Information processing system, information processing device, information processing method, and storage medium

Abstract

This information processing system comprises a first acquisition unit, a second acquisition unit, and a generation unit. The first acquisition unit acquires a first iris image. The second acquisition unit acquires a second iris image of which the quality is lower than that of the first iris image. On the basis of the first iris image and the second iris image, the generation unit generates a degraded iris image obtained by degrading the first iris image to approach the quality of the second iris image.

Description

Information processing system, information processing device, information processing method, and recording medium

[0001] The present disclosure relates to an information processing system, an information processing device, an information processing method, and a recording medium.

[0002] For example, Patent Literature 1 discloses a feature transformation learning device for solving the problem that the resolution of an image to be authenticated in iris authentication or the like is sometimes insufficient, and that applying super-resolution to increase the resolution can result in high computational costs. Super-resolution is a technology for converting an image into one with a higher resolution than the original image.

[0003] This feature transformation learning device includes an image acquisition means, an image reduction means, an image enlargement means, a feature extraction means, a feature transformation means, and a learning control means.

[0004] According to the description of Patent Document 1, the image acquisition means acquires a first image. The image reduction means reduces the first image to a second image having a lower resolution than the first image. The image enlargement means enlarges the second image to a third image having the same resolution as the first image. The feature extraction means extracts a first feature that is a feature of the first image and a second feature that is a feature of the third image. The feature conversion means converts the second feature into a third feature. The learning control means causes the feature conversion means to learn a feature conversion method based on a comparison result between the first feature and the third feature.

[0005] An image reducing unit described in Patent Document 1 as an example of image reducing means reduces an image by thinning out pixels of the image.

[0006] International Publication No. 2022 / 195819

[0007] The present disclosure aims to improve upon the techniques described in the prior art documents mentioned above.

[0008] The information processing system in the present disclosure includes a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image of lower quality than the first iris image; and a generation unit that generates a degraded iris image based on the first iris image and the second iris image by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0009] The information processing device of the present disclosure includes a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image of lower quality than the first iris image; and a generation unit that generates a degraded iris image by degrading the first iris image so as to approach the quality of the second iris image based on the first iris image and the second iris image.

[0010] The information processing method in the present disclosure includes one or more computers acquiring a first iris image, acquiring a second iris image of lower quality than the first iris image, and generating a degraded iris image based on the first iris image and the second iris image by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0011] The recording medium in the present disclosure has recorded thereon a program for causing one or more computers to execute the following operations: acquire a first iris image; acquire a second iris image of lower quality than the first iris image; and generate a degraded iris image by degrading the first iris image so as to approach the quality of the second iris image based on the first iris image and the second iris image.

[0012] 1 is a diagram illustrating an example of a first information processing system according to the present disclosure. FIG. 1 is a block diagram illustrating an example of a first information processing device according to the present disclosure. FIG. 2 is a flowchart illustrating an example of a processing operation of the first information processing device according to the present disclosure. FIG. 3 is a diagram illustrating, in a feature space, bringing a first iris image closer to the quality of a second iris image. FIG. 4 is a diagram illustrating an example of a physical configuration of a first information processing device according to the present disclosure. FIG. 5 is a block diagram illustrating an example of a first generation unit according to the present disclosure. FIG. 6 is a flowchart illustrating an example of a processing operation of the first generation unit according to the present disclosure. FIG. 7 is a diagram illustrating an example of first parameter defining information according to the present disclosure. FIG. 8 is a block diagram illustrating an example of a second information processing device according to the present disclosure. FIG. 9 is a flowchart illustrating an example of a processing operation of the second information processing device according to the present disclosure. FIG. 10 is a block diagram illustrating an example of a third information processing device according to the present disclosure. FIG. 11 is a flowchart illustrating an example of a processing operation of the third information processing device according to the present disclosure. FIG. 12 is a block diagram illustrating an example of a fourth information processing device according to the present disclosure. FIG. 13 is a flowchart illustrating an example of a processing operation of the fourth information processing device according to the present disclosure. FIG. 14 is a block diagram illustrating an example of a fifth information processing device according to the present disclosure. FIG. 15 is a flowchart illustrating an example of a processing operation of the fifth information processing device according to the present disclosure. FIG. 16 is a block diagram illustrating an example of a sixth information processing device according to the present disclosure. FIG. 17 is a flowchart illustrating an example of a processing operation of the sixth information processing device according to the present disclosure. FIG. 18 is a block diagram illustrating an example of a processing operation of the sixth information processing device according to the present disclosure. FIG. 19 is a block diagram illustrating an example of a processing operation of the sixth information processing device according to the present disclosure. FIG. 19 is Fig. 10 is a flowchart illustrating an example of a processing operation of an eighth information processing apparatus according to the present disclosure;Fig. 11 is a block diagram illustrating an example of a ninth information processing apparatus according to the present disclosure;Fig. 12 is a flowchart illustrating an example of a processing operation of the ninth information processing apparatus according to the present disclosure.

[0013] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In all drawings, similar components are denoted by similar reference numerals, and descriptions thereof will be omitted as appropriate. In the present disclosure, the drawings relate to one or more embodiments.

[0014] [First Embodiment] In general biometric authentication, for example, a target image corresponding to a target part (e.g., iris, face) used for biometric authentication of a target is generated from a captured image including the target part. This target is the subject of biometric authentication and may be a person or an animal such as a dog or snake. A person as used below is an example of such a target. Then, for example, image features extracted by inputting the target image into a learning model are used in matching to obtain an authentication result.

[0015] The target part is, for example, the iris in iris authentication, the face in face authentication, the veins in vein authentication, and the fingerprint in fingerprint authentication. In the case of multimodal authentication in which biometric authentication is performed using multiple target parts, the multiple target parts may be a combination of two or more target parts, such as the face and iris.

[0016] A photographed image is an image of a person or the like.

[0017] Generally, the quality of captured images is often lower than that of images captured in an ideal (e.g., stable) environment suitable for capturing images, or the quality of theoretically expected images. Furthermore, the quality of captured images often varies. For example, the quality of a captured image includes various elements (quality factors), such as the resolution of the area showing the target area in the captured image, focus blur, motion blur, eyeglass reflection, and noise. The quality of captured images may vary to various degrees with respect to one or more of these quality factors. Generally, when authentication is performed using such low-quality images, authentication accuracy often decreases compared to when high-quality images are used.

[0018] In this case, one possible method for improving authentication accuracy is to use learning images of similar quality to the target images obtained from actual captured images when training the learning model.

[0019] However, although the degradation and variation in quality of captured images are thought to be due to the specific shooting conditions, it is difficult to fully clarify the causes and processes. Therefore, it is difficult to obtain training images with the same quality as the target images used in actual authentication.

[0020] In particular, a large number of training images are often used in training. In such cases, it is extremely difficult to obtain a large number of training images that have the same quality or distribution as the target images used in actual authentication.

[0021] Furthermore, for example, Patent Document 1 describes reducing an image by thinning out pixels of the image, as described above. However, Patent Document 1 does not describe a method for degrading an image to a quality similar to that of a target image used in actual authentication (for example, to achieve authentication accuracy similar to that of a target image used in actual authentication).

[0022] Thus, there is a problem that it is difficult to obtain a degraded image of a quality that corresponds to the intended use of the image, etc. For example, when the intended use is authentication, there is a problem that it is difficult to obtain a degraded image of the same quality as an image used in actual authentication.

[0023] (Configuration of Information Processing System S1) As shown in FIG. 1, the information processing system S1 includes a first acquisition unit 101, a second acquisition unit 102, and a generation unit 103.

[0024] The first acquisition unit 101 acquires a first iris image. The second acquisition unit 102 acquires a second iris image of lower quality than the first iris image. The generation unit 103 generates a degraded iris image based on the first and second iris images by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0025] (Actions and Effects) According to this information processing system S1, by acquiring a second iris image according to the use etc. of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality is suited to the use etc. Therefore, it is possible to obtain a degraded image of quality suited to the use etc. of the image.

[0026] (Configuration of Information Processing Apparatus 100) As shown in FIG. 2, the information processing apparatus 100 includes a first acquisition unit 101, a second acquisition unit 102, and a generation unit 103.

[0027] The first acquisition unit 101 acquires a first iris image. The second acquisition unit 102 acquires a second iris image of lower quality than the first iris image. The generation unit 103 generates a degraded iris image based on the first and second iris images by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0028] (Actions and Effects) According to this information processing device 100, by acquiring a second iris image according to the use etc. of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality is suited to the use etc. Therefore, it is possible to obtain a degraded image of quality suited to the use etc. of the image.

[0029] (Example of Processing Operation of Information Processing Device 100) The information processing device 100 executes information processing as shown in FIG.

[0030] The first acquisition unit 101 acquires a first iris image (step S101). The second acquisition unit 102 acquires a second iris image of lower quality than the first iris image (step S102). The generation unit 103 generates a degraded iris image based on the first and second iris images by degrading the first iris image so that its quality approaches that of the second iris image (step S103).

[0031] (Actions and Effects) According to the information processing method, by acquiring a second iris image according to the use, etc. of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality is suited to the use. Therefore, it is possible to obtain a degraded image of quality suited to the use, etc. of the image.

