Information processing systems, information processing methods, and programs

Abstract

An information processing system (100) comprises a learning conversion unit (111) and a learning estimation unit (112). The learning conversion unit (111) converts a learning image on the basis of the learning image and a learning evaluation value, which is an evaluation value of the learning image, to generate a learning converted image. The learning estimation unit (112) uses the learning converted image and a group of learning conversion parameters used to generate the learning image to train a conversion estimation model for estimating a group of conversion parameters for converting a target image, thereby generating a trained conversion estimation model.

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 Document 1 discloses an image processing device that adaptively performs correction processing and event detection in response to the influence of disturbances on an input image. The image processing device described in Patent Document 1 performs correction processing on an input image in accordance with disturbance information. The disturbances described in Patent Document 1 include, for example, haze, heat haze, and noise. The disturbance information described in Patent Document 1 expresses the strength of the disturbance (the degree of degradation of the input image) in three levels (high, medium, low, etc.).

[0003] For example, Patent Literature 2 discloses an authentication device for solving the problem of insufficient resolution of an image to be authenticated, such as when performing iris authentication using a facial image. The authentication device described in Patent Literature 2 acquires an out-of-focus iris image as the authentication target image, and acquires an in-focus iris image with the same resolution as the authentication target image as a comparison image. The authentication device described in Patent Literature 2 also receives input of feature amounts of the out-of-focus iris image and converts the feature amounts so as to reduce the difference between the feature amounts of the out-of-focus iris image and the in-focus iris image.

[0004] International Publication No. WO 2017 / 047494 International Publication No. WO 2022 / 195819

[0005] The present disclosure aims to improve upon the techniques described in the prior art documents mentioned above.

[0006] According to one aspect of the present disclosure, an information processing system is provided, comprising: a training conversion means for converting a training image based on a training image and a training evaluation value, which is an evaluation value of the training image, to generate a training converted image; and a training estimation means for using the training converted image and a group of training transformation parameters used to generate the training image to train a transformation estimation model for estimating a group of transformation parameters for converting a target image, and generating the trained transformation estimation model.

[0007] According to one aspect of the present disclosure, there is provided an information processing device comprising: a training conversion means for converting a training image based on a training image and a training evaluation value that is an evaluation value of the training image, to generate a training converted image; and a training estimation means for using the training converted image and a group of training transformation parameters used to generate the training image to train a transformation estimation model for estimating a group of transformation parameters for converting a target image, and generating the trained transformation estimation model.

[0008] According to one aspect of the present disclosure, an information processing method is provided in which one or more computers transform a training image based on the training image and a training evaluation value that is an evaluation value of the training image to generate a training converted image, and use the training converted image and a group of training transformation parameters used to generate the training image to train a transformation estimation model for estimating a group of transformation parameters for transforming a target image, and generate the trained transformation estimation model.

[0009] According to one aspect of the present disclosure, a recording medium having recorded thereon a program for causing one or more computers to perform the following operations: transforming a training image based on a training image and a training evaluation value, which is an evaluation value of the training image, to generate a training converted image; training a conversion estimation model for estimating a set of conversion parameters for converting a target image using the training converted image and a set of training conversion parameters used to generate the training image; and generating the trained conversion estimation model.

[0010] 1 is a diagram illustrating a first example of a configuration of an information processing system according to the present disclosure. FIG. 1 is a diagram illustrating a first example of a functional configuration of an information processing device according to the present disclosure. FIG. 2 is a flowchart illustrating a first example of information processing according to the present disclosure. FIG. 3 is a diagram illustrating a third example of a configuration of an information processing system according to the present disclosure. FIG. 4 is a diagram illustrating a first example of an operation processing according to the present disclosure. FIG. 5 is a diagram illustrating a fifth example of a configuration of an information processing system according to the present disclosure. FIG. 6 is a diagram illustrating a functional configuration of a operation processing unit according to the present disclosure. FIG. 7 is a flowchart illustrating a second example of an operation processing according to the present disclosure.

[0011] Hereinafter, embodiments according to 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.

[0012] [Embodiment 1] For example, when performing iris authentication, the authentication accuracy may be reduced due to low image quality of a target image (described in detail later) including the iris of the person to be authenticated. Factors that cause the image quality of the target image to be low are not limited to haze, heat haze, and noise described in Patent Document 1 and defocus described in Patent Document 2, but include various other factors such as lighting conditions, camera control values, and subject movement. Furthermore, the factors that cause the image quality to be low are not limited to one of these, and may be multiple.

[0013] In the techniques described in Patent Documents 1 and 2, when the image quality of a target image is low due to one or more of various factors, there is a risk that the target image cannot be converted into an image of image quality appropriate for its intended use. In view of such circumstances, an object of the present disclosure is to provide an information processing system, an information processing device, an information processing method, a recording medium, etc. for converting a target image into an image of image quality appropriate for its intended use.

[0014] (Overview) As shown in FIG. 1 , an example of the functional configuration of an information processing system 100 according to the present disclosure includes a training conversion unit 111 and a training estimation unit 112 .

[0015] The conversion unit for learning 111 converts the training image based on the training image and a training evaluation value that is an evaluation value of the training image, to generate a converted training image.

[0016] The learning estimation unit 112 uses the learning transformation image and the group of learning transformation parameters used to generate the learning image to train a transformation estimation model for estimating a group of transformation parameters for transforming the target image, and generates a trained transformation estimation model.

[0017] According to this information processing system 100, it is possible to convert a target image into an image of a quality suitable for its intended use.

[0018] As shown in FIG. 2 , an example of the functional configuration of the information processing device 110 according to the present disclosure includes a training conversion unit 111 and a training estimation unit 112 .

[0019] The conversion unit for learning 111 converts the training image based on the training image and a training evaluation value that is an evaluation value of the training image, to generate a converted training image.

