Information processing system, control method, and program

Abstract

To reduce fraudulent insurance claims. [Solution] The system is characterized by comprising: a determination means for determining whether an image of an insured object, transmitted from a user terminal to an insurance company for the purpose of filing an insurance claim, is similar to an image publicly available on the internet; and an output means for outputting the determination result obtained by the determination means.

Description

Technical Field

[0004]

[0001] The present invention relates to an information processing system, a control method thereof, and a program, and particularly to a technique for reducing fraudulent insurance claims.

Background Art

[0002] Conventionally, techniques for preventing fraudulent insurance claims have been developed. For example, in Patent Document 1, there are provided means for acquiring inspection information indicating the content of building inspections, means for acquiring insurance information indicating the claim content of insurance related to the building, means for determining whether the claim content of the insurance is appropriate based on the inspection information and the insurance information, and means for presenting a warning regarding the insurance claim when it is determined that the claim content of the insurance is inappropriate. A computer system has been proposed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Disclosure of the Invention

Problems to be Solved by the Invention

[0004] When making an insurance claim online, a method is used in which the insured takes a photo of the insured object damaged and transmits the image to the insurance company. However, in reality, there is a possibility of fraudulently claiming insurance by misusing an image of a damaged insured object publicly available on the Internet even though the insured object is not actually damaged, and a technique for reducing the risk of fraudulent claims is desired.

[0005] An object of the present invention is to reduce fraudulent insurance claims.

Means for Solving the Problems

[0006] The present invention is An information processing system including a first server and a second server, wherein the first server is for insurance claims ApplicationInsured property Application image Based on the above, the second server comprises a first generation means for generating a feature vector of the application image, and a second generation means for generating text describing the application image based on the application image, wherein the second server provides instructions to search for an internet image, which is an image published on the internet, using the text as a search query, and a third generation means for generating a feature vector of the internet image found by the instructions, and based on the feature vector generated by the first generation means and the feature vector generated by the third generation means, the application image but, The aforementioned Internet to The system is characterized by comprising a determination means for determining whether an image is similar to another image, and an output means for outputting the determination result from the determination means. [Effects of the Invention]

[0007] According to the present invention, it becomes possible to reduce fraudulent insurance claims. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of an information processing system in an embodiment of the present invention. [Figure 2] This diagram shows an example of the hardware configuration of Server 101A, Server 101B, Server 102, User Terminal 103, Operation Terminal 104A, and Operation Terminal 104B. [Figure 3] This flowchart shows an example of processing in an information processing system according to an embodiment of the present invention. [Figure 4] This flowchart shows an example of the detailed processing steps for determining whether past images are being reused in S304. [Figure 5] This flowchart shows an example of the detailed processing steps for detecting internet image misuse in S305. [Figure 6] This flowchart shows an example of the detailed processing of the image processing judgment process in S306. [Figure 7] This flowchart shows an example of the detailed processing of the AI ​​image recognition process generated by S307. [Figure 8] This is an example of a functional block diagram related to the process for determining whether past images are being reused. [Figure 9] This is an example of a functional block diagram related to the process for detecting the misuse of images on the internet. [Figure 10] This is an example of a display screen 1001 that appears when an image included in an insurance claim is determined to be similar to an image previously submitted to another insurance company for insurance claims. [Figure 11] This is an example of the display screen 1101 that is displayed when it is determined that the image included in the insured insurance claim information is similar to an image publicly available on the Internet. [Figure 12] This is an example of the configuration of the insurance claim image database 801A, the feature vector database 806 of past application images, and the feature vector database 815 of Internet images.

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the present invention is not limited to the following embodiments. In each figure, the same member or element is denoted by the same reference numeral, and overlapping descriptions are omitted or simplified.

[0010] FIG. 1 is a diagram showing an example of an information processing system in an embodiment of the present invention.

[0011] The server 101A, the server 102, and the user terminal 103 are communicably connected via a network 105 including the Internet. Also, the server 101B, the server 102, and the user terminal 103 are communicably connected via the network 105.

[0012] The user terminal 103 is a terminal operated by a user such as an insured person or an insurance contract holder. Although only one user terminal is shown in FIG. 1, a plurality of user terminals operated by respective users may be connected to the network 105.

[0013] In addition, the server 101A and the operation terminal 104A are devices used by users (such as employees) of insurance company (Company A). The operation terminal 104A performs input and output of various information to and from the server 101A connected communicably. The server 101A and the operation terminal 104A are connected to the internal network of insurance company (Company A). Images and application information received from the user terminal 103 and stored in the server 101A can be viewed from the operation terminal 104A belonging to insurance company (Company A), and are configured not to be viewable from the servers and operation terminals of other insurance companies or from the server 102.

[0014] In addition, the server 101B and the operation terminal 104B are devices used by users (such as employees) of insurance company (Company B). The operation terminal 104B performs input and output of various information to and from the server 101B connected communicably. The server 101B and the operation terminal 104B are connected to the internal network of insurance company (Company B). Images and application information received from the user terminal 103 and stored in the server 101B can be viewed from the operation terminal 104B belonging to insurance company (Company B), and are configured not to be viewable from the servers and operation terminals of other insurance companies or from the server 102.

