A method, system, device, and medium for verifying the validity of a credential document.

Abstract

This invention discloses a method, system, device, and medium for identifying the validity of voucher documents, relating to the field of voucher document identification technology. The method includes: performing image processing and text recognition on business voucher images to construct a voucher image library; constructing a voucher seal extension library based on the official seal area location information of a benchmark voucher image in the voucher image library; constructing a voucher image point feature extension library based on the point feature information of the benchmark voucher image; and analyzing the voucher image to be identified based on the voucher seal extension library and / or the voucher image point feature extension library to obtain the identification result. This solution can accurately identify blank vouchers that have been stamped in color printing and photocopying, significantly improving the accuracy, reliability, and robustness of voucher document validity identification. It overcomes the problem that existing technologies rely on the verification of the authenticity of the seal, which cannot effectively avoid background interference on the voucher, leading to inaccurate and unreliable identification of the overall authenticity of the voucher document.

Description

Technical Field

[0001] This application relates to the field of document authentication technology, specifically to a method, system, device, and medium for authenticating the validity of documents. Background Technology

[0002] With the development of computer vision technology, various methods for identifying the authenticity of seals have emerged. For example, the patent "A method for identifying the authenticity of seal images based on Siamese feature extraction neural network" (publication number CN117765561B) discloses a Siamese model designed for the characteristics of seal images, making it suitable for identifying the authenticity of seal images. This invention can perform real-time intelligent identification of the seal image under test, determine whether it is a genuine seal, and achieve fast and effective identification of the authenticity of seals. However, this application suffers from low accuracy and insufficient robustness when identifying genuine and counterfeit seal images with subtle differences. Particularly for seal image identification tasks with complex background interference, the DenseNet network used is not accurate enough to meet practical application needs. Patent CN113705330B, "A Method and System for Identifying Genuine and Counterfeit Seals," describes a method that involves acquiring the original seal image and the seal image to be compared; performing a straightening operation on the seal image to be compared; conducting local comparison screening using a preset sampling area; and extracting special points from the seal image and using the lines connecting these special points for comparison. However, this invention only screens the seal image and does not analyze the influencing factors and material texture of the seal image, resulting in inaccurate identification of genuine and counterfeit seals.

[0003] In summary, existing seal verification methods suffer from the following shortcomings: First, manual verification is cumbersome and highly subjective, making it unsuitable for large-scale applications. Second, while existing deep learning methods have made some progress in seal verification, they still suffer from insufficient robustness, struggling to accurately verify seals in complex scenarios and easily affected by background interference. Furthermore, obtaining original seal information and images is challenging, hindering the promotion and application of seal authenticity verification. Finally, when faced with color-printed blank genuine documents containing authentic seal imprints, existing seal verification methods are clearly insufficient, necessitating a systematic solution for verifying the authenticity of counterfeit documents. Summary of the Invention

[0004] The purpose of this application is to address the problem that existing technologies rely on the verification of the authenticity of seals, which cannot effectively avoid interference from the background of the voucher, leading to inaccurate and unreliable identification of the overall authenticity of the voucher document. This application proposes a method, system, device, and medium for identifying the validity of voucher documents. By processing business voucher images to construct a voucher image library, and constructing an extended voucher seal image library and an extended voucher image point feature library based on the official seal area location information and point feature information of the voucher image library, the identification result of the voucher image to be identified is obtained based on the extended voucher seal image library and / or the extended voucher image point feature library. This solution can accurately identify blank vouchers that have been stamped in color photocopying, significantly improving the accuracy, reliability, and robustness of voucher document validity identification, and overcoming the problem that existing technologies rely on the verification of the authenticity of seals, which cannot effectively avoid interference from the background of the voucher, leading to inaccurate and unreliable identification of the overall authenticity of the voucher document.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0006] In a first aspect, embodiments of this application provide a method for verifying the validity of a credential document, the method comprising:

[0007] Image processing and text recognition are performed on business voucher images to construct a voucher image library; based on the official seal area location information of the benchmark voucher image in the voucher image library, a voucher seal text extension library is constructed; based on the point feature information of the benchmark voucher image, a voucher image point feature extension library is constructed; based on the voucher seal text extension library and / or the voucher image point feature extension library, the voucher image to be identified is analyzed to obtain the identification result.

