Evaluation device, control program, evaluation system, model construction device, and evaluation method

Abstract

To evaluate authenticity of a medium in accordance with an attribute, or the like, of the medium.SOLUTION: An evaluation apparatus (3) includes: an image acquisition unit (351) which acquires a target image (IMGO); an information acquisition unit (352) which acquires acquisition information; a model selection unit (353) which selects one of a first evaluation model (232) and a second evaluation model (233) in accordance with the acquisition information; and an index calculation unit (354) which calculates a first or second index using the evaluation model selected by the model selection unit (353).SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an evaluation device, a control program, an evaluation system, a model construction device, and an evaluation method for evaluating the authenticity of a medium displayed in an image. [Background technology]

[0002] Conventionally, there have been known techniques for evaluating the authenticity of media such as gift certificates, various coupons, and commuter passes using images of the media. "Evaluation of authenticity" refers to indicating how close the media to be evaluated (hereinafter referred to as "target media") is to being judged to be genuine. For example, Patent Document 1 discloses a system for determining the authenticity of gift certificates by comparing gift certificate characteristic information held by the gift certificate received from a customer with gift certificate characteristic information for verification. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-70380 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the authenticity determination system disclosed in Patent Document 1 merely determines the authenticity of gift certificates received from customers, that is, it uniformly evaluates the authenticity of the gift certificate with a binary choice of 100% or 0%. Here, "100% authenticity" means that there is a 100% probability that the gift certificate received from the customer is genuine, that is, the gift certificate is genuine. "0% authenticity" means that there is a 0% probability that the gift certificate received from the customer is genuine, that is, the gift certificate is a fake.

[0005] Therefore, the authenticity determination system disclosed in Patent Document 1 is not necessarily sufficient in terms of reflecting the attributes of the voucher received from the customer and the needs of the person determining the authenticity when assessing its authenticity. Here, the attributes of the voucher include the type of voucher, validity period, area where it can be used, and stores where it can be used.

[0006] One aspect of the present invention aims to evaluate the authenticity of media in general, including paper media such as gift certificates, according to the attributes of the media or the needs of the person assessing the authenticity. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, an evaluation device according to one embodiment of the present invention comprises an image acquisition unit that acquires a target image of a target medium, an information acquisition unit that acquires at least one of attribute information indicating the attributes of the target medium and precision information regarding the precision of the evaluation of the authenticity of the target medium, and an index calculation unit that calculates an index regarding the evaluation of the authenticity of the target medium by referring to the image acquired by the image acquisition unit and the information acquired by the information acquisition unit, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information.

[0008] In order to solve the above-mentioned problems, an evaluation system according to one embodiment of the present invention comprises an image acquisition unit that acquires a target image of a target medium, an information acquisition unit that acquires at least one of attribute information indicating the attributes of the target medium and precision information regarding the precision of the evaluation of the authenticity of the target medium, and an index calculation unit that calculates an index regarding the evaluation of the authenticity of the target medium by referring to the image acquired by the image acquisition unit and the information acquired by the information acquisition unit, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information.

[0009] In order to solve the above problem, a model construction device according to one embodiment of the present invention comprises a training data acquisition unit that acquires a dataset including an image of a medium as training data, and a model construction unit that performs machine learning using the training data to construct multiple evaluation models that use a target image of the target medium as input data and output an index related to the evaluation of the authenticity of the target medium at a predetermined number of levels, and the model construction unit constructs two or more evaluation models among the multiple evaluation models, each having a different number of levels for the index related to the evaluation of the authenticity of the target medium.

[0010] In order to solve the above-mentioned problems, an evaluation method according to one aspect of the present invention includes: an image acquisition step of acquiring a target image obtained by capturing an image of a target medium; an information acquisition step of acquiring at least one of attribute information indicating attributes of the target medium and precision information regarding the precision of an evaluation of the authenticity of the target medium; and an index calculation step of calculating an index relating to the evaluation of the authenticity of the target medium by referring to the acquired image acquired in the image acquisition step and the acquired information acquired in the information acquisition step, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information. [Effects of the Invention]

[0011] According to one aspect of the present invention, the authenticity of a medium can be evaluated according to the attributes of the medium or the needs of the evaluator. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a block diagram showing an example of a configuration of a main part of an evaluation system according to one embodiment of the present invention and first to fourth modified examples thereof. [Figure 2] FIG. 2 is a diagram illustrating an example of the surface of a reference medium. [Figure 3] FIG. 10 is a diagram illustrating an example of the back side of a reference medium. [Figure 4] 10 is a flowchart illustrating an example of processing by an evaluation device according to an embodiment of the present invention. [Figure 5] FIG. 4 is a diagram showing an example of an evaluation result displayed on a display unit of the evaluation device. [Figure 6] 10 is a diagram showing an example of a state in which a target image of a target medium that is an evaluation target of the evaluation result is displayed on the display unit. FIG. [Figure 7] 10 is a diagram showing another example of a state in which a non-target image of a non-target medium that is an evaluation target of the evaluation result is displayed on the display unit. FIG. [Figure 8] FIG. 10 is a diagram showing an example of a result of whether or not use is approved, displayed on the display unit. [Figure 9] 10 is a flowchart showing an example of processing by an evaluation device according to a first modified example of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0013] [Scenes where the evaluation system can be applied] In this embodiment, a case will be described in which the administrative office of a shopping mall uses the evaluation system 100 according to one embodiment of the present invention for the purpose of managing the use of paper premium gift certificates (hereinafter abbreviated as "gift certificates") that can be used at member stores in the shopping mall. Specifically, a case will be described in which an employee at the administrative office uses the evaluation system 100 to identify the authenticity of a medium that has been applied for use as a gift certificate at a member store and to decide whether to approve the use of the medium. Hereinafter, the medium that has been applied for use as a gift certificate at a member store will be referred to as the "target medium MO," and the employee at the administrative office will be referred to as the "evaluator."

[0014] The application of the evaluation system 100 is not limited to the case of gift certificates mentioned above. The evaluation system 100 can be used to manage the use of various media such as various coupons, local currencies, commuter passes, admission tickets, vaccination tickets, prepaid cards, and three-dimensional objects (for example, fresh foods such as fish and vegetables).

[0015] [Configuration of evaluation system] The configuration of the evaluation system 100 will be described with reference to Figs. 1 to 3. The evaluation system 100 is a system for evaluating the authenticity of a target medium MO (see Fig. 6). "Authenticity" is a concept that indicates how close the object being evaluated is to the real thing. As shown in Fig. 1, the evaluation system 100 includes an image generation device 1, a model construction device 2, and an evaluation device 3.

[0016] <Configuration of image generation device> The image generating device 1 generates a target image IMGO (see FIG. 6), which is an image captured of a target medium MO. The image generating device 1 also generates a non-target image IMGN (see FIG. 7), which is an image captured of a medium other than the target medium MO. Hereinafter, a medium other than the target medium will be referred to as a "non-target medium MN (see FIG. 7)." The image generating device 1 has a communication unit 11 for wireless communication with other devices. Communication units 24 and 34, which will be described later, also have the same functions as the communication unit 11.

[0017] In this embodiment, the image generating device 1 is a scanner that reads the front surface of the target medium MO and converts it into a target image IMGO as image data. The image generating device 1 also reads a specific surface of a non-target medium MN and converts it into a non-target image IMGN as image data. The scanner serving as the image generating device 1 is assumed to be installed in each affiliated store. Note that the image generating device 1 does not have to be a scanner, and may be a device capable of capturing images, such as a digital camera, smartphone, or tablet terminal, that is not installed in each affiliated store.

[0018] As mentioned above, the target image IMGO is composed only of images of the front side of the target medium MO, and the non-target image IMGN is composed only of images of a specific side of the non-target medium MN. However, both the target image IMGO and the non-target image IMGN may be datasets composed of multiple images. For example, the target image IMGO may be a dataset of images of the front side and the back side of the target medium MO, and the non-target image IMGN may be a dataset of images of the front side and the back side of the non-target medium MN.