[0015] The user terminal 103 has a function of receiving an operation from the user and transmitting insurance claim information including a photo of a damaged insured object (such as a building, household property, automobile, etc.) (also referred to as an image or an insured-claimed image in this embodiment) and application information necessary for online insurance claim (contract number, insured number (insured No.), accident occurrence date and time, situation description text, etc.) to the server (101A or 101B) of the insurance company with which the user has an insurance contract for online insurance claim. The image of this damaged insured object (also referred to as an insured object) is used in image appraisal for evaluating the damage situation of the object and calculating repair costs and insurance amounts.

[0016] Server 102 is a cloud server used by insurance companies when processing insurance claims online, and is operated by a different company (Company C) than insurance companies A and B. Server 102 works in conjunction with Servers 101A and 101B and has a function to determine whether an image of an insured object sent from User Terminal 103 is a fraudulent image. Specifically, Server 102 has a function to determine whether an image of an insured object sent from User Terminal 103 is an image that has been used to file an insurance claim with another insurance company in the past, or whether it is an image from the internet. Furthermore, Server 102 provides this determination function and other services as SaaS (Software as a Service) to Servers 101A and 101B of each insurance company.

[0017] Server 101A is an on-premise or cloud server of the insurance company (Company A). Server 101A receives and stores insurance claim information from user terminal 103, including an image of the damaged insured object and application information necessary for the insurance claim online, as a new insurance claim, and has a function to determine whether the image is fraudulent. If Server 101A determines that the image is fraudulent, it determines that the insurance claim from user terminal 103 is a fraudulent insurance claim and has a function to request user terminal 103 to retake the insured object or to instruct the dispatch of staff such as employees of the insurance company (Company A) to operate the terminal. Furthermore, if Server 101A determines that the image received from user terminal 103 is not fraudulent and determines that the insurance claim from user terminal 103 is a legitimate insurance claim, it has a function to accept the insurance claim, evaluate the damage from the image, calculate the repair costs and insurance amount, and process the payment of insurance money.

[0018] Server 101B is either an on-premise server or a cloud server of the insurance company (Company B). Server 101B receives and stores insurance claim information from user terminal 103, including an image of the damaged insured object and application information necessary for online insurance claims, as a new insurance claim, and has a function to determine whether the image is fraudulent. If Server 101B determines that the image is fraudulent, it determines that the insurance claim from user terminal 103 is a fraudulent insurance claim and has a function to request user terminal 103 to retake the insured object or to instruct the dispatch of staff such as employees of the insurance company (Company B) to operate the terminal. Furthermore, if Server 101B determines that the image received from user terminal 103 is not fraudulent and determines that the insurance claim from user terminal 103 is a legitimate insurance claim, it has a function to accept the insurance claim, evaluate the damage from the image, calculate the repair costs and insurance amount, and process the payment of insurance money. In this specification, the letters "A" and "B" at the end of the code are used to identify the company (Company A, Company B) to which the server or component belongs. If the letter preceding the final letter is the same, it indicates the same component or functional part. In this embodiment, two insurance companies, Company A and Company B, are described, but there may be three or more insurance companies. Furthermore, Server 101A, Server 101B, and Server 102 may be logically separate servers, or they may be servers built on the same cloud environment.

[0019] Figure 2 shows an example of the hardware configuration of Server 101A, Server 101B, Server 102, User Terminal 103, Operation Terminal 104A, and Operation Terminal 104B. Servers 101A, 101B, and 102, user terminal 103, operation terminal 104A, and operation terminal 104B are all information processing devices. As shown in Figure 2, the information processing device includes a CPU 201, ROM 202, RAM 203, external memory 204, input unit 205, display unit 206, communication interface 207 and I / O 208, and a system bus 209 that connects these devices to each other. I / F is an abbreviation for interface. I / O is an abbreviation for Input / Output.

[0020] ROM202 stores the BIOS program and boot program, and RAM203 is used as temporary storage for CPU201. External memory204 stores the computer program executed by CPU201, and by executing the computer program, CPU201 controls each device connected to the system bus209 and performs the processing in this embodiment.

[0021] The input unit 205 processes information input from a keyboard, mouse, etc. The keyboard, mouse, etc. may be provided by the information processing device, or they may be provided by an external device connected to the information processing device. The display unit 206 outputs the calculation results of the information processing device to the display device according to instructions from the CPU 201.

[0022] The display device may be a liquid crystal display device, a projector, an LED indicator, etc., and may be built into the information processing device or installed outside the information processing device.

[0023] Communication I / F207 facilitates communication between the information processing device and the outside world. The outside world includes devices on network 105 and servers on the internet.

[0024] Communication I / F207 is used for information communication over a network, and the communication interface can be Ethernet, USB, serial communication, wireless communication, etc., regardless of the type. I / O208 handles the input and output of each device connected to system bus 209.

[0025] Figure 3 is a flowchart showing an example of processing in an information processing system according to an embodiment of the present invention.

[0026] The processes shown in S301 and S302 in Figure 3 are executed by the CPU of the user terminal 103 executing a computer program stored in memory. The processes shown in S303 to S312 in Figure 3 are executed by the CPU of the server 101A executing a computer program stored in memory. Some of the detailed processes in S304 and S305 are executed by the CPU of the server 102 executing a computer program stored in memory.

[0027] In this embodiment, we will describe an example in which servers 101A and 102 determine whether an image of a damaged insured object, which has been submitted to an insurance company (Company A) by a user terminal 103, is a fraudulent image.

[0028] The user terminal 103 receives insurance claim information from users such as insured persons (S301). This insurance claim information includes images of the damaged insured property (buildings, household goods, automobiles, etc.) and application information necessary for the insurance claim (contract number, insured person number (insured person No.), date and time of accident, description of the situation, etc.).

