Authenticity verification system

Abstract

To reduce a possibility of false negatives while reducing the workload in authenticity verification.SOLUTION: An authenticity verification system includes a terminal device which captures an image of an article, and a learning model server which verifies the authenticity of the article. The terminal device generates the luminance image of the article image and transmits the luminance image to the learning model server, determines, on the basis of a determination result transmitted from the learning model server, whether additional determination is necessary or not, transmits, if the additional determination is required, a color image of a part of the article image to the learning model server. The learning model server executes, on receipt of the luminance image from the terminal device, determination including as to whether the article is an authentic product or a prohibited product, using luminance image learning models of the authentic product and the prohibited product, and executes, on receipt of a color image from the terminal device, a determination including as to whether the article is an authentic product or a prohibited product, using color image learning models of the authentic product and the prohibited product.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an authenticity determination system, specifically, a system for determining whether an item is a seizure target such as at a customs office, an authentic product, or something else.

Background Art

[0002] Conventionally, at customs, whether an item is subject to seizure has been visually determined by customs officers based on application documents. For example, for seizure targets during import, since the number of imported items is enormous, the burden on officers is large, and since a full inspection of all items is practically impossible, there is a possibility of detection omissions.

Summary of the Invention

Problems to be Solved by the Invention

[0003] The present invention has been made in view of the above-described circumstances, and an object thereof is to provide an authenticity determination system capable of reducing the work burden during authenticity determination (for example, when detecting seizure targets at customs, etc.) and reducing the possibility of determination or detection omissions.

Means for Solving the Problems

[0004] To achieve the above object, an authenticity determination system according to the present disclosure is an authenticity determination system having a terminal device that captures an image of an item, and a learning model server that is connected to the terminal device via a communication network and determines the authenticity of the item, wherein the terminal device generates a luminance image of the image of the item, transmits the luminance image to the learning model server, determines whether further determination is necessary based on the determination result sent from the learning model server, and if further determination is necessary, transmits a partial color image of the image of the item to the learning model server, and when receiving the determination result sent from the learning model server, displays the determination result. The learning model server includes a genuine luminance image learning model generated from the teacher data of the luminance image of a genuine product, a luminance image learning model of an item subject to seizure generated from the teacher data of the luminance image of the item subject to seizure, a genuine color image learning model generated from the teacher data of the color image of a genuine product, and a color image learning model of an item subject to seizure generated from the teacher data of the color image of the item subject to seizure. When receiving a luminance image from the terminal device, it uses the genuine luminance image learning model and the luminance image learning model of the item subject to seizure to perform a determination on whether the item is a genuine product or an item subject to seizure, and transmits the result of the determination to the terminal device. When receiving a color image from the terminal device, it uses the genuine color image learning model and the color image learning model of the item subject to seizure to perform a determination on whether the item is a genuine product or an item subject to seizure, and transmits the result of the determination to the terminal device.

Advantages of the Invention

[0005] As described above, in the counterfeit detection system of the present invention, the work load during counterfeit determination (for example, when detecting items subject to seizure at customs, etc.) can be reduced, and the possibility of omission in determination or detection can be reduced.

Brief Description of the Drawings

[0006]

Figure 1

Figure 2

Figure 3

Embodiments for Carrying Out the Invention

[0007] Embodiments of the counterfeit determination system according to the present invention will be described below with reference to the drawings. 1. System Configuration and Functional Outline The authenticity determination system according to this embodiment is composed of the following devices (1) to (3). Each device is connected by a communication network.

[0008] (1) Terminal (smartphone) This terminal (terminal device) is used by an operator who performs authenticity determination (determination of whether it is a genuine product or an object to be prohibited). When the operator takes a color image of a product (it is unknown whether it is a genuine product, an object to be prohibited, or something else) with the camera equipped on the terminal, the terminal generates a luminance image (which may also be a grayscale image or a brightness image). The generated luminance image is sent from the terminal to the learning model server via the communication network.

[0009] When the terminal receives the determination result (probability of correspondence to a genuine product (real thing), an object to be prohibited (counterfeit), and region information) sent from the learning model server, it determines whether a more detailed (i.e., highly accurate) determination is necessary. If a detailed determination is necessary, it sends the color image extracted based on the region information to the learning model server. Then, it receives the determination result based on the color image sent from the learning model server and displays it on the terminal screen.

