JP6925615B2 - Identity verification document authenticity system, method and program - Google Patents

Description

The present invention relates to a technique for determining the authenticity of identity verification documents such as a driver's license, passport, and health insurance card.

In modern society, identity verification documents are required to be presented in various places. Official certificates such as driver's licenses, passports, and health insurance cards are used as identity verification documents. In online transactions or contracts on websites of various services, it may be required to send images of identity verification documents.

On the other hand, in recent years, advances in copiers and digital devices have made it easier to forge identity verification documents in an elaborate manner. Such elaborate counterfeiting can be difficult to discern visually.

Therefore, there is a demand for a system that can be easily used in various situations and that can determine the authenticity of an identity verification document with high accuracy.

Patent Document 1 discloses a technique for determining the authenticity of a driver's license. The technique disclosed in Patent Document 1 compares an image of a specific part of a driver's license with an image of a real driver's license, and determines authenticity based on the comparison result.

Japanese Unexamined Patent Publication No. 2006-11866

As described above, the technique disclosed in Patent Document 1 compares an image of a specific part of a driver's license with an image of a real driver's license, and determines authenticity based on the comparison result. Therefore, if the specific part is forged elaborately or the image of the specific part is unclear, the accuracy of the authenticity determination may decrease.

An object of the present invention is to provide a technique for improving the determination accuracy of a forged identity verification document.

The identity verification document authenticity determination system according to one embodiment of the present invention is constructed by using the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input, and uses the determination target image as input to the determination target image. A model recording unit that stores a prediction model for calculating a forgery index indicating the degree to which the displayed identity verification document may be forged, and a judgment target by inputting the judgment target image into the prediction model. It has a authenticity determination unit for calculating the counterfeit index for an image.

According to the present invention, since the forgery determination of the identity verification document is performed using the prediction model generated by the learning, the forgery determination with high accuracy becomes possible.

The whole view of the identity verification document authenticity judgment system which concerns on Example 1. The schematic diagram of the driver's license which concerns on Example 1. FIG. FIG. 5 is a flow chart of the authenticity determination process according to the first embodiment. The whole view of the identity verification document authenticity judgment system which concerns on Example 2. FIG. 5 is a flow chart of the authenticity determination process according to the third embodiment. The flow chart of the classification process which concerns on Example 3. FIG. 5 is a flow chart of the authenticity determination process according to the fourth embodiment. The figure of the area information in the identity verification document which concerns on Example 4. FIG. The figure for demonstrating the state of the authenticity determination of the forged driver's license which concerns on Example 5. FIG. 5 is a relationship diagram between the angle of the edge and the number of edges according to the fifth embodiment. The figure for demonstrating the state of the authenticity determination of another forged driver's license which concerns on Example 5.

Hereinafter, some examples will be described with reference to the drawings.

FIG. 1 is an overall view of the identity verification document authenticity determination system in this embodiment, and FIG. 2 is a schematic view of a driver's license. The identity verification document authenticity determination system 1 is a system for determining whether or not the identity verification document in the input image is forged.

There are several types of official certificates used as identity verification documents. Depending on the type of identity verification document, the display content or display position such as photo, name, address, creator, etc. will differ. For example, a driver's license (hereinafter referred to as a license) 5 displays a face photo (blank in the center right in FIG. 2), but a health insurance card (not shown) does not display a face photo. In addition, the display position of the name is different between the driver's license 5 and the health insurance card.

Furthermore, even if the same type of identity verification document is created, the display may differ if the source is different. For example, the driver's license 5 may have different display formats or contents depending on the prefecture's Public Safety Commission (Tokyo Metropolitan Public Safety Commission, Kanagawa Prefecture Public Safety Commission, etc.). For example, if the printing machines used by the Tokyo Metropolitan Public Safety Commission and the Kanagawa Prefecture Public Safety Commission are different from each other, there may be slight differences in the display position, font, character size, etc. of the name. Further, if the photography equipment and the photography environment are different, a slight difference may occur in the photographic image including the background portion.

The identity verification document authenticity determination system 1 builds a prediction model for determining the authenticity of an identity verification document by deep learning using a legitimate identity verification document image and a forged identity verification document image as learning data. .. This prediction model takes an image of the identity verification document to be determined as an input, and outputs a forgery index indicating the degree to which the identity verification document can be forged. In constructing a prediction model, it is important to distinguish between a legitimate identity verification document and a forged identity verification document with high accuracy.