[0029] Then, when the user terminal 103 receives a transmission instruction from the user, it sends the entered insurance claim information to the insurance company's server 101A for the purpose of filing an insurance claim (S302).

[0030] When server 101A receives insurance claim information, it stores (registers) the images of the insured object included in the insurance claim information in the insurance claim image database 801A (S303). Figure 12 shows an example of the configuration of the insurance claim image database 801A. As shown in the figure, the insurance claim image database 801A stores application information such as the insured person number, image ID, and images included in the insurance claim information in association with each other.

[0031] Next, server 101A determines whether the image stored (registered) in the insurance claim image database 801A is similar (including identical) to an image previously claimed by another insurance company (Company B) (S304). In S304, server 101A can also determine whether the image stored (registered) in the insurance claim image database 801A is similar (including identical) to an image previously claimed by its own insurance company (Company A) and another insurance company (Company B). In this specification, "similar" is a concept that includes identical images. Figure 4 shows the details of the past image reuse determination process in S304, and Figure 8 shows a functional block diagram of the past image reuse determination process. Details of the past image reuse determination process will be described later using Figures 4 and 8.

[0032] If, in S304, server 101A determines that the image is not similar to an image previously claimed by another insurance company, it determines in S303 whether the image included in the insurance claim information is similar to an image publicly available on the internet (S305). Figure 5 shows the details of the internet image reuse determination process in S305, and Figure 9 shows a functional block diagram of the internet image reuse determination process. Details of the internet image reuse determination process will be described later using Figures 5 and 9.

[0033] If, in S305, server 101A determines that the image is not similar to any image publicly available on the internet, it determines in S303 whether the image included in the insurance claim information received has been altered (S306). Figure 6 shows the details of the image alteration determination process in S306. Details of the image alteration determination process will be described later using Figure 6.

[0034] In S306, if Server 101A determines that an image included in the insurance claim information has not been processed, it determines whether the image was generated by the generation AI (S307). Figure 7 shows the details of the generation AI image determination process in S307. Details of the generation AI image determination process will be described later with reference to Figure 7.

[0035] If, as a result of the AI ​​image recognition process in S307, server 101A determines that the image included in the insurance claim information is an image generated by the AI, it determines that the insurance claim is a fraudulent application (S308: YES) and proceeds to S311. If, as a result of the AI ​​image recognition process in S307, it determines that the image included in the insurance claim information is not an image generated by the AI, it determines that the insurance claim is a legitimate application and not a fraudulent application (S308: NO) and proceeds to S309.

[0036] Furthermore, as will be described later, if Server A determines in S304 that the image included in the insurance claim information received in S303 is similar to an image previously used in an insurance claim with another insurance company, or in S305 that it is similar to an image publicly available on the internet, or in S306 that it has been manipulated, Server A determines that the insurance claim is fraudulent and proceeds to processing in S311.

[0037] If server 101A determines that the insurance claim is a valid application (S308: NO), it accepts the insurance claim (insurance payment claim) from user terminal 103 (S309), evaluates the damage from the images included in the insurance claim information, calculates the repair costs and insurance amount, and processes the insurance payment (S310). S309 and S310 are examples of the application of processing related to insurance payment for insurance payments.

[0038] Furthermore, if server 101A determines that the insurance claim is a fraudulent application (S308: YES), it controls the operation terminal 104A to output the results of the fraudulent application determination in S304 to S307 and display them on the display unit 206 of the operation terminal 104A (S311). Figures 10 and 11 are examples of display screens showing the fraudulent application determination results displayed in step S311.

[0039] Figure 10 shows an example of the display screen 1001 that is displayed in S311 when, in S304, it is determined that the image included in the insurance claim information submitted is similar to an image previously submitted to another insurance company for insurance claims.

[0040] Figure 11 is an example of the display screen 1101 that is displayed in S311 when it is determined that an image included in the insurance claim information submitted is similar to an image that is publicly available on the internet.

[0041] Figures 10 and 11 show a re-photography reception unit 1005 that receives instructions to send a request to the user terminal for re-photography of the insured object. Figures 10 and 11 also show a dispatch request reception unit 1006 that receives instructions to send a request to dispatch an insurance officer (investigator) to the site to inspect the insured object on-site.

[0042] If the operating terminal 104 receives a request to re-photograph the insured object via the display screen showing the result of the fraudulent application (when the re-photograph request unit 1005 is pressed), it sends the re-photograph instruction to the server 101A. If the server 101A receives a re-photograph instruction from the operating terminal 104, it sends a request to the user terminal 103 to re-photograph the insured object.

[0043] Furthermore, if the operating terminal 104 receives an instruction to dispatch an officer to the site via the display screen showing the result of the fraudulent application determination (when the dispatch request reception unit 1006 is pressed), it sends the instruction to dispatch an officer to the site to the server 101A. When the server 101A receives an instruction to dispatch an officer to the site from the operating terminal 104, it sends a request to dispatch an insurance officer to the site to a terminal (not shown) operated by the insurance officer who will be inspecting the insured item on-site (S312). After executing the process in S310 or S312, the server 101A returns to S303 and waits for the receipt of new insurance claim information.