[0019] <Configuration of the model construction device> In this embodiment, the model construction device 2 is a management server, and is installed, for example, in a room in a building that houses the administrative office (hereinafter referred to as the "office"). The model construction device 2 does not have to be a management server, and may be built into, for example, an evaluation device 3 described below. The management server may be, for example, a server used in a known cloud service, or may be installed in the office of a company that has outsourced part or all of its operations from the administrative office. The model construction device 2 has an input unit 21, an output unit 22, a memory unit 23, a communication unit 24, and a control unit 25.

[0020] The input unit 21 accepts various operations from the evaluator. An input unit 31, which will be described later, has the same functions as the input unit 21. The output unit 22 outputs various information. The output unit 22 may be, for example, a display unit or a touch panel integrated with the input unit 21. Alternatively, for example, the output unit 22 may be a printer.

[0021] The memory unit 23 stores various information used by the model construction device 2. The memory unit 23 also stores training data 231, a first evaluation model 232 (evaluation model), and a second evaluation model 233 (evaluation model). The training data 231 is a data set consisting of training images (not shown) and correct answer data. The training images are images captured of training media (not shown). The training media consist of various media, such as media that are identical to the target media MO, media that are similar to the target media MO, and media that are clearly different from the target media MO. In this embodiment, the training media are media that have previously been applied for use as gift certificates at affiliated stores.

[0022] Examples of correct answer data include the authenticity of the training medium, whether the characteristic areas of the training medium match, and whether the identification areas of the training medium match. "Authenticity of the training medium" indicates whether the training medium is genuine or fake. In this embodiment, "genuine" refers to a gift certificate that the secretariat has authorized for use at affiliated stores and registered in the model construction device 2. Hereinafter, this gift certificate will be referred to as the "reference medium MS." The reference medium MS is registered in the model construction device 2 as a set with a label indicating authorization for use and a reference image (not shown) of the reference medium MS, and this data set is recorded in a dedicated database (not shown) in the memory unit 23, thereby completing the registration of the reference medium MS.

[0023] In this embodiment, the reference medium MS will be described using the examples shown in FIGS. 2 and 3. In this embodiment, the reference image is assumed to consist only of an image of the front surface MS-1 of the reference medium MS shown in FIG. 2. Note that, like the target image IMGO and the non-target image IMGN, the reference image does not have to consist of only one image, but may also be a data set consisting of multiple images. In the example shown in FIGS. 2 and 3, the reference image may be a data set of an image of the front surface MS-1 of the reference medium MS shown in FIG. 2 and an image of the back surface MS-2 of the reference medium MS shown in FIG. 3.

[0024] Furthermore, the training images are also assumed to consist of only images of the surface of the training medium, corresponding to the target image IMGO, the non-target image IMGN, and the reference image. The training images may also be a dataset consisting of multiple images, as with the target image IMGO.

[0025] A "characteristic area" is an area displayed on the surface of a learning medium that contains characteristic information that serves as a clue to identifying the authenticity of the learning medium. Examples of characteristic information include registration number, name, expiration date, logo, etc., and some or all of these may be lumped together to form a characteristic area. "Match / mismatch of characteristic area of ​​learning medium" indicates whether the characteristic area of ​​the learning medium matches or does not match the characteristic area of ​​the reference medium MS. Hereinafter, the characteristic area of ​​the learning medium will be referred to as the "learning characteristic area," and the characteristic area of ​​the reference medium MS will be referred to as the "reference characteristic area."

[0026] When the feature information is a character, a number, a symbol, or a combination thereof, the feature information in the training feature region and the feature information in the reference feature region are recognized, for example, by OCR (Optical Character Reader) processing. Alternatively, for example, this feature information may be recognized by AI-OCR (Artificial Intelligence-Optical Character Reader) processing. The feature information in the training feature region recognized by OCR processing (or AI-OCR processing) is compared with the feature information in the reference feature region to determine whether the feature regions of the training medium match.

[0027] In the example of Figure 2, the area MS-3 surrounded by a dashed line on the surface MS-1 of the reference medium MS is the reference feature area. The various pieces of information displayed within this area MS-3 surrounded by a dashed line are the feature information of the reference medium MS. In this case, the area on the surface of the learning medium corresponding to the area MS-3 surrounded by a dashed line in Figure 2 is the learning feature area. The various pieces of information displayed within this learning feature area are the feature information of the learning medium.

[0028] The "identification area" is an area displayed on the surface of the learning medium that contains identification information for identifying the authenticity of the learning medium, and is, for example, a rectangular area of ​​a predetermined size. Examples of identification information include a paper pattern ID. The paper pattern ID is a predetermined feature obtained from the identification area. If the learning medium is a paper voucher, an example of the feature would be a quantity that indicates the characteristics of the raw materials of the identification area, such as the degree of entanglement of the fibers in the identification area. "Match / mismatch of the identification area of ​​the learning medium" indicates whether the identification area of ​​the learning medium matches or does not match the identification area of ​​the reference medium MS. Hereinafter, the identification area of ​​the learning medium will be referred to as the "learning identification area," and the identification area of ​​the reference medium MS will be referred to as the "reference identification area."

[0029] In the example of Figure 2, the rectangular area MS-4 on the surface MS-1 of the reference medium MS is the reference identification area. The paper pattern ID, etc. displayed within this rectangular area MS-4 becomes the identification information of the reference medium MS. In this case, the rectangular area on the surface of the learning medium that corresponds to the rectangular area MS-4 in Figure 2 becomes the learning identification area. The paper pattern ID, etc. displayed within this learning identification area becomes the identification information of the learning medium.

[0030] In this embodiment, the teacher data 231 includes a plurality of training images and is configured with the same number of basic data sets as the number of training images. The basic data sets are data sets in which training images are associated with labels as ground truth data (such as whether the training medium is authentic, whether the characteristic regions of the training medium match, and whether the discrimination regions of the training medium match).

[0031] The teacher data 231 may be stored in advance in the storage unit 23, or may be input by the evaluator via the input unit 21 and then stored in the storage unit 23. For example, the teacher data 231 may be stored in a device other than the model construction device 2. Furthermore, the model construction device 2 may generate the teacher data 231 and store it in the storage unit 23.

[0032] The first and second evaluation models 232 and 233 are both trained models used to evaluate the authenticity of a target medium MO. Both the first and second evaluation models 232 and 233 use an acquired image IMGG (described below) as input data, and classify and output indices related to the evaluation of the authenticity of the target medium MO into a predetermined number of levels. Furthermore, the number of levels (hereinafter, "first level number") of the indices (hereinafter, "first index") output by the first evaluation model 232 is different from the number of levels (hereinafter, "second level number") of the indices (hereinafter, "second index") output by the second evaluation model. In this embodiment, the number of first levels is greater than the number of second levels.

[0033] Here, there are multiple types of gift certificates that can be used at affiliated stores depending on the amount and the type of product that can be used, so there are multiple types of reference medium MS, and training data 231 is also generated for each of the multiple types of reference medium MS. Therefore, first and second evaluation models 232 and 233 are also constructed for each of the multiple types of reference medium MS.

[0034] The number of trained models used to evaluate the authenticity of the target medium MO is not particularly limited. That is, three or more trained models, including the first and second evaluation models 232 and 233, may be used to evaluate the authenticity of the target medium MO. Details of how the evaluation device 3 uses the first and second evaluation models 232 and 233 will be described later.

[0035] The control unit 25 comprehensively controls each unit of the model construction device 2. The control unit 25 has a teacher data acquisition unit 251 and a model construction unit 252. The teacher data acquisition unit 251 acquires teacher data 231 from the storage unit 23 and transmits it to the model construction unit 252.