For example, as described above, the font of the name field of the driver's license 5 may differ depending on the Public Safety Commission that created it. In the license 5 created by the Tokyo Metropolitan Public Safety Commission, the font A is a legitimate font, but in the license created by the Kanagawa Prefecture Public Safety Commission, the font B may be a legitimate font. As described above, the mixture of the difference in the image due to forgery and the difference in the image due to the creator may hinder the improvement of the accuracy of the authenticity determination.

Regarding this point, in the identity verification document authenticity determination system 1 according to the present embodiment, a configuration is adopted in consideration of the fact that even if the identity verification document is a legitimate identity verification document, if the creation source is different, the image of each area may be different. ing. That is, in the identity verification document authenticity determination system 1 of this embodiment, a separate prediction model is prepared for each source of the identity verification document. As a result, the characteristics of the legitimate identity verification document and the forged identity verification document stand out for each creator, and the influence of the difference in the image due to the difference in the creator on the authenticity determination is reduced.

As shown in FIG. 1, the identity verification document authenticity determination system 1 of this embodiment includes an authenticity determination unit 2, a learning unit 3, and a model recording unit 4.

The authenticity determination unit 2 includes a classification determination unit 6, a model selection unit 7, and an authenticity determination calculation unit 8.

An image to be determined is input to the classification determination unit 6 from a user's terminal or the like via a telecommunication line. The classification determination unit 6 determines the classification to which the identification document displayed on the determination target image belongs from a plurality of classifications in which various identification documents are classified.

Specifically, the classification discrimination unit 6 has a major classification discrimination unit 6A and a minor classification discrimination unit 6a, and a plurality of classifications in which identity verification documents are classified by type are predetermined as major classifications, and the major classifications are defined. A plurality of classifications divided by attributes that may cause differences in the images of identity verification documents belonging to the classifications are predetermined as sub-classifications. Here, as an example, the classification determination unit 6 predetermines the type of the identity verification document as a major classification and the source of the identity verification document as a minor classification.

The major classification determination unit 6A extracts a partial image indicating the type from the determination target image, and determines the major classification to which the identity verification document displayed on the determination target image belongs. For example, the major categories are driver's license 5, passport, health insurance card (not shown), and the like.

The minor classification discriminating unit 6a discriminates the minor classification in the major classification to which the identity verification document displayed on the determination target image belongs according to the major classification discriminated by the major classification discriminating unit 6A. For example, the subclass is the source of the identity verification document. When the major classification is license 5, the minor classification is, for example, the prefectural public safety committee (Tokyo Metropolitan Public Safety Commission, Kanagawa Prefectural Public Safety Commission, etc.). If the major classification is a passport, the minor classification is, for example, the issuing country. If the major classification is a health insurance card, the minor classification is, for example, the insurer's name.

Return to FIG. The model selection unit 7 selects a prediction model for sub-classification to which the identity verification document displayed on the determination target image belongs from the prediction models 4a, 4b, ... Recorded by the model recording unit 4.

The authenticity determination calculation unit 8 calculates a forgery index for the determination target image by inputting the determination target image of the classified identity verification document into the prediction model selected by the model selection unit 7.

In the learning unit 3, a legitimate document drawing image and a counterfeit document drawing image of the identity verification document for each subcategory are input as learning data. The learning unit 3 calculates a forgery index indicating the degree to which the identity verification document displayed on the judgment target image may be forged for each subclass based on the input learning data for each subclass. Prediction models 4a, 4b, ... Are constructed by deep learning. The counterfeit index may be an index having a plurality of values, for example, a true / false binary index. For example, the learning unit 3 prepares the prediction model 4a for each sub-classification based on the learning data input for each sub-classification and the authenticity determination data of the known identity verification document separately input in addition to the learning data. 4b, ... may be constructed by deep learning.

The model recording unit 4 stores a plurality of prediction models 4a, 4b, ..., Which are different for each subclass constructed by the learning unit 3.

FIG. 3 is a flow chart of the authenticity determination process in this embodiment.