[0044] The flowchart in Figure 3 illustrates an example where, in S308, if it is determined that the insurance claim is not fraudulent, the insurance claim is automatically accepted and the payment process is carried out. However, the system may also be configured to output the determination result to the display unit 206 of the operation terminal 104A, and to execute the payment process upon receiving instructions from the user to accept the insurance claim and to make a payment.

[0045] Furthermore, the system can be configured to automatically send a request for reshooting or an instruction to dispatch to the site in S311 without displaying the judgment result. For example, even if the judgment is YES in S411, S506, S516, S602, and S702 and the process proceeds to S311, the system can be configured to automatically send a request for reshooting or an instruction to dispatch to the site in S311 without displaying the judgment result. In addition, the system can be configured in S312 to automatically send a request for reshooting if the insurance claim amount is less than a predetermined value, and to automatically issue an instruction to dispatch to the site if it is equal to or greater than the predetermined value.

[0046] In the flowchart shown in Figure 3, the following processes are executed: past image reuse detection process (S304), internet image reuse detection process (S305), image processing detection process (S306), and generated AI image detection process (S307). However, it is possible to configure the system to execute at least the internet image reuse detection process (S305) after the processing in S303. In this case, one or more of the processes of past image reuse detection process (S304), image processing detection process (S306), and generated AI image detection process (S307), or all of them, can be skipped, and the presence or absence of a fraudulent application can be determined in S308 based on the detection result of the internet image reuse detection process (S305). Server 101A can decide to perform processing related to the payment of insurance benefits related to the insurance claim, on the condition that the detection result of the internet image reuse detection process (S305) indicates that the image of the insured object, which was sent from the user terminal 103 to the insurance company (Company A) for the purpose of claiming insurance benefits, is not similar to an image published on the internet, and can perform said processing.

[0047] Furthermore, the display screens shown in Figures 10 and 11 may be configured to show a receiving unit that receives instructions to accept insurance claims, and to execute the processes in S309 and S310 when such instructions are received from the user.

[0048] Next, the detailed processing of the past image reuse determination process in S304 will be explained using Figures 4 and 8. Figure 4 is a flowchart showing an example of the detailed processing of the past image reuse determination process in S304. Figure 8 is an example of a functional block diagram related to the past image reuse determination process.

[0049] Server 101A comprises an insurance claim image database 801A, a feature vector generation unit 802A, a transmission unit 804A, a reception unit 805A, and a control unit 816A. Server 101B has a similar configuration, so its description is omitted.

[0050] Furthermore, the server 102 includes a feature vector database 806, a receiving unit 807, a similarity calculation unit 808, a determination unit 809, and a transmission unit 810.

[0051] The processes shown in S401, S402, S410, and S411 in Figure 4 are executed by the CPU of server 101A executing computer programs stored in memory. Furthermore, the processes shown in S403 to S409 in Figure 4 are executed by the CPU of server 102 executing computer programs stored in memory.

[0052] The insurance claim image database 801A is a storage means that stores images included in insurance claim information acquired (received) from the user terminal 103. These images are photographs of damaged insured items for which an insurance claim has been submitted from the user terminal 103 to the insurance company (Company A).

[0053] Furthermore, the insurance claim image database 801B is a storage means that stores images included in the insurance claim information acquired (received) from the user terminal 103. These images are photographs of damaged insured objects that have been claimed from the user terminal 103 to the insurance company (Company B).

[0054] This embodiment describes an example in which server 102 determines whether an image of a damaged insured object, for which an insurance claim has been submitted to insurance company (Company A) by user terminal 103, is similar to an image previously submitted for insurance claims to another insurance company (Company B), and outputs the determination result to server 101A of Company A. This embodiment describes an example in which server 101A executes the processing from S304 onwards triggered by receiving insurance claim information in S303. However, the processing from S304 onwards may also be triggered by receiving instructions from the insurance company's staff. Alternatively, the processing from S304 onwards may be executed at a predetermined time.

[0055] First, the feature vector generation unit 802A obtains an image of the insured object received in S303 from the insurance claim image database 801A, and generates a feature vector from that image (S401). The feature vector generation unit 802A can generate feature vectors from an image using a convolutional neural network (CNN), Vision Transformer, or local feature extraction algorithm (SIFT, SURF, etc.). It can generate feature vectors from an image using existing technologies for generating feature vectors from images. The transmitting unit 804A then transmits the generated feature vector to the server 102 (S402). The receiving unit 807 receives (acquires) the feature vector (S403) and stores (registers) it in the feature vector database 806 (S404).

[0056] The receiving unit 807 can also receive feature vectors transmitted from the transmitting unit 804B of another insurance company (Company B) and store (register) them in the feature vector database 806. Therefore, the feature vector database 806 acquires and stores feature vectors of images that have been claimed by insurance companies in the past from the respective servers (101A, 101B) of multiple insurance companies (Company A and Company B) that acquire images of damaged insured items from the user terminal 103 for insurance claims. Figure 12 shows an example of the feature vector database of past claim images. The feature vector database 806 of past claim images shown in Figure 12 stores the image ID of an image included in insurance claim information previously claimed by insurance company (Company A), the feature vector of that image, and the image ID of an image included in insurance claim information previously claimed by insurance company (Company B), the feature vector of that image.

[0057] In this embodiment, each insurance company needs to keep confidential the images submitted for insurance claims by users (such as insured persons), and therefore cannot send the images themselves to another company's server 102. Therefore, instead of sending the images themselves to server 102, the system is configured to send and share feature vectors generated (extracted) from the images to server 102.