[0010] Fig. 1 shows an example of the display on the terminal screen. This screen consists of an area for displaying the photographed image of the article, a status display area for displaying status such as whether the determination is in progress, and a determination result display area for displaying the determination result. Also, when the probability of correspondence to a genuine product or an object to be prohibited is equal to or greater than a certain value, the registration ID is displayed. The registration ID contains link information to the detailed data associated with the registration ID in the anti-counterfeiting product management server. By selecting the registration ID, the information of the link is displayed on a separate screen.

[0011] (2) Learning model server For each of the genuine product and the object to be prohibited, a learning model is generated based on the teacher data. When the learning model server receives the luminance image sent from the terminal, it obtains the object detection results {probability and region} for each of the genuine product and the object to be prohibited and returns them to the terminal. When receiving a color image of a partial area sent from the terminal, it obtains the {probability} of the object detection result for each of the genuine product and the product subject to injunction, and returns it to the terminal.

[0012] (3) Injunction Product Management Server It stores the injunction information (details) and the presence or absence of the color image for each registration ID (injunction application unit). In response to a request from the terminal, it returns the injunction information (details) and the information on the presence or absence of the color image to the terminal. The processing between the terminal and the injunction product management server is shown in Figure 3.

[0013] 2. Functions and Processing of the Learning Model Server (see Figure 2) (1) Since image data is transmitted between the terminal and the learning model server, the problem is how to reduce the communication load between them. (2) To solve this problem, the learning model of this embodiment generates a luminance image and a color image for each of the genuine product and the product subject to injunction.

[0014] Specifically, the following models are created for each registration ID (for example, the application unit to the customs). · Genuine Product Luminance Image Learning Model · Product Subject to Injunction Luminance Image Learning Model · Genuine Product Color Image Learning Model · Product Subject to Injunction Color Image Learning Model

[0015] (3) Processing Outline When the learning model server receives a command indicating the use of the luminance image learning model and the luminance image data sent from the terminal, it uses the genuine product luminance image learning model and the product subject to injunction luminance image learning model respectively to detect the corresponding probability and region coordinates and send them to the terminal. Note that the luminance image data sent from the terminal is the entire range of the photographed image.

[0016] When the learning model server receives a command to use the color image learning model and color image data sent from the terminal, it detects the corresponding probability using the genuine color image learning model and the color image learning model of the product to be blocked, respectively, and transmits it to the terminal. Note that the color image data is the image data of the area range detected by the luminance image learning model.

[0017] (4) Effect: According to this embodiment, first, using the luminance image learning model, the corresponding probabilities (0 to 1) of the genuine product and the product to be blocked are obtained, and only when there is no difference between the two, the corresponding probability is obtained using the color image data of a partial area, so that the communication traffic between the terminal and the server can be reduced.

Claims

【Claim 1】 A authenticity determination system comprising a terminal device for photographing an image of an article, and a learning model server connected to the terminal device via a communication network for determining the authenticity of the article, wherein the terminal device generates a luminance image of the image of the article, transmits the luminance image to the learning model server, determines whether further determination is necessary based on the determination result sent from the learning model server, and if further determination is necessary, transmits a partial color image of the image of the article to the learning model server, and when receiving the determination result sent from the learning model server, displays the determination result; the learning model server includes a genuine luminance image learning model generated from teacher data of a genuine luminance image, a luminance image learning model of an article subject to injunction generated from teacher data of a luminance image of the article subject to injunction, a genuine color image learning model generated from teacher data of a genuine color image, and a color image learning model of an article subject to injunction generated from teacher data of a color image of the article subject to injunction; when receiving a luminance image from the terminal device, the learning model server uses the genuine luminance image learning model and the luminance image learning model of the article subject to injunction to perform a determination as to whether the article is a genuine article or an article subject to injunction, and transmits the result of the determination to the terminal device, and when receiving a color image from the terminal device, uses the genuine color image learning model and the color image learning model of the article subject to injunction to perform a determination as to whether the article is a genuine article or an article subject to injunction, and transmits the result of the determination to the terminal device.

Images