In the authenticity determination process, first, the authenticity determination unit 2 corrects the input orientation, resolution, angle, brightness, and darkness of the input determination target image so as to be predetermined values (S301). Next, the major classification determination unit 6A extracted a partial image indicating the type from the corrected determination target image, and extracted which major classification the identity verification document displayed on the corrected determination target image belongs to. Determine based on the partial image. That is, the major classification determination unit 6A determines the major classification of the identity verification document from, for example, a driver's license 5, a passport, a health insurance card, and the like. The sub-classification discrimination unit 6a extracts a partial image indicating the creation source from the judgment target image according to the major classification discriminated by the large classification discrimination unit 6A, and the person displayed on the judgment target image based on the extracted partial image. It is determined which sub-category the confirmation document belongs to (S302). That is, the sub-classification determination unit 6a determines, for example, the source of the identity verification document as a sub-classification. The creator is an example of an attribute that may cause a difference in the image of the identity verification document.

Next, the model selection unit 7 selects a prediction model corresponding to the sub-classification determined by the classification discriminating unit 6 from the prediction models 4a, 4b, ... Recorded by the model recording unit 4 for each sub-classification.

Finally, the authenticity determination calculation unit 8 inputs the image to be determined of the identity verification document classified by the major classification determination unit 6A and the minor classification determination unit 6a into the prediction model selected by the model selection unit 7. A forgery index for the image to be determined is calculated, and the authenticity of the license 5 displayed on the image to be determined is determined (S303). When the forgery index is higher than a preset threshold value (for example, 10%), the authenticity determination calculation unit 8 presumes that the forgery license 5 is used.

According to this embodiment, the identity verification document authenticity determination system 1 has a model recording unit 4 and an authenticity determination unit 2. The model recording unit 4 is constructed by inputting the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input, and the identity verification document displayed on the determination target image is forged by inputting the determination target image. Store a predictive model for calculating a counterfeit index that indicates the extent to which it is possible. The authenticity determination unit 2 calculates a forgery index for the determination target image by inputting the determination target image into the prediction model. As a result, forgery of the identity verification document can be determined using the prediction model generated by the learning, and based on various features of the image to be determined by learning the forged document drawing image forged in various modes. Highly accurate counterfeiting judgment is possible.

The authenticity determination unit 2 determines the classification to which the identity verification document displayed in the determination target image belongs among the plurality of classifications in which the identity verification documents are classified, and uses the classification determined as the determination target image and the prediction model. Then, the forgery index of the image to be judged is calculated. As a result, the identity verification documents can be classified and the forgery determination can be made in consideration of the classification, and the forgery determination of various types of identity verification documents can be performed with high accuracy.

The authenticity determination unit 2 predetermines a plurality of classifications of identity verification documents by type as major classifications, and sets a plurality of classifications classified by attributes that may cause differences in the images of identity verification documents belonging to the major classifications as minor classifications. Determine in advance and determine the major classification and minor classification to which the identity verification document displayed on the judgment target image belongs. Then, the authenticity determination unit 2 calculates a forgery index for the determination target image by using the determination target image, the major classification and the minor classification to which the determination target image belongs, and the prediction model. As a result, the types of identity verification documents are classified into major categories, the attributes that may cause differences in the images to be judged are classified into minor categories, and the classification is gradually limited from the major classification to the minor classification to determine the authenticity. This makes it possible to determine the authenticity of various types of identity verification documents with high accuracy.

Since the prediction models 4a, 4b, ... For each classification are provided, the forgery determination of a plurality of categories can be performed with high accuracy by the prediction models 4a, 4b, ... For each classification.

If the source of the identity verification document is the same, it is usually created with the same equipment, so if you classify the identity verification document by type and source and consider them, it is possible to make a highly accurate forgery judgment. It becomes.

Since it is equipped with a learning unit that builds a prediction model using the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input, it is possible to build a highly reliable prediction model based on the input learning data, and the accuracy is high. Counterfeit judgment is possible.

In this embodiment, the identity verification document is the license 5, and the creator is the Public Safety Commission of the prefecture. Therefore, the license 5 created by the same Public Safety Commission has the same font and format. Therefore, it is possible to accurately determine forgery by making a forgery determination in consideration of which prefectural Public Safety Commission created the license 5.

In the case of driver's license 5, the "registered domicile" column may disappear or the font may be changed after 2010, so the subclass may be the time when the identity verification document is created. Further, the sub-classification may be whether or not it is a color image (color image or black-and-white image). Therefore, in the case of a color image, the amount of information input to the prediction model can be increased to improve the determination accuracy, while in the case of a black-and-white image, the amount of information input to the prediction model can be reduced and calculated. The capacity can be reduced. Further, the subclass may be the presence or absence of the excellent mark or the presence or absence of the AT-only mark.