[0058] The similarity calculation unit 808 obtains the feature vectors of each image that has been previously claimed by another insurance company (Company B) from the feature vector database 806 (S405), and calculates the similarity between the obtained feature vectors of each image and the feature vectors received in S403 (S406).

[0059] Then, the determination unit 809 determines whether the similarity calculated by the similarity calculation unit 808 is equal to or greater than a predetermined threshold, thereby determining whether the image of the feature vector received in S403 is similar to an image that has been claimed by another insurance company in the past (S407).

[0060] If the determination unit 809 determines that the value is above a predetermined threshold, it determines that the image of the feature vector received in S403 is similar to an image that has been claimed by another insurance company in the past. On the other hand, if it determines that the value is below the predetermined threshold, it determines that there is no image that is similar to the image of the feature vector received in S403 and that has been claimed by another insurance company in the past.

[0061] If the determination unit 809 determines that there is an image similar to the image of the feature vector received in S403 (an image for which an insurance claim was made with another insurance company in the past) (S408:YES), it proceeds to S409. If it determines that there is no similar image (S408:NO), it proceeds to S501.

[0062] Then, the transmission unit 810 transmits (outputs) the determination result from the determination unit 809 in S407 to the server 101A (S409).

[0063] The receiving unit 805A receives (acquires) the determination result (S410). Then, if the control unit 816A determines that the determination result indicates the existence of a similar image (the image of the insured object claimed from the user terminal 103 to the insurance company (Company A) is similar to an image previously claimed to another insurance company (Company B)) (S411: YES), the process moves to S311. If the determination result indicates that there is no similar image (the image of the insured object claimed from the user terminal 103 to the insurance company (Company A) is not similar to an image previously claimed to another insurance company (Company B)) (S411: NO), the process moves to S505.

[0064] The similarity calculation unit 808 can also be configured to obtain feature vectors from the feature vector database 806 for each image for which an insurance claim was previously filed with the insured company (Company A) and another insurance company (Company B), and to calculate the similarity between the feature vectors of each obtained image and the feature vectors received in S403. In this case, the determination unit 809 can determine whether the image of the feature vectors received in S403 is similar to images for which an insurance claim was previously filed with the insured company (Company A) and another insurance company (Company B) by determining whether the similarity calculated by the similarity calculation unit 808 is above a predetermined threshold (S407, S408).

[0065] Figure 10 shows the display screen 1001 that is displayed in S311 when the judgment result in S411 is determined to indicate that there are similar images (YES). The display screen 1001 shows the application information, the judgment result 1002, and the application image 1003 for which the insurance claim was made. An example of the image shown in application image 1003 is an image of a scratch on the exterior wall. In S311, the server 101A outputs (transmits) the information displayed on the display screen 1001 to the operation terminal 104A and controls it to be displayed on the display unit 206 of the operation terminal 104A.

[0066] By using a past image reuse detection process, it is possible to determine whether an image used in an insurance claim is similar to an image used in an insurance claim with another insurance company, thereby reducing fraudulent claims (duplicate claims with multiple insurance companies). In addition, insurance companies can check for similarity (match) with images used in past insurance claims across multiple insurance companies while keeping the images themselves confidential, further reducing fraudulent claims (duplicate claims with multiple insurance companies).

[0067] Next, the detailed processing of the internet image reuse detection process in S305 will be explained using Figures 5 and 9. Figure 5 is a flowchart showing an example of the detailed processing of the internet image reuse detection process in S305. Figure 9 is an example of a functional block diagram related to the internet image reuse detection process.

[0068] Server 101A, shown in Figure 9, is configured as shown in Figure 8, but further includes an image language processing unit 803A. Server 102, also shown in Figure 9, is configured as shown in Figure 8, but further includes a keyword image search unit 811, a data collection unit 813, a feature vector generation unit 814, and an internet image feature vector database 815.

[0069] The processes shown in S505-S508, S515, and S516 in Figure 5 are executed by the CPU of server 101A executing computer programs stored in memory. Furthermore, the processes shown in S501-S504, S504-1, and S509-514 in Figure 5 are executed by the CPU of server 102 executing computer programs stored in memory.

[0070] The collection unit 813 is a web crawler that automatically visits web pages and social networking services on the internet 812 and has the function of collecting images. The collection unit 813 collects images (internet images) that are publicly available on the internet from web servers. The images mainly collected here are images of insured objects that have been damaged, which are covered by insurance companies.

[0071] The feature vector generation unit 814 generates feature vectors for each image collected in advance by the collection unit 813 and stores (registers) them in the Internet image feature vector database 815. The feature vector generation unit 814 can generate feature vectors from collected images using a convolutional neural network (CNN), Vision Transformer, or local feature extraction algorithm (SIFT, SURF, etc.). It can generate feature vectors for images using existing techniques for generating image feature vectors. After generating the feature vectors for each image, the image is erased from memory.

[0072] An example of an internet image feature vector database 815 is shown in Figure 12. The feature vector database 815 shown in Figure 12 stores the image ID of the image collected by the collection unit 813, the feature vector of the image generated by the feature vector generation unit 814, and the URL of the source from which the image was collected.

[0073] The similarity calculation unit 808 obtains the feature vectors for each image from the feature vector database 815 (S501). Then, the similarity calculation unit 808 calculates the similarity between the feature vectors for each image obtained in S501 and the feature vectors received in S403 (S502).