[0008] This solution constructs a voucher image library by processing and recognizing business voucher images. The original vouchers are digitized and stored in the image library, providing standardized image data for subsequent analysis and reducing identification errors caused by original differences in voucher images. A voucher seal image extension library is constructed by acquiring the official seal area location information of the benchmark voucher image in the voucher image library. This accurately locates the official seal area and extracts seal image features, facilitating accurate identification of seal authenticity while avoiding interference from seal patterns on overall image feature analysis. A voucher image point feature extension library is constructed using the point feature information of the benchmark voucher image. This eliminates the influence of fixed patterns in text and seals, focusing only on subtle differences in the voucher background. By reflecting the inherent attributes of the voucher through feature points, it avoids the limitations of relying solely on seal authenticity identification and accurately identifies the authenticity of voucher content. The voucher image image to be identified is analyzed using targeted local features contained in the voucher seal image extension library and the voucher image point feature extension library. This achieves dual verification of seal and background features through regional processing and multi-dimensional feature fusion, significantly improving the accuracy and reliability of voucher document validity identification.

[0009] Preferably, the step of performing image processing and text recognition on the business voucher image to construct a voucher image library includes: extracting table borders from the business voucher image based on an edge detection algorithm and an adaptive threshold binarization algorithm to calculate the corner coordinates and attribute information of the largest rectangle; correcting and transforming the corner coordinates based on the rectangle attribute information to correct the business voucher image; performing text detection and recognition on the corrected business voucher image to extract the official seal unit information; and indexing historical valid voucher images based on the official seal unit information to construct the voucher image library containing several benchmark voucher images.

[0010] Preferably, the step of constructing an extended voucher imprint library based on the official seal area location information of the benchmark voucher images in the voucher image library includes: marking the official seal location in the business voucher images, obtaining an official seal location detection dataset, and constructing an official seal detection model; detecting the official seal area location information of each benchmark voucher image according to the official seal detection model; cropping the corresponding official seal area image according to the official seal area location information, performing corner point correction on the official seal area image, and constructing the extended voucher imprint library by combining it with the official seal unit information.

[0011] Preferably, the step of constructing an extended library of point features for voucher images based on the point feature information of the benchmark voucher image includes:

[0012] Based on text detection and recognition of the business voucher image, several text regions are labeled with text style categories to obtain text region image samples; a text style classification model is constructed based on the text region image samples; handwritten text regions in the benchmark voucher image are identified based on the text style classification model, and feature points and their corresponding feature descriptors in the handwritten text regions are removed to obtain the target feature points and feature descriptor set of the benchmark voucher image; the voucher image point feature extension library is constructed based on the target feature points and feature descriptor sets of all the benchmark voucher images.

[0013] Preferably, the step of analyzing the voucher image to be identified based on the voucher imprint extension library and / or the voucher image point feature extension library to obtain the identification result includes: obtaining the official seal area image, official seal unit information, and feature points of the image to be identified; searching for target feature points that are the same as the official seal unit information of the voucher image to be identified based on the voucher image point feature extension library; calculating the point matching result between the feature points of the voucher image to be identified and the target feature points, and obtaining the difference in the relative position of the official seal in the official seal area image; determining whether the voucher image to be identified is a suspected counterfeit voucher image based on the difference in the relative position of the official seal, performing imprint recognition on the suspected counterfeit voucher image, and obtaining the counterfeit voucher identification result.

[0014] Preferably, the step of performing seal recognition on the suspected counterfeit document image to obtain a counterfeit document identification result includes: based on the document seal extension library, taking multiple copies of the same document seal as combined samples and marking them as positive samples, and combining different document seals and marking them as negative samples; training a seal consistency recognition model based on the positive samples and the negative samples as input parameters of a Siamese network model; fusing the seal of the suspected counterfeit document image with the seal of the document image to be identified as input to the seal consistency recognition model to determine whether the suspected counterfeit document image is a counterfeit document image, and outputting a counterfeit document identification result.

