TWI777188B - Contract signature authentication method and device - Google Patents

Abstract

A contract signature verification device is used to verify a signature corresponding to a contract image of a target signatory, a storage unit stores a plurality of signature images corresponding to a plurality of signatories and the contract image, and a processing unit uses the contract image according to the contract image. A signature capture model for capturing the signature area of the image, obtaining a signature capture image, and according to the signature capture image and a target signature image corresponding to the target signatory in the signature images, using a Extracting the features of two images and calculating the similarity model of similarity, obtaining a similarity value, and extracting the image according to the signature, the target signature image, and the similarity value, and using a signature for classifying the authenticity of the signature The classification model obtains a classification result including a probability indicating that the signature is true.

Description

Contract signature authentication method and device

The present invention relates to an authentication method, in particular to a contract signature authentication method and device thereof.

In the insurance industry, insurance contracts such as request for guarantee, signature receipt, application for contract modification, and claims application are all signed on actual paper. In particular, insurance contracts are different from general contracts. Due to the transfer of the subject matter, the addition of surrendered insurance or the change of beneficiaries, etc., if the policy encounters a person with intention to modify it improperly, it can be changed as long as the signature is forged. Therefore, it is necessary to identify whether the signature on the contract is It is very important for the policyholder to sign in person, sign on his behalf or sign the wrong name (different name).

The authenticity of the signature on the existing contract is authenticated manually. The basic operation process is roughly that the internal affairs personnel scan the signature on the contract, and then hand it over to the underwriting personnel for signature identification. However, the existing signature authentication process is cumbersome, and the signature of the same person may be slightly different due to different writing environments or tools, etc., making it time-consuming and labor-intensive to manually authenticate signatures.

Therefore, the purpose of the present invention is to provide a contract signature authentication method that can automatically authenticate the authenticity of the contract signature.

Therefore, the contract signature verification method of the present invention, for verifying a signature corresponding to a contract image of a target signatory, is implemented by a signature verification device, and the signature verification device stores a plurality of signature images corresponding to a plurality of signatories and the contract. Image, the method includes a step (A), a step (B), and a step (C).

In the step (A), the signature authentication device obtains a signature capture image according to the contract image using a signature capture model of the signature area for capturing the image

In the step (B), the signature identification device extracts the image according to the signature and a target signature image corresponding to the target signatory in the signature images, and uses a method for extracting the features of the two images and calculating the similarity. Similarity model to obtain a similarity value.

In the step (C), the signature identification device obtains a classification result according to the signature capture image, the target signature image, and the similarity value, and uses a signature classification model for classifying the authenticity of the signature. The result includes a probability indicating that the signature is true.

The purpose of the present invention is to provide a contract signature identification device that can automatically identify the authenticity of a contract signature.

Therefore, the contract signature verification device of the present invention is used for verifying a signature corresponding to a contract image of a target signatory, and the device includes a storage unit and a processing unit.

The storage unit stores a plurality of signature images corresponding to a plurality of signatories and the contract image respectively.

The processing unit is electrically connected to the storage unit, and according to the contract image, the processing unit uses a signature capture model for capturing a signature area of the image to obtain a signature capture image, and captures the image and a signature capture image according to the signature Among the signature images corresponding to the target signature image of the target signatory, a similarity value is obtained by using a similarity model for extracting the features of the two images and calculating the similarity, and the image and the target signature are extracted according to the signature. The image, and the similarity value, utilize a signature classification model for classifying authenticity of signatures to obtain a classification result including a probability indicating that the signature is authentic.

The effect of the present invention is: the processing unit uses the signature capture model to capture the signature area of the contract image to obtain the signature capture image, and uses the similarity model to extract the signature capture image and the target The characteristics of the signature image are used to obtain the similarity, and finally the classification result is obtained by using the signature classification model to identify the authenticity of the signature captured image, which greatly saves the time and time of capturing the signature on the contract and identifying the authenticity of the signature. manpower.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are designated by the same reference numerals.

Referring to FIG. 1 , an embodiment of the contract signature verification apparatus 1 of the present invention is used to verify a signature of a contract image corresponding to a target signatory, and includes a storage unit 11 and a processing unit 12 electrically connected to the storage unit 11 .