[0074] Then, the determination unit 809 determines whether the image of the feature vector received in S403 is similar to an image published on the internet by determining whether the similarity calculated by the similarity calculation unit 808 is equal to or greater than a predetermined threshold (S503).

[0075] If the determination unit 809 determines that the value is above a predetermined threshold, it determines that the image of the feature vector received in S403 is similar to an image published on the internet and that a similar image exists. On the other hand, if it determines that the value is below the predetermined threshold, it determines that there is no image published on the internet that is similar to the image of the feature vector received in S403.

[0076] Then, the transmission unit 810 transmits (outputs) the determination result from the determination unit 809 in S503 to the server 101A (S504). In addition, if the transmission unit 810 determines in S503 that there is a similar image, it also transmits (outputs) the URL of the similar image (an image published on the internet) to the server 101A.

[0077] After server 102 sends (outputs) the judgment result to server 101A in S504, if it is determined in S503 that there are similar images (S504-1: YES), it terminates the process; if it is determined that there are no similar images (S504-1: NO), it moves the process to S509.

[0078] The receiving unit 805A receives (acquires) the determination result (S505). Then, the control unit 816A determines whether the determination result indicates the existence of a similar image (S506). If the determination result indicates that a similar image exists (i.e., the image of the insured object for which an insurance claim was submitted to the insurance company (Company A) from the user terminal 103 is similar to an image published on the internet) (S506: YES), the process proceeds to S311. If the determination result indicates that there is no similar image (i.e., the image of the insured object for which an insurance claim was submitted to the insurance company (Company A) from the user terminal 103 is not similar to an image published on the internet) (S506: NO), the process proceeds to S507.

[0079] Next, the image language processing unit 803A acquires the image of the insured object obtained in S401 and generates text describing the image based on that image (S507). The image language processing unit 803A can generate text describing the image based on the image using image caption generation technology, a visual language model (VLM), a multimodal large language model (LLM), or a small language model (SLM). Here, text includes words (keywords) or sentences.

[0080] Furthermore, when generating text describing the image using a cloud-based image caption generation technology, visual language model (VLM), multimodal large language model (LLM), or small language model (SLM) service, the image language processing unit 803A may generate the text by sending the image to the service, the service generating text describing the image, and the image language processing unit 803A acquiring the text.

[0081] Then, the image language conversion unit 803A transmits the text generated in S507 to the keyword image search unit 811 (S508).

[0082] When the keyword image search unit 811 receives (acquires) the text from the image language conversion unit 803A (S509), it uses the text as a search query and instructs a web search engine server on the internet 812 (for example, Google's search engine server) to perform a text search for images published on the internet, thereby performing an image search (S510).

[0083] When the keyword image search unit 811 obtains each image (an image published on the internet) found in S510, the feature vector generation unit 814 generates a feature vector for each image from that image (S511) and stores (registers) it in the internet image feature vector database 815.

[0084] Then, the similarity calculation unit 808 obtains the feature vectors of each image generated in S511 from the feature vector database 815 and calculates the similarity between the feature vectors of each image and the feature vectors received in S403 (S512).

[0085] Then, the determination unit 809 determines whether the image of the feature vector received in S403 is similar to an image published on the internet by determining whether the similarity calculated by the similarity calculation unit 808 is equal to or greater than a predetermined threshold (S513).

[0086] If the determination unit 809 determines that the value is above a predetermined threshold, it determines that the image of the feature vector received in S403 is similar to an image published on the internet and that a similar image exists. On the other hand, if it determines that the value is below the predetermined threshold, it determines that there is no image published on the internet that is similar to the image of the feature vector received in S403.

[0087] Then, the transmission unit 810 transmits (outputs) the determination result from the determination unit 809 in S513 to the server 101A (S514). In addition, if the transmission unit 810 determines in S513 that there is a similar image, it also transmits (outputs) the URL of the similar image (an image published on the internet) to the server 101A.

[0088] The receiving unit 805A receives (acquires) the determination result (S515). Then, the control unit 816A determines whether the determination result indicates the existence of a similar image (S516). If the determination result indicates that a similar image exists (i.e., the image of the insured object for which an insurance claim was submitted to the insurance company (Company A) from the user terminal 103 is similar to an image published on the internet) (S516: YES), the process proceeds to S311. If the determination result indicates that there is no similar image (i.e., the image of the insured object for which an insurance claim was submitted to the insurance company (Company A) from the user terminal 103 is not similar to an image published on the internet) (S516: NO), the process proceeds to S601.

[0089] Figure 11 shows the display screen 1101 that is displayed in S311 when the judgment result in S506 or S516 indicates that there is a similar image (YES). The display screen 1101 shows the application information, the judgment result 1102, the application image 1003 for which the insurance claim was made, the image 1104 that was determined to be similar and is publicly available on the internet, and the URL 1105 from which the image was obtained. The publicly available image 1104 is obtained and displayed by the operation terminal 104A by accessing URL 1105.

[0090] The above internet image reuse detection process allows insurance companies to determine whether an image of an insured object submitted for insurance claims is similar to an image publicly available on the internet, thereby reducing fraudulent claims (insurance claims using images publicly available on the internet).

[0091] Furthermore, the internet image misuse detection process makes it possible to check for similarity (match) between the feature vectors of images obtained from the internet and the feature vectors of images of insured objects for which insurance claims have been filed. This allows insurance companies to reduce fraudulent claims (insurance claims using images publicly available on the internet) while keeping the images themselves confidential.