[0020] The learning estimation unit 112 uses the learning transformation image and the group of learning transformation parameters used to generate the learning image to train a transformation estimation model for estimating a group of transformation parameters for transforming the target image, and generates a trained transformation estimation model.

[0021] According to this information processing device 110, it is possible to convert a target image into an image of a quality suitable for its intended use.

[0022] The information processing according to the present disclosure includes the following processes, as shown in FIG. 3, which is an example of a flowchart.

[0023] The learning conversion unit 111 converts the learning image based on the learning image and the learning evaluation value, which is the evaluation value of the learning image, to generate a converted learning image (step S101).

[0024] The learning estimation unit 112 uses the learning transformation image and the group of learning transformation parameters used to generate the learning image to train a transformation estimation model for estimating a group of transformation parameters for transforming the target image, and generates a trained transformation estimation model (step S102).

[0025] This information processing makes it possible to convert the target image into an image of a quality suitable for its intended use.

[0026] (Detailed Example of First Embodiment) The information processing system 100 includes an information processing device 110 that performs learning of a transformation estimation model, as shown in a detailed example of the functional configuration of the information processing system 100 in Fig. 4. The transformation estimation model is a machine learning model for estimating a group of transformation parameters for transforming a target image.

[0027] 4 shows an example in which the information processing system 100 is configured from one information processing device 110. Note that the information processing system 100 may be configured from multiple information processing devices. In this case, the multiple information processing devices may be connected, for example, via a communication network that is wired, wireless, or a combination of these, and may communicate with each other via the communication network. The multiple information processing devices may all have the same functions as the information processing system 100 according to the present disclosure.

[0028] The information processing device 110 functionally includes, for example, a training conversion unit 111 , a training estimation unit 112 , and a training information storage unit 113 .

[0029] The training transformation unit 111 transforms the training image based on the training image and the training evaluation value to generate a training transformed image. For example, the training transformation unit 111 performs transformation to improve the image quality of the training image based on the training image and the training evaluation value to generate a training transformed image. For this transformation, a machine learning model (image transformation model) for performing transformation to improve the image quality of the target image may be used, or a group of transformation formulas composed of one or more predetermined functions may be used. If the image transformation model is a machine learning model, it is preferable that the image transformation model be configured using a neural network (e.g., a convolutional neural network).

[0030] Both the training image and the target image are images that include an object that corresponds to the intended use of the target image. The training image is an image that is prepared in advance for training a machine learning model such as an image conversion model. In contrast, the target image is an image that is used in the operation of a machine learning model that has trained using the training image or the like. In detail, for example, the target image is an image of a subject that includes an object that corresponds to the intended use, photographed with a camera or the like (not shown).

[0031] When the purpose of use is iris authentication, the object is an iris. When the purpose of use is iris authentication, each of the learning image and the target image is an image including an iris, such as a monocular image including one eye of a person or the like, a binocular image including both eyes of a person or the like, or a facial image. When the purpose of use is iris authentication, the target image is, for example, an image including the iris of a person or the like to be authenticated.

[0032] The purpose of use is not limited to iris authentication, and the object is not limited to the iris.

[0033] For example, the purpose of use may be biometric authentication such as face authentication, fingerprint authentication, vein authentication, etc. in addition to iris authentication. When the purpose of use is face authentication, fingerprint authentication, or vein authentication, the object is a face, fingerprint, vein, etc. Furthermore, biometric authentication is typically used to authenticate people, but it may also be used to authenticate animals such as dogs and snakes.

[0034] Furthermore, the purpose of use is not limited to biometric authentication, and may be, for example, object detection for detecting general objects. Object detection may be, for example, detection or classification of objects for autonomous driving, detection or classification of products lined up on a store shelf, etc. When the purpose of use is detection or classification of objects for autonomous driving, the object is, for example, at least one of a vehicle, a person, a building, etc. When the purpose of use is detection or classification of products, the object is, for example, at least one of a product, a store shelf, etc. The purpose of use is not limited to those exemplified here, and may be determined as appropriate. Furthermore, the object may be determined according to the purpose of use.

[0035] The learning evaluation value is an evaluation value of the learning image, and is, for example, a value obtained using reference information to be compared.

[0036] For example, the reference information may be information that serves as a reference for the learning image and is used for comparison with the learning image. In this case, the reference information may include, for example, at least one of a reference image prepared in advance, a feature amount extracted from the reference image, and the like.

[0037] Furthermore, for example, the reference information may be information corresponding to registration information that is registered in advance when authentication is performed using a target image. In this case, the registration information is information that is used for comparison with the target image when authentication is performed using the target image, and includes, for example, at least one of the registered image that is registered in advance, feature amounts extracted from the registered image, etc.

[0038] For example, the learning image and the target image are images of lower image quality than the reference image and the registered image, respectively.

[0039] The evaluation value includes at least one of the similarity with the information to be compared and the classification evaluation value.

[0040] The information to be compared is the reference information for the evaluation value of the learning image, and is the registered information for the evaluation value of the target image, which will be described later.

[0041] The similarity is, for example, the similarity between the feature extracted from the training image and the feature included in the reference information, or the similarity between the feature extracted from the target image and the feature included in the registered information. The feature may be at least one of the final feature finally output in the neural network used to extract the feature, the intermediate feature generated in the intermediate layer of the neural network, etc. The similarity between the feature may be a value calculated using an appropriate method, such as cosine similarity.

[0042] The classification evaluation value is a value indicating an evaluation of the classification result. The classification evaluation value may be, for example, at least one of an error estimate (loss), a reliability, an accuracy rate, a precision, a recall rate, etc. Examples of methods for acquiring such evaluation values ​​will be described in other embodiments. Note that the evaluation value may be a score (probability indicating the identity or object-likeness) output by a machine learning model used for biometric authentication or object detection.