[0032] (Detailed Example) The information processing system S1 is a system for generating a degraded iris image by degrading the first iris image depending on the use of the first iris image, etc. Image degradation corresponds to a decrease in image quality, and an example of this is a decrease in image resolution (i.e., reducing the image).

[0033] Examples of uses of the first iris image include training of a machine learning model, iris authentication, etc. That is, the degraded iris image is used for training of a machine learning model, iris authentication, etc.

[0034] The machine learning model that is trained using the degraded iris image is a machine learning model used for iris recognition. Here, the machine learning model is, for example, a model configured by a neural network, and the same applies hereinafter.

[0035] Examples of such machine learning models include a feature extraction model and a super-resolution model, as will be described in detail below. The feature extraction model is a machine learning model for extracting image features of an image, and when an image is input, it outputs the image features of the image. The image features are represented by one or more vector components and are also referred to as a feature vector, etc. The super-resolution model is a machine learning model for increasing the resolution of an image, and when an image is input, it outputs a super-resolution image with increased resolution of the image.

[0036] The machine learning model that learns using the degraded iris image may be a machine learning model used for purposes other than iris authentication. Also, the feature extraction model is an example of a means (feature extraction unit) for extracting image features from an image, and the means for extracting image features is not limited to using a machine learning model.

[0037] As an example of using a degraded iris image for iris authentication, as will be described in detail below, when a pre-registered image is a high-resolution image and a captured image obtained by photography is a low-resolution image, the registered image is degraded to approach the quality of the captured image. In general iris authentication, the image features of the registered image and the captured image are compared, and authentication is performed based on the comparison result. The authentication accuracy of such iris authentication often increases the closer the quality of the registered image and the captured image. Therefore, authentication accuracy can be improved by comparing the image features of the captured image with the degraded iris image obtained by degrading the registered image.

[0038] The iris authentication may be performed in a specific location. In this case, the second iris image may be, for example, an image captured in a test environment corresponding to the specific location.

[0039] (Regarding the First Iris Image and the Second Iris Image) The first acquisition unit 101 and the second acquisition unit 102 each acquire at least one first iris image and at least one second iris image. The first acquisition unit 101 and the second acquisition unit 102 each acquire the first iris image and the second iris image, for example, from one or more storage units (not shown) that have one or both of the first iris image and the second iris image stored in advance. Note that the first acquisition unit 101 or the second acquisition unit 102 may each acquire the first iris image or the second iris image from a camera, another information processing device, or the like (not shown) via a wired, wireless, or combination of these networks.

[0040] Each of the first iris image and the second iris image is an image (iris image) that includes a person's iris. Each of the first iris image and the second iris image may be an image obtained by capturing an image in advance.

[0041] When a plurality of first iris images and a plurality of second iris images are acquired, the number of second iris images may be fewer than the number of first iris images.

[0042] The first iris image is, for example, an image captured in the ideal environment described above. The second iris image is, for example, an image captured in the test environment described above, and its quality may be low due to various factors. Therefore, acquiring a large number of second iris images with various qualities is more difficult than acquiring a large number of first iris images. By having fewer second iris images than first iris images, the first iris image and the second iris image can be easily prepared.

[0043] Furthermore, the first iris image and the second iris image may be iris images of the same person, or may be iris images of different people.

[0044] The iris image may include a person's iris, and may be, for example, an iris region image showing only the iris region, a monocular image including one predetermined eye of the left or right eye, or a binocular image including both eyes.

[0045] The first and second iris images have different qualities, with the quality of the second iris image being lower than the quality of the first iris image.

[0046] The quality may include one or more of various elements (quality elements), such as the resolution of the region showing the target part in the captured image, the degree of focus blur, the degree of motion blur, the degree of eyeglass reflection, and the degree of noise. The quality may be represented, for example, by a vector composed of values ​​of one or more quality elements. Note that the quality may also be represented by a score according to the quality element (for example, the greater the focus blur, the smaller the score).

[0047] The resolution may be, for example, the number of pixels of the iris diameter or iris radius. The degree of focus blur and the degree of motion blur may be values ​​representing the magnitude of focus blur and the magnitude of motion blur, respectively. The degree of eyeglass reflection may include at least one of the intensity of light reflected by the eyeglasses, the area of ​​the target area used for biometric authentication that is affected by the reflection, etc. The degree of noise may include at least one of the magnitude of the noise (e.g., the total amount of overall noise, the average of overall noise, etc.) and a value representing the noise distribution (variance, standard deviation, etc.).

[0048] In particular, for example, the quality of each of the first iris image and the second iris image may be represented by one or more predetermined quality element values.

[0049] (Regarding Degradation Processing) Based on the first iris image and the second iris image, the generation unit 103 performs the above-described image degradation processing (hereinafter, this processing is referred to as "degradation processing"). That is, the degradation processing is processing for generating a degraded iris image by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0050] In the degradation process, for example, the resolution of the first iris image may be reduced so that the quality of the first iris image approaches the quality of the second iris image.

[0051] 4 is a diagram illustrating, in a feature space, the process of bringing the quality of the first iris image closer to that of the second iris image. The feature space is a space for representing image features (feature vectors) extracted from an image.

[0052] Bringing the quality of the first iris image closer to that of the second iris image means bringing a first position in the feature space of the first image feature closer to a second position in the feature space of the second image feature. The first image feature is an image feature extracted from the first iris image. The second image feature is an image feature extracted from the second iris image.

[0053] The degradation process may, for example, use one or more common image processes and may use an image degradation model to degrade the image.

[0054] Examples of common types of image processing include resizing using the Bicubic method, adding Gaussian blur, convolution using a median filter, convolution using a sinc filter, adding Gaussian noise, adding Poisson noise, geometric transformation, etc. Note that the image processing used for degradation processing is not limited to these.

[0055] When one or more image processes are used in the degradation process, the type of one or more image processes, the order in which they are performed, and the number of times they are performed may each be predetermined, selected randomly, or selected according to a predetermined probability.

[0056] The image degradation model will be described in detail later.

[0057] Here, to perform degradation processing to degrade the first iris image so that its quality approaches that of the second iris image, it is necessary to define a value (distance) representing the difference in quality between the first iris image and the second iris image. This is because, for example, a high-resolution image that is aligned with a low-resolution image obtained by photographing generally does not exist. For example, even if the similarity between a low-resolution image obtained by photographing and a degraded image generated by the degradation processing is evaluated, the differences between these images arise due to various factors, such as the degree of eye opening and the lighting conditions. Therefore, it is difficult to quantify the degree of quality degradation caused by the degradation processing by comparing the images.

[0058] In the present disclosure, the distance between the first iris image and the second iris image is the distance between the images when image features extracted from each image using a feature extraction model are expressed in a feature space, as schematically shown in Fig. 4. The distance may be, for example, an angle θ such as the L2 norm or cosine similarity. The above-mentioned "bringing the first position closer to the second position" corresponds to reducing the distance between the first position and the second position.

[0059] In general, a feature extraction model is trained so that, for example, image features of the same ID (identifier), i.e., image features obtained from iris images of the same person, are embedded at the same position in a feature space. Therefore, iris images of the same ID are embedded at the same position robustly regardless of image quality other than resolution, and differences in image quality due to resolution tend to appear in the distance in the feature space. Therefore, by using the feature space, such as by comparing positions in the feature space embedded by the feature extraction model, differences in image quality can be quantified.

[0060] (Example of Physical Configuration of Information Processing Device 100) The information processing device 100 physically includes 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, as shown in FIG.

[0061] The bus 1010 is a data transmission path for transmitting and receiving data among the processor 1020, memory 1030, storage device 1040, network interface 1050, input interface 1060, and output interface 1070. However, the method of connecting the processor 1020 and the like to each other is not limited to bus connection.

[0062] The processor 1020 is implemented by a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or the like.

[0063] The memory 1030 is a main storage device realized by a RAM (Random Access Memory) or the like.

[0064] The storage device 1040 is an auxiliary storage device realized by a hard disk drive (HDD), a solid state drive (SSD), a memory card, a read only memory (ROM), or the like. The storage device 1040 stores program modules for realizing the functions of the information processing device 100. The processor 1020 reads each of these program modules into the memory 1030 and executes them to realize the function corresponding to the program module.

[0065] The network interface 1050 is an interface for connecting the information processing device 100 to a network. The network is a communication network for transmitting and receiving information to and from other devices (not shown), and may be configured as a wired or wireless network or a combination of these.

[0066] The input interface 1060 is an interface for the user to input information, and is composed of, for example, a touch panel, a keyboard, a mouse, and the like.

[0067] The output interface 1070 is an interface for presenting information to the user, and is configured, for example, by a liquid crystal panel, an organic EL (Electro-Luminescence) panel, or the like.

[0068] Although the information processing system S1 and the information processing device 100 are physically configured from one device (e.g., a computer, etc.) in the above example, the information processing device 100 may be configured from multiple devices (e.g., computers, etc.). In this case, the multiple devices may cooperate to execute information processing by transmitting and receiving information to and from each other via a network NT, for example.

[0069] As described above, according to this embodiment, it is possible to obtain a degraded image of a quality that corresponds to the intended use of the image. Furthermore, if the intended use is iris authentication, the degraded image can be used for iris authentication or for training a learning model for iris authentication. Therefore, it is possible to improve the accuracy of iris authentication.

[0070] According to this embodiment, bringing the quality of the first iris image closer to that of the second iris image means bringing the first position in the feature space of the first image feature extracted from the first iris image closer to the second position in the feature space of the second image feature extracted from the second iris image.