[0036] The model construction unit 252 constructs the first and second evaluation models 232 and 233 by performing machine learning with reference to the teacher data 231 acquired from the teacher data acquisition unit 251. The teacher data used to construct the first evaluation model 232 may be different from the teacher data used to construct the second evaluation model 233. Furthermore, the machine learning method and the type of the first evaluation model 232 are not particularly limited as long as they are capable of evaluating the authenticity of the target medium MO. For example, the model construction unit 252 may construct the first and second evaluation models 232 and 233 using a convolutional neural network. The model construction unit 252 records the constructed first and second evaluation models 232 and 233 in the storage unit 23.

[0037] <Configuration of evaluation device> In this embodiment, the evaluation device 3 is a desktop PC (personal computer) installed in an office. The evaluation device 3 does not have to be a desktop PC and may be, for example, a tablet terminal. The evaluation device 3 has an input unit 31, a display unit 32, a storage unit 33, a communication unit 34, and a control unit 35.

[0038] The display unit 32 displays various information. The display unit 32 also displays a second index related to the evaluation of the authenticity of the target medium MO calculated by the index calculation unit 354. The storage unit 33 stores various information used by the evaluation device 3. The storage unit 33 may store some or all of the various images acquired by the image acquisition unit 351 described below, the various information acquired by the information acquisition unit 352 described below, and the calculation results of the index calculation unit 354.

[0039] The control unit 35 comprehensively controls each unit of the evaluation device 3. The control unit 35 has an image acquisition unit 351, an information acquisition unit 352, a model selection unit 353, an index calculation unit 354, and a display control unit 355. The image acquisition unit 351 acquires a target image IMGO or a non-target image IMGN from the image generation device 1. Hereinafter, the target image IMGO and the non-target image IMGN acquired by the image acquisition unit 351 from the image generation device 1 will be collectively referred to as the "acquired image IMGG (see FIGS. 6 and 7)." Here, an example of a case in which the image acquisition unit 351 acquires a non-target image IMGN as an acquired image IMGG is when an employee of a franchise store equipped with the image generation device 1 mistakenly transmits a non-target image IMGN to the image generation device 1.

[0040] The information acquisition unit 352 acquires at least one of attribute information indicating the attributes of the target medium MO and precision information regarding the evaluation of the authenticity of the target medium MO. Examples of attribute information include the type of the target medium MO, expiration date, usable area, usable store, serial number, and, if the target medium MO is a security, the amount. The precision information indicates whether the authenticity of the target medium MO is evaluated using a fine scale or a coarse scale. For example, the authenticity of the target medium MO is evaluated using a four-scale scale of "◎, ○, △, ×" or a two-scale scale of "○, ×." The information acquisition unit 352 may acquire this information from the storage unit 33 where it is pre-stored, or may acquire this information by having the evaluator input this information via the input unit 31. The information acquisition unit 352 may also acquire information other than the attribute information and the precision information.

[0041] The model selection unit 353 selects either the first or second evaluation model 232 or 233 according to the information acquired by the information acquisition unit 352. The acquired information is mainly assumed to be composed of three cases: a case where it is composed of only attribute information, a case where it is composed of only refined or coarse information, and a case where it is composed of attribute information and refined or coarse information. However, the acquired information may be composed of information other than attribute information and refined or coarse information, for example, it may be composed of various input information of the evaluator acquired by the information acquisition unit 352 via the input unit 31.

[0042] The index calculation unit 354 calculates a first or second index by referring to the image IMGG acquired by the image acquisition unit 351 and the information acquired by the information acquisition unit 352. When calculating the first or second index, the index calculation unit 354 uses either the first or second evaluation model 232 or 233 selected by the model selection unit 353. In this embodiment, the index calculation unit 354 evaluates the authenticity of the target medium MO by evaluating how closely the surface of the target medium MO resembles the surface MS-1 of the reference medium MS. The acquired image IMGG includes not only the target image IMGO but also images of media not intended for use as gift certificates, such as non-target images IMGN. The display control unit 355 controls the display of the display unit 32 in accordance with various operations received by the input unit 31. Details of the control of the display unit 32 by the display control unit 355 will be described later.

[0043] [Processing flow (calculation of the first or second index)] The flow of processing up to when the evaluation device 3 calculates the first or second index will be described with reference to FIG. 4. The flowchart shown in FIG. 4 is an example of an evaluation method according to one aspect of the present invention. As shown in FIG. 4, in step S101, the image generation device 1 generates a target image IMGO or a non-target image IMGN. The image generation device 1 transmits the generated target image IMGO or non-target image IMGN to the image acquisition unit 351 via the communication unit 11. In step S102 (image acquisition step), the image acquisition unit 351 acquires the target image IMGO or non-target image IMGN transmitted from the image generation device 1 via the communication unit 34. The target image IMGO or non-target image IMGN acquired by the image acquisition unit 351 becomes an acquired image IMGG. The image acquisition unit 351 transmits the acquired image IMGG to the index calculation unit 354.

[0044] In step S103 (information acquisition step), the information acquisition unit 352 acquires at least one of attribute information and refined / coarse information. At least one of the attribute information and refined / coarse information acquired by the information acquisition unit 352 becomes acquired information. The information acquisition unit 352 transmits the acquired information to the model selection unit 353 and the index calculation unit 354. Note that the order of the processes of steps S102 and S103 is not limited to the example in FIG. 4. For example, the process of step S102 may be executed after the process of step S103 is completed, or the processes of step S102 and step S103 may be executed simultaneously.

[0045] In step S104 (index calculation step), the model selection unit 353 determines whether to select the first or second evaluation model 232 or 233 according to the acquired information acquired from the information acquisition unit 352. Specifically, the model selection unit 353 reads the reference data table from the storage unit 33 and determines whether to select the first or second evaluation model 232 or 233 by comparing the acquired information with the reference data table. The reference data table is composed of attributes of the reference medium MS for each of multiple types of reference medium MS and a level number associated with each of the attributes. This level number is either the first or second level number. The reference data table may be generated by referring to, for example, past attribute information and past refinement information, and may be updated as appropriate according to the accumulation of this information.

[0046] Below, several specific examples of the process of determining an evaluation model to be selected by the model selection unit 353 will be described. In the following explanation, the first level number is "5" and the second level number is "3". First, a first case will be described in which the amount of the target medium MO included in the attribute information constituting the acquired information is 500 yen. It is assumed that the acquired information does not include precision information. In this case, if the reference data table associates "type of reference medium MO: gift certificate, usable products (attribute): some products at affiliated stores, amount (attribute): less than 1000 yen, number of levels: 3", the model selection unit 353 will determine to select the second evaluation model 233.

[0047] Next, we will explain the second case where the amount of the target medium MO included in the attribute information constituting the acquired information is 10,000 yen. It is assumed that the acquired information is composed of attribute information and refinement information indicating "fine classification." In this case, if the reference data table associates "type of reference medium MO: gift certificate, usable products (attribute): all products at affiliated stores, amount (attribute): 5,000 yen or more, number of levels: 5," the model selection unit 353 determines to select the first evaluation model 232.

[0048] Here, if the number of stages once identified based on the reference data table differs from the directionality of the number of stages indicated in the precision / coarseness information constituting the acquired information, the model selection unit 353 re-identifies the number of stages in accordance with the directionality indicated in the precision / coarseness information. Specifically, the "direction of precision / coarseness of the number of stages" refers to either indicating a fine number of stages or indicating a coarse number of stages. By performing such a re-identification process, the model selection unit 353 can make the indicators output from the evaluation model more suited to the needs of the evaluator.

[0049] However, when determining an evaluation model to be selected, the model selection unit 353 may prioritize the number of stages specified based on the reference data table over the aforementioned directionality indicated by the precision / coarseness information. For example, in the second case, if the acquired information is made up of attribute information and precision / coarseness information indicating "coarse classification," the model selection unit 353 determines to select the second evaluation model 233 even if the corresponding item in the reference data table is "second number of stages: 5."

