning device 101 according to Embodiment 1. The learning device 101 is a device for learning an extraction model. Functionally, the learning device 101 includes, for example, a learning information acquisition unit 111, an extraction unit 112, a calculation unit 113, and an update unit 114.

[0051] (Example of the functions of the learning information acquisition unit 111) The learning information acquisition unit 111 acquires learning information. This learning information is used to train the extraction model. The learning information may be prepared in advance, for example.

[0052] The learning information includes, for example, training images, the resolution of the iris region, and the correct answer information.

[0053] A training image is a target image used to train an extraction model. That is, a training image includes the iris of the target (training target) used to train the extraction model. The overall size (number of pixels) of the training image is, for example, a predetermined value. Figure 8 shows an example of a training image. Note that while Figure 8 shows an example where the training image is rectangular, it may be changed to an appropriate shape.

[0054] The iris region is the area in the target image (including the training image) that shows the iris, that is, the sub-region of the target image that shows the iris. In Figure 8, the iris region is marked with a dot.

[0055] The resolution of the iris region is, as mentioned above, the size of the iris region. The size of the iris region is, for example, the number of pixels that represent the iris diameter. The iris diameter is the number of pixels of at least one dimension, such as the diameter D and radius R of the outer edge of the iris region (i.e., the outer edge of the iris in the target image (including the training image)) (see Figure 8).

[0056] The resolution of the iris region included in the training information is the same as the resolution of the iris region included in the training image.

[0057] The correct information includes the correct output values ​​from the extraction model, i.e., the correct information obtained from the target image using the extraction model.

[0058] The ground truth information may include, for example, information corresponding to at least a portion of the iris information output by the extraction model. The ground truth information may include, for example, at least one of the iris labels or classes included in the training images, images of the iris region, and the locations of feature regions.

[0059] Labels are information used to identify the class to which the irises contained in training images belong. Labels are represented using, for example, predetermined letters, symbols, or numbers.

[0060] A class might represent, for example, the iris of the left or right eye of a learning target. A class can be represented using a vector quantity, such as a one-hot vector.

[0061] A feature area is at least one distinctive location in the target image, including the iris. A feature area may include at least one feature point, such as the pupil center, outer corner of the eye, inner corner of the eye, the highest point of the upper eyelid, or the lowest point of the lower eyelid. Alternatively, a feature area may include not only feature points but also geometric information such as the radius of a circle or the length and width of a rectangle. Such feature areas may be predetermined or automatically determined using a learning model.

[0062] In this embodiment, the correct answer information will be explained using an example that includes a class.

[0063] (Example of the function of the extraction unit 112) Refer to Figure 7 again. The extraction unit 112 is configured to include an extraction model. The extraction model is a machine learning model for extracting iris information using a target image as input. In other words, the extraction unit 112 according to this embodiment extracts iris information using an extraction model that learns using training images as input. The extraction unit 112 may output the extracted iris information to the calculation unit 113.

[0064] Iris information is information about the iris. Iris information includes, for example, at least one of the following: class, label, iris feature, image of the iris region, and location of feature points. Iris feature is a feature extracted from the iris contained in the target image (including the training image). Iris feature is, for example, a vector quantity.

[0065] In this embodiment, an example is used in which iris information includes iris features and classes.

[0066] More specifically, the extraction model includes, for example, a feature extraction model and a classification model. Each of these models is, for example, a machine learning model constructed using a neural network, and outputs the extracted information. Corresponding to the configuration of such an extraction model, the extraction unit 112 functionally includes a feature extraction unit 112a and a classification unit 112b, as shown in Figure 7.

[0067] The feature extraction unit 112a is configured to include a feature extraction model. The feature extraction model is a machine learning model for extracting iris features from a target image. The feature extraction unit 112a uses the feature extraction model to extract iris features using a training image as input.

[0068] The classification unit 112b includes a classification model. The classification model is a machine learning model that uses iris features to extract classes corresponding to irises contained in a target image. The classification unit 112b uses the classification model to extract classes corresponding to irises contained in the training image, taking the iris features extracted from the training image by the feature extraction unit 112a as input. The classification model may, for example, extract labels instead of classes, or in addition to classes.

[0069] (Example of the functions of the calculation unit 113) The calculation unit 113 calculates the loss between iris information and ground truth information using the resolution of the iris region included in the training image.

[0070] For calculating the loss, ArcFace or CosFace are suitable loss functions. However, the loss function is not limited to ArcFace or CosFace; for example, L2 softmax loss, cosine loss, MagFace, T-center, SphereFace, AdaCos, etc., may also be used.

[0071] In detail, for example, the calculation unit 113 functionally includes a parameter calculation unit 113a and a loss calculation unit 113b, as shown in Figure 7.

[0072] The parameter calculation unit 113a determines the parameter values ​​related to the parameters for calculating the loss.

[0073] The parameter value includes, for example, at least one of the following (1) to (3). In this embodiment, an example in which the parameter value includes the following (1) to (3) will be used for explanation.

[0074] (1) Value of margin parameter (2) Value of the weight decay parameter (3) Size of iris features

[0075] Here, the margin parameter in (1) is a hyperparameter included in the loss function used to calculate the loss. Including the value of the margin parameter in the parameter values ​​calculated by the parameter calculation unit 113a is preferable, for example, when using ArcFace, CosFace, etc., as the loss function.

[0076] (2) The weight decay parameter is a hyperparameter of weight decay included in the loss function used to calculate the loss.

