Signature verification methods, devices, equipment and storage media

Abstract

This invention relates to the field of artificial intelligence technology, specifically to a signature verification method. The method filters acquired document images to be processed using a preset classification model to obtain valid document images; performs template matching on each valid document image to obtain handwritten and printed areas; determines whether a signature exists in the handwritten area; if no signature exists, identifies the printed area to obtain text content; pushes the document to be processed corresponding to the valid document image based on the text content, and obtains a signed document image after the first target person has signed it; verifies the signed document image using a preset feature tree; if two signatures are identical, the verification result is determined to be a correct signature. This invention is applied to signature verification in financial or insurance business. This invention achieves the determination of whether a signature exists in the signature area, pushes the document to unsigned personnel, and verifies the correctness of the signature in the signature area.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a signature verification method, apparatus, device, and storage medium. Background Technology

[0002] With the advent of the Internet era, handwritten signatures have become an important means of verifying a user's true intention to sign in industries such as insurance and finance. The rapid development of Internet applications such as Internet finance, B2B e-commerce, tourism, and education has driven the demand for online electronic signatures. In order to ensure that Internet business activities are legally compliant and traceable, more and more financial institutions are actively seeking legal and effective online handwritten signature verification solutions.

[0003] Currently, in existing technologies, if a document in industries such as insurance or finance lacks a handwritten signature, the verification fails, requiring a re-upload of the signed document for verification. However, re-uploading the signed document necessitates notifying the signatories, and reviewing a single document requires signatures from multiple individuals, resulting in a lengthy verification process that impacts subsequent reviews. Summary of the Invention

[0004] This invention provides a signature verification method, apparatus, device, and storage medium to solve the problem of long signature verification cycles caused by the absence of a signature in the handwritten area in the prior art.

[0005] A signature verification method, comprising:

[0006] At least one document image to be processed is obtained, and each document image to be processed is filtered through a preset classification model to obtain valid document images;

[0007] Template matching is performed on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas;

[0008] Determine whether a signature exists in the handwritten area. If no signature exists in the handwritten area, identify the printed area to obtain the text content.

[0009] Based on the text content, the document to be processed corresponding to the valid document image is pushed, and the signed document image after the first target person signs is obtained;

[0010] The signature document image is verified by a preset feature tree. When the signature in the preset feature tree is the same as the signature in the signature document image, the verification result is determined to be a correct signature.

[0011] A signature verification device, comprising:

[0012] The document filtering module is used to acquire at least one document image to be processed, and to filter each document image to be processed using a preset classification model to obtain valid document images.

[0013] The template matching module is used to perform template matching on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas.

[0014] The content recognition module is used to determine whether there is a signature in the handwritten area. When there is no signature in the handwritten area, the printed area is recognized to obtain the text content.

[0015] The document push module is used to push the document to be processed corresponding to the valid document image according to the text content, and to obtain the signed document image after the first target person signs it;

[0016] The signature verification module is used to verify the signature document image using a preset feature tree. When the signature in the preset feature tree is the same as the signature in the signature document image, the verification result is determined to be a correct signature.

[0017] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described signature verification method.

[0018] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described signature verification method.

[0019] This invention provides a signature verification method, apparatus, device, and storage medium. The method acquires at least one document image to be processed and filters each document image using a preset classification model to obtain valid document images. Template matching is performed on each valid document image to obtain a handwritten area and a printed area corresponding to each valid document image. It determines whether a signature exists in the handwritten area; if no signature exists, the printed area is identified to obtain text content. The document to be processed corresponding to the valid document image is pushed based on the text content, and a signed document image after being signed by a first target person is acquired. The signed document image is verified using a preset feature tree; if the signature in the preset feature tree matches the signature in the signed document image, the verification result is determined to be a correct signature.