[0043] Specifically, the training conversion unit 111 stores, in advance, initial values ​​of a group of conversion parameters to be used for conversion. The group of conversion parameters may be composed of one or more conversion parameters to be used for conversion to improve image quality.

[0044] The training conversion unit 111 executes a process of generating training converted images by changing the conversion parameter set so as to increase the training evaluation value. This process may be executed once or multiple times until each of the training images satisfies the generation conditions, for example. The training conversion unit 111 then outputs a training converted image that satisfies a predetermined condition from among the one or more generated training converted images, and the training conversion parameter set used to generate the training converted image. For example, the training conversion unit 111 outputs the training converted image and the training conversion parameter set to the training information storage unit 113, where they are stored.

[0045] The predetermined condition may be, for example, that the learning evaluation value is maximized or that the learning evaluation value is equal to or greater than a predetermined threshold. The above-mentioned generation condition may be a predetermined number of times or that a learning converted image that satisfies the predetermined condition is generated. Note that the predetermined condition is not limited to these examples.

[0046] A common method such as backpropagation may be applied to change the set of transformation parameters. When backpropagation is used, for example, an update amount calculated from a learning evaluation value such as a loss using a gradient descent method or the like may be backpropagated to a neural network that constitutes the image transformation model.

[0047] The learning information storage unit 113 is a storage unit for storing learning information 113a.

[0048] The training information 113a is information used to train a transformation estimation model for estimating a group of transformation parameters for transforming a target image. The training information 113a is information that associates at least one training transformed image with a group of training transformation parameters, as shown in Fig. 5, an example of the configuration of the training information 113a.

[0049] As described above, for example, the learning information 113a is stored in the learning information storage unit 113 by the learning conversion unit 111. In this case, the learning information 113a associates a training converted image that satisfies a predetermined condition with a group of training conversion parameters used to generate the training converted image.

[0050] The training estimation unit 112 trains a transformation estimation model using the training transformed image and the group of training transformation parameters used to generate the training image, and generates a trained transformation estimation model.

[0051] For example, the training estimation unit 112 may train a transformation estimation model using the training transformed image and the training transformation parameter group included in the training information 113a, and generate a trained transformation estimation model.

[0052] In detail, for example, when training a transformation estimation model, it is advisable to classify the training transformation parameter group into clusters using a general clustering method such as k-means, and perform training so as to output the training transformation parameter group corresponding to the cluster to which the training image belongs.

[0053] By inputting a target image into the transformation estimation model that has undergone such training, it is possible to, for example, estimate the cluster to which the target image belongs and estimate a set of transformation parameters corresponding to the cluster (for example, a set of transformation parameters corresponding to the cluster center).The transformation estimation model may then output the estimated set of transformation parameters as a set of transformation parameters for the target image.

[0054] (Example of Hardware Configuration) The information processing apparatus 110 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.

[0055] 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.

[0056] The processor 1020 is implemented as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0057] The memory 1030 is a main storage device realized by a RAM (Random Access Memory) or the like.

[0058] 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 device that includes the storage device 1040. The processor 1020 loads each of these program modules into the memory 1030 and executes them to realize the function corresponding to that program module.

[0059] The network interface 1050 is an interface for connecting a device equipped with the network.

[0060] The input interface 1060 is an interface for the user to input information, and is configured from, for example, a touch panel, a keyboard, a mouse, and the like.

[0061] 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.

[0062] The information processing system may be configured from a plurality of information processing devices as described above. In this case, each of the plurality of information processing devices may have a physical configuration such as that shown in FIG.

[0063] (Operations and Effects) As described above, according to this embodiment, the information processing system 100 includes the training conversion unit 111 and the training estimation unit 112 .

[0064] The conversion unit for learning 111 converts the training image based on the training image and a training evaluation value that is an evaluation value of the training image, to generate a converted training image.

[0065] The learning estimation unit 112 uses the learning transformation image and the group of learning transformation parameters used to generate the learning image to train a transformation estimation model for estimating a group of transformation parameters for transforming the target image, and generates a trained transformation estimation model.

[0066] This makes it possible to generate a trained transformation estimation model using training images converted based on training evaluation values ​​according to the intended use of the target images and the training transformation parameter set used to generate the training images. Therefore, using this trained transformation estimation model, it is possible to estimate a transformation parameter set for converting the target images to image quality according to the intended use of the target images.

[0067] Therefore, it is possible to convert the target image into an image of a quality suitable for the intended use.

[0068] According to this embodiment, the learning evaluation value is a value obtained using reference information to be compared with the learning image.

[0069] This allows a learning evaluation value to be generated using reference information according to the intended use of the target image. Then, a trained transformation estimation model can be generated using the learning image converted based on the learning evaluation value and the set of learning transformation parameters used to generate the learning image. Therefore, using this trained transformation estimation model, a set of transformation parameters for converting the target image to image quality according to the intended use of the target image can be estimated.

[0070] Therefore, it is possible to convert the target image into an image of a quality suitable for the intended use.

[0071] According to this embodiment, the evaluation value includes at least one of the similarity with the information to be compared and a value indicating an evaluation of the classification result.

[0072] This allows an evaluation value according to the intended use of the target image to be used as the training evaluation value. Then, a trained transformation estimation model can be generated using the training image converted based on the training evaluation value and the set of training transformation parameters used to generate the training image. Therefore, using this trained transformation estimation model, a set of transformation parameters for converting the target image to image quality according to the intended use of the target image can be estimated.

[0073] Therefore, it is possible to convert the target image into an image of a quality suitable for the intended use.