[0077] (3) The magnitude of the iris feature is, for example, the L2 norm of the feature vector representing the iris feature.

[0078] The Lp norm is the positive p-th root of the sum of the p-th powers of the absolute values ​​of each component in the feature vector. For example, the L1 norm is the sum of the absolute values ​​of each component in the feature vector. The L2 norm is the Euclidean norm.

[0079] The parameter calculation unit 113a has functions for calculating each of (1) to (3), for example. Figure 9 shows an example of the functional configuration of the parameter calculation unit 113a. Functionally, the parameter calculation unit 113a includes, for example, a margin calculation unit 113a_1, a weight decay calculation unit 113a_2, and a norm calculation unit 113a_3.

[0080] The margin calculation unit 113a_1 calculates the margin parameter using the resolution of the iris region. The weight attenuation calculation unit 113a_2 calculates the weight attenuation value using the resolution of the iris region. The norm calculation unit 113a_3 determines the magnitude of the iris feature.

[0081] Refer to Figure 7 again. The loss calculation unit 113b calculates the loss between the iris information and the ground truth information using the parameter values ​​obtained by the parameter calculation unit 113a.

[0082] (Function of update unit 114) The update unit 114 updates the parameter values ​​of the extraction model using the loss calculated by the calculation unit 113.

[0083] More specifically, as shown in Figure 7, the update unit 114 functionally includes a gradient calculation unit 114a and a parameter update unit 114b.

[0084] The gradient calculation unit 114a calculates the gradient with respect to the parameters of the extraction model using the loss calculated by the loss calculation unit 113b.

[0085] The parameter update unit 114b updates the parameter values ​​of the extraction model using the gradient calculated by the gradient calculation unit 114a.

[0086] (Example of functional configuration of the imaging device 102 according to Embodiment 1) The imaging device 102 is a camera or similar device that photographs the subject. The imaging device 102 generates an image of the subject by taking a photograph, for example, when it receives a shooting instruction.

[0087] The shooting instructions may be output to the shooting device 102 from, for example, a sensor (not shown) for detecting when the subject is located in a predetermined shooting area, or a detection device (not shown) for detecting a face from an image of the subject taken by another shooting device (not shown).

[0088] The target image only needs to include the iris of one of the subject's eyes; it can be a binocular image including both of the subject's eyes, a monocular image including either the left or right eye, a facial image including the subject's face, or a full-body image including the subject's entire body.

[0089] (Example of functional configuration of the matching device 103 according to Embodiment 1) As described above with reference to Figure 3, the matching device 103 includes a matching extraction unit 131 and a matching unit 132.

[0090] The matching extraction unit 131 is configured to include, for example, an extraction model learned using the learning device 101. The matching extraction unit 131 takes a target image containing the target iris as input and uses the learned extraction model to extract the iris information of the target from the target image. The matching extraction unit 131 may output the extracted iris information to the matching unit 132.

[0091] The matching unit 132 compares the iris information extracted by the matching extraction unit 131 with pre-registered registration data. The result of this matching can be used, for example, to authenticate whether or not the target is pre-registered. However, the use of the matching result is not limited to this.

[0092] Up to this point, we have mainly described an example of the functional configuration of the information processing system 100 according to Embodiment 1. From here, we will describe an example of the physical configuration of the information processing system 100 according to this embodiment.

[0093] (Physical configuration of information processing system 100) The information processing system 100 consists of a learning device 101, an imaging device 102, and a matching device 103, which are physically connected via the network NT. Each of the learning device 101, imaging device 102, and matching device 103 consists of, for example, a single, physically different device.

[0094] Furthermore, some or all of the learning device 101, the imaging device 102, and the matching device 103 may be physically composed of a single device. In addition, the learning device 101, the imaging device 102, and the matching device 103 may be composed of multiple different devices connected via an appropriate communication line, such as a network NT, for each of the one or more functions they possess.

[0095] The first computer group may consist of one or more devices equipped with the functions of the learning device 101, and in this embodiment, it consists of the learning device 101. The second computer group may consist of one or more devices equipped with the functions of the matching device 103, and in this embodiment, it consists of the matching device 103.

[0096] The learning device 101 and the matching device 103 should each be physically configured similarly. Here, we will explain a physical configuration example using the learning device 101 as an example, with reference to the diagram.

[0097] Figure 10 shows an example of the physical configuration of the learning device 101 according to Embodiment 1. Physically, the learning device 101 includes, for example, a bus 1010, a processor 1020, a memory 1030, a storage device 1040, a network interface 1050, an input interface 1060, and an output interface 1070.

[0098] Bus 1010 is a data transmission path for the processor 1020, memory 1030, storage device 1040, user interface 1050, network interface 1060, camera 1070, and microphone 1080 to send and receive data to and from each other. However, the method of connecting the processor 1020 and other components to each other is not limited to bus connection.

[0099] The 1020 processor is a processor implemented in components such as the CPU (Central Processing Unit) and GPU (Graphics Processing Unit).

[0100] Memory 1030 is a main memory device implemented using RAM (Random Access Memory), etc.

[0101] The storage device 1040 is an auxiliary storage device implemented as an HDD (Hard Disk Drive), SSD (Solid State Drive), memory card, or ROM (Read Only Memory). The storage device 1040 stores program modules for realizing the functions of the device equipped with it (in the example in Figure 10, the learning device 101). The processor 1020 reads each of these program modules into memory 1030 and executes them, thereby realizing the functions corresponding to those program modules.