[0015] Preferably, the step of analyzing the voucher image to be identified based on the voucher imprint extension library and / or the voucher image point feature extension library to obtain the identification result further includes: constructing a voucher image imprint set of the same unit by filtering voucher image imprints of the same unit based on the voucher imprint extension library; constructing a same official seal imprint recognition model based on the same official seal imprint set of voucher image imprints of the same unit; searching for imprints in the voucher imprint extension library that are identical to the official seal information of the voucher image to be identified, and fusing them with the imprints of the voucher image to be identified as the input of the same official seal imprint recognition model, so as to determine whether the imprints of the voucher image to be identified and the imprints of the reference voucher image belong to the same official seal, and outputting the imprint authenticity identification result.

[0016] Secondly, embodiments of this application provide a system for verifying the validity of voucher documents, comprising: an image detection module for performing image processing and text recognition on business voucher images to construct a voucher image library; a seal analysis module for constructing a voucher seal extension library based on the official seal area location information of a reference voucher image in the voucher image library; a feature analysis module for constructing a voucher image point feature extension library based on the point feature information of the reference voucher image; and a voucher verification module for analyzing the voucher image to be verified based on the voucher seal extension library and / or the voucher image point feature extension library to obtain verification results.

[0017] Thirdly, embodiments of this application provide a computer device, including: a processor, a memory, and a network interface. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory through the network interface, and the processor executes the machine-readable instructions to perform the steps of the authentication method for the validity of credential documents as described in the first aspect above.

[0018] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the authentication method for the validity of a credential document as described in the first aspect above.

[0019] The beneficial effects of this application are:

[0020] 1. By accurately identifying the table area in the voucher to clarify the baseline range of the content layout, the interference of unstructured background is eliminated to focus on the core format features of the voucher, ensuring the coordinate accuracy of subsequent text detection and seal positioning, and avoiding misjudgment of areas due to image deformation; at the same time, building a voucher image library can effectively avoid the subjectivity of the manual verification process, overcome the problems of low robustness in the existing technology and the need for a large amount of data support in the implementation process.

[0021] 2. By accurately locating and extracting regions, an extended seal database is constructed to separate the official seal from the complex background, enabling precise modeling and efficient management of the seal's features. This transforms the authentication of official seals from "solely relying on manual verification" to "data-driven automated comparison," reducing reliance on human experience and improving authentication efficiency and objectivity.

[0022] 3. The constructed document classification model can quickly locate all handwritten text areas in a voucher without manual intervention, improving processing efficiency. It removes feature points and their corresponding feature descriptors from handwritten text areas, eliminating interference from variable content, thereby enhancing the robustness of feature points. This solves the problem of identifying invalid vouchers when the official seal itself is authentic, achieving efficient and accurate identification of voucher validity. For example, in cases where a blank voucher with a seal is forged by color copying and combining it with image processing software, the official seal itself may be genuine, but the voucher is missing and invalid. Attached Figure Description

[0023] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0024] Figure 1 A flowchart illustrating a method for verifying the validity of a credential document, as provided in this application embodiment.

[0025] Figure 2 An embodiment of this application provides the above-described method. Figure 1 A flowchart illustrating steps S101-S103.

[0026] Figure 3 An embodiment of this application provides the above-described method. Figure 1 A flowchart illustrating the method of step S104.

[0027] Figure 4A schematic diagram of a certificate document validity authentication system module provided in this application embodiment.

[0028] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] Example 1: As Figure 1 As shown, a method for verifying the validity of a voucher document includes:

[0031] S101. Perform image processing and text recognition on business voucher images to build a voucher image library.

[0032] As an optional implementation, S101 specifically includes:

[0033] The table borders in the business voucher image are extracted based on the edge detection algorithm and the adaptive threshold binarization algorithm, so as to calculate the corner coordinates and attribute information of the largest rectangle.