[0050] Next, a third case will be described in which the acquired information consists of only refinement and coarseness information. In the third case, if the acquired information consists of refinement and coarseness information such as "classifying into three levels," the model selection unit 353 determines to select the second evaluation model 233 regardless of the attributes of the target medium MO or non-target medium MN displayed in the acquired image IMGG.

[0051] Once it has been decided whether to select the first or second evaluation model 232 or 233, the model selection unit 353 reads and acquires the evaluation model that has been decided to be selected from the storage unit 23 and transmits it to the index calculation unit 354. In the following explanation, it is assumed that the model selection unit 353 selects the first evaluation model 232 and transmits it to the index calculation unit 354.

[0052] In step S105 (index calculation step), the index calculation unit 354 inputs the acquired image IMGG acquired from the image acquisition unit 351 to the first evaluation model 232 acquired from the model selection unit 353. Then, the index calculation unit 354 acquires the first index output from the first evaluation model 232 and calculates the first index.

[0053] Here, the original function of the first evaluation model 232 (or the second evaluation model 233 if the second evaluation model 233 is selected) is to evaluate the authenticity of the target medium MO. However, the first evaluation model 232 (or the second evaluation model 233 if the second evaluation model 233 is selected) evaluates the authenticity of the medium displayed in the input image, which is the input data, regardless of whether the input image is a target image IMGO or a non-target image IMGN.

[0054] The first and second numbers of stages may remain unchanged regardless of updates to the first and second evaluation models 232 and 233. For example, the first and second numbers of stages may change each time the first and second evaluation models 232 and 233 are updated by machine learning that takes into account at least one of past attribute information and past refined / coarse information.

[0055] The index calculation unit 354 records the calculated first index in the storage unit 33. When the process of step S105 ends, the series of processes up to the time when the evaluation device 3 calculates the first or second index ends.

[0056] [Display mode of display unit] In this embodiment, after the process of step S105 in Fig. 4 is completed, when a keyboard (not shown) constituting the input unit 31 accepts a login operation, the display control unit 355 reads out the first or second indicator from the storage unit 33 and transmits it to the display unit 32. Then, the display control unit 355 causes the first or second indicator to be displayed on the display unit 32. However, the display control unit 355 may automatically transmit the indicator to the display unit 32 after the process of step S105 in Fig. 4 is completed, regardless of whether a login operation via the keyboard is accepted.

[0057] There are no particular limitations on the manner in which the first and second indicators are displayed by the display unit 32, and various variations are conceivable for this display manner. In this embodiment, the examples of FIGS. 5 to 8 will be described. As a premise, it is assumed that a home screen (not shown) is displayed on the display unit 32 during the period from the end of the processing of step S105 in FIG. 4 until the keyboard accepts a login operation. Furthermore, in the examples of FIGS. 5 to 8, it is assumed that the keyboard and a mouse (not shown) constitute the input unit 31. However, instead of a mouse, another pointing device such as a trackpad may constitute the input unit 31.

[0058] First, when the keyboard accepts a login operation, the display control unit 355 displays a "gift certificate usage status" screen as shown in Fig. 5 on the display unit 32. Hereinafter, the "gift certificate usage status" screen will be referred to as the "usage status screen 32a." The usage status screen 32a is composed of a search screen 32b for searching for target media MO and non-target media MN that the evaluator wants to search for, and a result display screen 32c that displays the search results obtained by operating the search screen 32b.

[0059] The search screen 32b consists of the fields "Usage Status," "AI Judgment," "Application Date," and "Registration Number," as well as the buttons "Search" and "Bulk Update." In the "Usage Status" field, the evaluator selects the history of the target medium MO or non-target medium MN that they wish to search. Specifically, by accepting the movement operation of the mouse pointer displayed on the usage status screen 32a, the display control unit 355 superimposes the mouse pointer on the "Usage Status" field.

[0060] Next, when the mouse receives a click operation while the histories are overlapping, the display control unit 355 displays a menu (not shown) above the "Usage Status" item, allowing the user to select one of the four histories. In the examples of FIGS. 5 and 8, the four histories selectable in the menu displayed in the "Usage Status" item are "Application in Progress, Approved, Rejected, and Paid." Next, when the mouse receives a mouse pointer movement operation, the display control unit 355 causes the mouse pointer to overlap one of the four histories displayed in the menu. Next, when the mouse receives a click operation while the histories are overlapping, the display control unit 355 selects the history overlapping with the mouse pointer and displays it in the "Usage Status" item. The display control unit 355 also records the history displayed in the "Usage Status" item in the storage unit 33.

[0061] This series of processes from accepting operations via the keyboard and mouse to display control by the display control unit 355 is all well-known. Therefore, in the following explanation, we will omit the explanation of this series of processes and will basically explain the operations of the evaluator and what is displayed on the display unit 32 in response to the operations of the evaluator.

[0062] "History" refers to the history of the target medium MO or non-target medium MN from the time a usage application is made to the affiliated store until payment is completed after use. In this specification, "the time a usage application is made to the affiliated store" refers to the time when the image acquisition unit 351 acquires the acquired image IMGG. In addition, cases where "a usage application is made to the affiliated store" also include cases where the image acquisition unit 351 acquires a non-target image IMGN as the acquired image IMGG.

[0063] "Application in progress" refers to the period from the time when the calculation process of the first or second index by the index calculation unit 354 is completed (the time when the processing of step S105 is completed) to the time when the evaluator finishes identifying the authenticity of the target medium MO or non-target medium MN that has been evaluated. "Approval" refers to the period from the time when the evaluator identifies that the target medium MO that has been evaluated is the reference medium MS to the time when the amount displayed on the reference medium MS is deposited into the secretariat's designated account. "Rejection" refers to the period from the time when the evaluator identifies that the target medium MO that has been evaluated is different from the reference medium MS. "Deposited" refers to the period from the time when the amount displayed on the reference medium MS is deposited into the secretariat's designated account.

[0064] In the "AI Judgment" field, the evaluator selects the index (first or second index) they wish to search for. In the examples of Figures 5 and 8, a menu (not shown) is displayed above the "AI Judgment" field, displaying the index (first or second index) in four levels: ◎, ○, △, ×. The evaluator selects the index (first or second index) they wish to search for from the menu displayed above the "AI Judgment" field. In the "Application Date" field, the evaluator inputs the application date from the keyboard, and the application date is displayed in that field. The display control unit 355 records the application date displayed in the "Application Date" field in the memory unit 33. The application date is the date on which the image acquisition unit 351 acquired the acquired image IMGG.

[0065] In the "Registration Number" field, the evaluator inputs the registration number from the keyboard, and the registration number is displayed in that field. The display control unit 355 records the registration number displayed in the "Registration Number" field in the storage unit 33. The registration number is a number assigned to a specified medium when it is permitted for use at an affiliated store and registered in the model construction device 2 (i.e., when the specified medium becomes the reference medium IMGG).

[0066] After completing entry into at least one of the fields "Usage Status," "AI Judgment," "Application Date," and "Registration Number," the evaluator uses the mouse to press the "Search" button. Pressing this "Search" button displays the search results on the results display screen 32c. In the example of Figure 5, the evaluator selects "Application in progress" in the "Usage Status" field and enters the period from "February 25, 2021" to "April 23, 2021" in the "Application Date" field, and then presses the "Search" button.

[0067] The result display screen 32c displays the application number, registration number, target number, history, index (first or second index), etc. of the target medium MO and non-target medium MN searched by operating the search screen 32b. The "application number" is a number assigned to the target medium MO and non-target medium MN displayed in the acquired image IMGG, and is assigned in ascending order of the acquisition date and time of the acquired image IMGG by the image acquisition unit 351. The application number is displayed in the "Application No." column on the result display screen 32c.