[0102] The network interface 1050 is an interface for connecting a device equipped with it (in the example in Figure 10, the learning device 101) to the network NT.

[0103] The input interface 1060 is an interface for the user to input information. The input interface 1060 consists of, for example, a touch panel, a keyboard, a mouse, and so on.

[0104] The output interface 1070 is an interface for presenting information to the user. The output interface 1070 is composed of, for example, a liquid crystal panel, an organic EL (Electro-Luminescence) panel, etc.

[0105] The imaging device 102 may be configured to include a camera as described above. This camera may be equipped with a network interface 1050 for connecting to the network NT.

[0106] We have so far described an example configuration of the information processing system 100 according to Embodiment 1. From here, we will describe the operation of the information processing system 100 according to this embodiment.

[0107] (Operation of the information processing system 100 according to Embodiment 1) The information processing system 100 according to Embodiment 1 performs information processing that includes a learning process performed by the learning device 101 and a matching process performed by the matching device 103. Each of these processes will be described with reference to the figures.

[0108] (Example of learning process according to Embodiment 1) Figure 11 is a flowchart illustrating an example of the learning process according to Embodiment 1. The learning process is a process for training the extraction model. The learning device 101 starts the learning process when, for example, the learning information acquisition unit 111 acquires learning information.

[0109] The learning information acquisition unit 111 acquires the learning information of the mini-batch (step S101).

[0110] In detail, for example, the learning information acquisition unit 111 randomly acquires mini-batch learning information from the entire learning information acquired as an instruction to start the learning process. The mini-batch learning information is a part of the learning information and includes, for example, a number of training images corresponding to a predetermined batch size (e.g., 128 images), and the resolution of the iris region and correct answer information corresponding to each training image. The mini-batch learning information may include training images with different iris region resolutions.

[0111] The correct answer information according to this embodiment includes, for example, the iris class included in the training image, as described above.

[0112] The batch size is not limited to 128 and may be predetermined as appropriate. Furthermore, the learning information acquisition unit 111 may perform augmentation processing on the learning images.

[0113] The extraction unit 112 uses the extraction model with the training image acquired in step S101 as input to extract iris information (step S102).

[0114] In detail, for example, the feature extraction unit 112a uses a feature extraction model with a training image as input to extract iris features (step S102a).

[0115] The classification unit 112b uses the iris features extracted in step S102a as input and employs a classification model to extract the class corresponding to the iris contained in the training image (step S102b).

[0116] The calculation unit 113 uses the resolution of the iris region acquired in step S101 to calculate the loss between the iris information extracted in step S102 and the ground truth information acquired in step S101 (step S103).

[0117] In detail, for example, the parameter calculation unit 113a determines the parameter values ​​related to the parameters for calculating the loss (step S103a).

[0118] Figure 12 is a flowchart showing an example of the parameter calculation process (step S103a) according to Embodiment 1.

[0119] The margin calculation unit 113a_1 calculates the value of the margin parameter using the resolution of the iris region acquired in step S101 (step S103a_1).

[0120] For example, the margin calculation unit 113a_1 calculates a larger value for the margin parameter as the resolution of the iris region decreases. In other words, the margin calculation unit 113a_1 increases the value of the margin parameter for training images with low resolution of the iris region and decreases the value of the margin parameter for training images with high resolution of the iris region.

[0121] The weight attenuation calculation unit 113a_2 calculates the value of the weight attenuation parameter using the resolution of the iris region acquired in step S101 (step S103a_2).

[0122] For example, the weight attenuation calculation unit 113a_2 determines a smaller value for the weight attenuation parameter as the resolution of the iris region decreases. That is, the weight attenuation calculation unit 113a_2 decreases the value of the weight attenuation parameter for training images with low resolution of the iris region, and increases the value of the weight attenuation parameter for training images with high resolution of the iris region.

[0123] For example, the weight attenuation calculation unit 113a_2 calculates the weight attenuation parameter value wd(r) using wd(r) = wd0 * (min(r,100) / 100). wd0 is the value of the weight attenuation parameter before modification, and is, for example, a constant. r is the resolution of the iris region (for example, the number of pixels in the iris diameter). min(a,b) is a function that represents the smaller of a and b. In this formula, the weight attenuation parameter value wd(r) decreases according to the resolution of the iris region when the resolution of the iris region is less than 100, and becomes a constant when the resolution of the iris region is 100 or greater.

[0124] The norm calculation unit 113a_3 calculates the magnitude of the iris feature (step S103a_3). The magnitude of the iris feature is, for example, the iris diameter (e.g., diameter or radius) of the iris included in the training image.

[0125] Refer to Figure 11 again. The loss calculation unit 113b calculates the loss between the iris information and the correct information using the parameter values ​​obtained in step S103a (step S103b).

[0126] For example, in a typical CosFace model, the losses expressed by equations (1) and (2) are used. Also, in a typical ArcFace model, the losses expressed by equation (3) are used.

[0127]

number

[0128]

number

[0129]

number

[0130] Here, N is the batch size. θyi,i is the angle between the feature (Fi) and the yi-th column of the weights. S is the scaling parameter. m is the margin. W is the weight vector. The weight vector W consists of components with dimensions of the number of classes CN × the feature vector F. 。 F is a feature vector (a vector representing iris features). y is a one-hot vector representing the class included in the ground truth information (ground truth class). By minimizing these losses, the distance between the weights of positive examples and their classes (vector quantities) can be reduced.