[0071] In this way, by using the feature space, it is possible to quantify the difference in image quality. Therefore, it is possible to obtain a degraded image of a quality that corresponds to the purpose of the image. Furthermore, if the purpose is iris recognition, the degraded image can be used for iris recognition or for training a learning model for iris recognition. Therefore, it is possible to improve the accuracy of iris recognition.

[0072] Note that iris authentication, the first iris image, and the second iris image are examples of biometric authentication, the first biometric image, and the second biometric image. Biometric authentication is authentication performed using an image of a target part of a person or an animal (e.g., a snake, a dog, etc.), and may be facial authentication, vein authentication, fingerprint authentication, etc. The first biometric image and the second biometric image are biometric images including the target part, such as a facial image including a face in facial authentication, a vein image including veins in vein authentication, and a fingerprint image or finger image including a fingerprint in fingerprint authentication.

[0073] [Embodiment 2] As described in embodiment 1, general image processing can be used for degradation processing. In embodiment 2, an example will be described in which one or more general image processing processes are used for degradation processing, among the detailed examples of the generation unit 103 and the generation processing (step S103) described above.

[0074] As shown in FIG. 6 , the generation unit 103 includes an estimation unit 104 and a degradation processing unit 105 .

[0075] The estimation unit 104 estimates a first parameter to be used for generating a degraded iris image (degradation processing) by degrading the first iris image so as to approach the quality of the second iris image, based on the first iris image acquired by the first acquisition unit 101 and the second iris image acquired by the second acquisition unit 102.

[0076] In detail, for example, the estimation unit 104 includes a first quality estimation unit 104a, a second quality estimation unit 104b, and a first parameter estimation unit 104c.

[0077] The first quality estimation unit 104a estimates a first quality, which is the quality of the first iris image, based on the first iris image.

[0078] The second quality estimation unit 104b estimates a second quality, which is the quality of the second iris image, based on the second iris image.

[0079] The first parameter estimation unit 104c estimates the first parameter using the first quality estimated by the first quality estimation unit 104a and the second quality estimated by the second quality estimation unit 104b.

[0080] The degradation processing unit 105 generates a degraded iris image using the first parameter estimated by the first parameter estimation unit 104c.

[0081] (Example of Processing Operation of Generator 103) The generator 103 executes a generation process (step S103) as shown in FIG.

[0082] The estimation unit 104 estimates a first parameter based on the first iris image acquired in step S101 and the second iris image acquired in step S102 (step S104).

[0083] The first parameter is used to generate a degraded iris image by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0084] In detail, for example, in the estimation process (step S104), the first quality estimation unit 104a estimates a first quality, which is the quality of the first iris image, based on the first iris image acquired in step S101 (step S104a).

[0085] The second quality estimation unit 104b estimates a second quality, which is the quality of the second iris image, based on the second iris image (step S104b).

[0086] The first parameter estimation unit 104c estimates the first parameter using the first quality estimated in step S104a and the second quality estimated in step S104b (step S104c).

[0087] The degradation processor 105 generates a degraded iris image using the first parameter estimated in step S104c (step S105).

[0088] (Regarding the first and second quality)

[0089] The first quality estimation unit 104 a estimates the first quality by, for example, inputting the first iris image into a quality estimation model. The quality estimation model is a machine learning model for estimating the quality of an image, and when an image is input, it outputs the quality of the image.

[0090] The first quality may be represented by a vector consisting of values ​​of one or more predetermined quality elements, for example, as described above.

[0091] The second quality estimation unit 104b estimates the second quality by, for example, inputting the second iris image into a quality estimation model similar to that of the first quality estimation unit 104a.

[0092] The second quality may be represented by a vector consisting of one or more predetermined quality element values, as described above. The first quality and the second quality may each be composed of the same type of quality element.

[0093] (First Parameter) The first parameter may include a parameter (computation parameter) used in the computation of each image processing for one or more of the image processing methods described above used in the degradation processing.

[0094] The first parameter may be estimated further using first parameter defining information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality.

[0095] 8 is a diagram showing an example of first parameter definition information. The diagram shows an example in which n*m first parameters V_1*1 to V_n*m are defined for each of the combinations obtained by dividing the first quality and the second quality into n and m parts from the minimum value to the maximum value without overlapping each other. Each of the first parameters V_1*1 to V_n*m is, for example, a vector including one or more values ​​in a predetermined order.

[0096] The first parameter may be estimated by further using a first parameter estimation model that has been trained to estimate the first parameter, instead of the first parameter-defining information. The first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input.

[0097] A method for learning the first parameter estimation model will be described in another embodiment.

[0098] (Operations and Effects) As described above, according to this embodiment, the generation unit 103 includes the estimation unit 104 and the degradation processing unit 105 .

[0099] The estimation unit 104 estimates a first parameter to be used to generate a degraded iris image by degrading the first iris image so as to bring it closer to the quality of the second iris image, based on the first iris image acquired by the first acquisition unit 101 and the second iris image acquired by the second acquisition unit 102.

[0100] The degradation processing unit 105 generates a degraded iris image using the first parameter estimated by the first parameter estimation unit 104c.

[0101] This allows for the acquisition of a second iris image according to the intended use of the first iris image, thereby generating a degraded iris image by degrading the first iris image so that the quality is suited to that intended use. Therefore, it is possible to obtain a degraded image of a quality suited to the intended use of the image.

[0102] Furthermore, when the application is iris recognition, the degraded image can be used for iris recognition or for training a learning model for iris recognition, thereby improving the accuracy of iris recognition.

[0103] According to this embodiment, the estimation unit 104 includes a first quality estimation unit 104a, a second quality estimation unit 104b, and a first parameter estimation unit 104c.

[0104] The first quality estimation unit 104a estimates a first quality, which is the quality of the first iris image, based on the first iris image. The second quality estimation unit 104b estimates a second quality, which is the quality of the second iris image, based on the second iris image. The first parameter estimation unit 104c estimates a first parameter using the first quality estimated by the first quality estimation unit 104a and the second quality estimated by the second quality estimation unit 104b.

[0105] This allows for the acquisition of a second iris image according to the intended use of the first iris image, thereby generating a degraded iris image by degrading the first iris image so that the quality is suited to that intended use. Therefore, it is possible to obtain a degraded image of a quality suited to the intended use of the image.

[0106] Furthermore, when the application is iris recognition, the degraded image can be used for iris recognition or for training a learning model for iris recognition, thereby improving the accuracy of iris recognition when using low-quality images.

[0107] According to this embodiment, the first parameter is estimated further using first parameter defining information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality.

[0108] This allows for the acquisition of a second iris image according to the intended use of the first iris image, thereby generating a degraded iris image by degrading the first iris image so that the quality is suited to that intended use. Therefore, it is possible to obtain a degraded image of a quality suited to the intended use of the image.

[0109] According to this embodiment, the first parameter is estimated using a first parameter estimation model that has been trained to estimate the first parameter. The first parameter estimation model is a machine learning model that receives the first quality and the second quality as input and outputs the first parameter.

[0110] This allows for the acquisition of a second iris image according to the intended use of the first iris image, thereby generating a degraded iris image by degrading the first iris image so that the quality is suited to that intended use. Therefore, it is possible to obtain a degraded image of a quality suited to the intended use of the image.

[0111] Third Embodiment In a third embodiment, a first example of a method for learning a first parameter estimation model will be described.

[0112] As shown in FIG. 9, the information processing device 200 includes a first acquisition unit 101, a second acquisition unit 102, and a generation unit 103, as well as a first extraction unit 201, a second extraction unit 202, a similarity calculation unit 203, and a second parameter update unit 204.

[0113] The first extraction unit 201 extracts degraded image features, which are image features of the degraded iris image, by inputting the degraded iris image generated by the generation unit 103 into a feature extraction model for extracting image features.

[0114] The second extraction unit 202 inputs the second iris image acquired by the second acquisition unit 102 into a feature extraction model, thereby extracting second image features that are image features of the second iris image.

[0115] The similarity calculation unit 203 calculates the similarity between the degraded image feature extracted by the first extraction unit 201 and the second image feature extracted by the second extraction unit 202 .

[0116] The second parameter update unit 204 updates the second parameters applied to the first parameter estimation model for estimating the first parameters, based on the similarity calculated by the similarity calculation unit 203 .

[0117] (Example of Processing Operation of Information Processing Device 200) The information processing device 200 executes a first learning process as shown in FIG.

[0118] The above-described steps S101 to S103 are executed.

[0119] The first extraction unit 201 extracts degraded image features, which are image features of the degraded iris image, by inputting the degraded iris image generated in S103 into a feature extraction model for extracting image features (step S201).

[0120] The second extraction unit 202 inputs the second iris image acquired in step S102 into a feature extraction model, thereby extracting second image features that are image features of the second iris image (step S202).

[0121] The similarity calculation unit 203 calculates the similarity between the degraded image feature extracted in step S201 and the second image feature extracted in step S202 (step S203).

[0122] The second parameter update unit 204 updates the second parameters to be applied to the first parameter estimation model for estimating the first parameters, based on the similarity calculated in step S203 (step S204).

[0123] This information processing may be repeatedly executed until a predetermined termination condition is satisfied, thereby enabling the first parameter estimation model to be trained.