[0092] Furthermore, because a feature vector database 815 that stores feature vectors of previously collected internet images is used, the processing speed related to fraud detection in S501 to S503 can be improved.

[0093] Furthermore, the internet image misuse detection process is configured such that server 101A generates text describing the image of the insured object for which an insurance claim has been filed, and server 102 uses this text as a search query to search for images publicly available on the internet. As a result, insurance companies can check for similarity (match) with images publicly available on the internet while keeping the image itself confidential, thereby reducing fraudulent claims (insurance claims using images publicly available on the internet).

[0094] Furthermore, if a feature vector for a similar image is not found in the feature vector database 815, which stores feature vectors of previously collected internet images, an image search is performed using text describing the image of the insured object for which an insurance claim was filed. This allows for an internet image search for images similar to the image in question, thereby improving the accuracy of detecting fraudulent claims (insurance claims using images publicly available on the internet).

[0095] Figure 5 shows a series of processes 1 (S501~, S506) which calculate the similarity between the feature vectors of each internet image pre-collected by the collection unit 813 and the feature vectors of the images included in the insurance claim information received in S303, and determine whether the images are similar. Figure 5 also shows a series of processes 2 (S508~, S516) which calculate the similarity between the feature vectors of the images included in the insurance claim information received in S303 and the feature vectors of each internet image retrieved using the text of the image's description as a search query, and determine whether the images are similar. It is possible to execute only one of these two series of processes, or to perform them in parallel simultaneously.

[0096] Next, the detailed processing of the image processing judgment process in S306 will be explained using Figure 6. Figure 6 is a flowchart showing an example of the detailed processing of the image processing judgment process in S306.

[0097] The processes shown in S601 and S602 of Figure 6 are executed by the CPU of server 101A executing a computer program stored in memory.

[0098] Server 101A analyzes the image of the insured object received in S303 and determines the degree of processing, which is the extent to which the image has been manipulated (S601). Server 101A can calculate the degree of image processing using pixel statistical feature analysis, frequency domain analysis, or machine learning models. Server 101A then determines whether the degree of processing is above a predetermined threshold (S602). If it is determined to be above the predetermined threshold, it is determined that the image has been manipulated (S602: YES), and the process moves to S311. On the other hand, if it is determined to be below the predetermined threshold (S602: NO), it is determined that the image has not been manipulated, and the process moves to S701. If it is determined that the image has been manipulated, in S311, the degree of processing from S601 and the determination that the image has been manipulated are displayed on the display screen indicating the result of the fraudulent application determination.

[0099] Next, the detailed processing of the AI ​​image recognition process in S307 will be explained using Figure 7. Figure 7 is a flowchart showing an example of the detailed processing of the AI ​​image recognition process in S307.

[0100] The processes shown in S701 and S702 in Figure 7 are executed by the CPU of server 101A executing computer programs stored in memory.

[0101] Server 101A analyzes the image of the insured object received in S303 and determines whether the image was generated by a generative AI (S701). In S701, it determines the probability / percentage of the image being likely to be generated by a generative AI. To determine the possibility that the image was generated by a generative AI, Server 101A can calculate the probability / percentage of it being a generative AI image using image feature analysis or a learning model. Techniques such as deep learning models (CNN, Vision Transformer), generative image detection models (GAN detection), frequency domain analysis, or noise pattern analysis can be used for this determination. Existing image checkers that determine whether an image was generated by a generative AI may also be used. Server 101A then determines whether the image was generated by a generative AI by determining whether the probability / percentage is above a predetermined threshold.

[0102] If server 101A determines that the probability / percentage is above a predetermined threshold and that the image was generated by the generation AI (S701:YES), it proceeds to S311. If server 101A determines that the probability / percentage is below a predetermined threshold and that the image was not generated by the generation AI (S701:NO), it proceeds to S308. If it is determined that the image was generated by the generation AI, in S311, the percentage from S701 and the determination that the image was generated by the generation AI are displayed on the display screen indicating the result of the fraudulent application determination.

[0103] The processes are executed in the following order: past image reuse detection process (S304), internet image reuse detection process (S305), image processing detection process (S306), and generated AI image detection process (S307). In this embodiment, the detection processes are executed in order of least processing load or fastest processing speed, and if a fraudulent image is detected in any of the processes, the subsequent detection processes are not executed. This reduces the overall processing time required to detect fraudulent applications.

[0104] As described above, this embodiment makes it possible to reduce fraudulent insurance claims.

[0105] The present invention can take the form of, for example, a system, apparatus, method, program, or recording medium. Specifically, it may be applied to a system consisting of multiple devices, or to an apparatus consisting of a single device.

[0106] Furthermore, the various controls described above, which are performed by CPU201, may be performed by a single piece of hardware, or multiple pieces of hardware (for example, multiple processors or circuits) may share the processing to control the entire device.

[0107] Furthermore, the program in this invention is a program that allows a computer to execute the processing methods shown in the flowcharts from Figures 3 to 7, and the storage medium of this invention stores a program that allows a computer to execute the processing methods from Figures 3 to 7. Note that the program in this invention may also be a program for each processing method of each device in Figure 1.

[0108] As described above, it goes without saying that the object of the present invention can also be achieved by supplying a recording medium containing a program that realizes the functions of the embodiments described above to a system or device, and by having the computer (or CPU or MPU) of that system or device read and execute the program stored on the recording medium.

[0109] In this case, the program read from the recording medium itself realizes the novel function of the present invention, and the recording medium on which that program is recorded constitutes the present invention.