[0131] In step S103b, such losses are corrected using the parameter values ​​obtained in step S103a.

[0132] In detail, for example, the loss calculation unit 113b uses the value of the margin parameter calculated in step S103a_1 to calculate the loss.

[0133] For example, the loss calculation unit 113b uses the value of the weight attenuation parameter calculated in step S103a_2 to calculate the loss.

[0134] For example, if the resolution of the iris region acquired in step S101 is less than or equal to a predetermined value, the loss calculation unit 113b calculates the loss so that the magnitude of the iris feature calculated in step S103a_3 approaches a predetermined reference value. Here, the predetermined reference value is the magnitude of the iris feature extracted from a training image in which the resolution of the iris region is a predetermined value.

[0135] Here, if the reference value is D0 and the magnitude of the iris feature calculated in step S103a_3 is D, then for example, the loss calculation unit 113b uses |max(D0-D),0| for the loss E. max(A,B) is a function that represents the larger of A and B.

[0136] The update unit 114 updates the parameter values ​​of the extraction model using the loss calculated in step S103 (step S104).

[0137] In detail, for example, the gradient calculation unit 114a calculates the gradient with respect to the parameters of the extracted model using the loss calculated in step S103b (step S104a).

[0138] In step S104a, for example, the gradient calculation unit 114a uses the loss calculated in step S103b to calculate the gradient with respect to the parameters of the extracted model by backpropagation.

[0139] The parameter update unit 114b updates the parameter values ​​of the extracted model using the gradient calculated in step S104a (step S104b), and then terminates the learning process.

[0140] In step S104b, for example, the parameter update unit 114b updates the parameters of the extraction model using the gradient calculated in step S104a and a preset learning rate.

[0141] Such a learning process can be repeated, for example, according to a predetermined number of times. The classes extracted in step S102b will be random values ​​(e.g., vector quantities) in the early stages of learning, but as learning progresses, they will become closer to the values ​​of the one-hot vector representing the correct class. This allows the extraction model to be trained.

[0142] Furthermore, the loss calculation unit 113b may terminate the learning process even if the number of iterations of the learning process is less than a predetermined number, if the change in loss calculated in step S103 falls below a predetermined threshold.

[0143] (Example of matching process according to Embodiment 1) The matching process according to Embodiment 1 includes, for example, the process described above with reference to Figure 5. The matching process is a process for matching a target image with a trained extraction model (or feature extraction model). The matching device 103 starts the matching process when, for example, the matching extraction unit 131 acquires a target image that includes the iris of the target.

[0144] The matching extraction unit 131 takes a target image containing the target iris as input and uses an extraction model to extract the iris information of the target from the target image (step S201).

[0145] The extraction model used here is one that has been trained by repeatedly executing the training process. Note that in step S201, only the feature extraction model may be applied.

[0146] Furthermore, the iris information extracted in step S201 may include, for example, iris features.

[0147] The matching unit 132 compares the iris information extracted in step S201 with pre-registered registration data (step S202), and then terminates the matching process.

[0148] For example, the matching unit 132 may output, as a result of the matching, the similarity between the extracted iris features and the iris features contained in the registered data. Alternatively, as a result of the matching, the matching unit 132 may output, for the registered data in which the similarity of the iris features contained in the registered data is above a threshold or is the highest, information for identifying an individual, such as the personal ID (Identifier) ​​and personal name assigned to each individual, contained in the registered data.

[0149] (Effects / Actions) As described above, according to this embodiment, the learning device 101 comprises an extraction unit 112, a calculation unit 113, and an update unit 114. The extraction unit 112 extracts iris information by using an extraction model that learns from a training image including the iris as input. The calculation unit 113 calculates the loss between the iris information and the correct information using the resolution of the iris region included in the training image. The update unit 114 updates the parameter values ​​of the extraction model using the calculated loss.

[0150] This allows the values ​​of the extraction model parameters to be updated using a loss corresponding to the resolution of the iris region included in the training image. Therefore, even if the resolution of the iris region is low in the target image containing the target iris, it becomes possible to obtain accurate information from the image of that iris region.

[0151] According to this embodiment, the calculation unit 113 includes a parameter calculation unit 113a and a loss calculation unit 113b. The parameter calculation unit 113a calculates parameter values ​​for parameters used to determine the loss. The loss calculation unit 113b uses the parameter values ​​to calculate the loss between the iris information and the correct information.

[0152] This allows the values ​​of the extraction model parameters to be updated using a loss corresponding to the resolution of the iris region contained in the training image. Therefore, even if the resolution of the iris region contained in the target image is low, it becomes possible to obtain accurate information from the image of that iris region.

[0153] According to this embodiment, the parameter value includes at least one of (1) the value of the margin parameter, which is a hyperparameter included in the loss function for calculating the loss, (2) the value of the weight decay parameter, and (3) the magnitude of the iris feature.

[0154] By including the value of the margin parameter (1), it is possible to construct an extraction model that has excellent recognition performance for target images containing the irises of different people, even when the resolution of the iris region included in the target image is low.

[0155] By including the (2) weight decay parameter value in the parameter values, it is possible to construct an extraction model that has excellent recognition performance for target images containing the irises of different people, even when the resolution of the iris region included in the target image is low.

[0156] By including the (3) magnitude of the iris feature in the parameter value, it is possible to calculate a loss corresponding to the resolution of the iris region even when the resolution of the iris region included in the target image is low.