The identity verification document authenticity determination system 1 according to the second embodiment will be described. Each of the following examples including this example corresponds to a modified example of the first embodiment. In the first embodiment, a prediction model was constructed for each source of the identity verification document, and a forgery index was calculated using the prediction model corresponding to the source to which the image to be determined belongs. As a result, the influence of the difference in the image due to the difference in the creation source on the authenticity judgment is reduced. On the other hand, in the second embodiment, by inputting the information indicating the creation source into the prediction model, the influence of the difference in the image due to the difference in the creation source on the authenticity determination is reduced. Hereinafter, the second embodiment will be described mainly on the differences from the first embodiment.

FIG. 4 is an overall view of the identity verification document authenticity determination system according to the second embodiment.

In the learning unit 23 of this embodiment, the legitimacy of the identity verification document for each of the information of the major classification (for example, license 5, passport, health insurance card, etc.) and the information of the minor classification (for example, the creator of the identity verification document). The document drawing image and the forged document drawing image are input as training data. The learning unit 23 builds a single prediction model 20 by deep learning based on the input learning data for each of the major classification information and the minor classification information.

The model recording unit 4 of this embodiment stores the prediction model 10 constructed by the learning unit 23. The prediction model 10 of the present embodiment is different from the prediction models 4a, 4b, ... Of the first embodiment in that the information of the major classification and the information of the minor classification are input in addition to the image to be determined. Further, in the first embodiment, the model storage unit 4 stores a plurality of prediction models for each creator, but in the second embodiment, a single prediction model that can be commonly used for the identity verification documents of all the creators. Memorize 20.

The classification determination unit 6 of this embodiment is the same as that of the first embodiment shown in FIG.

The authenticity determination process of this embodiment differs from that of Example 1 shown in FIG. 3 only in step S303. In the authenticity determination process, in step S302, as in the first embodiment, the large classification determination unit 6A extracts a partial image indicating the type from the corrected determination target image and displays it on the corrected determination target image. Which major category the identity verification document belongs to is determined based on the extracted partial image. The sub-classification discriminating unit 6a extracts a partial image indicating the creation source from the judgment target image based on the large classification discriminated by the large classification discriminating unit 6A, and displays it on the judgment target image of the identity verification document belonging to the major classification. It is determined based on the extracted partial image whether it was created by the creator. Finally, in step S303, unlike the first embodiment, the authenticity determination calculation unit 18 inputs the major classification and minor classification determined by the classification determination unit 6 and the determination target image into the prediction model 20, and determines the determination target. Calculate the counterfeit index for the image.

According to this embodiment, the prediction model 20 for which the classification is input can be constructed, and the forgery determination by the prediction model 20 can be performed with high accuracy.

If the source of the identity verification document is the same, it is usually created with the same equipment, so if you classify the identity verification document by type and source and consider them, it is possible to make a highly accurate forgery judgment. It becomes.

The identity verification document authenticity determination system 1 according to the third embodiment will be described. The process of correcting the image to be determined in Example 1 did not use the outer edge of the identity verification document. On the other hand, in the third embodiment, the information on the outer edge of the identity verification document is used for the process of correcting the determination target image input to the authenticity determination unit 2 to a predetermined direction and angle. Therefore, in the third embodiment, it is determined whether or not the outer edge of the identity verification document can be detected in the image to be determined, and if the outer edge cannot be detected, the outer edge is estimated.

The overall view of the identity verification document authenticity determination system 1 of the third embodiment is the same as that of the first embodiment shown in FIG. However, the authenticity determination process executed by the authenticity determination unit 2 in the third embodiment uses the process of detecting the outer edge of the identity verification document, the process of estimating the outer edge of the identity verification document, and the outer edge of the identity verification document. It differs from that of Example 1 in that it includes processing.

FIG. 5 is a flow of the authenticity determination process according to the third embodiment.

The authenticity determination unit 2 of this embodiment determines whether or not the outer edge of the driver's license 5 is displayed on the image to be determined.

(Processing to detect the outer edge of the identity verification document)
In the authenticity determination process, the authenticity determination unit 2 detects the outer edge of the identity verification document from the determination target image before the image correction (S301), and whether the outer edge of the identity verification document is displayed on the determination target image. It is determined whether or not (S500). If the outer edge of the identity verification document is displayed (S500: YES), the process proceeds to the image correction process (S301).