[0124] (Regarding Training Images) The first and second iris images used for training the first parameter estimation model may be, for example, iris images of the same person. In this case, the second parameter update unit 204 may update the second parameters so that the degree of similarity increases, i.e., so that the degraded image feature and the second image feature have more similar values.

[0125] (Regarding the Feature Extraction Model) The feature extraction model is a machine learning model for extracting image features. The first extraction unit 201 and the second extraction unit 202 may use the same feature extraction model. The feature extraction model may be trained in advance using, for example, high-quality iris images. This training may involve, for example, supervised learning using correct answer data including correct answer labels such as personal IDs.

[0126] (Regarding Similarity) The similarity calculated by the similarity calculation unit 203 is, for example, cosine similarity, L2 norm, etc. However, the similarity is not limited to these.

[0127] (Regarding the method for updating the second parameter) The second parameter may be updated using a general optimization method such as a search method such as a gradient method or a grid search. The gradient method is a technique for calculating the gradient of a parameter to be updated from a loss, and updating the parameter using the gradient.

[0128] (Operations and Effects) As described above, according to this embodiment, the information processing device 200 further includes the first extraction unit 201 , the second extraction unit 202 , the similarity calculation unit 203 , and the second parameter update unit 204 .

[0129] The first extraction unit 201 inputs the generated degraded iris image into a feature extraction model for extracting image feature amounts, thereby extracting degraded image feature amounts that are image feature amounts of the degraded iris image.

[0130] The second extraction unit 202 inputs the second iris image into a feature extraction model to extract second image features that are image features of the second iris image.

[0131] The similarity calculation unit 203 calculates the similarity between the extracted degraded image feature amount and the extracted second image feature amount.

[0132] The second parameter update unit 204 updates the second parameters applied to the first parameter estimation model for estimating the first parameters, based on the calculated similarity.

[0133] This allows the first parameter estimation model to be trained. As a result, by acquiring a second iris image according to the purpose of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality is suited to the purpose. Therefore, it is possible to obtain a degraded image of quality suited to the purpose of the image.

[0134] [Fourth Embodiment] In a fourth embodiment, an example will be described in which a feature extraction model is trained together with a first parameter estimation model. In this embodiment, an example will be described in which the same training method as in the third embodiment is used for training the first parameter estimation model.

[0135] 11 , the information processing device 300 includes a first acquisition unit 101, a second acquisition unit 102, a generation unit 103, a first extraction unit 201, a second extraction unit 202, a similarity calculation unit 203, and a second parameter update unit 204. The information processing device 300 further includes an authentication loss calculation unit 301 and a third parameter update unit 302.

[0136] The authentication loss calculation unit 301 calculates the authentication loss using the degraded image feature amount extracted by the first extraction unit 201 and the correct answer data.

[0137] The third parameter update unit 302 updates the third parameter applied to the feature extraction model based on the authentication loss calculated by the authentication loss calculation unit 301 .

[0138] (Example of Processing Operation of Information Processing Device 300) The information processing device 300 executes a second learning process as shown in FIG.

[0139] The above-described steps S101 to S103 and S201 to S204 are executed.

[0140] The authentication loss calculation unit 301 calculates an authentication loss using the degraded image feature amount extracted in step S201 and the correct answer data (step S301).

[0141] The third parameter update unit 302 updates the third parameter applied to the feature extraction model based on the authentication loss calculated in step S301 (step S302).

[0142] This information processing may be repeatedly executed until a predetermined termination condition is satisfied. As described in the third embodiment, steps S201 to S204 are processes for training a first parameter estimation model. Steps S301 and S302 are processes for training a feature extraction model. Therefore, training of the first parameter estimation model and the feature extraction model is alternately performed.

[0143] As described in the third embodiment, the first extraction unit 201 and the second extraction unit 202 may use a common feature extraction model. Therefore, in step S302, the third parameter applied to these feature extraction models may be updated.

[0144] In steps S301 and S302, the feature extraction model is trained using the degraded image features, which allows image features suited to the application, i.e., image features that enable accurate authentication, to be extracted from the degraded image.

[0145] (Operations and Effects) As described above, according to this embodiment, the information processing device 300 further includes the authentication loss calculation unit 301 and the third parameter update unit 302 .

[0146] The authentication loss calculation unit 301 calculates an authentication loss using the correct data and the degraded image features extracted by the first extraction unit 201. The third parameter update unit 302 updates a third parameter applied to the feature extraction model based on the authentication loss calculated by the authentication loss calculation unit 301.

[0147] This makes it possible to extract image features appropriate for the intended use, i.e., image features that enable accurate authentication, from the degraded image, thereby improving the accuracy of the intended use, i.e., the accuracy of iris authentication, for example.

[0148] [Fifth Embodiment] In the fifth embodiment, a second example of a training method for a first parameter estimation model will be described. In the third embodiment, an example was described in which the first parameter estimation model is trained so that the similarity between the degraded image feature and the second image feature increases. In this embodiment, an example is described in which the first parameter estimation model is trained using an adversarial learning technique so that the degraded image feature and the second image feature cannot be distinguished by a classifier (i.e., so that the classification score of the classifier becomes small).

[0149] 13 , the information processing device 400 includes a first acquisition unit 101, a second acquisition unit 102, a generation unit 103, a first extraction unit 201, and a second extraction unit 202. Furthermore, the information processing device 400 includes a discrimination score estimation unit 401, an adversarial loss calculation unit 402, a second parameter update unit 403, and a fifth parameter update unit 404.

[0150] The discrimination score estimation unit 401 includes a so-called discriminator, and when an image feature amount of either the degraded image feature amount or the second image feature amount is input, calculates a discrimination score for the input image feature amount.

[0151] The classification score includes a first probability and a second probability. The first probability is the probability that the input image feature is a degraded image feature. The second probability is the probability that the input image feature is a second image feature.

[0152] The adversarial loss calculation unit 402 calculates the adversarial loss based on the discrimination score calculated by the discrimination score estimation unit 401 .

[0153] The second parameter update unit 403 updates the second parameter applied to the first parameter estimation model for estimating the first parameter based on the adversarial loss calculated by the adversarial loss calculation unit 402.

[0154] The fifth parameter update unit 404 updates a fifth parameter for calculating the discrimination score based on the adversarial loss calculated by the adversarial loss calculation unit 402.

[0155] (Example of Processing Operation of Information Processing Device 400) The information processing device 400 executes a third learning process as shown in FIG.

[0156] The above-described steps S101 to S103 and S201 to S202 are executed.

[0157] When the discrimination score estimation unit 401 receives either the degraded image feature extracted in S201 or the second image feature extracted in S202 as input, it calculates a discrimination score including the first probability and the second probability as described above (step S401).

[0158] The adversarial loss calculation unit 402 calculates the adversarial loss based on the discrimination score calculated in step S401 (step S402).

[0159] The second parameter update unit 403 updates the second parameters to be applied to the first parameter estimation model for estimating the first parameters, based on the adversarial loss calculated in step S402 (step S403).

[0160] The fifth parameter update unit 404 updates the fifth parameter for calculating the discrimination score based on the adversarial loss calculated in step S402 (step S404).

[0161] This information processing may be repeatedly executed until a predetermined termination condition is satisfied, thereby enabling the first parameter estimation model to be trained.

[0162] (Regarding Training Images) The first and second iris images used for training the first parameter estimation model may be, for example, iris images of the same person or may be iris images of different people.

[0163] (Regarding the Identification Score) As described above, either the degraded image feature amount or the second image feature amount is input to the identification score estimation unit 401. In this input, for example, the degraded image feature amount and the second image feature amount may be input alternately. Note that whether the degraded image feature amount or the second image feature amount is input does not have to be input alternately, and may be determined, for example, according to a predetermined rule.

[0164] The discrimination score estimation unit 401 calculates a discrimination score including the probability (first probability and second probability) that the input image feature is the degraded image feature and the second image feature, respectively.

[0165] As described above, the classification score estimation unit 401 includes a classifier. This classifier may be a general classifier configured using a machine learning model, for example.

[0166] (Regarding Adversarial Loss AL) The adversarial loss AL includes, for example, a generation loss AL1 and an identification loss AL2.

[0167] The generation loss AL1 is a loss for updating a parameter used in the generation unit 103. The parameter used in the generation unit 103 may be the second parameter described above, or may be the second parameter and the fourth parameter.

[0168] The classification loss AL2 is a loss for updating the fifth parameter. Here, as described above, the fifth parameter is a parameter for calculating the classification score. In other words, it is a parameter used by the classification score estimation unit 401 (more specifically, applied to, for example, a machine learning model that constitutes a classifier).

[0169] For example, assume that the classification score estimation unit 401 outputs "1" for the degraded image feature and "0" for the second image feature. In this case, the generation loss AL1 and the classification loss AL2 are expressed by, for example, the following equations (1) and (2).

[0170]

[0171]

[0172] In equations (1) and (2), D(G(Z)) represents the classification score for the degraded image feature G(Z) extracted from the degraded iris image generated by the generation unit 103. In equation (2), D(x) represents the classification score for the second image feature.

[0173] (Parameter Update Method Using Adversarial Loss AL) For updating parameters using the adversarial loss AL, a general optimization method such as a search method such as a gradient method or a grid search may be used.