[0157] Therefore, even if the resolution of the iris region in the target image, which includes the iris, is low, it becomes possible to obtain accurate information from the image of that iris region.

[0158] According to this embodiment, the parameter values ​​include (1) the value of the margin parameter, which is a hyperparameter included in the loss function. The parameter calculation unit 113a includes a margin calculation unit 113a_1 that calculates a larger value for the margin parameter as the resolution of the iris region decreases.

[0159] By imposing constraints on the margin parameters in this way, it is possible to construct an extraction model that excels at distinguishing between target images containing the irises of different individuals, even when the resolution of the iris region in the target image is low. Therefore, even when the resolution of the iris region in a target image containing the iris of a target person is low, it becomes possible to obtain accurate information from the image of that iris region.

[0160] According to this embodiment, the parameter value includes (2) the value of the weight attenuation parameter. The parameter calculation unit 113a includes a weight attenuation calculation unit 113a_2 that determines a smaller value for the weight attenuation parameter as the resolution of the iris region decreases.

[0161] By imposing constraints on the weight decay parameter in this way, it is possible to construct an extraction model that excels at distinguishing target images containing the irises of different individuals, even when the resolution of the iris region in the target image is low. Therefore, even when the resolution of the iris region in a target image containing the target iris is low, it becomes possible to obtain accurate information from the image of that iris region.

[0162] According to this embodiment, the parameter values ​​include (3) the size of the iris feature. The parameter calculation unit 113a includes a norm calculation unit 113a_3 that determines the size of the iris feature. The loss calculation unit 113b calculates the loss between the iris information and the ground truth information so that when the resolution of the iris region is less than or equal to a predetermined value, the size of the iris feature approaches a predetermined reference value. The predetermined reference value is the size of the iris feature extracted from a training image in which the resolution of the iris region is a predetermined value.

[0163] This allows for the setting of constraints to maintain a constant resolution in the iris region of the training image, even when the resolution of the iris region is low. Therefore, even when the resolution of the iris region in the target image is low, it is possible to construct an extraction model with excellent recognition performance for target images containing the irises of different individuals. Consequently, even when the resolution of the iris region in the target image containing the target iris is low, it becomes possible to obtain accurate information from the image of that iris region.

[0164] According to this embodiment, the iris information includes at least one of the following: iris features extracted from a training image; an image of the iris region in the training image; the position of the feature area; and a class corresponding to the iris included in the training image.

[0165] This allows various iris information to be obtained using the extraction model. Therefore, even if the resolution of the iris region included in the target image is low, it becomes possible to obtain various information with high accuracy from the image of that iris region.

[0166] (Variation 1) As described above, the parameter values ​​should include at least one of the following: (1) the value of the margin parameter, (2) the value of the weight decay parameter, or (3) the magnitude of the iris feature. In other words, the parameter values ​​may be one or two of any of (1) to (3).

[0167] In this case, the parameter calculation unit 113a may include one or more of the margin calculation unit 113a_1, weight attenuation calculation unit 113a_2, and norm calculation unit 113a_3 to determine which of (1) to (3) the parameter value includes. Furthermore, the parameter calculation process (step S103a) may include one or more of the steps 103a_1 to 103a_3 to calculate which of (1) to (3) the parameter value includes. In step S103b, the loss may be corrected using the parameter value calculated in one of the steps 103a_1 to 103a_3.

[0168] This modified form also produces the same effects as Embodiment 1.

[0169] (Modification 2) This modified example describes an example of a loss function for calculating the loss using the resolution of the iris region included in the training image, when the iris information includes an image of the iris region.

[0170] The loss E2 in this modified example is, for example, the loss E (see Equation 4) obtained by dividing the loss E using the L1 norm between the extracted map showing the image of the iris region and the ground truth map showing the image of the iris region included in the ground truth information by the resolution of the iris region included in the training image (for example, the number of pixels of the iris diameter).

[0171]

number

[0172] Here, j is information (image ID) used to identify the training images included in the batch. M is the number of samples. jy is the resolution of the iris region in the j-th training image. i is the value representing each pixel. N is the total number of pixels in the training image. i f(x) is the pixel value of the ground truth map. x is a vector quantity composed of the pixel values ​​of the training images. i This represents the pixel values ​​of the extracted map. Note that the loss E2 is not limited to this and may be other than the L1 norm, for example.

[0173] This modified form also produces the same effects as Embodiment 1.

[0174] (Variation 3) This modified example describes an example of a loss function for calculating loss using the resolution of the iris region included in the training image, when the iris information includes the location of feature regions.

[0175] The loss E3 in this modified example is, for example, the loss E obtained by dividing the loss E, which uses the L1 norm between the extracted position (the location of the extracted feature) and the ground truth position (the location of the feature included in the ground truth information), by the resolution of the iris region included in the training image (for example, the number of pixels in the iris diameter). Here, the loss E can be the loss E obtained using equation (4), as in Modified Example 2.

[0176] However, in this modified example, j is information (image ID) for identifying the training images included in the batch. M is the number of samples. Rj is the resolution of the iris region included in the j-th training image. i is a value representing each pixel. N is the total number of pixels in the training image. yi is the correct position. x is a vector quantity composed of the pixel values ​​of the training image. f(x)i is the extraction position. For example, , special If there are three points of reference, the elements that make up their positions will be three x-coordinates and three y-coordinates, for a total of six elements. Note that the loss E3 is not limited to this and may be other than the L1 norm, for example.