[0034] The corner coordinates are corrected and transformed based on the rectangular frame attribute information to correct the business voucher image; text detection and recognition are performed on the corrected business voucher image to extract the official seal unit information;

[0035] Based on the official seal unit information, historical valid voucher images are indexed to construct the voucher image library containing several benchmark voucher images.

[0036] In some embodiments, the above-mentioned business voucher image is a standard, authentic voucher image. The table borders in the business voucher image can be extracted based on the Canny edge detection algorithm and the OTSU binarization algorithm with adaptive thresholding. Then, the coordinates of the four corner points of the largest rectangle can be calculated, for example (top left, top right, bottom left, bottom right): [(x0,y0),(x1,y1),(x2,y2),(x3,y3)], and the average width of the largest rectangle can be calculated. high: Construct the coordinates of the four corner points of the corrected target rectangle using the width and height of the rectangle:

[0037] [(0,0),(w m -1,0),(0,hm -1),(w m -1,h m -1)];

[0038] The homography transformation matrix is ​​estimated using singular value decomposition, and then the business voucher image is transformed into a corrected voucher image using left-hand transformation and interpolation algorithms, such as the warpPerspective() function in OpenCV.

[0039] Furthermore, such as Figure 2 As shown, an optical character recognition algorithm can be used to extract text content from a corrected business voucher image as input, and combined with a seal recognition model to obtain the unit information in the seal. Based on the seal unit information, historically valid voucher images that have passed authentication are gradually indexed and saved to obtain a voucher image library.

[0040] Among them, the official seal recognition model can be trained based on OCR optical character recognition. The OCR model is a text recognition model that can recognize and extract printed and handwritten text from digital images and PDFs. The official seal recognition model built based on the OCR model can efficiently and accurately obtain the unit information in the official seal in the voucher image.

[0041] In this embodiment, pixel gradient changes are identified through edge detection algorithms to accurately identify the table area in the voucher, and the clarity of the table border is enhanced through adaptive threshold binarization. The maximum rectangle of the table border is extracted to determine the "standard size range" of the voucher, providing a geometric reference for subsequent image correction. The voucher is unified into a positive rectangular view through corner coordinate correction and image transformation to ensure the coordinate accuracy of subsequent text detection and seal positioning, and to avoid misjudgment of areas caused by image deformation.

[0042] Understandably, by using the official seal and unit information as keywords, images of vouchers from the same unit are selected from historical valid vouchers to form a dedicated image library. Invalid samples are eliminated to ensure that all voucher images in the image library belong to the same unit. Subsequent vouchers to be identified only need to be compared with the image library samples of the same unit, without having to traverse the entire dataset, thus reducing computational complexity.

[0043] S102. Based on the official seal area location information of the reference voucher image in the voucher image library, construct an extended voucher seal library.

[0044] As an optional implementation, S102 specifically includes:

[0045] The location of the official seal in the business voucher image is marked to obtain the official seal location detection dataset, and an official seal detection model is constructed.

[0046] The official seal area location information of each of the benchmark voucher images is detected based on the official seal detection model;

[0047] Based on the location information of the official seal area, the corresponding image of the official seal area is extracted. After corner correction of the image of the official seal area, the voucher seal image extension library is constructed by combining it with the official seal unit information.

[0048] In some embodiments, since the official seal in the voucher image is red and consists of a circular outline, the official seal unit text, and a five-pointed star in the center, the positions of the official seals in a large number of voucher images can be marked, and an official seal position detection dataset can be constructed by combining electronic official seal synthesis methods; then, based on typical image semantic segmentation models, such as the U-net network model, an official seal detection model for official seal detection and localization is trained.

[0049] Furthermore, such as Figure 2 As shown, each voucher image in the voucher image library is processed by the official seal detection model to obtain the location information of the official seal area in each voucher image, and a local image of the official seal area is extracted from the voucher image; the local image of the official seal area is detected by the corner detection algorithm to obtain the five corner points of the pentagram in the center area of ​​the official seal image; the image correction algorithm is used to correct the image of the official seal area based on the corner coordinates, and then the official seal unit information of the voucher image is combined to construct an extended voucher seal image library.