[0174] In more detail, for example, when the gradient method is used, the second parameter update unit 403 obtains a gradient using the generation loss AL1 and updates the second parameter (or the second parameter and the fourth parameter) so as to bring the gradient closer to zero. Also, for example, when the gradient method is used, the fifth parameter update unit 404 obtains a gradient using the discrimination loss AL2 and updates the fifth parameter so as to bring the gradient closer to zero.

[0175] (Actions and Effects) As described above, according to this embodiment, the information processing device 400 further includes a first extraction unit 201, a second extraction unit 202, a discrimination score estimation unit 401, an adversarial loss calculation unit 402, and a second parameter update unit 403.

[0176] The first extraction unit 201 inputs the generated degraded iris image into a feature extraction model for extracting image feature amounts, thereby extracting degraded image feature amounts that are image feature amounts of the degraded iris image.

[0177] The second extraction unit 202 inputs the second iris image into a feature extraction model to extract second image features that are image features of the second iris image.

[0178] When an image feature, either a degraded image feature or a second image feature, is input, the identification score estimation unit 401 calculates an identification score including a first probability that the input image feature is a degraded image feature and a second probability that the input image feature is the second image feature.

[0179] The adversarial loss calculation unit 402 calculates the adversarial loss based on the calculated discrimination score.

[0180] The second parameter update unit 403 updates the second parameters applied to the first parameter estimation model for estimating the first parameters, based on the calculated adversarial loss.

[0181] The fifth parameter update unit 404 updates the fifth parameter for calculating the discrimination score based on the calculated adversarial loss.

[0182] This allows the first parameter estimation model to be trained. As a result, by acquiring a second iris image according to the purpose of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality distribution is appropriate for that purpose. Therefore, it is possible to obtain a degraded image with a quality distribution appropriate for the purpose of the image.

[0183] Sixth Embodiment In the sixth embodiment, a third example of a training method for a first parameter estimation model will be described. In this embodiment, as in the fifth embodiment, an example will be described in which the first parameter estimation model is trained using an adversarial learning technique. In this embodiment, instead of the degraded image feature amount and the second image feature amount, an example will be described in which the first parameter estimation model is trained so that the degraded iris image and the second iris image cannot be distinguished by the classifier (i.e., so that the classification score of the classifier becomes small). This embodiment roughly corresponds to a case in which the image feature amount in the fifth embodiment is replaced with an image, i.e., the degraded image feature amount and the second image feature amount in the fifth embodiment are replaced with the degraded iris image and the second iris image, respectively.

[0184] As shown in FIG. 15 , the information processing device 500 includes a first acquisition unit 101, a second acquisition unit 102, a generation unit 103, as well as an identification score estimation unit 501, an adversarial loss calculation unit 502, a second parameter update unit 503, and a fifth parameter update unit 504.

[0185] The identification score estimation unit 501 includes a classifier, and when either the degraded iris image or the second iris image is input, calculates an identification score for the input iris image.

[0186] The identification score includes a first probability, which is the probability that the input iris image is an iris image, and a second probability, which is the probability that the input iris image is a second iris image.

[0187] The adversarial loss calculation unit 502 calculates the adversarial loss based on the discrimination score calculated by the discrimination score estimation unit 501.

[0188] The second parameter update unit 503 updates the second parameter applied to the first parameter estimation model for estimating the first parameter based on the adversarial loss calculated by the adversarial loss calculation unit 502.

[0189] The fifth parameter update unit 504 updates a fifth parameter for calculating the discrimination score based on the adversarial loss calculated by the adversarial loss calculation unit 502.

[0190] (Example of Processing Operation of Information Processing Device 500) The information processing device 500 executes a fourth learning process as shown in FIG.

[0191] The above-described steps S101 to S103 are executed.

[0192] When the identification score estimation unit 501 receives either the degraded iris image acquired in S103 or the second iris image acquired in S102 as input, it calculates an identification score including the first probability and second probability as described above (step S501).

[0193] The adversarial loss calculation unit 502 calculates the adversarial loss based on the discrimination score calculated in step S501 (step S502).

[0194] The second parameter update unit 503 updates the second parameters applied to the first parameter estimation model for estimating the first parameters, based on the adversarial loss calculated in step S502 (step S503).

[0195] The fifth parameter update unit 504 updates the fifth parameter for calculating the discrimination score based on the adversarial loss calculated in step S502 (step S504).

[0196] This information processing may be repeatedly executed until a predetermined termination condition is satisfied, thereby enabling the first parameter estimation model to be trained.

[0197] (Regarding Training Images) The first and second iris images used for training the first parameter estimation model may be, for example, iris images of the same person or may be iris images of different people.

[0198] (Regarding Identification Score and Adversarial Loss AL) The identification score and the adversarial loss AL may be the same as those described in the fifth embodiment.

[0199] (Operations and Effects) As described above, according to this embodiment, the information processing device 500 further includes the discrimination score estimation unit 501 , the adversarial loss calculation unit 502 , the second parameter update unit 503 , and the fifth parameter update unit 504 .

[0200] When the identification score estimation unit 501 receives an image, either a degraded iris image or a second iris image, it calculates an identification score that includes a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image.

[0201] The adversarial loss calculation unit 502 calculates the adversarial loss based on the calculated discrimination score.

[0202] The second parameter update unit 503 updates the second parameters applied to the first parameter estimation model for estimating the first parameters, based on the calculated adversarial loss.

[0203] The fifth parameter update unit 504 updates the fifth parameter for calculating the discrimination score based on the calculated adversarial loss.

[0204] This allows the first parameter estimation model to be trained. As a result, by acquiring a second iris image according to the purpose of the first iris image, it is possible to generate a degraded iris image by degrading the first iris image so that the quality distribution is appropriate for that purpose. Therefore, it is possible to obtain a degraded image with a quality distribution appropriate for the purpose of the image.

[0205] [Embodiment 7] As described in embodiment 1, a machine learning model can be used for the degradation process. In embodiment 7, among detailed examples of the above-mentioned generation unit 103 and generation process (step S103), an example will be described in which an image degradation model, which is a machine learning model for degrading an image, is used for the degradation process.

[0206] As shown in FIG. 17, the generation unit 103 includes a second quality estimation unit 104 b, a first parameter estimation unit 604 c, and a degradation processing unit 605 .

[0207] As described above, the second quality estimation unit 104b estimates the second quality, which is the quality of the second iris image, based on the second iris image.

[0208] The first parameter estimation unit 604c estimates a first parameter to be used to generate a degraded iris image by degrading the first iris image so that the quality approaches that of the second iris image, based on the second quality estimated by the second quality estimation unit 104b.

[0209] The degradation processing unit 605 applies the first parameter estimated by the first parameter estimation unit 604c to an image degradation model for degrading an image, and inputs the first iris image into the image degradation model, thereby generating a degraded iris image.

[0210] (Example of Processing Operation of Generator 103) The generator 103 executes a generation process (step S103) as shown in FIG.

[0211] As described above, the second quality estimation unit 104b estimates the second quality, which is the quality of the second iris image, based on the second iris image (step S104b).

[0212] The first parameter estimation unit 604c estimates a first parameter to be used to generate a degraded iris image by degrading the first iris image so as to bring it closer to the quality of the second iris image, based on the second quality estimated in step S104b (step S604c).

[0213] The degradation processing unit 605 applies the first parameters estimated in step S604c to an image degradation model for degrading an image, and inputs the first iris image into the image degradation model to generate a degraded iris image (step S605).

[0214] (Regarding the method for learning the image degradation model) For learning the image degradation model, the second parameter update unit 204, 403, 503 according to the third to seventh embodiments may estimate the second parameter and the fourth parameter as a parameter update unit. The fourth parameter is a parameter applied to the image degradation model.

[0215] The second parameter may be estimated by the same method as described above.

[0216] That is, for example, the second parameter updating unit 204 may function as a parameter updating unit and further update the fourth parameter based on the similarity calculated by the similarity calculating unit 203. At this time, the second parameter updating unit 204 may function as a parameter updating unit and further update the fourth parameter so that the similarity increases, that is, so that the degraded image feature amount and the second image feature amount have more similar values.

[0217] For example, the second parameter update unit 403 may function as a parameter update unit and further update the fourth parameter based on the adversarial loss calculated by the adversarial loss calculation unit 402 .

[0218] For example, the second parameter update unit 503 may function as a parameter update unit and update the fourth parameter based on the adversarial loss calculated by the adversarial loss calculation unit 502 .

[0219] (Operations and Effects) As described above, according to this embodiment, the generation unit 103 includes the second quality estimation unit 104 b , the first parameter estimation unit 604 c , and the degradation processing unit 605 .

[0220] The second quality estimation unit 104b estimates a second quality, which is the quality of the second iris image, based on the second iris image.

[0221] The first parameter estimation unit 604c estimates, based on the second quality, a first parameter used to generate a degraded iris image by degrading the first iris image so as to approach the quality of the second iris image.

[0222] The degradation processing unit 605 applies the estimated first parameters to an image degradation model for degrading an image, and inputs the first iris image to the image degradation model, thereby generating a degraded iris image.

[0223] This allows for the acquisition of a second iris image according to the intended use of the first iris image, thereby generating a degraded iris image by degrading the first iris image so that the quality is suited to that intended use. Therefore, it is possible to obtain a degraded image of a quality suited to the intended use of the image.