[0020] In this invention, after template matching to extract handwritten and printed areas from each valid document image, the presence of a signature in the handwritten area is first detected, thus determining whether a signature exists in the handwritten area of ​​a document in financial or insurance transactions. Then, if no signature is found, the printed area is identified to recognize the text content within it, and the document image to be processed is pushed to the target personnel, enabling the delivery of unsigned documents to be processed in financial or insurance transactions. Simultaneously, this invention uses a preset feature tree to verify the signature on the signed document image, thus detecting whether the signature in the handwritten area of ​​a document in financial or insurance transactions is correct. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the application environment of the signature verification method in one embodiment of the present invention;

[0023] Figure 2 This is a flowchart of a signature verification method in one embodiment of the present invention;

[0024] Figure 3 This is a flowchart of step S20 of the signature verification method in one embodiment of the present invention;

[0025] Figure 4 This is a flowchart of step S40 of the signature verification method in one embodiment of the present invention;

[0026] Figure 5 This is a schematic block diagram of a signature verification device according to an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The signature verification method provided in this embodiment of the invention can be applied to, for example... Figure 1The application environment shown. Specifically, this signature verification method is applied in a signature verification device, which includes, as shown in the example. Figure 1 The client and server shown communicate via a network to address the problem of long signature verification cycles caused by the absence of a signature in the handwritten area in existing technologies. The server can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The client, also known as the user terminal, refers to the program that provides categorized services to customers, corresponding to the server. The client can be installed on, but is not limited to, various computers, laptops, smartphones, tablets, and portable wearable devices.

[0030] In one embodiment, such as Figure 2 As shown, a signature verification method is provided, which is applied to... Figure 1 Taking the server in the example, the following steps are included:

[0031] S10: Obtain at least one document image to be processed, and filter each document image to be processed using a preset classification model to obtain valid document images.

[0032] Understandably, the document images to be processed are those requiring signature verification, such as images uploaded by users in a claims scenario, which may contain upload errors. The preset classification model is trained using a large amount of sample data. Valid document images are those that conform to the preset scenario; for example, in a claims scenario, landscape photos and selfies uploaded by users are invalid.

[0033] Specifically, at least one image of a document to be processed is acquired. A preset classification model is then invoked, and the images are filtered using this model. In other words, the preset classification model selects all images of documents to be processed, retaining valid images and deleting invalid ones, thus obtaining valid document images. For example, in an insurance claims scenario, the images of documents to be processed might be images of signatures from various supervisors reviewing the claims materials. The preset classification model filters out other images mixed in with these documents, thus obtaining valid images of documents with supervisor signatures. Alternatively, the images of documents to be processed might be images of vehicles at the scene, images of accident reports, as well as landscape photos and selfies. The preset classification model filters and classifies these images, resulting in valid images of vehicles at the scene and accident reports.

[0034] S20: Perform template matching on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas.

[0035] In essence, the handwritten area refers to the area where a signature is required. The printed area refers to the area where text is printed. For example, in an insurance contract, the printed names of Party A and Party B are the printed area, and the area where signatures are required is the handwritten area.

[0036] Specifically, after obtaining valid document images, template matching is performed on each valid document image. This involves first recognizing the title of each valid document image to obtain the corresponding image title. Then, the image titles are matched with preset templates in the template dataset, and the successfully matched preset templates are identified as target templates. Next, the handwritten and printed areas in each valid document image are extracted using the target template to obtain the corresponding handwritten areas and printed areas. For example, in the claims review process, the review document images are matched with preset templates. When the title of the review document image matches the title of the preset template, it is identified as the target template. The corresponding areas in the review document image are then extracted according to the handwritten and printed areas marked in the target template to obtain the handwritten areas and the corresponding printed areas in the review document image.

[0037] S30, determine whether there is a signature in the handwritten area. If there is no signature in the handwritten area, identify the printed area to obtain the text content.

[0038] Understandably, text content refers to the text within the printed area. For example, in the claims process, the text reviewed by each supervisor, i.e., the first-level supervisor's review, the second-level supervisor's review, and so on.

[0039] Specifically, after obtaining the handwritten and printed areas, the process determines whether a signature exists in the handwritten area. This involves scanning and recognizing the handwritten area, specifically detecting the presence of text. If text is present, a signature is confirmed. If no text is present, no signature is confirmed. When no signature is found, the printed area is then identified using OCR technology. This involves scanning the text document, analyzing the image file, and obtaining the corresponding text content for each printed area. The OCR process itself is not detailed here. For example, in the claims review process, each level of supervisor needs to sign after review. In this case, the supervisor's signature is determined by detecting the handwritten area; if no signature is found, the corresponding supervisor is identified by recognizing the printed area.