[0068] The "target number" is a number corresponding to the registration number of the reference medium MS displayed in the acquired image IMGG. In this embodiment, the display unit 32 displays, as the target number, the read number read by an image analysis unit (not shown) through image analysis of the acquired image IMGG. The image analysis unit is provided in the control unit 35. The display unit 32 may display, for example, a number included in attribute information acquired from the information acquisition unit 352 as the target number. Note that since non-target medium MN includes media other than gift certificates, when the acquired image IMGG is a non-target image IMGN, there may be cases where the number itself is not displayed on the non-target medium MN. In this case, the registration number and target number are not displayed on the result display screen 32c.

[0069] The history displayed in the "Usage Status" column on the result display screen 32c can be updated. Specifically, when the evaluator selects a specific "Usage Status" cell with a mouse, a menu allowing the evaluator to select one of four histories is displayed above the "Usage Status" cell, as shown in FIG. 5. Then, when the evaluator selects the appropriate latest history from the menu, the latest history is displayed in the "Usage Status" column, as shown in FIG. 8. After performing this series of mouse operations for all target media MO and non-target media MN whose histories need to be updated, the evaluator presses the "Bulk Update" button. When the "Bulk Update" button is pressed, the display control unit 355 records the latest history displayed in the "Usage Status" column in the memory unit 33.

[0070] In the "AI Judgment" column, an index is displayed on a four-level scale of "◎, ○, △, ×" for each of the target medium MO and non-target medium MN displayed in the acquired image IMGG. In this embodiment, the evaluator refers to the latest history displayed in the "Usage Status" column and the index displayed in the "AI Judgment" column to finally identify the authenticity of the target medium MO and non-target medium MN displayed in the acquired image IMGG.

[0071] Specifically, when the evaluator selects a cell for a specific "Application No." using the mouse, the display control unit 355 reads out from the storage unit 33 an acquired image IMGG that shows the target medium MO or non-target medium MN that corresponds to the application number displayed in that specific "Application No." Then, the display control unit 355 stops displaying the usage status screen 32a and instead displays the read-out acquired image IMGG on the display unit 32. When displaying this acquired image IMGG, the display control unit 355 also displays the application number, registration number, target number, history, and index (first or second index) on the display unit 32.

[0072] For example, in the example of FIG. 5, when the evaluator selects the cell for "No. 642" on the result display screen 32c, the display control unit 355 causes the display unit 32 to display the acquired image IMGG corresponding to the application number "642," as shown in FIG. 6. In the example of FIG. 6, the "usage status (history)" is "approved," while the "AI judgment (first or second index)" is "△." Taking these two display contents into consideration, the evaluator visually checks the acquired image IMGG displayed on the display unit 32 to finally identify its authenticity.

[0073] The medium displayed in the acquired image IMGG in FIG. 6 is the target medium MO, which is clearly a gift certificate even when visually inspected. In other words, the acquired image IMGG in FIG. 6 is the target image IMGO. In addition to the fact that the "gift certificate number (registration number)" and the "read number (target number)" match, the "usage status (history)" is "approved." Based on these facts, the evaluator is likely to ultimately approve the medium. In other words, the evaluator is likely to ultimately determine that the target medium MO displayed in the acquired image IMGG in FIG. 6 is the reference medium MS.

[0074] If the evaluator finally approves, he or she performs a mouse operation to re-display the usage status screen 32a on the display unit 32. This mouse operation causes the display control unit 355 to terminate the display of the acquired image IMGG and re-display the usage status screen 32a on the display unit 32. In the example of FIG. 6, there is no history update before and after the evaluator's final approval. Therefore, as in the example of FIG. 9, on the re-displayed usage status screen 32a, the "usage status" of "No. 642" on the result display screen 32c remains "approved."

[0075] Also, for example, in the example of FIG. 5, when the evaluator selects the cell for "No. 638" on the result display screen 32c, the display control unit 355 causes the display unit 32 to display the acquired image IMGG corresponding to the application number "638," as shown in FIG. 7. In the example of FIG. 7, the "usage status" is "application in progress," and the "AI judgment" is "x." Furthermore, neither the "gift certificate number" nor the "read number" is displayed. The evaluator visually checks the acquired image IMGG displayed on the display unit 32 while taking these displayed contents into consideration, and finally identifies the authenticity.

[0076] The medium displayed in the acquired image IMGG in FIG. 7 is a non-target medium MN, which is clearly not a gift certificate even when visually inspected. In other words, the acquired image IMGG in FIG. 7 is a non-target image IMGN. Therefore, the evaluator is likely to ultimately reject it. In other words, the evaluator is likely to ultimately determine that the non-target medium MN displayed in the acquired image IMGG in FIG. 6 is not the reference medium MS.

[0077] Even if the evaluator ultimately rejects the application, the evaluator causes the display unit 32 to redisplay the usage status screen 32a, just as if the evaluator ultimately approved the application. In the example of FIG. 7, the history is updated from "Application in progress" to "Rejected" before and after the evaluator's final rejection. Therefore, as in the example of FIG. 9, on the redisplayed usage status screen 32a, the "Usage Status" of "No. 632" on the result display screen 32c has been updated to "Rejected."

[0078] [Summary] In the evaluation system 100, the model selection unit 353 of the evaluation device 3 selects either the first or second evaluation model 232 or 233 depending on the acquired information. Also, in the evaluation system 100, the index calculation unit 354 of the evaluation device 3 calculates the first or second index using the evaluation model selected by the model selection unit 353. Because the acquired information consists of at least one of attribute information and refinement information, the evaluation device 3 and evaluation system 100 can present the evaluation results of the authenticity of the target medium MO using an evaluation model that corresponds to the attributes of the target medium or the needs of the evaluator.

[0079] Furthermore, in the evaluation system 100, the model construction unit 252 of the model construction device 2 constructs the first and second evaluation models 232 and 233 so that the number of stages of the first index (first number of stages) and the number of stages of the second index (second number of stages) are different from each other. As a result, the evaluation device 3 and the evaluation system 100 can present the evaluation result of the authenticity of the target medium MO with a number of stages that corresponds to the attributes of the target medium or the needs of the evaluator, by having the model selection unit 353 select an evaluation model according to the acquired information.

[0080] Furthermore, in the evaluation system 100, the evaluation device 3 is equipped with a display unit 32 and a display control unit 355, and the first or second index can be displayed on the display unit 32 in a display mode according to the control of the display control unit 355. This allows the evaluator to easily understand the index, and the convenience of the evaluation system 100 is improved.

[0081] [Modification] <First Modification> A first modified example of an evaluation system 100 according to an embodiment of the present invention will be described with reference to Figures 1, 5, and 9. For ease of explanation, components having the same functions as those described in the above embodiment are denoted by the same reference numerals, and their description will not be repeated. An evaluation system 200 according to the first modified example of the present invention shown in Figure 1 differs from the evaluation system 100 in that the index calculation unit 354a determines whether or not to execute a calculation process depending on at least one of the acquired image IMGG and the acquired information.

[0082] The following describes characteristic processing of the evaluation system 200. The processing of steps S201 to S203 shown in Fig. 9 is the same as the processing of steps S101 to S103 shown in Fig. 4. In addition, in the processing of step S207 shown in Fig. 9, the model selection unit 353 selects the first evaluation model 232 and transmits it to the index calculation unit 354, similar to the processing of step S104 shown in Fig. 4.

[0083] In step S204, the index calculation unit 354a determines whether the acquired information includes processing end information. The processing end information is, for example, information indicating that the target medium MO or non-target medium MN displayed in the acquired image IMGG is different from the reference medium (i.e., is a counterfeit), and corresponds to attribute information such as "price not displayed." Another example of the processing end information is input information acquired by the information acquisition unit 352 via the input unit 31 indicating that "authenticity will be identified visually only."

[0084] If it is determined that the processing end information is included (YES in step S204), the index calculation unit 354a decides not to input the acquired image IMGG to the first evaluation model 232, and transmits the decision result to the display control unit 355. In step S205, the display control unit 355 causes the display unit 32 to display an "x" indicator indicating that the target medium MO or non-target medium MN displayed in the acquired image IMGG is a fake. In addition, the display control unit 355 causes a history of "not performed" (not shown) to be displayed in the corresponding cell of "AI judgment" on the result display screen 32c shown in FIG. 5. Other display aspects on the display unit 32 are the same as those in the examples of FIGS. 5 to 8.