[0050] The image correction algorithm is the same as the method described in step S101 above, which uses singular value decomposition to estimate the homography transformation matrix and uses left-hand transformation and interpolation algorithms to correct the image.

[0051] In this embodiment, a unique official seal location label is generated for each voucher to facilitate subsequent analysis of the fixed position of the official seal in the voucher and to quickly detect abnormal displacement. Based on the detected official seal location coordinates, a sub-image of the official seal area is cropped from the voucher image. Perspective transformation or affine transformation and corner correction are then applied to the official seal sub-image to ensure that different official seal samples from the same unit maintain consistency in size, angle, and clarity. This facilitates the subsequent extraction of a unified feature vector, thereby improving the matching accuracy of subsequent voucher images to be identified.

[0052] In this embodiment, the voucher seal extension library not only stores static templates, but also can be self-updated by continuously incorporating new valid vouchers to adapt to reasonable changes in the official seal style, such as the new seal after the unit changes its name. This solves the problem that static matching templates cannot cope with dynamic changes in conventional technologies, and improves the adaptability and flexibility of the identification method.

[0053] S103. Based on the point feature information of the benchmark voucher image, construct an extended voucher image point feature library.

[0054] As an optional implementation, S103 specifically includes:

[0055] Based on text detection and recognition of the business voucher image, several text regions are labeled with text style categories to obtain text region image samples.

[0056] A text classification model is constructed based on the image samples of the text region.

[0057] Based on the text classification model, the handwritten text region in the benchmark voucher image is identified, and the feature points and their corresponding feature descriptors in the handwritten text region are removed to obtain the target feature points and feature descriptor set of the benchmark voucher image.

[0058] The extended feature library of the voucher image points is constructed based on the target feature points and their feature descriptor subsets of all the benchmark voucher images.

[0059] In some embodiments, such as Figure 2 As shown, multiple text regions in the business voucher image can be obtained through the text detection algorithm in the optical character recognition algorithm. These text regions are labeled to obtain a sample library of printed and handwritten text regions. Based on the convolutional neural network structure, a network model for classifying handwritten and printed text is constructed, and then a style classification model for classifying printed and handwritten text types is trained.

[0060] In some embodiments, the point feature information of each reference voucher image in the voucher image library can be calculated based on image feature point detection and feature description algorithms, such as SIFT scale-invariant features, to obtain the feature point set P. n and the feature descriptor subset F for each feature point n Based on the problem classification model, the handwritten text regions of all benchmark voucher images were identified; feature point set P was removed. n The feature points falling within the handwritten text area and their corresponding feature descriptors are used to obtain the set of feature points and feature descriptors (P, F) of the benchmark voucher image. m Where m≤n represents the number of points. The above processing is performed on each baseline voucher image in the voucher image library, ultimately constructing an extended voucher image point feature library.

[0061] The text region can be labeled manually or by algorithm to create labeled text region image samples, thereby enabling automatic recognition of handwritten and printed text. This process removes feature points from handwritten text regions, preventing the inclusion of handwritten content feature points in the identification criteria.

[0062] In this embodiment, a text classification model is used to identify the handwritten text region in the benchmark voucher image, and all feature points and their descriptors in the region are removed to eliminate the interference of variable handwritten content and ensure that the identification error is not caused by changes in handwritten content. After removing the variable region, the remaining feature points can better reflect the inherent format and printing characteristics of the voucher, reducing the feature matching deviation caused by differences in the filled content.

[0063] Understandably, the voucher image point feature extension library stores stable feature points of fixed areas on similar vouchers. Since it's difficult to accurately replicate all feature points in all fixed areas, the uniqueness of the feature points in the library effectively identifies tampered or forged vouchers. In subsequent voucher document authentication, the voucher to be authenticated can be quickly determined whether its fixed structure matches that of a genuine voucher by matching it with feature points in the library. This avoids overall misjudgment caused by tampering with handwriting or official seal areas, thereby improving authentication accuracy and anti-interference capabilities. It effectively solves the problem of existing technologies being unable to effectively avoid background interference on vouchers.