[0074] [Embodiment 2] An information processing device 110 constituting an information processing system 100 according to the present disclosure may include a learning matching unit 114 in addition to the above-described learning conversion unit 111, learning estimation unit 112, and learning information storage unit 113, as shown in an example of its functional configuration in FIG. 7 .

[0075] The learning comparison unit 114 compares the learning image with the reference information and calculates a learning evaluation value.

[0076] In detail, for example, the learning collation unit 114 includes a learning extraction unit 114a and a learning evaluation unit 114b, as shown in an example of the functional configuration of FIG.

[0077] The learning extraction unit 114a extracts learning features, which are feature quantities of the learning images, from the learning images. The learning extraction unit 114a extracts learning features from the learning images using a machine learning model (feature extraction model) configured by, for example, a neural network.

[0078] The feature extraction model may be, for example, a general machine learning model that extracts and outputs features from input images such as training images. Such a feature extraction model may include, for example, a neural network used to extract the above-mentioned features. This neural network may be, for example, a convolutional neural network (CNN), but is not limited to this. Furthermore, in training the feature extraction model, for example, supervised learning may be performed using correct answer data including training images and correct answer labels based on the training images (e.g., matching results). Note that the training method for the feature extraction model is not limited to this.

[0079] The training features may include, for example, at least one of features generated in an intermediate layer of a neural network and features output from a final layer of the neural network.

[0080] The learning evaluation unit 114b compares the learning feature amount with the reference information to calculate a learning evaluation value.

[0081] For example, the reference information may include features extracted from a reference image. In this case, the learning evaluation unit 114b may calculate the similarity between the learning features and the features included in the reference information as the learning evaluation value. The learning evaluation unit 114b may also calculate the classification evaluation value as described above as the learning evaluation value.

[0082] The information processing according to the present disclosure may include the following processing, as shown in the second example of the flowchart in Figure 9. The information processing may be started, for example, in response to a user instruction, but the trigger for starting the information processing is not limited to this example.

[0083] The learning comparison unit 114 acquires the previously prepared learning image and reference information from a memory unit (not shown) or the like, and compares the learning image and the reference information to calculate a learning evaluation value (step S103).

[0084] In detail, for example, the learning evaluation value calculation process (step S103) may include the following processes as shown in FIG.

[0085] When acquiring a training image prepared in advance, the training extraction unit 114a extracts training features, which are feature amounts of the training image, from the training image (step S103a).

[0086] When the learning evaluation unit 114b acquires the previously prepared reference information, it compares the learning features extracted in step S103a with the reference information to calculate a learning evaluation value (step S103b), and returns to information processing.

[0087] The learning conversion unit 111 converts the learning image based on the learning image and the learning evaluation value calculated for the learning image in step S103, to generate a converted learning image (step S101).

[0088] For example, when the training conversion unit 111 uses an image conversion model, the image conversion model can be trained using a general method such as backpropagation from the neural network constituting the training extraction unit 114a. That is, when changing the set of conversion parameters included in the image conversion model, for example, an update amount calculated using a gradient descent method or the like from the training evaluation value such as the loss calculated in step S103b may be backpropagated to the neural network constituting the image conversion model.

[0089] The training conversion unit 111 determines whether each of the training images satisfies a predetermined generation condition (step S104). For example, if at least one of the training images does not satisfy the generation condition (step S104; No), steps S103 and S101 are executed again.

[0090] For example, if all of the training images satisfy the generation condition (step S104; Yes), the training estimation unit 112 trains a transformation estimation model using the training transformed images and the training transformation parameter group generated for each of the training images in step S103. Then, the training estimation unit 112 generates the trained transformation estimation model (step S102), and ends the information processing.

[0091] Here, the feature extraction model constituting the learning extraction unit 114a (i.e., the feature extraction model that extracts learning features in step S103a) may be a trained machine learning model.

[0092] By training the feature extraction model separately from the image transformation model, the feature extraction model can be trained to extract features with high accuracy. Then, by training the image transformation model using the highly accurate features extracted by such a feature extraction model, it becomes possible to train the image transformation model to transform images so that highly accurate features can be extracted.

[0093] The feature extraction model constituting the learning extraction unit 114a may be an untrained machine learning model. The update amount calculated using a gradient descent method or the like from the learning evaluation value, such as the loss calculated in step S103b, may be back-propagated to the neural networks constituting the feature extraction model and the image transformation model. This allows the neural networks constituting the feature extraction model and the image transformation model to be trained together. This reduces the effort required for training.

[0094] As described above, according to this embodiment, the information processing system 100 includes the learning comparison unit 114 that compares the learning image with the reference information and calculates the learning evaluation value. The learning comparison unit 114 includes the learning extraction unit 114 a and the learning evaluation unit 114 b.

[0095] The learning extraction unit 114a extracts learning features from the learning images, which are features of the learning images. The learning evaluation unit 114b compares the learning features with reference information to calculate a learning evaluation value.

[0096] This makes it possible to calculate a learning evaluation value according to the intended use of the target image. Then, a trained transformation estimation model can be generated using the training image converted based on the learning evaluation value and the set of training transformation parameters used to generate the training image. Therefore, using this trained transformation estimation model, it is possible to estimate a set of transformation parameters for converting the target image to image quality according to the intended use of the target image.

[0097] Therefore, it is possible to convert the target image into an image of a quality suitable for the intended use.

[0098] [Embodiment 3] An information processing device 110 constituting an information processing system 100 according to the present disclosure may include an estimation unit 121 and a conversion unit 122 in addition to the above-described training conversion unit 111, training estimation unit 112, and training information storage unit 113, as shown in an example of its functional configuration in FIG. 11 .

[0099] When a target image is input, the estimation unit 121 uses the trained transformation estimation model to output a group of transformation parameters for the target image.

[0100] The transformation unit 122 transforms the target image using the target image and a group of transformation parameters for the target image to generate a transformed image.