[0224] Furthermore, when the application is iris recognition, the degraded image can be used for iris recognition or for training a learning model for iris recognition, thereby improving the accuracy of iris recognition.

[0225] [Embodiment 8] As described above, examples of applications of degraded iris images include training of machine learning models, iris authentication, etc. In embodiment 4, as a first example, an example of training a feature extraction model is described. In embodiment 8, as a second example, an example of training a super-resolution model for increasing the resolution of an image is described.

[0226] First, an example of an information processing device 700 that performs authentication using a super-resolution model will be described.

[0227] As shown in FIG. 19 , the information processing device 700 includes a first acquisition unit 101, a second acquisition unit 102, an image super-resolution unit 701, a third extraction unit 702, a fourth extraction unit 703, and an authentication unit 704.

[0228] The image super-resolution unit 701 generates a high-resolution image by increasing the resolution of the second iris image by inputting the second iris image acquired by the second acquisition unit 102 into a super-resolution model for increasing the resolution of the image. The super-resolution model used here is a trained super-resolution model.

[0229] The third extraction unit 702 and the fourth extraction unit 703 input the first iris image acquired by the first acquisition unit 101 into a feature extraction model, thereby extracting first image features that are image features of the first iris image.

[0230] The fourth extraction unit 703 inputs the high-resolution image generated by the image super-resolution unit 701 into a feature extraction model, thereby extracting high-resolution image features that are image features of the high-resolution image.

[0231] The first iris image here may be, for example, a registered image that is registered in advance. The feature extraction models used by each of the third extraction unit 702 and the fourth extraction unit 703 are machine learning models for extracting image features, and may be, for example, feature extraction models similar to the feature extraction models used by the first extraction unit 201 and the second extraction unit 202.

[0232] The authentication unit 704 performs authentication using the first image feature and the high-resolution image feature extracted by the third extraction unit 702 and the fourth extraction unit 703. A general technique may be used for this authentication.

[0233] (Example of Processing Operation of Information Processing Device 700: Authentication Using Super-Resolution Model) The information processing device 700 executes a first authentication process as shown in FIG.

[0234] Steps S101 and S102 are executed.

[0235] The image super-resolution unit 701 generates a high-resolution image with increased resolution of the second iris image by inputting the second iris image acquired in step S102 into a super-resolution model for increasing the resolution of the image (step S701).

[0236] The third extraction unit 702 inputs the first iris image acquired in step S101 into a feature extraction model to extract a first image feature (step S702).

[0237] The fourth extraction unit 703 inputs the high-resolution image generated in step S701 into a feature extraction model, thereby extracting high-resolution image features (step S703).

[0238] The authentication unit 704 performs authentication using the first image feature amount and the high-resolution image feature amount extracted in steps S702 and S703 (step S704).

[0239] This allows authentication to be performed using a super-resolution model. Because the resolutions of both the first iris image and the high-resolution image are high, it is possible to improve the accuracy of iris authentication.

[0240] Hereinafter, an example in which the information processing device 700 performs learning of a super-resolution model will be described.

[0241] As shown in FIG. 21, for example, the information processing device 700 further includes a generation unit 103, an image distance loss calculation unit 705, and a sixth parameter update unit 706 in order to train the super-resolution model.

[0242] As described above, the generating unit 103 generates a degraded iris image based on the first and second iris images by degrading the first iris image so as to approach the quality of the second iris image.

[0243] The image super-resolution unit 701 inputs the degraded iris image generated by the generation unit 103 into a super-resolution model for increasing the resolution of the image, thereby generating a high-resolution image in which the resolution of the degraded iris image is increased.

[0244] The image distance loss calculation unit 705 calculates the image distance loss by comparing the first iris image used by the generation unit 103 to generate the degraded iris image with the high-resolution image generated by the image super-resolution unit 701.

[0245] The sixth parameter update unit 706 updates the sixth parameter applied to the super-resolution model based on the image distance loss calculated by the image distance loss calculation unit 705 .

[0246] (Example of Processing Operation of Information Processing Device 700: Learning of Super-Resolution Model) The information processing device 700 executes a fifth learning process as shown in FIG. 22 .

[0247] Steps S101 to S103 are executed.

[0248] The image super-resolution unit 701 inputs the degraded iris image generated in step S103 into a super-resolution model for increasing the resolution of the image, thereby generating a high-resolution image with increased resolution of the degraded iris image (step S705).

[0249] The image distance loss calculation unit 705 calculates the image distance losses of the degraded iris image and the high-resolution image generated in steps S103 and S705, respectively (step S706).

[0250] The sixth parameter update unit 706 updates the sixth parameter applied to the super-resolution model based on the image distance loss calculated by the image distance loss calculation unit 705 (step S707).

[0251] This information processing may be repeated until a predetermined termination condition is met, thereby enabling the super-resolution model to be trained.

[0252] (Regarding Training Images) The first and second iris images used for training the first parameter estimation model may be, for example, iris images of the same person or may be iris images of different people.

[0253] (Regarding Image Distance Loss) The image distance is the distance between images when image features extracted from each image using a feature extraction model are expressed in a feature space, for example, as described above. This distance may be the sum of absolute values ​​of differences, the sum of squares of the differences, or the like. In this case, the information processing device 700 may further include a first extraction unit 201. The image distance loss calculation unit 705 may then calculate the image distance loss of the degraded iris image and the high-resolution image using the degraded image features extracted by the first extraction unit 201 and the high-resolution image features extracted by the fourth extraction unit 703.

[0254] Note that the image distance is not limited to the distance between images in the feature space. The image distance may be, for example, a value obtained by comparing pixel values ​​for pixels whose positions correspond in the images. More specifically, the image distance may be, for example, the sum of absolute values ​​of differences in pixel values ​​of pixels whose positions correspond, the sum of squares of the differences, or the like.

[0255] For the image distance loss, a general loss function such as a sum of squares error or a cross entropy error may be applied.

[0256] (Regarding the Method of Updating the Sixth Parameter) The sixth parameter may be updated using a general optimization method such as a search method, such as a gradient method or a grid search.

[0257] (Operations and Effects) As described above, according to this embodiment, the information processing device 700 includes the generation unit 103 , the image distance loss calculation unit 705 , and the sixth parameter update unit 706 .

[0258] The generating unit 103 generates a degraded iris image based on the first iris image and the second iris image by degrading the first iris image so that the quality of the first iris image approaches that of the second iris image.

[0259] The image super-resolution unit 701 generates a high-resolution image by increasing the resolution of the degraded iris image by inputting the generated degraded iris image into a super-resolution model for increasing the resolution of the image.

[0260] The image distance loss calculation unit 705 calculates the image distance loss of the generated degraded iris image and high-resolution image.

[0261] The sixth parameter update unit 706 updates the sixth parameter applied to the super-resolution model based on the calculated image distance loss.

[0262] This makes it possible to train a super-resolution model using images with degraded quality according to the intended use of the images, etc.

[0263] Generally, in the training of a super-resolution model, image processing such as thinning out pixels of a high-resolution test image is performed, and a reduced image generated by this is sometimes used. However, this method often results in a discrepancy in the quality distribution between the quality of the reduced image generated using image processing and the quality of the degraded image obtained by shooting. Therefore, when a super-resolution model that has undergone such training is used, the quality of the high-resolution image output from the super-resolution model is often lower than the quality of the test image used to generate the reduced image.

[0264] Furthermore, image super-resolution generally corresponds to the inverse transformation of the image degradation process. However, there are still many unknowns about the degradation process that occurs during photography.

[0265] According to this embodiment, when the second iris image is a low-resolution image, the process of resolution degradation (degradation process) can be accurately estimated and the inverse transformation of the degradation process can be learned. Therefore, it is possible to generate a high-resolution image with high accuracy from a low-resolution image of similar quality to the second iris image. Furthermore, by performing iris authentication using such a super-resolution image, it is possible to improve the accuracy of iris authentication, as described above.

[0266] Ninth Embodiment In a ninth embodiment, an example in which a degraded iris image is used for iris authentication will be described as a third example of the use of a degraded iris image.

[0267] 23, the information processing device 800 includes a first acquisition unit 101, a second acquisition unit 102, a generation unit 103, a first extraction unit 201, a second extraction unit 202, and an authentication unit 704. The functions of each of these components are generally as described above.

[0268] (Example of Processing Operation of Information Processing Device 800) The information processing device 800 executes the second authentication process as shown in FIG.

[0269] Steps S101 to S103 and S201 to S202 are executed.

[0270] The authentication unit 704 performs authentication using the degraded image feature amount and the second image feature amount extracted in steps S201 and S202, respectively. For this authentication, a general technique may be used.

[0271] (Actions and Effects) As described above, according to this embodiment, the information processing device 800 includes the first acquisition unit 101, the second acquisition unit 102, the generation unit 103, the first extraction unit 201, the second extraction unit 202, and the authentication unit 704. The functions of each of these components are generally as described above.

[0272] This allows authentication to be performed using the degraded image feature amount and the second image feature amount. The resolution of both the degraded image feature amount and the second image feature amount is low. Generally, in authentication processing, the smaller the difference in quality of the images used for authentication, the higher the authentication accuracy can be. Therefore, according to this embodiment, it is possible to improve the authentication accuracy.