[0064] S104. Analyze the voucher image to be identified based on the voucher imprint extension library and / or the voucher image point feature extension library to obtain the identification result.

[0065] As an optional implementation, S104 specifically includes: determining the point feature set of the image to be identified based on the extended library of voucher image point features, identifying the voucher image to be identified, and obtaining the identification result; specifically as follows: obtaining the official seal area image, official seal unit information, and feature points of the image to be identified;

[0066] Based on the document image point feature expansion library, find target feature points that are the same as the official seal unit information of the document image to be identified;

[0067] Calculate the point matching results between the feature points of the image of the document to be identified and the target feature points, and obtain the difference in the relative position of the official seal in the image of the official seal area;

[0068] Based on the difference in the relative position of the official seal, determine whether the image of the document to be identified is a suspected counterfeit document image, perform seal text recognition on the suspected counterfeit document image, and obtain the counterfeit document identification result.

[0069] Specifically, the step of identifying the seal text of the suspected counterfeit voucher image and obtaining the counterfeit voucher identification result includes: based on the voucher seal text expansion library, taking multiple copies of the same voucher seal text as combined samples and marking them as positive samples, and combining different voucher seal texts and marking them as negative samples.

[0070] A seal consistency recognition model is trained based on the positive and negative samples as input parameters of the Siamese network model. The seal of the suspected counterfeit document image is fused with the seal of the document image to be identified as the input of the seal consistency recognition model to determine whether the suspected counterfeit document image is a counterfeit document image, and the counterfeit document identification result is output.

[0071] In some embodiments, such as Figure 3 As shown, based on the extended library of seal impressions in voucher images, two copies of the same voucher seal impression are combined and labeled as positive samples, while samples combining seal impressions from two different vouchers are labeled as negative samples. This establishes a training dataset for a seal consistency recognition model to determine whether the seal impressions in two voucher documents are identical. The model can be trained using a Siamese network model that calculates the feature similarity between two input images. For example, a residual convolutional neural network structure with shared parameters between the two main branches can be used to train the seal consistency recognition model.

[0072] Specifically, the official seal area image and the official seal unit name are extracted from the image of the document to be authenticated based on the official seal detection model. The handwritten text area of ​​the image of the document to be authenticated is identified based on the text classification model. The point features in the image of the document to be authenticated are calculated based on the image feature point detection and feature description algorithm. Based on the point features, the point feature set with the same official seal unit name is searched in the document image point feature extension library. The target feature points that fall into the handwritten text area and the official seal area in the image of the document to be authenticated are obtained.

[0073] Specifically, the point feature matching algorithm is used to calculate the point matching results between the point features of the image of the document to be identified and the point features of each official seal image region in the point feature set of the document images of the same unit. Then, the homography transformation matrix between the two official seal images is calculated to obtain the difference in the relative position of the official seals in the two official seal images. The document images to be identified with differences exceeding the relative position difference threshold are regarded as the set of suspected counterfeit document images.

[0074] Furthermore, the imprint of each suspected forged document image in the suspected forged document image set constitutes the input of the imprint consistency recognition model with the imprint of the image to be identified. The imprint consistency recognition model can determine whether there is a case where two document images are forged by color copying a blank document that has been stamped, combined with image processing software.

[0075] In this embodiment, the geometric positional relationship of the official seal in two images can be quantified using the homography transformation matrix. If the positions are highly consistent, it indicates possible forgery such as copying and pasting or template application, thus filtering out suspected forgery samples, narrowing the scope of subsequent identification, and focusing computational resources on suspicious samples to improve identification efficiency. Analyzing the differences in the seal impressions between the document to be identified and the suspected forgery document using the seal impression consistency recognition model is effectively applicable to scenarios involving "color-printed and photocopied blank documents with seals + forged content from image processing software," filling the gap that single optical character recognition cannot detect the source of the seal impression and improving the universality of identification.