[0040] S40: Push the document to be processed corresponding to the valid document image according to the text content, and obtain the signed document image after the first target person signs it.

[0041] Understandably, a signed document image refers to an image of a document after the handwritten area has been signed.

[0042] Specifically, after obtaining the text content, the system pushes the corresponding document to be processed to the image of the document to be processed based on the text content. This involves obtaining a preset signature dictionary and matching the text content using the dictionary to identify the first target person corresponding to the text content. The document to be processed is then pushed to the first target person, and an image of the signed document is obtained after the first target person signs it. If multiple text contents exist, the same document to be processed can be pushed to different target persons simultaneously, and the signatures of each target person are verified separately. Alternatively, the documents can be pushed to the matched target persons sequentially according to the order of the identified text content. For example, in the claims process, if the second-level supervisor and the fifth-level supervisor do not have signatures for the document review, the document is first pushed to the second-level supervisor for signature, then to the fifth-level supervisor for signature, and then the feedback is sent to the server.

[0043] S50: Verify the signed document image using a preset feature tree. If the signature in the preset feature tree is the same as the signature in the signed document image, determine that the verification result is that the signature is correct.

[0044] Understandably, the preset feature tree refers to the feature tree obtained by extracting features from all signatures.

[0045] Specifically, after obtaining the image of the signed document, features are extracted from the handwritten area within the image to obtain image features of the handwritten area. Then, the image features are retrieved using the Annoy search algorithm within a preset feature tree. This involves calculating the similarity between the image features and preset image features in the preset feature tree. If the similarity value is less than a preset threshold, it indicates a signature error, meaning the signature in the preset feature tree and the signature in the signed document image are different. The search continues until the similarity value is greater than or equal to the preset threshold, indicating a signature match, meaning the signature in the preset feature tree and the signature in the signed document image are the same, and the signature is considered correct. If no corresponding preset image feature is found in the preset feature tree, the signature is considered incorrect. When the signature is incorrect, a verification failure message is displayed. For example, in a claims review process, if the signatures of each level of supervisor are correct, the review is considered complete, and subsequent steps can proceed.

[0046] In one embodiment of the present invention, a signature verification method is proposed. This method involves template matching of valid document images to extract handwritten and printed areas. First, it detects the presence of a signature in the handwritten area, thus determining whether a signature exists in the handwritten area of ​​a document used in financial or insurance transactions. Then, if no signature is found, it identifies the text content within the printed area and pushes the unsigned document image to the target personnel, enabling the delivery of unsigned documents to the intended recipients in financial or insurance transactions. Simultaneously, the present invention verifies the signature of the signed document image using a preset feature tree, thereby detecting the correctness of the signature in the handwritten area of ​​the document in financial or insurance transactions.

[0047] In one embodiment, before step S10, that is, before filtering each of the document images to be processed using a preset classification model, the process includes:

[0048] S101, Obtain a sample dataset, the sample dataset including at least one sample data and sample labels corresponding to each of the sample data.

[0049] Understandably, sample data consists of images from historical data. The sample dataset includes at least one sample data point and corresponding sample labels. For example, in an insurance claims scenario, sample data might be images from claims, such as medical records including signatures of attending physicians, etc. Sample labels are used to characterize the validity of the sample data; for example, detailed medical expense lists, original medical expense vouchers, and outpatient or emergency room medical records. Furthermore, a sample dataset is constructed based on all sample data and the corresponding sample labels. Additionally, some negative samples can be added to train a pre-defined classification model, thereby improving the accuracy of the pre-defined classification model.

[0050] S102, Obtain a preset training model, and classify all the sample data using the preset training model to obtain predicted labels.

[0051] Understandably, the predicted label is the valid image obtained by the pre-trained model classifying the sample data.