(Processing to estimate the outer edge of the identity verification document)
When the outer edge of the identity verification document is not displayed (S500: NO), the outer edge of the identity verification document is estimated (S504). For example, if a portion of the outer edge of an identity verification document is not displayed, a portion of the undisplayed outer edge can be estimated based on most of the displayed outer edge. If most of the outer edge of the identity verification document is not displayed, a reference position is extracted from the displayed identity verification document, and the display is displayed based on the relationship between the extracted reference position and the displayed part of the outer edge. It is possible to estimate the outer edge of the identity verification document that does not exist.

(Processing using the outer edge of the identity verification document)
In the image correction process (301), the estimated outer edge of the identity verification document can be used as a reference line for correcting the direction and angle of the image to be determined input to the authenticity determination unit 2. For example, if the outer edge of the identity verification document displayed on the image to be determined is trapezoidal, the trapezoidal correction is performed. If it is determined that the background displayed outside the outer edge of the image to be determined is the same as the background when the forgery is determined in the past, the forgery index may be increased.

Then, the large classification determination unit 6A determines the classification of the identity verification document displayed on the determination target image based on the image in the outer edge of the identity verification document displayed on the determination target image (S302). The classification determination process of this embodiment will be specifically described below.

FIG. 6 is a flow chart of the classification process in this embodiment.

The major classification discrimination unit 6A of this embodiment discriminates the type of the identity verification document in which a part of the outer edge is not displayed in the judgment target image, and the creator of the identity verification document is displayed from the identity verification document based on the discriminated type. Extract two or more areas, one for the area to be used and the other for the area. The sub-classification determination unit 6a classifies the identity verification document by determining the creation source of the identity verification document based on the area extracted by the large classification determination unit 6A where the creation source of the identity verification document is displayed. Determine.

According to this embodiment, if the outer edge of the identity verification document is not displayed in the determination target image, the outer edge can be estimated, and the orientation and angle of the determination target image can be corrected using the estimated outer edge, and the identity verification document can be used. Authenticity can be determined with high accuracy.

Based on the type of identity verification document, the image to be judged is divided into a plurality of areas including at least the presence / absence of the display of the creation source, and the creation source is selected from the area in which the creation source of the identity verification document is displayed among the divided multiple areas. Since the creation source can be determined by the determination procedure, the creation source can be determined with high accuracy.

The identity verification document authenticity determination system 1 according to the fourth embodiment will be described. In Example 1, a prediction model for each creator was provided, and a single counterfeit index was calculated. On the other hand, in the fourth embodiment, a prediction model corresponding to a plurality of regions of the image to be determined is provided, individual counterfeit indexes for the plurality of regions are calculated, and the calculated individual counterfeit indexes for each region are integrated. Calculate the forgery index of the image to be judged.

In the learning unit 43 of this embodiment, the information on the type of the identity verification document (for example, license 5, passport, health insurance card, etc.) and the legitimate document drawing image of the identity verification document for each information of the source of the identity verification document are provided. And the forged document drawing image is input as training data. Based on the input major classification information and learning data for each minor classification information, the learning unit 43 has a plurality of areas included in the identity verification documents for each type and each creator of the identity verification documents. Build a prediction model corresponding to each.

The model storage unit 44 of this embodiment stores a plurality of prediction models 44a, 44b, ... Constructed by the learning unit 43.

FIG. 7 is a flow of the authenticity determination process according to the fourth embodiment.

The authenticity determination unit 2 of the present embodiment holds in advance the area information in the identity verification document indicating what kind of area exists in the identity verification document for each type of the identity verification document.

FIG. 8 is a diagram showing an example of area information in the identity verification document. Referring to FIG. 8, the area information in the identity verification document stores the type of the identity verification document, the area names of a plurality of areas included in the type of identity verification document, and the coordinate information of those areas. There is. For example, as shown in FIG. 2, the license 5 has a name area, a date of birth area, an address area, a delivery information area, an expiration date area, a condition area, a good driver display area, a driver's license number area, and a photo area. , License acquisition date area, and creation source area. Therefore, in FIG. 8, the area names and coordinate information of the areas of the driver's license 5 are stored.