[0177] This modified form also produces the same effects as Embodiment 1.

[0178] (Embodiment 2) Embodiment 2 describes an example of changing the parameter value when the parameter value includes (3) the magnitude of the iris feature. In this embodiment, in order to simplify the explanation, explanations that overlap with Embodiment 1 will be omitted as appropriate.

[0179] The parameter values ​​in this embodiment include (3) the size of the iris feature. The parameter values ​​in this embodiment differ from those in Embodiment 1 in that they include the sizes of multiple iris feature quantities. The multiple iris feature quantities include, for example, low-resolution feature quantities and high-resolution feature quantities extracted from low-resolution training images and high-resolution training images, respectively, which share common ground truth information but have different iris region resolutions.

[0180] Low-resolution training images and high-resolution training images are training images with low resolution of the iris region and high resolution of the iris region, respectively. Low-resolution features and high-resolution features are iris features extracted from low-resolution training images and iris features extracted from high-resolution training images, respectively.

[0181] In this embodiment, the norm calculation unit 113a_3 is better configured to determine the magnitudes of both the low-resolution feature and the high-resolution feature.

[0182] In this embodiment, the loss calculation unit 113b may calculate the loss between the iris information and the ground truth information such that the size of the low-resolution feature approaches the size of the high-resolution feature. Here, if the size of the high-resolution feature is DH and the size of the low-resolution feature is DL, for example, the loss calculation unit 113b may further use |DH-DL| for the loss E.

[0183] However, using this loss function may cause the size DH of the high-resolution feature vector to approach the size DL of the low-resolution feature vector. To mitigate this possibility, a minimum value D0 may be predetermined for the size DH of the high-resolution feature vector. In this case, the loss function E may be further calculated using, for example, |max(DH,D0)-DL|.

[0184] (Effects / Actions) As described above, according to this embodiment, the parameter value includes the magnitudes of multiple iris features. The multiple iris features include low-resolution features and high-resolution features extracted from low-resolution training images and high-resolution training images, respectively, which share common ground truth information but have different iris region resolutions.

[0185] The parameter calculation unit 113a includes a norm calculation unit 113a_3 that determines the magnitudes of the low-resolution features and the high-resolution features, respectively. The loss calculation unit 113b calculates the loss between the iris information and the ground truth information so that the magnitude of the low-resolution features approaches the magnitude of the high-resolution features.

[0186] By imposing such constraints on the loss function, the extraction model can be trained to bring the iris information extracted from low-resolution iris region images closer to the iris information extracted from high-resolution iris region images. Therefore, even when the resolution of the iris region in the target image is low, it is possible to construct an extraction model with excellent recognition performance for target images containing the irises of different individuals. Consequently, even when the resolution of the iris region in a target image containing the target iris is low, it becomes possible to obtain accurate information from the image of that iris region.

[0187] (Embodiment 3) Embodiment 3 describes an example of changing the learning rate for updating the parameters of the extraction model. In this embodiment, in order to simplify the explanation, explanations that overlap with Embodiment 1 will be omitted as appropriate.

[0188] In this embodiment, the parameter update unit 114b applies different learning rates to the feature extraction model and the classification model. For example, the learning rate of the feature extraction model may be a real number multiple of the learning rate of the classification model (where this real number is greater than 1) or an integer multiple such as 2, 3, 4, or 5.

[0189] In the parameter update process (step S104b) according to this embodiment, the parameter update unit 114b may update the parameters of the extraction model using the gradient calculated in step S104a and different learning rates for the feature extraction model and the classification model.

[0190] (Effects / Actions) As described above, according to this embodiment, the extraction model includes a feature extraction model that extracts iris features from a training image as input, and a classification model that extracts classes corresponding to irises contained in the training image from the iris features as input. The update unit 114 applies different learning rates to the feature extraction model and the classification model.

[0191] This allows the feature extraction model and the classification model to be trained at different amplitudes. For example, the feature extraction model can be trained at a larger amplitude than the classification model. As the size of the iris features increases, the learning effect using the loss calculated by the loss calculation unit 113b can be further increased, making it less likely for the size of high-resolution features to decrease and more likely for the size of low-resolution features to increase.

[0192] Therefore, even if the resolution of the iris region in the target image is low, it is possible to construct an extraction model with excellent recognition performance for target images containing the irises of different individuals. Consequently, even if the resolution of the iris region in the target image containing the iris is low, it becomes possible to obtain accurate information from the image of that iris region.

[0193] (Embodiment 4) Embodiment 4 describes an example in which the gradient of the feature extraction model is multiplied by a constant. In this embodiment, in order to simplify the explanation, explanations that overlap with Embodiment 1 will be omitted as appropriate.

[0194] The gradient calculation unit 114a in this embodiment multiplies the gradient obtained for the feature extraction model by a constant. This constant may be a real number greater than 1, or an integer such as 2, 3, 4, or 5.

[0195] In the gradient calculation process (step S104a) according to this embodiment, the gradient calculation unit 114a calculates the gradients for the feature extraction model, the classification model, and their respective parameters. Then, the gradient calculation unit 114a multiplies the gradient obtained for the feature extraction model by a constant.

[0196] In the parameter update process (step S104b) according to this embodiment, the parameter update unit 114b updates the parameters of the feature extraction model using a gradient multiplied by a constant. The parameter update unit 114b may update the parameters of the extraction model using the gradient obtained in step S104a, similar to Embodiment 1. In addition, in this embodiment, the learning rates of the feature extraction model and the classification model may be the same as in Embodiment 1, or they may be different as in Embodiment 3.