[0076] As an optional implementation, S104 further includes: searching for seals identical to the official seal information of the document image to be authenticated based on the document seal extension library, performing seal recognition on the document image to be authenticated, and obtaining the authentication result, as follows:

[0077] Based on the aforementioned voucher seal image extension library, voucher image seals from the same unit are selected to construct a voucher image seal set from the same unit; based on the voucher image seal set from the same unit, a model for recognizing the same official seal is constructed.

[0078] The system searches for seals in the extended database that are identical to the seal information of the document image to be authenticated. It then merges these seals with the seal information of the document image to be authenticated and uses this fusion as input to the identical seal seal recognition model. This determines whether the seal information of the document image to be authenticated and the seal information of the reference document image belong to the same seal, and outputs the seal authenticity authentication result.

[0079] In some embodiments, such as Figure 3 As shown, n voucher image imprints from the same unit are selected and filtered from the voucher image imprint extended library to construct a subset of voucher image imprints from the same unit. in, Let i represent a template library of n document seals belonging to the same unit i. A sample combining any one document from unit i with the other documents is labeled as a positive sample. A sample combining any one document from unit i with the template library of the unit is labeled as a negative sample. After end-to-end training, a model for identifying identical official seals that determines whether any document seal belongs to the same seal of the unit is obtained.

[0080] Furthermore, the system searches the document seal extension library for seals with the same unit name as the official seal on the document image to be authenticated, and randomly selects a specified number of seals from this library to construct a set of seals for documents from the same unit. This set of seals, along with the seals on the document to be authenticated, serves as input to the same official seal seal recognition model. This model then determines whether the seals on the document to be authenticated and the seals in the same unit document library originate from the same official seal, thus completing the official seal authentication of the document to be authenticated.

[0081] It should be noted that the use of physical official seals may produce slight differences, but the core features are unique; the same seal imprint recognition model can measure the cosine distance or Euclidean distance of the feature vectors to determine whether the imprints come from the same source; in particular, by comparing multiple samples of the same official seal, the authenticity of the imprint is verified from the level of individual uniqueness, which solves the problem that it is difficult to judge from a single sample.

[0082] As an optional implementation, S104 further includes: comprehensively analyzing the forgery risk level of the image of the document to be identified by integrating the forgery identification result and the seal authenticity identification result, and triggering a manual review procedure based on the forgery risk level; specifically as follows:

[0083] When the suspected counterfeit document image is determined to be a counterfeit document image and the imprint of the document image to be identified is a counterfeit seal, the document image to be identified is determined to be a purely counterfeit document.

[0084] When the suspected forged document image is determined to be a forged document image and the seal of the document image to be identified is a genuine seal, the document image to be identified is determined to be a high-risk forged document, triggering a manual review procedure to determine the validity of the content of the document to be identified.

[0085] In this embodiment, the act of altering content on a blank voucher with a genuine seal is identified through the forgery identification result. The authenticity of the seal itself is further verified through the seal imprint authenticity identification result, effectively preventing direct forgery of "fake seal + genuine content" or "fake seal + fake content". Based on the comprehensive identification of the authenticity of the voucher document to be identified, a three-dimensional protection system of "anti-copying and tampering + anti-seal forgery" is formed. For example, in the case of forging voucher documents by color copying a blank voucher with a seal and combining it with image processing software, the seal imprint itself is genuine, but the voucher document is missing and invalid. This significantly improves the ability to identify various complex forgery methods and ensures the comprehensiveness and reliability of the voucher authenticity judgment.

[0086] Based on the same inventive concept, this application also provides a rail transit business operation and maintenance system corresponding to the rail transit business operation and maintenance method, such as... Figure 4 As shown, the system includes:

[0087] Image detection module 401 is used to perform image processing and text recognition on business voucher images to build a voucher image library;

[0088] Seal analysis module 402 is used to construct a voucher seal extension library based on the official seal area location information of the reference voucher image in the voucher image library;

[0089] The feature analysis module 403 is used to construct a voucher image point feature extension library based on the point feature information of the reference voucher image; the voucher identification module 404 is used to analyze the voucher image to be identified based on the voucher imprint extension library and / or the voucher image point feature extension library to obtain the identification result.