[0052] Specifically, a pre-set training model is obtained, and all sample data and sample labels are input into the pre-set training model. The pre-set training model classifies all sample data, that is, iteratively trains the pre-set training model with sample data, enabling the pre-set training model to classify the sample data as valid and invalid, thereby obtaining the predicted label corresponding to each sample data. For example, in an insurance claims scenario, images are classified, specifically landscape photos, ID card photos, accident liability determination letters, medical expense details, and outpatient or emergency medical records. ID card photos, accident liability determination letters, medical expense details, and outpatient or emergency medical records are identified as valid images, while landscape photos are identified as invalid images and deleted, thus obtaining the predicted label.

[0053] S103, determine the prediction loss value of the preset training model based on the sample label and prediction label corresponding to the same sample data.

[0054] Understandably, the prediction loss is generated during the feature extraction process of the sample training data.

[0055] Specifically, after obtaining the predicted labels, all predicted labels corresponding to the sample data are arranged according to the order of the sample data in the sample dataset. Then, the predicted labels associated with the sample data are compared with the sample labels of sample data in the same sequence. That is, according to the sample data sorting, the sample label corresponding to the first sample data is compared with the predicted label corresponding to the first sample data, and the loss value between the sample label and the predicted label is determined by the loss function. The sample label corresponding to the second sample data is compared with the predicted label corresponding to the second sample data, and the loss value between the sample label and the predicted label is determined by the loss function. This process continues until all sample labels and predicted labels have been compared, and the predicted loss value of the preset training model can be obtained.

[0056] S104, when the predicted loss value reaches the preset convergence condition, the preset training model after convergence is determined as the preset classification model.

[0057] Understandably, the convergence condition can be either the predicted loss value being less than a set threshold, or the predicted loss value being very small after 500 calculations and no longer decreasing, at which point training can stop.

[0058] Specifically, after obtaining the predicted loss value, if the predicted loss value does not reach the preset convergence condition, the initial parameters of the preset training model are adjusted based on the predicted loss value. All sample data and sample labels are then re-inputted into the preset training model with adjusted initial parameters, and iterative training is performed on the model to obtain the predicted loss value corresponding to the model with adjusted initial parameters. Then, if the predicted loss value does not reach the preset convergence condition, the initial parameters of the preset training model are adjusted again based on this predicted loss value, until the predicted loss value of the model with adjusted initial parameters reaches the preset convergence condition. In this way, the predicted result continuously approaches the correct result, and the accuracy of the preset training model increases until the predicted loss value of the preset training model reaches the preset convergence condition. At this point, the converged preset training model is determined as the preset classification model.

[0059] This invention iteratively trains a pre-defined training model using a large amount of sample data and calculates the overall loss value of the pre-defined training model using a loss function, thereby determining the predicted loss value of the pre-defined training model. The initial parameters of the pre-defined training model are adjusted based on the predicted loss value until the model converges, thus determining the pre-defined classification model and ensuring its high accuracy.

[0060] In one embodiment, such as Figure 3 As shown, in step S20, which involves template matching for each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to the handwritten area, the process includes:

[0061] S201, perform title recognition on each of the valid document images to obtain the image title corresponding to each of the valid document images.

[0062] Understandably, the image title is the title in the document, such as an insurance contract or a service contract.

[0063] Specifically, a stroke-width equalization algorithm is used to analyze the pages of valid document images. This involves scanning the pages of valid document images to check if the content is printed text, and identifying the valid document images corresponding to printed text as standard images. Then, the standard images are scanned and divided using a projection histogram method. This involves scanning the image from the top left corner, progressively from left to right and top to bottom, and then using projection to break down the content of the standard image into line-by-line content, thus obtaining decomposed data images. Next, connected component analysis is used to segment all decomposed data images with equal width, that is, to cut the divided line-by-line content into characters, thus obtaining character content images. Width recognition is performed on the character content images, and the widest or largest character area is selected as the title area of ​​the valid document image. The corresponding image title is then extracted from the text in the title area. For example, in insurance claims review, the printed text in the selected claims data images is recognized, and the content with the widest character width or largest font size is selected as the image title.

[0064] S202, obtain the template dataset, match the image title with the preset template in the template dataset to obtain the target template.

[0065] S203, the handwritten area and printed area in each of the valid document images are extracted using the target template to obtain the handwritten area and the printed area corresponding to the handwritten area.