[0197] (Effects / Actions) As described above, according to this embodiment, the extraction model includes a feature extraction model that extracts iris features from training images as input, and a classification model that extracts classes corresponding to irises contained in training images from the iris features as input. The update unit 114 includes a gradient calculation unit 114a and a parameter update unit 114b. The gradient calculation unit 114a calculates the gradient with respect to the parameters of the extraction model using the calculated loss. The parameter update unit 114b calculates and updates the parameters of the extraction model using the calculated gradient. In calculating the parameters of the extraction model, the parameter update unit 114b uses a value obtained by multiplying the calculated gradient by a predetermined constant.

[0198] This allows the feature extraction model and the classification model to be trained with different amplitudes, similar to Embodiment 3. For example, the feature extraction model can be trained with a larger amplitude than the classification model. As the size of the iris features increases, the effect of loss correction can be further enhanced, making it less likely for the size of high-resolution features to decrease and more likely for the size of low-resolution features to increase.

[0199] Therefore, even if the resolution of the iris region in the target image is low, it is possible to construct an extraction model with excellent recognition performance for target images containing the irises of different individuals. Consequently, even if the resolution of the iris region in the target image containing the iris is low, it becomes possible to obtain accurate information from the image of that iris region.

[0200] The embodiments and modifications of the present invention have been described above with reference to the drawings, but these are merely examples of the present invention, and various other configurations can also be adopted.

[0201] Furthermore, while the flowcharts used in the above description show multiple steps (processes) in sequence, the execution order of the steps performed in each embodiment is not limited to the order in which they are described. In each embodiment, the order of the illustrated steps can be changed to the extent that it does not impede the content. Also, the above embodiments and modifications can be combined to the extent that their content is not contradictory.

[0202] Some or all of the above embodiments may also be described as follows, but are not limited to the following:

[0203] 1. An extraction means for extracting iris information related to the iris by using an extraction model that learns using training images including the iris as input, A calculation means for calculating the loss between the iris information and the correct information using the resolution of the iris region included in the training image, The system includes an update means for updating the parameter values ​​of the extraction model using the calculated loss. Information processing system. 2. The calculation means is: A parameter calculation means for determining parameter values ​​related to the parameters for determining the aforementioned loss, Includes a loss calculation means that calculates the loss between the iris information and the correct answer information using the parameter values. The information processing system described in 1. 3. The parameter value includes at least one of the following: (1) the value of the margin parameter, which is a hyperparameter included in the loss function for calculating the loss; (2) the value of the weight decay parameter included in the loss function for calculating the loss; and (3) the magnitude of the iris feature. The information processing system described in 2. 4. The parameter values ​​include the value of the margin parameter, which is a hyperparameter included in the loss function. The parameter calculation means includes a margin calculation means that determines a larger value for the margin parameter as the resolution of the iris region decreases. The information processing system described in 3. 5. The parameter value includes the value of the weight attenuation parameter, The parameter calculation means includes a weight attenuation calculation means that determines a smaller value for the weight attenuation parameter as the resolution of the iris region decreases. The information processing system described in 3. or 4. 6. The parameter values ​​include the magnitude of the iris feature, The parameter calculation means includes a norm calculation means for determining the magnitude of the iris feature quantity. The loss calculation means calculates the loss between the iris information and the ground truth information such that the size of the iris feature approaches a predetermined reference value when the resolution of the iris region is below a predetermined value. The predetermined reference value is the magnitude of the iris feature extracted from the training image, where the resolution of the iris region is a predetermined value. An information processing system described in any one of items 3 through 5. 7. The parameter value includes the magnitudes of a plurality of iris features, The plurality of iris features include low-resolution features and high-resolution features extracted from low-resolution training images and high-resolution training images, respectively, which share the same ground truth information but have different resolutions for the iris region. The parameter calculation means includes a norm calculation means for determining the magnitudes of the low-resolution feature and the high-resolution feature, The loss calculation means calculates the loss between the iris information and the ground truth information such that the size of the low-resolution feature approaches the size of the high-resolution feature. An information processing system described in any one of items 3 through 5. 8. The iris information includes at least one of the following: iris features extracted from the training image; an image of the iris region in the training image; the location of the feature; and a class corresponding to the iris included in the training image. An information processing system described in any one of items 1 through 7. 9. The extraction model described above is: A feature extraction model that extracts the iris features using the aforementioned training image as input, The system includes a classification model that takes the iris features as input and extracts a class corresponding to the iris contained in the training image, The update means applies different learning rates to the feature extraction model and the classification model. An information processing system described in any one of items 1 through 8. 10. The extraction model described above is A feature extraction model that extracts iris information using the aforementioned training image as input, The system includes a classification model that takes the iris information as input and extracts a class corresponding to the iris contained in the training image, The update means is, A gradient calculation means that uses the calculated loss to calculate the gradient with respect to the parameters of the extraction model, The system includes a parameter update means for calculating and updating the parameters of the extraction model using the calculated gradient, The parameter update means uses a value obtained by multiplying the calculated gradient by a predetermined constant when calculating the parameters of the extraction model. An information processing system described in any one of items 1 through 8. 11. An extraction means for extracting iris information related to the iris by using an extraction model with a training image including the iris as input, A calculation means for calculating the loss between the iris information and the correct information using the resolution of the iris region included in the training image, The system includes an update means for updating the parameter values ​​of the extraction model using the calculated loss. Learning device. 12. A matching extraction means that takes a target image including the target iris as input and extracts iris information relating to the target iris using the extraction model trained using the learning device described in 11., The system includes a matching means for comparing the iris information of the target with pre-registered registration data. Verification device. 13. The first group of computers, consisting of one or more computers, By using an extraction model that learns using training images including the iris as input, iris information related to the iris is extracted. Using the resolution of the iris region included in the aforementioned training image, the loss between the iris information and the correct information is calculated. The calculated loss is used to update the parameter values ​​of the extraction model. Information processing methods. 14. A second group of computers consisting of one or more computers, Using a target image including the target iris as input, the first group of computers extracts iris information related to the target iris using the extraction model learned by executing the information processing method described in 13. The iris information of the subject is compared with pre-registered data. Information processing methods. 15. In the first computer group consisting of one or more computers, By using an extraction model that learns using training images including the iris as input, iris information related to the iris is extracted. Using the resolution of the iris region included in the aforementioned training image, the loss between the iris information and the correct information is calculated. A recording medium on which a program is stored that causes the system to update the parameter values ​​of the extraction model using the calculated loss. 16. A second group of computers consisting of one or more computers, Using the target image including the target iris as input, the extraction model, which has been trained by having the first group of computers execute the program recorded on the recording medium described in 15., is used to extract iris information relating to the target iris. A recording medium on which a program is stored that causes the subject to compare the iris information with pre-registered data. [Explanation of Symbols]