[0273] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0274] In addition, although the flowcharts used in the above description show a sequence of steps (processes), the order of steps executed in each embodiment is not limited to the sequence shown in the flowcharts. In each embodiment, the order of steps shown in the diagrams can be changed as long as it does not cause any problems in terms of the content.

[0275] Some or all of the above embodiments can be described as in the following supplementary notes, but are not limited to the following. 1. An information processing system comprising: a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image of lower quality than the first iris image; and a generation unit that generates a degraded iris image based on the first and second iris images, by degrading the first iris image so that its quality approaches that of the second iris image. 2. The information processing system described in 1., wherein the generation unit comprises: an estimation unit that estimates first parameters used in generating a degraded iris image based on the first and second iris images, by degrading the first iris image so that its quality approaches that of the second iris image; and a degradation processing unit that generates the degraded iris image using the estimated first parameters. 3. The information processing system described in 2., wherein bringing the first iris image closer to the quality of the second iris image means bringing a first position in a feature space of a first image feature extracted from the first iris image closer to a second position in the feature space of a second image feature extracted from the second iris image. 4. The generation unit includes: a second quality estimation unit that estimates a second quality that is the quality of the second iris image based on the second iris image; a first parameter estimation unit that estimates first parameters used to generate a degraded iris image by degrading the first iris image so as to bring it closer to the quality of the second iris image, based on the second quality; and a degradation processing unit that applies the estimated first parameters to an image degradation model for image degradation and inputs the first iris image to the image degradation model to generate the degraded iris image. 5. The information processing system described in any one of 1. to 4., wherein the degraded iris image is used for training a machine learning model or for iris authentication.6. The information processing system described in 2. or 3., wherein the estimation unit includes: a first quality estimation unit that estimates a first quality that is the quality of the first iris image based on the first iris image; a second quality estimation unit that estimates a second quality that is the quality of the second iris image based on the second iris image; and a first parameter estimation unit that estimates the first parameter using the estimated first quality and the estimated second quality. The information processing system described in 3., further comprising: a first extraction unit that extracts degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image to a feature extraction model for extracting image features, a second extraction unit that extracts second image features that are image features of the second iris image by inputting the second iris image to the feature extraction model, a similarity calculation unit that calculates a similarity between the extracted degraded image features and the extracted second image features, and a second parameter update unit that updates second parameters that are applied to a first parameter estimation model for estimating the first parameters based on the calculated similarity.8. The information processing system described in 7., further comprising: an authentication loss calculation unit that calculates an authentication loss using the extracted degraded image features and correct data, and a third parameter update unit that updates a third parameter that is applied to the feature extraction model based on the calculated authentication loss.9. The information processing system described in 3., further comprising: a first extraction unit that extracts degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image into a feature extraction model for extracting image features; a second extraction unit that extracts second image features that are image features of the second iris image by inputting the second iris image into the feature extraction model; an identification score estimation unit that, when one of the degraded image features or the second image feature is input, calculates an identification score including a first probability that the input image feature is the degraded image feature and a second probability that the input image feature is the second image feature; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and a fifth parameter update unit that updates a fifth parameter for calculating the identification score based on the calculated adversarial loss. 3. The information processing system according to 3., further comprising: an identification score estimation unit that, when either the degraded iris image or the second iris image is input, calculates an identification score including a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and a fifth parameter update unit that updates a fifth parameter for calculating the identification score based on the calculated adversarial loss. 11. The information processing system according to 6., wherein the first parameter is estimated further using first parameter definition information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality.12. The information processing system described in 6., wherein the first parameter is estimated further using a first parameter estimation model that has trained to estimate the first parameter, and the first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input. 13. An information processing device comprising: a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image of lower quality than the first iris image; and a generation unit that generates a degraded iris image, based on the first iris image and the second iris image, by degrading the first iris image so that its quality approaches that of the second iris image. 14. The information processing device described in 13., wherein the generation unit comprises: an estimation unit that estimates first parameters used in generating a degraded iris image, based on the first iris image and the second iris image, by degrading the first iris image so that its quality approaches that of the second iris image; and a degradation processing unit that generates the degraded iris image using the estimated first parameters. 15. The information processing device according to 14., wherein bringing the first iris image closer to the quality of the second iris image means bringing a first position in a feature space of a first image feature extracted from the first iris image closer to a second position in the feature space of a second image feature extracted from the second iris image. 16. The information processing device according to 13., wherein the generation unit comprises: a second quality estimation unit that estimates a second quality that is the quality of the second iris image based on the second iris image; a first parameter estimation unit that estimates first parameters used to generate a degraded iris image by degrading the first iris image so as to bring it closer to the quality of the second iris image, based on the second quality; and a degradation processing unit that applies the estimated first parameters to an image degradation model for image degradation and inputs the first iris image to the image degradation model to generate the degraded iris image. 17. The information processing device according to any one of 13. to 16., wherein the degraded iris image is used for training a machine learning model or for iris authentication.18. The information processing device described in 14. or 15., wherein the estimation unit includes: a first quality estimation unit that estimates a first quality that is the quality of the first iris image based on the first iris image; a second quality estimation unit that estimates a second quality that is the quality of the second iris image based on the second iris image; and a first parameter estimation unit that estimates the first parameter using the estimated first quality and the estimated second quality. 19. The information processing device according to 15., further comprising: a first extraction unit that extracts degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image to a feature extraction model for extracting image features; a second extraction unit that extracts second image features that are image features of the second iris image by inputting the second iris image to the feature extraction model; a similarity calculation unit that calculates a similarity between the extracted degraded image features and the extracted second image features; and a second parameter update unit that updates a second parameter that is applied to a first parameter estimation model for estimating the first parameter based on the calculated similarity. 20. The information processing device according to 19., further comprising: an authentication loss calculation unit that calculates an authentication loss using the extracted degraded image features and correct data; and a third parameter update unit that updates a third parameter that is applied to the feature extraction model based on the calculated authentication loss.21. The information processing device according to 15., further comprising: a first extraction unit that extracts degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image into a feature extraction model for extracting image features; a second extraction unit that extracts second image features that are image features of the second iris image by inputting the second iris image into the feature extraction model; a discrimination score estimation unit that, when one of the degraded image features and the second image feature is input, calculates a discrimination score including a first probability that the input image feature is the degraded image feature and a second probability that the input image feature is the second image feature; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated discrimination score; a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and a fifth parameter update unit that updates a fifth parameter for calculating the discrimination score based on the calculated adversarial loss. 15. The information processing device according to 15., further comprising: an identification score estimation unit that, when either the degraded iris image or the second iris image is input, calculates an identification score including a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and a fifth parameter update unit that updates a fifth parameter for calculating the identification score based on the calculated adversarial loss. 23. The information processing device according to 18., wherein the first parameter is estimated further using first parameter definition information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality.24. The information processing device described in 18., wherein the first parameter is estimated further using a first parameter estimation model that has trained to estimate the first parameter, and the first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input. 25. An information processing method, wherein one or more computers: acquire a first iris image; acquire a second iris image of lower quality than the first iris image; and generate a degraded iris image based on the first iris image and the second iris image, by degrading the first iris image so that its quality approaches that of the second iris image. 26. The information processing method described in 25., wherein generating the degraded iris image includes: estimating, based on the first iris image and the second iris image, a first parameter used to generate a degraded iris image by degrading the first iris image so that its quality approaches that of the second iris image; and generating the degraded iris image using the estimated first parameter. 27. The information processing method according to 26., wherein bringing the first iris image closer to the quality of the second iris image means bringing a first position in a feature space of a first image feature extracted from the first iris image closer to a second position in the feature space of a second image feature extracted from the second iris image. 28. The information processing method according to 25., wherein generating the degraded iris image includes estimating a second quality that is the quality of the second iris image based on the second iris image, estimating first parameters used to generate a degraded iris image that has degraded the first iris image so as to bring it closer to the quality of the second iris image based on the second quality, and applying the estimated first parameters to an image degradation model for image degradation, and inputting the first iris image to the image degradation model to generate the degraded iris image. 29. The information processing method according to any one of 25. to 28., wherein the degraded iris image is used for training a machine learning model or for iris authentication.30. The information processing method described in 26. or 27., wherein generating the degraded iris image includes: estimating a first quality that is the quality of the first iris image based on the first iris image; estimating a second quality that is the quality of the second iris image based on the second iris image; and estimating the first parameter using the estimated first quality and the estimated second quality. 31. The information processing method described in 27., further comprising: inputting the generated degraded iris image into a feature extraction model for extracting image features to extract degraded image features that are image features of the degraded iris image; inputting the second iris image into the feature extraction model to extract second image features that are image features of the second iris image; calculating a similarity between the extracted degraded image features and the extracted second image features; and updating a second parameter that is applied to a first parameter estimation model for estimating the first parameter based on the calculated similarity. 32. 33. The information processing method according to 31., further comprising: calculating an authentication loss using the extracted degraded image feature and correct data; and updating a third parameter applied to the feature extraction model based on the calculated authentication loss. 27. The information processing method according to 27, further comprising: extracting degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image into a feature extraction model for extracting image features; extracting second image features that are image features of the second iris image by inputting the second iris image into the feature extraction model; calculating, when either the degraded image feature or the second image feature is input, an identification score including a first probability that the input image feature is the degraded image feature and a second probability that the input image feature is the second image feature; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; updating a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and updating a fifth parameter for calculating the identification score based on the calculated adversarial loss.34. The information processing method according to 27., further comprising: when either the degraded iris image or the second iris image is input, calculating an identification score including a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image; calculating an adversarial loss based on the calculated identification score; updating a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and updating a fifth parameter for calculating the identification score based on the calculated adversarial loss. 35. The information processing method according to 30., wherein the first parameter is estimated further using first parameter defining information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality. 36. The information processing method according to 30., wherein the first parameter is estimated further using a first parameter estimation model that has trained to estimate the first parameter, and the first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input. 37. A program causing one or more computers to execute the following steps: acquire a first iris image; acquire a second iris image of lower quality than the first iris image; and generate a degraded iris image based on the first and second iris images, by degrading the first iris image so as to approximate the quality of the second iris image. 38. The program described in 37., wherein generating the degraded iris image includes estimating, based on the first and second iris images, first parameters used in generating the degraded iris image by degrading the first iris image so as to approximate the quality of the second iris image, and generating the degraded iris image using the estimated first parameters. 39. The program described in 38., wherein bringing the quality of the first iris image closer to that of the second iris image means bringing a first position in feature space of a first image feature extracted from the first iris image closer to a second position in the feature space of a second image feature extracted from the second iris image.40. The program described in 37., wherein generating the degraded iris image includes: estimating a second quality, which is the quality of the second iris image, based on the second iris image; estimating a first parameter, based on the second quality, to be used in generating a degraded iris image by degrading the first iris image so as to approximate the quality of the second iris image; applying the estimated first parameter to an image degradation model for image degradation, and inputting the first iris image to the image degradation model, thereby generating the degraded iris image. 41. The program described in any one of 37. to 40., wherein the degraded iris image is used for training a machine learning model or for iris authentication. 42. The program described in 38. or 39., wherein generating the degraded iris image includes: estimating a first quality, which is the quality of the first iris image, based on the first iris image; estimating a second quality, which is the quality of the second iris image, based on the second iris image; and estimating the first parameter using the estimated first quality and the estimated second quality. 43. The program according to 39., further causing the program to execute the following: extracting degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image into a feature extraction model for extracting image features; extracting second image features that are image features of the second iris image by inputting the second iris image into the feature extraction model; calculating a similarity between the extracted degraded image features and the extracted second image features; and updating a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated similarity. 44. The program according to 43., further causing the program to execute the following: calculating an authentication loss using the extracted degraded image features and correct data; and updating a third parameter applied to the feature extraction model based on the calculated authentication loss.45. The program described in 39. is further configured to execute the following: extracting degraded image features that are image features of the degraded iris image by inputting the generated degraded iris image into a feature extraction model for extracting image features; extracting second image features that are image features of the second iris image by inputting the second iris image into the feature extraction model; calculating, when either the degraded image feature or the second image feature is input, an identification score including a first probability that the input image feature is the degraded image feature and a second probability that the input image feature is the second image feature; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; updating a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and updating a fifth parameter for calculating the identification score based on the calculated adversarial loss. 46. The program according to item 39. is further configured to execute the following when either the degraded iris image or the second iris image is input: calculating an identification score including a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image; calculating an adversarial loss based on the calculated identification score; updating a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and updating a fifth parameter for calculating the identification score based on the calculated adversarial loss. 47. The program according to item 42., in which the first parameter is estimated further using first parameter defining information that defines the first parameter according to each of a plurality of patterns of the first quality and the second quality. 48. The program according to item 42., in which the first parameter is estimated further using a first parameter estimation model that has trained to estimate the first parameter, and the first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input. 49. 37. 48. A recording medium on which the program according to any one of 47. to 48. is recorded.