[0066] Understandably, the target template is a preset template that is the same as the image title.

[0067] Specifically, a template dataset is obtained, and the image titles are matched with preset templates in the dataset. This involves matching the image titles with the titles of the preset templates until both titles match. A successful match is then identified, and the matched preset template is designated as the target template. The handwritten and printed areas in each valid document image are extracted using the target template. Specifically, according to the regions defined in the target template, the handwritten and printed areas in each valid document image are extracted, corresponding to the handwritten and printed areas in the target template. For example, in insurance claims review, the printed areas of the text content indicating first-level and second-level supervisor review in the target template are extracted, along with the corresponding handwritten areas, to obtain the handwritten and printed areas in the claims document image.

[0068] This invention, through title recognition of each valid document image, determines the image title. Matching the image title with preset templates in the template dataset allows for the selection of a target template. Extracting regions from the valid document images using regions within the target template enables the extraction of handwritten and printed areas.

[0069] In one embodiment, after step S30, that is, after recognizing the printed area and obtaining the text content, the method further includes:

[0070] S301, Obtain a preset error correction model, and embed the text content through the embedding layer of the preset error correction model to obtain the vector to be corrected.

[0071] S302, perform masking error correction on the vector to be corrected to obtain the target content.

[0072] Understandably, the target content is obtained after error correction and detection of the text content.

[0073] Specifically, after obtaining the text content, a preset error correction model is acquired. The text content corresponding to each printing area is input into the preset error correction model, and the text content is embedded through the embedding layer of the preset error correction model, that is, one-hot encoding of the text content, thereby generating content encoding. Then, all content encodings corresponding to the text content are quantized to obtain the error vector to be corrected for each text content. Then, masking error correction is performed on the error vector to be corrected, that is, MASK prediction is performed on the error vector to be corrected, thereby obtaining the error-corrected content. When the error-corrected content is the same as the text content, it means that all the recognized text content is correct, and the text content is determined as the target content. When the error-corrected content is different from the text content, it means that the recognized text content contains characters with recognition errors, and the error-corrected content is determined as the target content. For example, in the insurance review scenario, masking error correction is performed on the error vectors to be corrected corresponding to the first reviewer and the second reviewer, that is, the first [MASK] person and the [MASK] reviewer, thereby predicting the error-corrected content. Then, the predicted content and the recognized content are compared for similarity, thereby determining the target content. When the vector to be corrected is (h1, h2, ..., hn), masking is performed on (h1, h2, ..., hn), that is, the correction result is generated by masking (h1, [MASK], ..., hn).

[0074] This invention achieves the acquisition of the error-correction vector by embedding text content. By performing masking error correction on the error-correction vector, the target content is determined, thereby improving the accuracy of text correction.

[0075] In one embodiment, after step S30, that is, after determining whether a signature exists in the handwritten area, the method further includes:

[0076] S303, when a signature exists in the handwritten area, feature extraction is performed on the signature in the handwritten area to obtain signature features.

[0077] Understandably, signature features refer to the image features of a signature within a handwritten area.

[0078] Specifically, after determining whether a signature exists in the handwritten area, feature extraction is performed on the signature within that area. This involves obtaining an image feature extraction model. The document image to be processed is input into the image feature extraction model, which extracts features from the handwritten areas within the image to obtain the signature features. Algorithms such as SIFT, HOG, and LBP can also be used for extraction. For example, in an insurance review scenario, image features of the signatures of each level of management in the image are extracted to obtain the image features corresponding to the signatures in each handwritten area.

[0079] S304, the signature feature is verified by the preset feature tree. When the signature feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

[0080] Specifically, the signature features are validated using a preset feature tree. This involves retrieving signature features from the preset feature tree using the Annoy search algorithm, or calculating the similarity between the signature features and preset image features. If the similarity value is less than a preset threshold, the signature in the preset feature tree is determined to be different from the signature in the document image to be processed, resulting in a signature error and a validation failure message. Matching continues until the similarity value is greater than or equal to the preset threshold, indicating that the signature in the preset feature tree matches the signature in the valid document image, at which point the validation result is determined to be a correct signature. For example, in an insurance verification scenario, if the similarity between any preset feature in the preset feature tree and the extracted feature exceeds the threshold, the validation result is determined to be a correct signature.