[0204] 100 Information Processing Systems 101 Learning device 102 Imaging device 103 Verification device 111 Learning Information Acquisition Department 112 Extraction part 112a Feature extraction unit 112b Classification section 113 Calculation Section 113a Parameter calculation unit 113a_1 Margin Calculation Unit 113a_2 Weight Attenuation Calculation Unit 113a_3 Norm calculation unit 113b Loss calculation section 114 Update Department 114a Gradient Calculation Unit 114b Parameter update section 131. Extraction unit for verification. 132 Verification section

Claims

1. An extraction means for extracting iris information related to the iris by using an extraction model that learns using training images including the iris as input, A calculation means for calculating the loss between the iris information and the correct information using the resolution of the iris region included in the training image, The system includes an update means for updating the parameter values ​​of the extraction model using the calculated loss, The calculation means is A parameter calculation means for determining parameter values ​​related to the parameters for determining the aforementioned loss, Includes a loss calculation means that calculates the loss between the iris information and the correct answer information using the parameter values. Information processing system.

2. The parameter value includes at least one of the following: (1) the value of the margin parameter, which is a hyperparameter included in the loss function for calculating the loss; (2) the value of the weight decay parameter included in the loss function for calculating the loss; and (3) the magnitude of the iris feature. The information processing system according to claim 1.

3. The parameter value includes the value of the margin parameter, which is a hyperparameter included in the loss function. The parameter calculation means includes a margin calculation means that determines a larger value for the margin parameter as the resolution of the iris region decreases. The information processing system according to claim 2.

4. The parameter value includes the value of the weight decay parameter, The parameter calculation means includes a weight attenuation calculation means that determines a smaller value for the weight attenuation parameter as the resolution of the iris region decreases. The information processing system according to claim 2 or 3.

5. The parameter value includes the magnitude of the iris feature, The parameter calculation means includes a norm calculation means for determining the magnitude of the iris feature quantity. The loss calculation means calculates the loss between the iris information and the ground truth information such that the size of the iris feature approaches a predetermined reference value when the resolution of the iris region is below a predetermined value. The predetermined reference value is the magnitude of the iris feature extracted from the training image, where the resolution of the iris region is a predetermined value. The information processing system according to claim 2 or 3.

6. The parameter value includes the magnitudes of low-resolution features and high-resolution features respectively, extracted from low-resolution training images and high-resolution training images, respectively, which share the same ground truth information but have different iris region resolutions. The parameter calculation means includes a norm calculation means for determining the magnitudes of the low-resolution feature and the high-resolution feature, The loss calculation means calculates the loss between the iris information and the ground truth information such that the size of the low-resolution feature approaches the size of the high-resolution feature. The information processing system according to claim 2 or 3.

7. The iris information includes at least one of the following: iris features extracted from the training image; an image of the iris region in the training image; the location of the feature; and a class corresponding to the iris included in the training image. The information processing system according to any one of claims 1 to 3.

8. The first group of computers, consisting of one or more computers, By using an extraction model that learns using training images including the iris as input, iris information related to the iris is extracted. Using the resolution of the iris region included in the aforementioned training image, the loss between the iris information and the correct information is calculated. This includes updating the parameter values ​​of the extraction model using the calculated loss, Calculating the aforementioned loss is Determine the parameter values ​​for the parameters used to calculate the aforementioned loss, This includes calculating the loss between the iris information and the ground truth information using the aforementioned parameter values. Information processing methods.

9. The first group of computers, consisting of one or more computers, By using an extraction model that learns using training images including the iris as input, iris information related to the iris is extracted. Using the resolution of the iris region included in the aforementioned training image, the loss between the iris information and the correct information is calculated. The calculated loss is used to update the parameter values ​​of the extraction model. Calculating the aforementioned loss is Determine the parameter values ​​for the parameters used to calculate the aforementioned loss, A program that includes calculating the loss between the iris information and the correct information using the parameter values ​​mentioned above.

Images