The name area is the area where the name of the license holder is displayed. The date of birth area is the area where the date of birth of the license holder is displayed. The address area is the area where the license holder's address is displayed. The issuance information area is an area in which the date of issuance of the driver's license is displayed. The expiration date area is an area where the expiration date of the driver's license is displayed. The condition area is the area where the license conditions are displayed. The excellent driver display area is an area in which an excellent mark is displayed for an excellent driver. The driver's license number area is the area where the driver's license number is displayed. The photographic area is an area in which a photograph of the license holder's face is displayed. The license acquisition date area is an area in which the acquisition date of the two-wheeled license, the acquisition date of the four-wheeled license, and the acquisition date of the second-class license are displayed. The creation source area is the area where the name and mark of the Public Safety Commission of the prefecture where the license was created are displayed.

The authenticity determination unit 2 determines the type of the identity verification document displayed on the image to be determined, refers to the area information in the identity verification document based on the determined type, and has a plurality of areas included in the identity verification document of that type. To know. Subsequently, the authenticity determination unit 2 extracts a plurality of known regions from the image to be determined, calculates an individual counterfeit index using a prediction model selected for each of the plurality of regions, and integrates the individual counterfeit indexes for each region. To calculate the forgery index of the image to be judged.

The plurality of areas extracted from the determination target image include the text description and the drawing display of the identity verification document displayed on the determination target image. The text is, for example, the name. The drawing is, for example, a photograph.

Further, in the present embodiment, an example in which the forgery index of all the regions is calculated by the prediction model corresponding to each region is shown, but the present invention is not limited thereto. As another example, the counterfeit index may be calculated by the prediction model for some regions, and the counterfeit index may be calculated by another determination method for some other regions. Various existing methods can be applied as other determination methods. For example, the image of the area may be compared with the image of the area in the real identity verification document and the counterfeit index may be calculated based on the similarity of the images.

In the authenticity determination process, first, the authenticity determination unit 2 determines the type of the identity verification document displayed on the determination target image, extracts a plurality of areas from the determination target image based on the determined type, and extracts a plurality of extracted areas. One area is selected from the areas of (S701). Next, the authenticity determination unit 2 selects a determination method to be used in the region (S702). In this embodiment, since the determination method is based on the prediction model, the authenticity determination unit 2 selects the prediction model.

The authenticity determination unit 2 calculates an individual counterfeit index by the determination method (S703). Then, the authenticity determination unit 2 determines whether or not there is a remaining area (S704). If there is a remaining area (S704: YES), the process returns to S701 until the remaining area is exhausted.

When the remaining area is exhausted (S704: NO), the authenticity determination unit 2 calculates the forgery index for the image to be determined by integrating the individual forgery indexes for each area (705). For example, the counterfeit index for the image to be determined is calculated by weighting and adding the individual counterfeit indexes of each region.

According to this embodiment, since the counterfeit index of the judgment target image can be calculated by integrating the individual counterfeit indexes for each region, the counterfeit index for the judgment target image can be calculated with high accuracy.

The identity verification document authenticity determination system 1 according to the fifth embodiment will be described. In the first embodiment, when the authenticity determination process is started, the authenticity determination unit 2 corrects the input orientation, resolution, angle, brightness, and the like of the input determination target image to be predetermined values. On the other hand, in the fifth embodiment, the authenticity determination unit 2 executes a preliminary counterfeit determination process before starting the correction process shown in the first embodiment. The preliminary forgery determination process is a process for determining that the image is a forged identity verification document when there is a clear cut-and-paste mark on the image to be determined. For the image to be determined that is not determined to be the identity verification document that has been forged by the preliminary forgery determination process, the process shown in Example 1 is further executed.

Hereinafter, the preliminary forgery determination process will be described.

FIG. 9 is a diagram for explaining a state of authenticity determination of the forged driver's license according to the fifth embodiment.

The authenticity determination unit 2 of this embodiment detects the cut-and-paste mark based on the inclination of the boundary line of the cut-and-paste mark, and calculates a forgery index for the image to be determined based on the detected cut-and-paste mark.

In the preliminary forgery determination process, first, when the upper and lower edge lines 11 and 12 are not displayed, the outer edge detection process, the estimation process, and the use process of the identity verification document are performed as in the third embodiment to confirm the identity. Estimate the outer edge of the document. Next, the authenticity determination unit 2 detects a straight line or a curve from the image to be determined by edge detection processing. The image of the identity verification document has a line that should be originally included in the image depending on the type of the identity verification document. For example, if the identity verification document is a driver's license 5, there are upper and lower edges 11 and 12 of the driver's license 5 itself. In addition, there are upper and lower borders 13 and 14 in the name display column. The face photograph also has upper and lower borders 15 and 16. In addition, the image to be judged may have a line other than the line originally included in the identity verification document.