The storage unit 11 stores a plurality of contract training data, a plurality of signature training data, a plurality of classification training data, a plurality of signature images corresponding to a plurality of signatories, and the contract image. Each contract training data includes a contract training image and at least one signature area mark. Each signature training data includes a signature training image and a signatory annotation. Each classification training data includes a first signature training image, a second signature training image, a training similarity value associated with the similarity between the first signature training image and the second signature training image, and an indicator indicating the first signature training image. Whether the signature training image and the second signature training image are signed by the same person is the authenticity label.

，該寬度例如為

，則簽名區域的四個點分別為(x

,y

)、(x

,y

)、(x

,y

)，及(x

,y

)，但不以此為限。It is worth noting that, in this embodiment, the signature area is, for example, a rectangle, and each signature area label includes a center coordinate, a width, and a height of the signature area. For example, the center coordinate of the signature area is, for example, in A rectangular coordinate system, the value of which is for example ( x , y ), the height is for example

, the width is for example

, then the four points of the signature area are ( x

, y

), ( x

, y

), ( x

, y

), and ( x

, y

), but not limited thereto.

An embodiment of the contract signature authentication method of the present invention includes a modeling program and a signature authentication program, and is implemented by the embodiment of the contract signature authentication device 11 of the present invention shown in FIG. 1 .

Referring to FIGS. 1 and 2 , the following describes the steps of the modeling procedure of the embodiment of the contract signature authentication method of the present invention.

In step 21, the processing unit 12 uses a first deep learning algorithm to establish a signature capture model for capturing the signature region of the image according to the contract training data. It is worth noting that, in this embodiment, the first deep learning algorithm is such as but not limited to RetinaNet, and its detailed method is described in "Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár: Focal Loss for In Dense Object Detection. In ICCV, 2017”, their details are omitted here for brevity, but not limited to this.

In step 22, the processing unit 12 uses a second deep learning algorithm to establish a similarity model for extracting the features of the two images and calculating the similarity according to the signature training data. It is worth noting that, in this embodiment, the second deep learning algorithm such as but not limited to FaceNet, FaceNet is a solution proposed by Google in 2015 for recognizing face counts, which is not the focus of the present invention, and is not described here. to repeat.

In step 23, the processing unit 12 uses a machine learning algorithm to establish a signature classification model for classifying authenticity of signatures according to the classification training data. It should be noted that, in this embodiment, the machine learning algorithm is, for example, but not limited to, an extreme gradient boosting (eXtreme Gradient Boosting, XGBoost) algorithm.

Referring to FIG. 3 , step 23 further includes sub-steps 231 to 235 , and the sub-steps of step 23 are described below.

In step 231, for each classification training data, the processing unit 12 obtains a first training black pixel quantity associated with the first signature training image according to the first signature training image of the classification training data.

In step 232, for each classification training data, the processing unit 12 obtains a second training black pixel quantity associated with the second signature training image according to the second signature training image of the classification training data.

In step 233, for each classification training data, the processing unit 12 obtains a related first signature training image and the second signature training image according to the first signature training image and the second signature training image of the classification training data The training black pixel difference value for the image. It should be noted that, in this embodiment, the training black pixel difference value is calculated by converting the first signature training image and the second signature training image to grayscale through OpenCV (Open Source Computer Vision Library). Obtained by black pixels.

In step 234, for each classification training data, the processing unit 12 obtains a ratio value of the number of training black pixels according to the first number of training black pixels and the second number of training black pixels obtained in steps 231 and 232. For example, if the first training black pixel quantity is 2989 pixels, and the second training black pixel quantity is 2188 pixels, the ratio of the training black pixel quantity is 2989/2188=1.366088.

In step 235, the processing unit 12 obtains the first training black pixel number, the second training black pixel number, the training black pixel difference value, the training black pixel number ratio value, and the classification training data obtained in steps 231-234. The training similarity value and authenticity label of , and the machine learning algorithm is used to establish the signature classification model.

Referring to FIGS. 1 and 4 , the following describes the steps flow of the signature authentication procedure of the embodiment of the contract signature authentication method of the present invention.

In step 31, the processing unit 12 obtains a signature capture image by using the signature capture model according to the contract image.

In step 32, the processing unit 12 obtains a similarity value by using the similarity model according to the signature capture image and a target signature image corresponding to the target signatory in the signature images. It should be noted that, in this embodiment, the processing unit 12 obtains the target signature image by using the file name of the contract image. For example, the file name of the contract image is ACND341170_A72.jpg, then the processing unit 12 executes the The signature area of other files in case ACND341170 is the target signature image, and the filename of the target signature image is, for example, ACND341170_A01, but not limited thereto.