[0101] The group of transformation parameters used to generate this transformed image may be one or more.

[0102] The set of transformation parameters may be changed using the set of transformation parameters output from the estimation unit 121. In other words, the conversion unit 122 may be configured to be able to change the set of transformation parameters used for the conversion using the set of transformation parameters output from the estimation unit 121.

[0103] The information processing according to the present disclosure may include the operation processing shown in Fig. 12. For example, the operation processing may be started by acquiring a target image captured by a camera or the like (not shown) from the camera or the like, but the trigger for starting the operation processing is not limited to this example.

[0104] When a target image is input, the estimation unit 121 outputs a group of transformation parameters for the target image using the transformation estimation model generated in the above-mentioned step S102 (i.e., the transformation estimation model that has undergone training) (step S111).

[0105] The conversion unit 122 converts the target image using the target image and the group of conversion parameters generated in step S111 (i.e., the group of conversion parameters for the target image) to generate a converted image (step S112), and then terminates the operational processing.

[0106] (Actions and Effects) As described above, according to this embodiment, the information processing device 110 includes the estimation unit 121 and the conversion unit 122. When a target image is input, the estimation unit 121 outputs a set of conversion parameters for the target image using a trained conversion estimation model. The conversion unit 122 converts the target image using the target image and the set of conversion parameters for the target image to generate a converted image.

[0107] This makes it possible to estimate a group of conversion parameters for converting the target image to an image quality that suits the intended use of the target image, thereby enabling the target image to be converted to an image with an image quality that suits the intended use.

[0108] According to this embodiment, there are a plurality of sets of transformation parameters used to generate the transformed image, and the transformation parameters are changed using the outputted sets of transformation parameters.

[0109] This allows the target image to be converted using a group of conversion parameters for converting the target image to an image quality that suits the intended use of the target image, thereby making it possible to convert the target image to an image with an image quality that suits the intended use.

[0110] [Embodiment 4] An information processing device 110 constituting an information processing system 100 according to the present disclosure may include a matching unit 123 in addition to the above-described training conversion unit 111, training estimation unit 112, training information storage unit 113, estimation unit 121, and conversion unit 122, as shown in an example of its functional configuration in FIG. 13 .

[0111] The matching unit 123 matches the converted image with the registered information to be compared, and calculates a target evaluation value.

[0112] The target evaluation value is an evaluation value of the target image, and is, for example, the degree of similarity between the feature amount extracted from the target image and the feature amount included in the registered information.

[0113] In detail, for example, the collation unit 123 includes an extraction unit 123a and an evaluation unit 123b, as shown in an example of the functional configuration of FIG.

[0114] The extraction unit 123a extracts target features, which are features of the converted image, from the converted image. The extraction unit 123a performs processing using the converted image as input instead of the learning image in the learning extraction unit 114a. Except for this, the extraction unit 123a may be configured similarly to the learning extraction unit 114a. That is, the extraction unit 123a may be configured, for example, from a trained feature extraction model.

[0115] The evaluation unit 123b compares the target feature amount with the registered information to calculate a target evaluation value.

[0116] For example, assume that the registered information includes features extracted from the registered image. In this case, the evaluation unit 123b may calculate the similarity between the target features and the features included in the registered information as a target evaluation value. The target evaluation value is, for example, the similarity between the target features output from the final layer of the neural network constituting the extraction unit 123a and the features included in the registered information.

[0117] The target evaluation value may be at least one of a similarity based on the target feature amount output from the final layer of the neural network, a classification evaluation value, and the like.

[0118] The information processing according to the present disclosure may include the operation processing shown in Fig. 15. The operation processing shown in Fig. 15 includes step S113 in addition to steps S111 and S112 similar to those of the operation processing shown in Fig. 12.

[0119] The comparison unit 123 acquires pre-registered registration information, compares the converted image generated in step S112 with the acquired registration information, calculates the target evaluation value (step S113), and terminates the operational processing.

[0120] In detail, for example, the matching process (step S113) may include the following processes as shown in FIG.

[0121] The extraction unit 123a extracts target feature amounts, which are feature amounts of the converted image, from the converted image generated in step S112 (step S113a).

[0122] When the evaluation unit 123b acquires the pre-registered registered information, it compares the target feature extracted in step S113a with the registered information to calculate a target evaluation value (step S113b), and ends the operational processing.

[0123] (Operations and Effects) As described above, according to this embodiment, the information processing system 100 includes the matching unit 123 that matches the converted image with the registered information to be compared and calculates a target evaluation value that is an evaluation value of the target image.

[0124] This allows the target evaluation value to be calculated using the target image whose image quality has been converted to suit the intended use of the target image, thereby enabling the target image to be evaluated with high accuracy.

[0125] According to this embodiment, the collation unit 123 includes an extraction unit 123a and an evaluation unit 123b.

[0126] The extracting unit 123a extracts target features from the converted image, which are features of the converted image. The evaluating unit 123b compares the target features with registered information to calculate a target evaluation value.

[0127] This allows the target evaluation value to be calculated using the target image whose image quality has been converted to suit the intended use of the target image, thereby enabling the target image to be evaluated with high accuracy.

[0128] [Embodiment 5] An information processing device 110 constituting an information processing system 100 according to the present disclosure may include an authentication unit 124 in addition to the above-described training conversion unit 111, training estimation unit 112, training information storage unit 113, estimation unit 121, conversion unit 122, and matching unit 123, as shown in an example of its functional configuration in FIG. 17 .

[0129] The authentication unit 124 determines the identity of the person indicated by each of the target image and the registered information based on the target evaluation value. Hereinafter, "the identity of the person indicated by each of the target image and the registered information" will also be simply referred to as "person identity."