[0081] This invention extracts signature features directly from the handwritten area when a signature is present. The signature features are then validated using a preset feature tree to determine if the signature is correct. Furthermore, if the signature features match preset features in the preset feature tree, the validation result indicates that the signature is correct.

[0082] In one embodiment, such as Figure 4 As shown, in step S40, that is, pushing the document to be processed corresponding to the valid document image according to the text content, and obtaining the signed document image after the first target person signs, including:

[0083] S401, Obtain a preset signature dictionary, match the text content with the preset signature dictionary, and obtain the first target person corresponding to the text content.

[0084] S402, push the document to be processed corresponding to the valid document image to the first target person, and obtain the signed document image after the first target person signs it.

[0085] Understandably, the default signature dictionary is a signature database built based on different document materials. The default signature dictionary includes printed areas and corresponding handwritten areas, for example, Attending Physician: XXX; First Reviewer: XXX. The signed document image is an image of the document after the primary target person has signed it.

[0086] Specifically, after obtaining the text content, a preset signature dictionary is retrieved from the database. This dictionary is then used to match the text content; that is, the title of the valid document image is matched against the corresponding signature information in the dictionary. The printed content within each signature is then matched against the text content. When a matching printed content is found, the person's name in the corresponding signature is identified as the first target person. The document to be processed, corresponding to the valid document image, is then sent to the first target person via a preset channel. After the first target person signs in the designated handwritten area, the signed document is sent back to the server, and the signed document image is retrieved. For example, during a claims review process, if an unsigned supervisor is detected, the supervisor's name is determined using the preset signature dictionary, the valid document is then sent to that supervisor, and the signed image is retrieved.

[0087] This invention, through a preset signature dictionary, matches text content to identify the target person signing the handwritten area. By pushing a document image to the first target person, the signed document image is obtained, thus enabling the pushing of an unsigned document to the target person when no signature exists.

[0088] In one embodiment, before step S40, that is, before pushing the document to be processed corresponding to the valid document image according to the text content and before obtaining the signed document image after the first target person's signature, the process includes:

[0089] S403, when the first target person changes, push the document to be processed corresponding to the valid document image to the second target person according to the text content, and obtain the target document image after the second target person signs it.

[0090] S404, Perform feature extraction on the target document image to obtain target features.

[0091] Specifically, when the first target person changes, a preset signature dictionary is retrieved. The text content is then matched against this dictionary, specifically by matching the title of the valid document image with the corresponding signature information in the dictionary. The printed content within each signature is matched against the text content until a matching printed content is found. The person's name in the signature corresponding to this printed content is then identified as the second target person. The document to be processed, corresponding to the valid document image, is then pushed to the second target person, and the signed target document image is retrieved. Feature extraction is performed on the target document image, specifically by obtaining an image feature extraction model. The target document image is input into this model, and features are extracted from the handwritten areas within the image to obtain the target features. Feature extraction can also employ algorithms such as SIFT, SURF, HOG, and LBP.

[0092] S405, the target feature is verified by the preset feature tree. When the target feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

[0093] Specifically, the target features are validated using a preset feature tree. This involves retrieving the target features from the preset feature tree using the Annoy search algorithm, which calculates the similarity between the target features and preset image features. When the similarity value is less than a preset threshold, the signature in the preset feature tree is determined to be different from the signature in the target document image, resulting in a signature error and a validation failure message. Matching continues until the similarity value is greater than or equal to the preset threshold, indicating that the signature in the preset feature tree and the signature in the target document image are the same, thus determining the signature as correct. For example, in a claims review scenario, when the similarity between a preset feature in the preset feature tree and an extracted feature exceeds a preset threshold, the signature is determined to be correct. In this embodiment, when the first target person changes, the document to be processed is pushed to a second target person. Validating the target features using the preset feature tree determines whether the signature is correct; therefore, when the signature feature is the same as a preset feature in the preset feature tree, the validation result is determined to be a correct signature.