[0085] On the other hand, if it is determined that processing information is not included (NO in step S204), the process proceeds to step S206. In step S206, the index calculation unit 354a reads the reference image IMGG from the dedicated database in the storage unit 23 of the model construction device 2, and compares the identification information in the identification area displayed in the acquired image IMGG with the identification information in the reference identification area on the reference medium MS. Hereinafter, the identification information in the identification area displayed in the acquired image IMGG will be referred to as "target identification information," and the identification information in the reference identification area on the reference medium MS will be referred to as "reference identification information."

[0086] If it is determined that the object identifying information does not match the reference identifying information, or that the object identifying information itself does not exist (NO in step S206), the index calculation unit 354a decides not to input the acquired image IMGG to the first evaluation model 232, and transmits the decision result to the display control unit 355. After this transmission process is completed, the process proceeds to step S205.

[0087] On the other hand, if it is determined that the object identification information and the reference identification information match (YES in step S206), the index calculation unit 354a decides not to input the acquired image IMGG to the first evaluation model 232, and executes the processes of steps S207 and S208. The processes of steps S207 and S208 are similar to the processes of steps S104 and S105 shown in FIG.

[0088] The end of the process of step S208 completes all of the characteristic processes of the evaluation system 200. Note that, similar to the evaluation system 100, after the process of step S208 is completed, the display control unit 355 displays the first or second index on the display unit 32 when the keyboard accepts a login operation.

[0089] In this way, in the evaluation system 200, the index calculation unit 354a of the evaluation device 3 does not calculate the first or second index depending on the contents of the acquired image IMGG and the acquired information. This prevents the evaluation system 200 (specifically, the index calculation unit 354a) from performing unnecessary calculation processes. Furthermore, the evaluation device 3 and the evaluation system 200 can evaluate the authenticity of the target medium MO or non-target medium MN displayed in the acquired image IMGG in accordance with the attributes and the needs of the evaluator.

[0090] Note that the series of processes in steps S204 to S206 is merely an example, and the index calculation unit 354a may execute other processes. For example, the index calculation unit 354a may determine whether or not to execute the calculation process based only on the acquired image IMGG. Specifically, in the example of FIG. 9, the index calculation unit 354a may not execute the process of step S204, and may proceed to the process of step S206 immediately after the process of step S203 is completed. Also, for example, the index calculation unit 354a may determine whether or not to execute the calculation process based only on the acquired information. Specifically, in the example of FIG. 9, the index calculation unit 354a may not execute the process of step S206, and may proceed to the process of step S207 immediately after determining NO in step S204.

[0091] <Second Modification> A second variant of the evaluation system 100 according to one embodiment of the present invention will be described with reference to Fig. 1. The evaluation system 300 according to the second variant of the present invention shown in Fig. 1 differs from the evaluation systems 100 and 200 in that the information acquisition unit 352a acquires auxiliary information to assist in the evaluation of the authenticity of the target medium MO. Note that the evaluation system 300 also uses the auxiliary information to assist in the evaluation of the authenticity of non-target medium MN.

[0092] The auxiliary information may be any information that can assist in evaluating the authenticity of the target medium MO (and non-target medium MN). For example, the auxiliary information may be a comparison result obtained by comparing store information, including information such as the address of the affiliated store and the store owner, with the location information of the sender at the time the target image IMGO or the non-target image IMGN was sent. If the comparison result indicates a match / mismatch between the address of the affiliated store included in the store information and the address of the sender included in the location information, the comparison result can be used as reference information when evaluating authenticity. The store information may be pre-stored in the storage unit 23 of the model construction device 2. Alternatively, the evaluator may input the store information via the input unit 31 of the evaluation device 3. The comparison result described above as auxiliary information is obtained by the information acquisition unit 352a, for example, by acquiring store information and location information and comparing these pieces of information.

[0093] Furthermore, for example, the supplementary information may be suspicious information indicating facts that increase the likelihood that the target medium MO or non-target medium MN is different from the reference medium MS. For example, if multiple target media whose authenticity has been properly evaluated in the past were all requested for use during a specific time period, and the target medium MO or non-target medium MN is requested for use during a time period significantly different from the specific time period, the likelihood that they are different from the reference medium MS increases. Therefore, the fact that the target medium MO or non-target medium MN was requested for use during a time period significantly different from the specific time period can be reference information when evaluating authenticity.

[0094] In this case, the information acquisition unit 352a reads the time periods of all past usage requests (limited to those whose authenticity has been properly evaluated) from the storage unit 23 of the model construction device 2, and compares the read time periods with the time periods when usage requests for the target medium MO or non-target medium MN were made. If this comparison reveals that the read time periods are significantly different from the time periods when usage requests for the target medium MO or non-target medium MN were made, the information acquisition unit 352a acquires this fact as suspicious information.

[0095] For example, if multiple target media whose authenticity has been properly evaluated in the past have all been applied for use at a specific affiliated store, and the target image IMGO or non-target image IMGN is transmitted from a location far away from the specific affiliated store, the likelihood that it differs from the reference medium MS increases. Therefore, the fact that the location from which the target image IMGO or non-target image IMGN was transmitted is far away from the specific affiliated store can be reference information when evaluating authenticity. In this case, the method of acquiring suspicious information by the information acquisition unit 352a is the same as when the suspicious information is that the target medium MO or non-target medium MN was applied for use at a time period significantly different from a specific time period.

[0096] For example, the auxiliary information may be registration number information indicating the registration number of a gift certificate other than the reference medium MS. If the application number of the target medium MO or the application number of the non-target medium MN displayed in the acquired image IMGG matches the registration number included in the registration number information, the target medium MO or the non-target medium MN is not the reference medium MS. The registration number information may be pre-stored in the memory unit 23 of the model construction device 2. Alternatively, the evaluator may input the registration number information via the input unit 31 of the evaluation device 3. Furthermore, any of the various feature information displayed in the reference feature area of ​​the reference medium MS may be used as the auxiliary information.

[0097] The following describes the characteristic processing of the evaluation system 300. The information acquisition unit 352a acquires auxiliary information in addition to the attribute information and the refinement information, and transmits this information to the index calculation unit 354b. The index calculation unit 354b inputs the acquired image IMGG into the evaluation model selected by the model selection unit 353 to acquire a first or second index, and then changes the evaluation result indicated by the first or second index by referring to the auxiliary information.

[0098] For example, if an indicator "△" is acquired as the first or second indicator and auxiliary information is acquired indicating that the address of the affiliated store included in the store information does not match the address of the sender included in the location information, the index calculation unit 354b changes the first or second indicator from "△" to "×". Also, for example, if an indicator "○" is acquired as the first or second indicator and auxiliary information is acquired indicating that gift certificates with consecutive numbers and the application number of the target medium MO displayed in the acquired image are frequently used, the index calculation unit 354b changes the first or second indicator from "○" to "◎".

[0099] This characteristic processing of the evaluation system 300 (specifically, the index calculation unit 354b of the evaluation device 3) can improve the accuracy of the evaluation result of authenticity represented by the first or second index, thereby increasing the reliability of the evaluation result of the evaluation system 300.

[0100] The index calculation unit 354b may not execute the process of changing the first or second index by referring to the auxiliary information. In this case, for example, the information acquisition unit 352a may transmit the auxiliary information to the display control unit 355, and the display control unit 355 may display the auxiliary information on the display unit 32. This allows the evaluator to make a final determination of authenticity by referring to the display content of the result display screen 32c and the auxiliary information displayed on the display unit 32. The index calculation unit 354b may also determine whether to execute the calculation process depending on at least one of the acquired image IMGG and the acquired information.