[0130] For example, the target evaluation value is an evaluation value calculated by the matching unit 123, and is the degree of similarity between the feature amount extracted from the target image and the feature amount included in the registration information as described above. In this case, for example, the authentication unit 124 compares the target evaluation value with a predetermined authentication threshold and determines the person's identity based on the comparison result.

[0131] In more detail, for example, if the target evaluation value is equal to or greater than the authentication threshold, the authentication unit 124 determines that the person indicated by the target image and the registered information are the same. If the target evaluation value is less than the authentication threshold, the authentication unit 124 determines that the person indicated by the target image and the registered information are not the same (i.e., they are different people). Note that this determination method may be changed as appropriate.

[0132] The information processing according to the present disclosure may include the operation processing shown in Fig. 18. The operation processing shown in Fig. 18 includes step S114 in addition to steps S111 to S113 similar to those of the operation processing shown in Fig. 15.

[0133] The authentication unit 124 determines whether the target image and the registered information represent the same person based on the target evaluation value (step S114), and ends the operation process.

[0134] (Operations and Effects) As described above, according to this embodiment, the information processing system 100 includes the authentication unit 124 that determines the identity of the person indicated by each of the target image and the registered information based on the target evaluation value.

[0135] This allows authentication to be performed using a target image whose image quality has been converted according to the intended use of the target image, thereby enabling authentication to be performed with high accuracy.

[0136] Sixth Embodiment In training the transformation estimation model, a training transformation image and a training transformation parameter group may be used to perform training to estimate a transformation parameter group for transforming an image and a degradation factor of the image.

[0137] For example, in training the transformation estimation model, the training transformation parameter group may be classified into clusters as described above, and the training transformation parameter group for the cluster to which the training image belongs may be output. In this case, the transformation estimation model may classify the training transformation parameter group into clusters and assign a label indicating the degradation factor corresponding to each cluster. Then, the transformation estimation model may output the training transformation parameter group corresponding to the cluster to which the training image belongs and the label indicating the degradation factor.

[0138] The deterioration factors corresponding to a cluster may be one or a combination of multiple factors.

[0139] When a target image is input, the transformation estimation model that has undergone such learning may output a group of transformation parameters for the target image and a label indicating the degradation factor.

[0140] (Actions and Effects) As described above, according to this embodiment, in training of the transformation estimation model, learning is performed to estimate a group of transformation parameters for transforming an image and a label indicating a degradation factor of the image, using a training transformed image and a group of training transformation parameters. When a target image is input, the trained transformation estimation model outputs a group of transformation parameters for the target image and a label indicating a degradation factor.

[0141] This makes it possible to estimate the degradation factor of the target image, including the possibility of multiple degradation factors, not just one degradation factor.

[0142] Therefore, it is possible to accurately estimate the cause of degradation of the target image. Furthermore, it is possible to take measures such as improving the shooting environment for capturing the target image according to the estimated cause of degradation, thereby improving the image quality of the target image itself.

[0143] Although the embodiments and modifications according to the present disclosure have been described above with reference to the drawings, these are merely examples of the present disclosure, and various configurations other than those described above can also be adopted.

[0144] In addition, although the flowcharts used in the above description show multiple steps (processes) in a sequential order, the order of steps performed in each embodiment is not limited to the order shown. In each embodiment, the order of steps shown in the drawings can be changed as long as it does not cause any problems in terms of content. Furthermore, the above-described embodiments and variations can be combined as long as the content is not contradictory.

[0145] Furthermore, the present disclosure may provide a program that causes one or more computers to function as the information processing system 100 or information processing device 110 described in the above-mentioned embodiment, or that causes one or more computers to execute information processing, a recording medium on which the program is recorded, etc.