In step 33, the processing unit 12 obtains a classification result according to the signature capture image, the target signature image, and the similarity value, using the signature classification model, and the classification result includes a probability indicating that the signature is true .

Referring to FIG. 5 , step 33 further includes sub-steps 331 to 335 , and the sub-steps of step 33 are described below.

In step 331, the processing unit 12 captures the image according to the signature, and obtains a number of black pixels in the captured image associated with the signature captured image.

In step 332, the processing unit 12 obtains a target image black pixel quantity related to the target signature image according to the target signature image.

In step 333, the processing unit 12 obtains a black pixel difference value associated with the signature captured image and the target signature image according to the signature captured image and the target signature image.

In step 334, the processing unit 12 obtains a ratio of the number of black pixels according to the number of black pixels in the captured image and the number of black pixels in the target image.

In step 335, the processing unit 12 uses the signature classification model according to the number of black pixels in the captured image, the number of black pixels in the target image, the difference value of black pixels, the ratio value of the number of black pixels, and the similarity value, Obtain the classification result.

It should be noted that, in this embodiment, if the probability of the classification result is greater than a threshold of the signature classification model, such as 0.8, it means that the signature is true, but not limited to this.

To sum up, in the contract signature identification method and device of the present invention, the processing unit 12 uses the signature capture model to capture the signature area of the contract image to obtain the signature capture image, and uses the similarity The model extracts the features of the signature captured image and the target signature image to obtain the similarity, and finally uses the signature classification model to obtain the classification result to identify the authenticity of the signature captured image, which greatly saves the cost of the capture contract. The signature and the time and manpower to verify the authenticity of the signature, in addition, the signature extraction model can support different forms of contract images, and the signature classification model obtains the classification result according to the similarity obtained by the similarity model, which greatly improves The accuracy of the classification result can indeed achieve the purpose of the present invention.

However, the above are only examples of the present invention, and should not limit the scope of implementation of the present invention. Any simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the patent specification are still included in the scope of the present invention. within the scope of the invention patent.

1: Contract signature authentication device 11: Storage unit 12: Processing unit 21~23: Steps 231~235: Steps 31~33: Steps 331~335: Steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: 1 is a block diagram illustrating an embodiment of a contract signature authentication device of the present invention; 2 is a flowchart illustrating a modeling procedure of an embodiment of the contract signature authentication method of the present invention; Fig. 3 is a flow chart to assist in explaining the sub-steps of step 23 of Fig. 2; 4 is a flow chart illustrating a signature authentication procedure of this embodiment of the contract signature authentication method of the present invention; and FIG. 5 is a flowchart to assist in explaining the sub-steps of step 33 of FIG. 4 .

1: Contract signature authentication device

11: Storage unit

12: Processing unit

Claims (8)