[0101] <Third Modification> A third modified example of evaluation system 100 according to an embodiment of the present invention will be described with reference to Fig. 1. Evaluation system 400 according to the third modified example of the present invention shown in Fig. 1 differs from evaluation systems 100 to 300 in that model construction unit 252a constructs third evaluation model 234 (evaluation model) and fourth evaluation model 235 (evaluation model). Evaluation system 400 also differs from evaluation systems 100 to 300 in that index calculation unit 354c calculates the third or fourth index.

[0102] The following describes the characteristic processing of the evaluation system 400. The model construction unit 252a constructs a third evaluation model 234 by performing machine learning with reference to the first teacher data 231a. The model construction unit 252a also constructs a fourth evaluation model 235 by performing machine learning with reference to the second teacher data 231b. The third and fourth evaluation models 234 and 235 are similar to the first and second evaluation models 232 and 233 in that the input data for both is the acquired image IMGG.

[0103] The first and second teacher data 231a and 231b are similar to the teacher data 231 in that they are data sets composed of training images and correct answer data. However, the correct answer data of the first teacher data 231a is labeled with the authenticity of the training medium, while the correct answer data of the second teacher data 231b is labeled with the match / mismatch of the characteristic regions of the training medium. In other words, the correct answer data of the first teacher data 231a has a higher evaluation accuracy of the authenticity of the training medium than the correct answer data of the second teacher data 231b. Therefore, the index (third index) output by the third evaluation model 234 has a higher evaluation accuracy of the authenticity of the target medium MO (or, in some cases, the non-target medium MN) than the index (fourth index) output by the fourth evaluation model 235.

[0104] The correct answer data of the first teacher data 231a may be labeled, for example, with a match / mismatch of the identification region of the training medium. In this case, since the correct answer data of the first teacher data 231a has an extremely high evaluation accuracy of the authenticity of the training medium, the correct answer data of the second teacher data 231b may be labeled with a match / mismatch of the characteristic region of the training medium, or may be labeled with the authenticity of the training medium. Furthermore, the model construction unit 252a may construct three or more evaluation models, including third and fourth evaluation models 234 and 235, which output indices with different evaluation accuracy of the authenticity of the target medium MO.

[0105] For example, when accuracy information is included in the information acquired by the information acquisition unit 352, or when the acquired information consists only of accuracy information, the model selection unit 353a selects either the third or fourth evaluation model 234 or 235 depending on the accuracy information. The accuracy information is information regarding the evaluation accuracy of the authenticity of the target medium MO, such as "evaluate with high accuracy" or "evaluate with low accuracy."

[0106] Furthermore, for example, when attribute information is included in the information acquired by the information acquisition unit 352, or when the acquired information consists of attribute information alone, the model selection unit 353a selects either the third or fourth evaluation model 234 or 235 according to the attribute information. For example, when the price of the target medium MO or non-target medium MN included in the attribute information constituting the acquired information is 10,000 yen, the model selection unit 353a selects the third evaluation model 234. For example, when the price of the target medium MO or non-target medium MN included in the attribute information constituting the acquired information is 500 yen, the model selection unit 353a selects the fourth evaluation model 235.

[0107] The index calculation unit 354c inputs the acquired image IMGG acquired from the image acquisition unit 351 to the third or fourth evaluation model 234 or 235 selected by the model selection unit 353a. Then, the index calculation unit 354c calculates the third or fourth index by acquiring the third index output from the third evaluation model 234 or the fourth index output from the fourth evaluation model 235.

[0108] Such characteristic processing of the evaluation system 400 allows the evaluation results of the authenticity of the target medium MO (or in some cases, non-target medium MN) to be presented with accuracy according to the attributes of the target medium MO or the needs of the evaluator.

[0109] The model construction unit 252a may construct, for example, at least one of the first and second evaluation models 232 and 233, in addition to the third and fourth evaluation models 234 and 235. When the model construction unit 252a constructs, for example, the first to fourth evaluation models 232 to 235, the model selection unit 353a selects one of the first to fourth evaluation models 232 to 235 in accordance with the acquired information acquired from the information acquisition unit 352.

[0110] Furthermore, the information acquisition unit 352 of the evaluation device 3 of the evaluation system 400 may acquire auxiliary information. The index calculation unit 354c may determine whether or not to execute the calculation process depending on at least one of the acquired image IMGG and the acquired information.

[0111] <Fourth Modification> A fourth modified example of evaluation system 100 according to one embodiment of the present invention will be described with reference to Fig. 1. Evaluation system 500 according to the fourth modified example of the present invention shown in Fig. 1 differs from evaluation systems 100 to 400 in that model construction unit 252b constructs fifth evaluation model 236 (evaluation model) in addition to first and second evaluation models 232 and 233. Evaluation system 500 also differs from evaluation systems 100 to 400 in that model selection unit 353b selects two evaluation models from first and second evaluation models 232 and 233 and fifth evaluation model 236. Evaluation system 500 also differs from evaluation systems 100 to 400 in that index calculation unit 354d finally calculates secondary indexes.

[0112] The characteristic processing of the evaluation system 500 will be described below. For convenience, in the following description, the first and second evaluation models 232 and 233 and the fifth evaluation model 236 will be collectively referred to as the "first to fifth evaluation models 232 to 236." The model construction unit 252b constructs the first to fifth evaluation models 232 to 236 by performing machine learning with reference to the training data 231. The fifth evaluation model 236 outputs a fifth index that classifies the evaluation results of the authenticity of the target medium MO using a third level number. In a fourth modified example, the third level number is "4." The input data of the fifth evaluation model 236 is the acquired image IMGG, as is the input data of the first to fourth evaluation models 232 to 235.

[0113] The model selection unit 353b selects two evaluation models from the first to fifth evaluation models 232 to 236 according to the information acquired by the information acquisition unit 352. For example, if the acquired information includes precision / coarseness information indicating "rough classification," the model selection unit 353b selects the second evaluation model 233 with the fewest number of stages and the fifth evaluation model 236 with the second fewest number of stages. Also, for example, if the price of the target medium MO or non-target medium MN included in the attribute information constituting the acquired information is 10,000 yen, the model selection unit 353b selects the first evaluation model 232 with the most number of stages and the fifth evaluation model 236 with the second fewest number of stages.

[0114] The index calculation unit 354d inputs the acquired image IMGG acquired from the image acquisition unit 351 to each of the two evaluation models selected by the model selection unit 353b. Then, the index calculation unit 354d calculates a first primary index by acquiring the first primary index output from one of the two evaluation models. Also, the index calculation unit 354d calculates a second primary index by acquiring the second primary index output from the other of the two evaluation models.

[0115] The first and second primary indicators are both primary indicators related to the evaluation of the authenticity of the target medium MO or the non-target medium MN. In the following description, it is assumed that the model selection unit 353b selects the first and fifth evaluation models 232 and 236. It is also assumed that the index calculation unit 354d calculates the first primary indicator by acquiring the first indicator output from the first evaluation model 232, and calculates the second primary indicator by acquiring the fifth indicator output from the fifth evaluation model 236. In other words, the first indicator becomes the first primary indicator, and the fifth indicator becomes the second primary indicator.

[0116] Next, the index calculation unit 354d calculates a secondary index using the first and second primary indexes. The secondary index is a secondary index related to the evaluation of the authenticity of the target medium MO or the non-target medium MN. Specifically, the index calculation unit 354d determines the final index as the secondary index by comprehensively considering the evaluation result of the authenticity indicated by the first primary index and the evaluation result of the authenticity indicated by the second primary index.