[0110] For recording media used to supply programs, examples include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, CD-Rs, DVD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, EEPROMs, silicon disks, and the like.

[0111] Furthermore, it goes without saying that the functions of the aforementioned embodiments are realized not only by the computer executing the program it has read, but also by the operating system (OS) running on the computer performing some or all of the actual processing based on the instructions of that program, thereby realizing the functions of the aforementioned embodiments.

[0112] Furthermore, it goes without saying that this also includes cases where, after a program read from a recording medium is written to the memory of a function expansion board inserted into a computer or a function expansion unit connected to a computer, the CPU or other components of the function expansion board or function expansion unit perform some or all of the actual processing based on the instructions of the program code, and the functions of the aforementioned embodiments are realized through that processing.

[0113] Furthermore, the present invention may be applied to a system consisting of multiple devices or to a device consisting of a single device. It goes without saying that the present invention can also be applied when the results are achieved by supplying a program to a system or device. In this case, by reading a recording medium containing a program for achieving the present invention into the system or device, the system or device can enjoy the effects of the present invention.

[0114] Furthermore, by downloading and reading the program for achieving the present invention from a server, database, etc. on a network using a communication program, the system or device can enjoy the effects of the present invention.

[0115] Although the present invention has been described in detail based on its preferred embodiments, the present invention is not limited to the above embodiments, and various modifications and combinations of the above embodiments are possible in accordance with the spirit of the present invention, and these are not excluded from the scope of the present invention. 。 [Explanation of symbols]

[0116] 101A Server (Company A) 101B Server (Company B) 102 Server (Company C) 103 User terminal 104A Operating Terminal 104B Operating Terminal 105 Network

Claims

1. An information processing system including a first server and a second server, The first server is, A first generation means that generates feature vectors of an application image of an insured object, based on the application image of the insured object submitted for insurance claim purposes, A second generation means generates text describing the application image based on the application image, Equipped with, The second server is, An instruction means that gives instructions to search for internet images, which are images published on the internet, using the aforementioned text as a search query, A third generation means for generating feature vectors of the internet images retrieved according to the above instructions, A determination means that determines whether the application image is similar to the internet image based on the feature vector generated by the first generation means and the feature vector generated by the third generation means, An output means for outputting the determination result by the determination means, An information processing system characterized by comprising the following features.

2. The text includes words or sentences. The information processing system according to claim 1, characterized by the following:

3. The first server is The information processing system according to claim 1, further comprising control means for controlling the display of a screen including the application image and the Internet image determined to be similar by the determination means, based on the determination result obtained from the second server.

4. The output means outputs the determination result by the determination means and the URL of the internet image determined to be similar by the determination means to the first server. The control means controls the display of the screen including the URL obtained from the second server. The information processing system according to claim 3, characterized by the following:

5. The determination means calculates the similarity between the feature vector generated by the first generation means and the feature vector generated by the third generation means, and determines that the application image is similar to the internet image if the calculated similarity is determined to be equal to or greater than a predetermined threshold. The information processing system according to claim 1, characterized by the following:

6. The first server comprises storage means for storing the application image, The first generation means generates feature vectors of the application image stored in the storage means, The second generation means generates text describing the application image based on the application image stored in the storage means. The first server includes a transmission means for transmitting the feature vector generated by the first generation means and the text generated by the second generation means to the second server. The second server includes an acquisition means for acquiring the feature vectors and text of the application image transmitted by the transmission means. The instruction means issues an instruction to search for the internet image using the text obtained by the acquisition means as a search query. The determination means determines whether the application image is similar to the internet image based on the feature vector of the application image acquired by the acquisition means and the feature vector generated by the third generation means. The information processing system according to claim 1, characterized by the following:

7. The transmission means transmits to the second server the feature vector generated by the first generation means and the text generated by the second generation means, without transmitting the application image to the second server. The information processing system according to claim 6, characterized by the following:

8. The first server is a server of an insurance company, The second server is a server belonging to a different company than the aforementioned insurance company. The information processing system according to claim 1, characterized by the following:

9. The second generation means generates text describing the application image based on the application image, using image caption generation technology, a visual language model (VLM), a multimodal large language model (LLM), or a small language model (SLM). The information processing system according to claim 1, characterized by the following:

10. The first generation means generates feature vectors of the application image based on the application image using a convolutional neural network (CNN), a Vision Transformer, or a local feature extraction algorithm, The third generation means generates feature vectors of the internet image based on the internet image using a convolutional neural network (CNN), Vision Transformer, or local feature extraction algorithm. The information processing system according to claim 1, characterized by the following:

11. A control method for an information processing system including a first server and a second server, The first generation means of the first server performs a first generation step of generating a feature vector of an application image of an insured object that has been applied for for an insurance claim, The second generation means of the first server performs a second generation step of generating text that describes the application image based on the application image, The instruction means of the second server gives an instruction to search for an internet image, which is an image published on the internet, using the text as a search query. The third generation means of the second server includes a third generation step of generating feature vectors of the Internet image retrieved by the instruction, The determination means of the second server includes a determination step in which it determines whether the application image is similar to the internet image based on the feature vector generated in the first generation step and the feature vector generated in the third generation step, The output means of the second server includes an output step that outputs the determination result in the determination step, A control method characterized by including

12. A program for causing at least one computer to function as one of the means of an information processing system described in any one of claims 1 to 10.

Images