[0146] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0147] 1. An information processing system comprising: a conversion means for learning that converts a training image based on a training image and a training evaluation value that is an evaluation value of the training image to generate a converted training image; and an estimation means for learning that uses the training converted image and a set of training transformation parameters used to generate the training image to train a transformation estimation model for estimating a set of transformation parameters for converting a target image, to generate the trained transformation estimation model. 2. The information processing system described in 1., in which the training evaluation value is a value obtained using reference information to be compared with the training image. 3. The information processing system described in 1. or 2., further comprising: an estimation means that, when a target image is input, uses the trained transformation estimation model to output the set of transformation parameters for the target image; and a conversion means that transforms the target image using the target image and the set of transformation parameters for the target image to generate a converted image. 4. The information processing system according to 3. further comprises: a matching means for matching the converted image with registered information to be compared and calculating a target evaluation value which is an evaluation value of the target image; and an authentication means for determining the identity of the person indicated by the target image and the registered information based on the target evaluation value. 5. The information processing system according to 3. or 4., wherein in training the transformation estimation model, the training is performed using the training converted image and the training transformation parameter group to estimate a transformation parameter group for transforming an image and a label indicating a degradation factor of the image, and the trained transformation estimation model outputs the transformation parameter group for the target image and the label indicating the degradation factor when the target image is input. 6. The information processing system according to 2. to 5. further comprises: a learning matching means for matching the training image with the reference information and calculating the learning evaluation value, the learning matching means including: a learning extraction means for extracting training features which are feature amounts of the training image from the training image, and a learning evaluation means for matching the training features with the reference information to calculate a learning evaluation value. 10. The information processing system according to claim 9, wherein:7. The information processing system described in 4., wherein the matching means includes: extraction means for extracting, from the transformed image, target features that are features of the transformed image; and evaluation means for calculating a target evaluation value by matching the target features with the registered information. 8. The information processing system described in any one of 3. to 5., wherein the transformation parameter sets used to generate the transformed image are plural and are changed using the output transformation parameter sets. 9. The information processing system described in any one of 2. to 8., wherein the evaluation value includes at least one of a similarity with the information to be compared and a value indicating an evaluation regarding the classification result. 10. An information processing device comprising: training conversion means for converting a training image based on a training image and a training evaluation value that is an evaluation value of the training image to generate a training transformed image; and training estimation means for training a transformation estimation model for estimating a transformation parameter set for transforming a target image using the training transformed image and the training transformation parameter set used to generate the training image, and generating the trained transformation estimation model. 11. The information processing device according to 10., wherein the learning evaluation value is a value obtained using reference information to be compared with the learning image. 12. The information processing device according to 10. or 11., further comprising: estimation means that, when a target image is input, outputs the set of transformation parameters for the target image using the transformation estimation model that has undergone the training; and conversion means that converts the target image using the target image and the set of transformation parameters for the target image to generate a converted image. 13. The information processing device according to 12., further comprising: comparison means that compares the converted image with registered information to be compared and calculates a target evaluation value that is an evaluation value for the target image; and authentication means that determine the identity of the person indicated by each of the target image and the registered information based on the target evaluation value.14. The information processing device according to 12. or 13., wherein the training of the transformation estimation model uses the training transformed image and the training transformation parameter group to perform training to estimate a transformation parameter group for transforming an image and a label indicating a degradation factor of the image, and when the target image is input, the trained transformation estimation model outputs the transformation parameter group for the target image and the label indicating the degradation factor. 15. The information processing device according to any one of 11. to 14., further comprising a learning comparison means that compares the training image with the reference information to calculate the learning evaluation value, and the learning comparison means includes: a learning extraction means that extracts, from the training image, learning features that are features of the training image, and a learning evaluation means that compares the learning features with the reference information to calculate the learning evaluation value. 16. 13. The information processing device according to 13, wherein the matching means includes: extraction means for extracting, from the transformed image, target features that are features of the transformed image; and evaluation means for matching the target features with the registered information to calculate a target evaluation value. 17. The information processing device according to any one of 12. to 14., wherein the transformation parameter sets used to generate the transformed image are plural and are changed using the output transformation parameter sets. 18. The information processing device according to any one of 11. to 17., wherein the evaluation value includes at least one of a similarity with the information to be compared and a value indicating an evaluation regarding the classification result. 19. An information processing method, wherein one or more computers: transform the training image based on the training image and a training evaluation value that is the evaluation value of the training image to generate a training transformed image; and train a transformation estimation model for estimating a transformation parameter set for transforming a target image using the training transformed image and the training transformation parameter set used to generate the training image, to generate the trained transformation estimation model. 20. 19. The information processing method according to 18., wherein the learning evaluation value is a value obtained using reference information that is to be compared with the learning image.21. The information processing method according to 19. or 20., further comprising: when a target image is input, outputting the set of transformation parameters for the target image using the trained transformation estimation model; and transforming the target image using the target image and the set of transformation parameters for the target image to generate a transformed image. 22. The information processing method according to 21., further comprising: comparing the transformed image with registered information to be compared, calculating a target evaluation value that is an evaluation value for the target image, and determining the identity of the person represented by each of the target image and the registered information based on the target evaluation value. 23. The information processing method according to 21. or 22, wherein, in training the transformation estimation model, learning is performed using the training transformed image and the set of training transformation parameters to estimate a set of transformation parameters for transforming an image and a label indicating a degradation factor of the image, and when the target image is input, the trained transformation estimation model outputs the set of transformation parameters for the target image and the label indicating the degradation factor. 24. The information processing method according to any one of 20. to 23., further comprising comparing the training image with the reference information to calculate the training evaluation value, wherein calculating the training evaluation value comprises: extracting training features from the training image, which are features of the training image, and comparing the training features with the reference information to calculate the training evaluation value. 25. The information processing method according to 22., wherein calculating the target evaluation value comprises: extracting target features from the converted image, which are features of the converted image, and comparing the target features with the registered information to calculate the target evaluation value. 26. The information processing method according to any one of 21. to 23., wherein the set of transformation parameters used to generate the converted image is plural and is changed using the outputted set of transformation parameters. 27. The information processing method according to any one of 20. to 25., wherein the evaluation value comprises at least one of a similarity with the information to be compared and a value indicating an evaluation of the classification result.28. A program causing one or more computers to execute the following: transforming a training image based on a training image and a training evaluation value that is an evaluation value of the training image to generate a training converted image; training a transformation estimation model for estimating a set of transformation parameters for transforming a target image using the training converted image and a set of training transformation parameters used to generate the training image, and generating the trained transformation estimation model. 29. The program described in 28., in which the training evaluation value is a value obtained using reference information with which the training image is compared. 30. The program described in 28. or 29., further causing the program to execute the following: when a target image is input, outputting the set of transformation parameters for the target image using the trained transformation estimation model; and transforming the target image using the target image and the set of transformation parameters for the target image to generate a converted image. 31. The program according to 30., further causing the program to execute the steps of: comparing the converted image with registered information to be compared, and calculating a target evaluation value that is an evaluation value of the target image; and determining the identity of the person indicated by each of the target image and the registered information based on the target evaluation value. 32. The program according to 30. or 31., wherein in training the transformation estimation model, the training is performed using the training converted image and the training transformation parameter group to estimate a group of transformation parameters for transforming an image and a label indicating a degradation factor of the image, and the trained transformation estimation model outputs the group of transformation parameters for the target image and the label indicating the degradation factor when the target image is input. 33. The program according to any one of 29. to 32., further causing the program to calculate the learning evaluation value by comparing the training image with the reference information, wherein calculating the learning evaluation value includes extracting, from the training image, training features that are features of the training image, and comparing the training features with the reference information to calculate the learning evaluation value.34. The program described in 31., wherein calculating the target evaluation value includes: extracting target features that are features of the converted image from the converted image; and calculating the target evaluation value by matching the target features with the registered information. 35. The program described in any one of 30. to 32., wherein the group of transformation parameters used to generate the converted image is plural and is changed using the output group of transformation parameters. 36. The program described in any one of 29. to 35., wherein the evaluation value includes at least one of a similarity with the information to be compared and a value indicating an evaluation related to the classification result. 37. A recording medium having recorded thereon a program for causing one or more computers to execute the following: transforming the training image based on the training image and a training evaluation value that is an evaluation value of the training image to generate a training converted image; training a transformation estimation model for estimating a group of transformation parameters for transforming a target image using the training converted image and the training transformation parameter group used to generate the training image, and generating the trained transformation estimation model. 38. 39. A recording medium having recorded thereon the program described in 37., wherein the learning evaluation value is a value obtained using reference information with which the learning image is compared. 39. A recording medium having recorded thereon the program described in 37. or 38., which further causes a computer to execute the following: when a target image is input, outputting the set of transformation parameters for the target image using the transformation estimation model that has performed the training, and transforming the target image using the target image and the set of transformation parameters for the target image to generate a transformed image. 40. A recording medium having recorded thereon the program described in 39., which further causes a computer to execute the following: collating the transformed image with registered information with which it is compared, calculating a target evaluation value that is an evaluation value for the target image, and determining the identity of the person indicated by each of the target image and the registered information based on the target evaluation value.41. A recording medium having recorded thereon the program described in 39. or 40., wherein the training of the transformation estimation model uses the training transformed image and the training transformation parameter group to perform training to estimate a transformation parameter group for transforming an image and a label indicating a degradation factor of the image, and when the target image is input, the trained transformation estimation model outputs the transformation parameter group for the target image and the label indicating the degradation factor. 42. A recording medium having recorded thereon the program described in any one of 38. to 41., further causing the program to compare the training image with the reference information to calculate the learning evaluation value, wherein calculating the learning evaluation value includes extracting, from the training image, learning features that are features of the training image, and comparing the learning features with the reference information to calculate the learning evaluation value. 43. Calculating the target evaluation value includes extracting, from the transformed image, target features that are features of the transformed image, and comparing the target features with the registered information to calculate the target evaluation value. 44. A recording medium on which the program described in any one of 39. to 41. is recorded, in which the group of transformation parameters used to generate the transformed image is plural and is changed using the output group of transformation parameters. 45. A recording medium on which the program described in any one of 38. to 44. is recorded, in which the evaluation value includes at least one of a similarity with the information to be compared and a value indicating an evaluation of the classification result.