A contract signature verification method for verifying a signature corresponding to a contract image of a target signatory, implemented by a signature verification device, the signature verification device storing a plurality of signature images corresponding to a plurality of signatories and the contract image, and A plurality of classification training data, each classification training data includes a first signature training image, a second signature training image, and a training similarity value related to the similarity between the first signature training image and the second signature training image, and a authenticity label indicating whether the first signature training image and the second signature training image are signed by the same person, the method includes the following steps: (A) according to the contract image, using a signature for capturing images A signature capture model of the region to obtain a signature capture image; (B) according to the signature capture image and a target signature image corresponding to the target signatory in the signature images, use one for extracting two image features and calculate the similarity model of the similarity to obtain a similarity value; (C) according to the classification training data, use a machine learning algorithm to establish a signature classification model for classifying the authenticity of the signature, wherein, step (C) ) includes the following sub-steps, (C-1) for each classification training data, according to the first signature training image of the classification training data, obtain a first training black pixel quantity related to the first signature training image; (C -2) For each classification training data, obtain a second training black pixel number associated with the second signature training image according to the second signature training image of the classification training data; (C-3) For each classification training data, training according to the classification The first signature training image and the second signature training image of the data, obtain a training black pixel difference value associated with the first signature training image and the second signature training image; (C-4) For each classification training data, According to the first training black pixel number and the second training black pixel number obtained in steps (C-1) and (C-2), a ratio value of the number of training black pixels is obtained; and (C-5) according to step (C- 1) ~ (C-4) obtained first training black pixel number, second training black pixel number, training black pixel difference value, training black pixel number ratio value, and the training similarity value of these classification training data and Authenticity labeling, using the machine learning algorithm to establish the signature classification model; and (D) extracting an image according to the signature, the target signature image, and the similarity value, and using the signature classification model to obtain a classification result, The classification result includes a probability indicating that the signature is true. According to the contract signature identification method according to claim 1, the signature identification device further stores a plurality of contract training data, each contract training data includes a contract training image and at least one signature area mark, and before step (A), further comprising: The following steps: (E) According to the contract training data, use a first deep learning algorithm to establish the signature capture model. According to the contract signature authentication method described in claim 1, the signature authentication device further stores a plurality of signature training data, each signature training data includes a signature training image and a signer's mark, and before step (B), it also includes the following step: (F) Using a second deep learning algorithm to establish the similarity model according to the signature training data. The contract signature authentication method according to claim 1, wherein step (D) includes the following sub-steps: (D-1) Capture an image according to the signature, and obtain a captured image black pixel related to the signature captured image (D-2) According to the target signature image, obtain a target image black pixel number related to the target signature image; (D-3) According to the signature capture image and the target signature image, obtain a target image related to the signature image. The difference value of black pixels between the signature captured image and the target signature image; (D-4) according to the captured image black pixel quantity and the target image black pixel quantity, obtain a black pixel quantity ratio value; and (D-5) According to the number of black pixels in the captured image, the number of black pixels in the target image, the difference value of the black pixels, the ratio value of the number of black pixels, and the similarity value, the signature classification model is used to obtain the classification result. A contract signature identification device for verifying a signature corresponding to a contract image of a target signatory, the device comprising: a storage unit storing a plurality of signature images corresponding to a plurality of signatories and the contract image, and a plurality of classifications training data, each classified training data includes a first signature training image, a second signature training image, a training similarity value related to the similarity between the first signature training image and the second signature training image, and an indication output the first signature training image and the Whether the second signature training image is signed by the same person is the authenticity mark; and a processing unit, electrically connected to the storage unit, obtains a signature according to the contract image using a signature capture model for capturing the signature area of the image Capture images, and capture images according to the signature and a target signature image corresponding to the target signatory in the signature images, and obtain a similarity using a similarity model for extracting features of the two images and calculating the similarity degree value, and according to the classification training data, use a machine learning algorithm to establish a signature classification model for classifying the authenticity of the signature, and extract the image, the target signature image, and the similarity value according to the signature, Using the signature classification model, a classification result is obtained, the classification result includes a probability indicating that the signature is true, wherein, for each classification training data, the processing unit obtains a first signature training image according to the classification training data The number of first training black pixels related to the first signature training image, and for each classification training data, the processing unit obtains a The training black pixel difference value between the first signature training image and the second signature training image, and for each classification training data, the processing unit obtains a The training black pixel number ratio value, and then based on the obtained first training black pixel number, second training black pixel number, training black pixel difference value, training black pixel number ratio value, and the training similarity value of the classified training data The signature classification model is established by using the machine learning algorithm with the authenticity label. The contract signature authentication device according to claim 5, wherein the storage unit Also stores a plurality of contract training data, each contract training data includes a contract training image and at least one signature area mark, the processing unit uses a first deep learning algorithm to establish the signature retrieval model according to the contract training data . The contract signature identification device according to claim 5, wherein the storage unit further stores a plurality of signature training data, each signature training data includes a signature training image and a signer's label, and the processing unit is based on the signature training data , using a second deep learning algorithm to establish the similarity model. The contract signature authentication device according to claim 5, wherein the processing unit captures an image according to the signature, obtains a number of black pixels in the captured image related to the signature captured image, and obtains a black pixel count according to the target signature image. The number of black pixels in the target image related to the target signature image, and according to the signature capture image and the target signature image, obtain a black pixel difference associated with the signature capture image and the target signature image, and then according to the capture image The number of black pixels in the image and the number of black pixels in the target image are obtained, and a ratio value of the number of black pixels is obtained. Finally, according to the number of black pixels in the captured image, the number of black pixels in the target image, the difference value of black pixels, and the ratio value of the number of black pixels, and the similarity value, and use the signature classification model to obtain the classification result.

Images