[0094] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0095] In one embodiment, a signature verification device is provided, which corresponds one-to-one with the signature verification methods described in the above embodiments. For example... Figure 5 As shown, the signature verification device includes a document filtering module 11, a template matching module 12, a content recognition module 13, a document push module 14, and a signature verification module 15. Detailed descriptions of each functional module are as follows:

[0096] The document filtering module 11 is used to acquire at least one document image to be processed, and to filter each document image to be processed through a preset classification model to obtain a valid document image.

[0097] The template matching module 12 is used to perform template matching on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas.

[0098] Content recognition module 13 is used to determine whether there is a signature in the handwritten area. When there is no signature in the handwritten area, the printed area is recognized to obtain the text content.

[0099] The document push module 14 is used to push the document to be processed corresponding to the valid document image according to the text content, and to obtain the signed document image after the first target person signs it.

[0100] The signature verification module 15 is used to verify the signature document image through a preset feature tree. When the signature in the preset feature tree is the same as the signature in the signature document image, the verification result is determined to be a correct signature.

[0101] In one embodiment, the device further includes:

[0102] The signature feature unit is used to extract features from the signature in the handwritten area when a signature exists therein, and obtain signature features.

[0103] The second verification unit is used to verify the signature feature through the preset feature tree. When the signature feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

[0104] In one embodiment, the document push module 14 includes:

[0105] The content matching unit is used to obtain a preset signature dictionary, match the text content with the preset signature dictionary, and obtain the first target person corresponding to the text content;

[0106] The document push unit is used to push the document to be processed corresponding to the valid document image to the first target person, and to obtain the signed document image after the first target person signs it.

[0107] In one embodiment, the device further includes:

[0108] The information replacement unit is used to push a document to be processed corresponding to the valid document image to a second target person according to the text content when the first target person changes, and to obtain the target document image after the second target person signs it.

[0109] The target feature unit is used to extract features from the target document image to obtain target features;

[0110] The third verification unit is used to verify the target feature through the preset feature tree. When the target feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

[0111] In one embodiment, the device further includes:

[0112] An embedding processing unit is used to obtain a preset error correction model, and to embed the text content through the embedding layer of the preset error correction model to obtain the error vector to be corrected.

[0113] The masking error correction unit is used to perform masking error correction on the vector to be corrected to obtain the target content.

[0114] In one embodiment, the template matching module 12 includes:

[0115] The title recognition unit is used to recognize the titles of each of the valid document images to obtain the image titles corresponding to each of the valid document images.

[0116] The template matching unit is used to acquire a template dataset, match the image title with a preset template in the template dataset, and obtain the target template.

[0117] The region extraction unit is used to extract the handwritten and printed regions from each of the valid document images using the target template, so as to obtain the handwritten region and the printed region corresponding to the handwritten region.

[0118] In one embodiment, the document filtering module 11 includes:

[0119] A sample acquisition unit is used to acquire a sample dataset, the sample dataset including at least one sample data and sample labels corresponding to each of the sample data;

[0120] The label prediction unit is used to acquire a preset training model, classify all the sample data using the preset training model, and obtain predicted labels.

[0121] The loss prediction unit is used to determine the predicted loss value of the preset training model based on the sample label and predicted label corresponding to the same sample data.

[0122] The model convergence unit is used to determine the preset classification model by the preset training model after convergence when the predicted loss value reaches the preset convergence condition.

[0123] For specific limitations regarding the signature verification device, please refer to the limitations on the signature verification method above, which will not be repeated here. Each module in the aforementioned signature verification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the operations corresponding to each module.

[0124] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data used in the signature verification method described in the above embodiments. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a signature verification method.

[0125] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described signature verification method.

[0126] In one embodiment, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described signature verification method.