The authenticity determination unit 2 puts the upper and lower edge lines 11 and 12 of the license 5, the upper and lower frame lines 13 and 14 of the name display column, and the upper and lower frame lines 15 and 16 of the face photograph on the image to be judged displaying the license 5. When the inclined line segments 17 and 18 that are inclined with respect to the above are detected, the inclined line segments 17 and 18 are extracted. In this embodiment, two upper and lower inclined line segments 17 and 18 that incline downward to the right across the upper and lower frame lines 13 and 14 in the name display column can be confirmed.

FIG. 10 is a diagram showing the relationship between the angle and the number of edges according to this embodiment.

Here, in a normal driver's license 5, the upper and lower edge lines 11 and 12 and the upper and lower frame lines 13 to 16 extend in parallel to each other on the left and right. Therefore, it can be judged that the inclined line segments 17 and 18 that are not orthogonal to the left-right direction and greatly deviate from other angular distributions are highly likely to be clear cut-and-paste marks. For the distortion of the inclined line segments 17 and 18, the inclined line segments 17 and 18 are extracted, the number of the extracted inclined line segments 17 and 18 for each angle is counted, and the forgery index for the image to be determined is calculated. Thereby, it is possible to determine the forgery of the driver's license 5 including the cut-and-paste mark. The inclined line segments 17 and 18 may detect angles with respect to the vertical direction parallel to the left and right edge lines and the left and right frame lines.

As a method of detecting an edge from an image, a method using a Sobel filter or a Hough transform is generally known. Specifically, the input image is converted into a set of edge points by a Sobel filter, and the edge is detected by performing a Hough transform on the converted set of edge points.

Further, as shown in FIG. 11, the authenticity determination unit 2 may detect the cut / paste mark based on the distortion of the edge of the cut / paste mark. For example, the authenticity determination unit 2 may detect two upper and lower curves 27 and 28 that are distorted vertically between the upper and lower edges 23 and 24 of the name display field and the name.

Not limited to the angle and distortion of the line segment, the forgery index of the image to be determined may be calculated based on the length and thickness of the line segment, the position of the maximum or minimum, or the difference in color from the background. Therefore, this method can be applied even if a new license 5 including a inclined line segment or a curve is issued.

The embodiments of the present invention described above are examples for the purpose of explaining the present invention, and the scope of the present invention is not limited to those embodiments. One of ordinary skill in the art can practice the present invention in various other aspects without departing from the gist of the present invention.

1 ... Identity verification document authenticity judgment system, 2 ... Authenticity judgment unit, 3 ... Learning unit, 4 ... Model recording unit, 4a, 4b ... Prediction model, 5 ... License, 20 ... Prediction model, 23 ... Learning unit

Claims (15)