[0117] For example, if the first primary index is "B" out of five levels "A, B, C, D, E" and the second primary index is "△" out of four levels "◎, ○, △, ×," the index calculation unit 354d comprehensively considers these primary indexes and determines the final index to be "50%." The index calculation unit 354d then sets the determined final index "50%" as the secondary index. Here, of the first primary indexes "A, B, C, D, E," the index "A" is evaluated as the highest level of authenticity, and the index "E" is evaluated as the lowest level of authenticity. In addition, in the fifth modified example, the secondary index is expressed as a "number + percentage" between "100%" and "0%." The number of levels of the secondary index may be set in advance, or it may be expressed as an "arbitrary number + percentage" between "100%" and "0%" depending on the result of comprehensively considering the first and second primary indexes.

[0118] This characteristic processing of the evaluation system 500 allows the presentation of a secondary index that takes into account the first and second primary indexes, which differ in the degree of accuracy of the evaluation of authenticity, as the final evaluation result of the authenticity of the target medium MO (or in some cases, non-target medium MN). This increases the reliability of the evaluation results of the evaluation system 500.

[0119] The model construction unit 252b may construct at least one of the first and second evaluation models 232 and 233, and at least one of the third and fourth evaluation models 234 and 235. In this case, the model selection unit 353b selects one or more evaluation models from at least one of the first and second evaluation models 232 and 233 according to the information acquired by the information acquisition unit 352. The model selection unit 353b also selects one or more evaluation models from at least one of the third and fourth evaluation models 234 and 235 according to the information acquired by the information acquisition unit 352.

[0120] Furthermore, in this case, the index calculation unit 354d calculates a first primary index using one or more evaluation models selected from at least one of the first and second evaluation models 232 and 233. Similarly, the index calculation unit 354d calculates a second primary index using one or more evaluation models selected from at least one of the third and fourth evaluation models 234 and 235. If the evaluation system 500 performs such processing, it can present a secondary index that takes into account the first and second primary indexes, which have different perspectives on authenticity evaluation, as the final evaluation result of the authenticity of the target medium MO (or, in some cases, non-target medium MN).

[0121] Furthermore, the information acquisition unit 352 of the evaluation device 3 of the evaluation system 500 may acquire auxiliary information. The index calculation unit 354d may determine whether or not to execute the calculation process depending on at least one of the acquired image IMGG and the acquired information.

[0122] <Other variations> Another modified example of the evaluation system 100 according to an embodiment of the present invention will be described. First, the image acquisition unit 351 does not have to be included in the control unit 35, but may be included in, for example, the image generation device 1. In this case, the image acquisition unit 351 transmits the target image IMGO to the index calculation unit 354 via the communication unit 11. Furthermore, the information acquisition unit 352 does not have to be included in the control unit 35, but may be included in, for example, the model construction device 2 or another device (not shown) constituting the evaluation system 100. In this case, the information acquisition unit 352 transmits at least one of the attribute information and the refinement information to the index calculation unit 354 via the communication unit 24.

[0123] In other words, it is sufficient that one or more devices constituting the evaluation system 100 include the image acquisition unit 351, the information acquisition unit 352, the model selection unit 353, and the index calculation unit 354, and there are no limitations on which devices include each of these units. Therefore, for example, the teacher data acquisition unit 251 and the model construction unit 252 of the model construction device 2 may be included in the control unit 35 of the evaluation device 3. Here, the model selection unit 353 is not an essential component of the evaluation system 100, and for example, the index calculation unit 354 may select either the first or second evaluation model 232 or 233.

[0124] [Software implementation example] The functions of the evaluation device 3 (hereinafter referred to as the "device") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control unit 35).

[0125] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above-described embodiments.

[0126] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0127] In addition, some or all of the functions of the aforementioned control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the aforementioned control blocks is formed is also included in the scope of the present invention. In addition, the functions of the aforementioned control blocks can also be realized by, for example, a quantum computer.

[0128] [Additional Notes] The present invention is not limited to the above-described embodiment and each modification, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiment and each modification are also included in the technical scope of the present invention. [Explanation of symbols]

[0129] 2. Model construction equipment 3 Evaluation equipment 100, 200, 300, 400, 500 rating system 231 Teacher Data 231a First teacher data (teacher data) 231b Second training data (training data) 232 First Evaluation Model (Evaluation Model) 233 Second Evaluation Model (Evaluation Model) 234 Third Evaluation Model (Evaluation Model) 235 Fourth Evaluation Model (Evaluation Model) 236 5th Evaluation Model (Evaluation Model) 251 Teacher Data Acquisition Department 252, 252a, 252b Model Construction Section 351 Image Acquisition Unit 352, 352a Information acquisition section 354, 354a, 354b, 354c, 354d Indicator calculation section IMGG Acquired image IMGO target image MO target media

Claims

1. an image acquisition unit that acquires a target image obtained by capturing an image of the target medium; an information acquisition unit that acquires at least one of attribute information indicating the attributes of the target medium and precision information regarding the precision of an evaluation of the authenticity of the target medium; and an index calculation unit that calculates an index relating to an evaluation of the authenticity of the target medium by referring to the image acquired by the image acquisition unit and the information acquired by the information acquisition unit, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information; The acquired information includes the fine and rough information.

2. The evaluation device according to claim 1 , wherein the plurality of evaluation models include two or more evaluation models with different numbers of levels of indexes relating to evaluation of the authenticity of the target medium.

3. The plurality of evaluation models are It is constructed by performing machine learning using a dataset containing images of the medium as training data, The evaluation device according to claim 1 or 2, wherein the correct answer data included in the data set includes two or more evaluation models that differ from each other depending on the accuracy of evaluation of the authenticity of the target medium.

4. The index calculation unit For each of the two or more evaluation models selected according to the acquired information, calculate a primary indicator for evaluating the authenticity of the target medium using the evaluation model; The evaluation device according to claim 1 , wherein the evaluation device calculates a secondary index relating to an evaluation of the authenticity of the target medium using two or more calculated primary indexes.

5. The evaluation device according to claim 1 , wherein the index calculation unit determines whether or not to execute processing by the index calculation unit in accordance with at least one of the acquired image and the acquired information.

6. The evaluation device according to claim 1 , wherein the information acquisition unit acquires auxiliary information for assisting in evaluation of the authenticity of the target medium.

7. The evaluation device according to claim 1 , further comprising a display unit that displays the index relating to the evaluation of the authenticity of the target medium calculated by the index calculation unit.

8. A control program for causing a computer to function as the evaluation device according to claim 1, the control program causing a computer to function as the image acquisition unit, the information acquisition unit, and the index calculation unit.

9. an image acquisition unit that acquires a target image obtained by capturing an image of the target medium; an information acquisition unit that acquires at least one of attribute information indicating the attributes of the target medium and precision information regarding the precision of an evaluation of the authenticity of the target medium; and an index calculation unit that calculates an index relating to an evaluation of the authenticity of the target medium by referring to the image acquired by the image acquisition unit and the information acquired by the information acquisition unit, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information; An evaluation system, wherein the acquired information includes the refined and coarse information.

10. a training data acquisition unit that acquires a data set including an image of the medium as training data; a model construction unit that constructs a plurality of evaluation models by performing machine learning with reference to the training data, using target images of the target medium as input data, and outputs an index relating to the evaluation of the authenticity of the target medium in a predetermined number of stages; The model construction unit constructs two or more evaluation models among the plurality of evaluation models, each having a different number of levels of indexes related to the evaluation of the authenticity of the target medium.

11. The model construction device according to claim 10, wherein the model construction unit constructs two or more evaluation models among the plurality of evaluation models, the evaluation models being different from each other depending on the accuracy of the evaluation of the authenticity of the target medium using the correct answer data included in the dataset.

12. an image acquisition step of acquiring a target image of the target medium; an information acquisition step of acquiring at least one of attribute information indicating attributes of the target medium and precision information regarding the precision of an evaluation of the authenticity of the target medium; an index calculation step of calculating an index relating to an evaluation of the authenticity of the target medium by referring to the acquired image acquired in the image acquisition step and the acquired information acquired in the information acquisition step, using one or more evaluation models selected from a plurality of different evaluation models according to the acquired information; The evaluation method, wherein the acquired information includes the refined and coarse information.

Images