[0276] 100, 200, 300, 400, 500, 700, 800 Information processing device 101 First acquisition unit 102 Second acquisition unit 103 Generation unit 104 Estimation unit 104a First quality estimation unit 104b Second quality estimation unit 104c First parameter estimation unit 105 Degradation processing unit 201 First extraction unit 202 Second extraction unit 203 Similarity calculation unit 204 Second parameter update unit 301 Authentication loss calculation unit 302 Third parameter update unit 401 Discrimination score estimation unit 402 Adversarial loss calculation unit 403, 503 Second parameter update unit 404, 504 Fifth parameter update unit 501 Discrimination score estimation unit 502 Adversarial loss calculation unit 604c First parameter estimation unit 605 Degradation processing unit 701 Image super-resolution unit 702 Third extraction unit 703 Fourth extraction unit 704 Authentication unit 705 Image distance loss calculation unit 706 Sixth parameter update unit

Claims

1. An information processing system comprising: a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image having lower quality than the first iris image; and a generation unit that generates a degraded iris image obtained by degrading the first iris image so as to approximate the quality of the second iris image based on the first iris image and the second iris image.

2. The information processing system according to claim 1, wherein the generation unit includes: an estimation unit that estimates a first parameter used for generating a degraded iris image obtained by degrading the first iris image so as to approximate the quality of the second iris image based on the first iris image and the second iris image; and a degradation processing unit that generates the degraded iris image using the estimated first parameter.

3. The information processing system according to claim 2, wherein approximating the quality of the first iris image to that of the second iris image means bringing a first position in a feature amount space of a first image feature amount extracted from the first iris image closer to a second position in the feature amount space of a second image feature amount extracted from the second iris image.

4. The information processing system according to claim 1, wherein the generation unit includes: a second quality estimation unit that estimates a second quality which is the quality of the second iris image based on the second iris image; a first parameter estimation unit that estimates a first parameter used for generating a degraded iris image obtained by degrading the first iris image so as to approximate the quality of the second iris image based on the second quality; and a degradation processing unit that generates the degraded iris image by applying the estimated first parameter to an image degradation model for degrading an image and inputting the first iris image to the image degradation model.

5. The degraded iris image according to any one of claims 1 to 4 is used for training a machine learning model or for iris authentication.

6. The information processing system according to claim 2 or 3, wherein the estimation unit includes: a first quality estimation unit that estimates a first quality which is the quality of the first iris image based on the first iris image; a second quality estimation unit that estimates a second quality which is the quality of the second iris image based on the second iris image; and a first parameter estimation unit that estimates the first parameter using the estimated first quality and the estimated second quality.

7. A first extraction unit that extracts a degraded image feature amount, which is an image feature amount of the generated degraded iris image, by inputting the generated degraded iris image into a feature amount extraction model for extracting an image feature amount; a second extraction unit that extracts a second image feature amount, which is an image feature amount of the second iris image, by inputting the second iris image into the feature amount extraction model; a similarity calculation unit that calculates a similarity between the extracted degraded image feature amount and the extracted second image feature amount; and a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated similarity. The information processing system according to claim 3, further comprising the above components.

8. An authentication loss calculation unit that calculates an authentication loss using the extracted degraded image feature amount and correct data; and a third parameter update unit that updates a third parameter applied to the feature amount extraction model based on the calculated authentication loss. The information processing system according to claim 7, further comprising the above components.

9. A first extraction unit that extracts a degraded image feature amount, which is an image feature amount of the generated degraded iris image, by inputting the generated degraded iris image into a feature amount extraction model for extracting an image feature amount; a second extraction unit that extracts a second image feature amount, which is an image feature amount of the second iris image, by inputting the second iris image into the feature amount extraction model; an identification score estimation unit that calculates an identification score including a first probability that the input image feature amount is the degraded image feature amount and a second probability that the input image feature amount is the second image feature amount when any one of the degraded image feature amount and the second image feature amount is input; an adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; a second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; and a fifth parameter update unit that updates a fifth parameter for calculating the identification score based on the calculated adversarial loss. The information processing system according to claim 3, further comprising the above components.

10. When any one of the degraded iris image and the second iris image is input, an identification score estimation unit that calculates an identification score including a first probability that the input image is the degraded iris image and a second probability that the input image is the second iris image; An adversarial loss calculation unit that calculates an adversarial loss based on the calculated identification score; A second parameter update unit that updates a second parameter applied to a first parameter estimation model for estimating the first parameter based on the calculated adversarial loss; A fifth parameter update unit that updates a fifth parameter for calculating the identification score based on the calculated adversarial loss; The information processing system according to claim 3, further comprising:

11. The information processing system according to claim 6, wherein the first parameter is further estimated using first parameter specification information that specifies the first parameter corresponding to each of a plurality of patterns of the first quality and the second quality.

12. The first parameter is further estimated using a first parameter estimation model that has been trained to estimate the first parameter, and the first parameter estimation model is a machine learning model that outputs the first parameter when the first quality and the second quality are input. The information processing system according to claim 6.

13. An information processing apparatus comprising: a first acquisition unit that acquires a first iris image; a second acquisition unit that acquires a second iris image having a lower quality than the first iris image; and a generation unit that generates a degraded iris image obtained by degrading the first iris image so as to approach the quality of the second iris image based on the first iris image and the second iris image.

14. An information processing method in which one or more computers acquire a first iris image, acquire a second iris image having a lower quality than the first iris image, and generate a degraded iris image obtained by degrading the first iris image so as to approach the quality of the second iris image based on the first iris image and the second iris image.

15. A recording medium having recorded thereon a program for causing one or more computers to acquire a first iris image, acquire a second iris image having a lower quality than the first iris image, and generate a degraded iris image obtained by degrading the first iris image so as to approach the quality of the second iris image based on the first iris image and the second iris image.

Images