[0148] REFERENCE SIGNS LIST 100 Information processing system 110 Information processing device 111 Conversion unit for learning 112 Estimation unit for learning 113 Learning information storage unit 113a Learning information 114 Matching unit for learning 114a Extraction unit for learning 114b Evaluation unit for learning 121 Estimation unit 122 Conversion unit 123 Matching unit 123a Extraction unit 123b Evaluation unit 124 Authentication unit

Claims

1. A training conversion means that converts a training image to generate a training converted image based on a training image and a training evaluation value which is an evaluation value of the training image, The system comprises a learning estimation means that uses the aforementioned training transformation image and the training transformation parameter set used to generate the training image to train a transformation estimation model for estimating a set of transformation parameters for transforming a target image, and thereby generates the trained transformation estimation model. Information processing system.

2. The aforementioned learning evaluation value is a value obtained using reference information to be compared with the learning image. The information processing system according to claim 1.

3. When a target image is input, the estimation means outputs the set of transformation parameters for the target image using the transformation estimation model that has been trained, The system further comprises a conversion means that uses the target image and the set of conversion parameters for the target image to convert the target image and generate a converted image. The information processing system according to claim 1 or 2.

4. A comparison means that compares the converted image with the registered information to be compared and calculates a target evaluation value, which is the evaluation value of the target image. The system further comprises authentication means for determining the identity of the person represented by the target image and the registration information, based on the target evaluation value. The information processing system according to claim 3.

5. In the training of the transformation estimation model, the training transformation image and the training transformation parameter set are used to perform training to estimate a set of transformation parameters for transforming an image and labels indicating the degradation factors of the image. When the trained transformation estimation model receives the target image as input, it outputs the transformation parameter set for the target image and a label indicating the degradation factor. The information processing system according to claim 3.

6. The system further comprises a learning comparison means that compares the learning image with the reference information and calculates the learning evaluation value, The aforementioned learning matching means is A learning extraction means for extracting learning features, which are the features of the learning image, from the aforementioned learning image, Includes a learning evaluation means that calculates a learning evaluation value by comparing the learning features with the reference information. The information processing system according to claim 2.

7. The aforementioned matching means is An extraction means for extracting target features, which are feature quantities of the converted image, from the converted image, Includes an evaluation means that calculates a target evaluation value by comparing the target feature quantity with the registered information. The information processing system according to claim 4.

8. The group of transformation parameters used to generate the transformed image is a plurality and is modified using the outputted group of transformation parameters. The information processing system according to claim 3.

9. One or more computers, Based on the training image and the training evaluation value which is the evaluation value of the training image, the training image is transformed to generate a transformed training image. Using the aforementioned training transformation image and the training transformation parameter set used to generate the training image, a transformation estimation model for estimating a set of transformation parameters for transforming a target image is trained to generate the trained transformation estimation model. Information processing methods.

10. On one or more computers, Based on the training image and the training evaluation value which is the evaluation value of the training image, the training image is transformed to generate a transformed training image. A program for performing the following actions: training a transformation estimation model to estimate a set of transformation parameters for transforming a target image, using the aforementioned training transformation image and the set of training transformation parameters used to generate the training image; and generating the trained transformation estimation model.