[0127] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0129] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A signature verification method, characterized in that, include: At least one document image to be processed is obtained, and each document image to be processed is filtered through a preset classification model to obtain valid document images; Valid document images are those that match the preset scenario; Template matching is performed on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas; Determine whether a signature exists in the handwritten area. If no signature exists in the handwritten area, identify the printed area to obtain the text content. Based on the text content, the document to be processed corresponding to the valid document image is pushed, and the signed document image after the first target person signs is obtained; The signature document image is verified by a preset feature tree. When the signature in the preset feature tree is the same as the signature in the signature document image, the verification result is determined to be a correct signature. The step of performing template matching on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to the handwritten area includes: Perform title recognition on each of the valid document images to obtain the image title corresponding to each of the valid document images; Obtain a template dataset, match the image title with the preset templates in the template dataset, and obtain the target template; The handwritten and printed areas in each of the valid document images are extracted using the target template to obtain the handwritten area and the printed area corresponding to the handwritten area. The step of pushing the document to be processed corresponding to the valid document image based on the text content, and obtaining the signed document image after the first target person signs it, includes: Obtain a preset signature dictionary, and match the text content with the preset signature dictionary to obtain the first target person corresponding to the text content; The document to be processed, corresponding to the valid document image, is pushed to the first target person, and the signed document image after the first target person signs it is obtained.

2. The signature verification method as described in claim 1, characterized in that, Before pushing the document to be processed corresponding to the valid document image based on the text content, and obtaining the signed document image after the first target person's signature, the process includes: When the first target person changes, the document to be processed corresponding to the valid document image is pushed to the second target person according to the text content, and the target document image after the second target person signs it is obtained. Feature extraction is performed on the target document image to obtain target features; The target feature is verified by the preset feature tree. When the target feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

3. The signature verification method as described in claim 1, characterized in that, After identifying the printed area to obtain the text content, the process further includes: Obtain a preset error correction model, and embed the text content through the embedding layer of the preset error correction model to obtain the error vector to be corrected; The target content is obtained by performing masking error correction on the vector to be corrected.

4. The signature verification method as described in claim 1, characterized in that, After determining whether a signature exists in the handwritten area, the method further includes: When a signature is present in the handwritten area, feature extraction is performed on the signature in the handwritten area to obtain signature features; The signature feature is verified by the preset feature tree. When the signature feature is the same as the preset feature in the preset feature tree, the verification result is determined to be a correct signature.

5. The signature verification method as described in claim 1, characterized in that, Before filtering the images of each document to be processed using a preset classification model, the process includes: Obtain a sample dataset, which includes at least one sample data and sample labels corresponding to each of the sample data; Obtain a preset training model, and classify all the sample data using the preset training model to obtain predicted labels; The prediction loss value of the preset training model is determined based on the sample label and prediction label corresponding to the same sample data. When the predicted loss value reaches the preset convergence condition, the preset training model after convergence is determined as the preset classification model.

6. A signature verification device, characterized in that, include: The document filtering module is used to acquire at least one document image to be processed, and to filter each document image to be processed using a preset classification model to obtain valid document images. Valid document images are those that match the preset scenario; The template matching module is used to perform template matching on each of the valid document images to obtain the handwritten area corresponding to each of the valid document images and the printed area corresponding to each of the handwritten areas. The content recognition module is used to determine whether there is a signature in the handwritten area. When there is no signature in the handwritten area, the printed area is recognized to obtain the text content. The document push module is used to push the document to be processed corresponding to the valid document image according to the text content, and to obtain the signed document image after the first target person signs it; The signature verification module is used to verify the signature document image through a preset feature tree. When the signature in the preset feature tree is the same as the signature in the signature document image, the verification result is determined to be a correct signature. The template matching module includes: The title recognition unit is used to recognize the titles of each of the valid document images to obtain the image titles corresponding to each of the valid document images. The template matching unit is used to acquire a template dataset, match the image title with a preset template in the template dataset, and obtain the target template. The region extraction unit is used to extract the handwritten region and the printed region from each of the valid document images using the target template, so as to obtain the handwritten region and the printed region corresponding to the handwritten region; The document push module includes: The content matching unit is used to obtain a preset signature dictionary, match the text content with the preset signature dictionary, and obtain the first target person corresponding to the text content; The document push unit is used to push the document to be processed corresponding to the valid document image to the first target person, and to obtain the signed document image after the first target person signs it.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the signature verification method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the signature verification method as described in any one of claims 1 to 5.

Images