Indicates the degree to which the identity verification document displayed on the determination target image, which is constructed by inputting the legitimate document drawing image and the forged document drawing image of the identity verification document as the learning data input, can be forged. A model recording unit that stores a prediction model for calculating counterfeit indicators,
An identity verification document authenticity determination system including a authenticity determination unit that calculates the counterfeit index for the determination target image by inputting the determination target image into the prediction model. The authenticity determination unit
Among the plurality of classifications in which the identity verification documents are classified, the classification to which the identity verification document displayed in the determination target image belongs is determined.
Using the determination target image, the classification, and the prediction model, the counterfeit index for the determination target image is calculated.
The identity verification document authenticity determination system according to claim 1. The authenticity determination unit
A plurality of classifications in which the above-mentioned identity verification documents are divided by type are defined in advance as major classifications.
A plurality of classifications divided by attributes that may cause differences in the images of identity verification documents belonging to the above major classifications are predetermined as minor classifications.
The major classification and the minor classification to which the identity verification document displayed on the judgment target image belongs are discriminated.
Using the determination target image, the major classification and the minor classification to which the determination target image belongs, and the prediction model, the counterfeit index for the determination target image is calculated.
The identity verification document authenticity determination system according to claim 2. The model's rating record unit stores predictive models provided for each of the classification,
The authenticity determination unit
It is determined which classification the identity verification document displayed on the determination target image is an image of the identity verification document.
Using the determined prediction model of the classification, the counterfeit index for the image to be determined is calculated.
The identity verification document authenticity determination system according to claim 2. The types of the identity verification documents are classified into major categories, and the sources of the identity verification documents are classified into minor categories.
The model's rating record unit stores the prediction models of the small classification basis,
The authenticity determination unit
A partial image indicating the type is extracted from the determination target image, and which major classification the identity verification document represented in the determination target image belongs to is determined based on the partial image.
A partial image indicating the creator is extracted from the determination target image according to the determined major classification, and which minor classification the identity verification document displayed on the determination target image belongs to is based on the partial image. Determine and
Using the prediction model of the sub-classification to which the identity verification document displayed on the determination target image belongs, the counterfeit index for the determination target image is calculated.
The identity verification document authenticity determination system according to claim 4. The model's rating record unit stores a prediction model to input the classification,
The authenticity determination unit
It is determined which classification the identity verification document displayed on the determination target image belongs to.
The discriminated classification is input to the prediction model, and the counterfeit index for the image to be determined is calculated.
The identity verification document authenticity determination system according to claim 2. The types of the identity verification documents are classified into major categories, and the sources of the identity verification documents are classified into minor categories.
The model's rating record unit stores the prediction model to enter the large classification and the minor class,
The authenticity determination unit
A partial image indicating the type is extracted from the determination target image, and which major classification the identity verification document represented in the determination target image belongs to is determined based on the partial image.
A partial image indicating the creator is extracted from the determination target image according to the determined major classification, and which minor classification the identity verification document displayed on the determination target image belongs to is based on the partial image. Determine and
The major classification and the minor classification to which the identity verification document displayed on the judgment target image belongs are input to the prediction model, and the counterfeit index for the judgment target image is calculated.
The identity verification document authenticity determination system according to claim 6. The authenticity determination unit
It is determined whether or not the outer edge of the identity verification document is displayed on the image to be determined.
If there is no outer edge, the outer edge is estimated and
Based on the image in the outer edge displayed on the determination target image, the classification of the identity verification document displayed on the determination target image is determined.
The identity verification document authenticity determination system according to claim 2. The authenticity determination unit is in the process of determining the classification.
Determine the type of identity verification document displayed on the image to be determined,
A plurality of regions are extracted from the determination target image based on the determined type, and a plurality of regions are extracted.
Based on the area in which the creation source of the identity verification document is displayed among the extracted plurality of areas, the attribute that may cause a difference in the image of the identity verification document is determined.
The identity verification document authenticity determination system according to claim 8. The authenticity determination unit
Determine the type of identity verification document displayed on the image to be determined,
A plurality of regions are extracted from the determination target image based on the determined type, and a plurality of regions are extracted.
An individual counterfeit index is calculated by the determination method selected for each of the plurality of areas, and the individual counterfeit index is calculated.
The counterfeit index of the image to be determined is calculated by integrating the individual counterfeit indexes for each region.
The identity verification document authenticity determination system according to claim 2. A learning unit for constructing the prediction model by using the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input is further provided.
The identity verification document authenticity determination system according to claim 1. The identity verification document is a driver's license,
The creator is the Prefectural Public Safety Commission,
The identity verification document authenticity determination system according to claim 5 or 7. The authenticity determination unit
The cut-and-paste mark of the identity verification document displayed on the judgment target image is detected, and the forgery index for the judgment target image is calculated based on the detected cut-and-paste mark.
The identity verification document authenticity determination system according to claim 1. The model recording means provided in the computer is constructed by inputting the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input, and the identity verification document displayed on the determination target image is forged by inputting the determination target image. Record a predictive model to calculate a counterfeit index that indicates the extent to which it can be
An identity verification document authenticity determination method in which a authenticity determination means provided in a computer calculates the forgery index for the determination target image by inputting the determination target image into the prediction model. The model recording means provided in the computer is constructed by inputting the legitimate document drawing image and the forged document drawing image of the identity verification document as learning data input, and the identity verification document displayed on the determination target image is forged by inputting the judgment target image. Have them record a predictive model to calculate a counterfeit index that indicates the extent to which they can be
An identity verification document authenticity determination program that causes a computer-equipped authenticity determination means to calculate the forgery index for the determination target image by inputting the determination target image into the prediction model.

Images