[0090] As an optional implementation, the image detection module 401 is specifically used to: extract the table borders in the business voucher image based on the edge detection algorithm and the adaptive threshold binarization algorithm, so as to calculate the corner coordinates and rectangle attribute information of the largest rectangle;

[0091] The corner coordinates are corrected and transformed based on the rectangular frame attribute information to correct the business voucher image; text detection and recognition are performed on the corrected business voucher image to extract the official seal unit information; based on the official seal unit information, historical valid voucher images are indexed to construct the voucher image library containing several benchmark voucher images.

[0092] As an optional implementation, the seal analysis module 402 is specifically used to: mark the position of the official seal in the business voucher image, obtain the official seal position detection dataset, and construct an official seal detection model; detect the official seal area position information of each of the reference voucher images according to the official seal detection model; extract the corresponding official seal area image according to the official seal area position information, perform corner correction on the official seal area image, and construct the voucher seal extended library by combining it with the official seal unit information.

[0093] As an optional implementation, the feature analysis module 403 is specifically used for: labeling several text regions obtained by text detection and recognition of the business voucher image with text style categories to obtain text region image samples; constructing a text style classification model based on the text region image samples; recognizing handwritten text regions in the benchmark voucher image based on the text style classification model, removing feature points and their corresponding feature descriptors in the handwritten text regions to obtain the target feature points and feature descriptor set of the benchmark voucher image; and constructing the voucher image point feature extension library based on the target feature points and feature descriptor sets of all the benchmark voucher images.

[0094] As an optional implementation, the credential authentication module 404 specifically includes:

[0095] The first document authentication unit is used to acquire the official seal area image, official seal unit information, and feature points of the image to be authenticated; based on the document image point feature expansion library, search for target feature points that are the same as the official seal unit information of the document image to be authenticated; calculate the point matching result between the feature points of the document image to be authenticated and the target feature points, and obtain the difference in the relative position of the official seal in the official seal area image; determine whether the document image to be authenticated is a suspected counterfeit document image based on the difference in the relative position of the official seal, perform seal text recognition on the suspected counterfeit document image, and obtain a counterfeit document authentication result, wherein: based on the document seal text expansion library, multiple copies of the same document seal text are used as combined samples and marked as positive samples, and different document seal texts are combined and marked as negative samples; a seal text consistency recognition model is trained based on the positive samples and the negative samples as input parameters of a Siamese network model; the seal text of the suspected counterfeit document image and the seal text of the document image to be authenticated are fused as input to the seal text consistency recognition model to determine whether the suspected counterfeit document image is a counterfeit document image, and output the counterfeit document authentication result.

[0096] The second voucher authentication unit is used to construct a set of voucher image seals of the same unit by filtering voucher image seals of the same unit based on the voucher seal seal extension library; construct a model for recognizing the same official seal seal based on the set of voucher image seals of the same unit; search for seals in the voucher seal seal extension library that are identical to the official seal information of the voucher image to be authenticated, merge them with the seals of the voucher image to be authenticated as input to the model for recognizing the same official seal seal, so as to determine whether the seals of the voucher image to be authenticated and the seals of the reference voucher image belong to the same official seal, and output the seal authenticity authentication result.

[0097] The third document authentication unit is used to determine that the document image to be authenticated is a pure forgery when the suspected forgery document image is determined to be a forgery document image and the imprint of the document image to be authenticated is a forgery seal; and to determine that the document image to be authenticated is a high-risk forgery document when the suspected forgery document image is determined to be a forgery document image and the imprint of the document image to be authenticated is a genuine seal, and to trigger a manual review procedure to determine the validity of the content of the document to be authenticated.

[0098] This application also provides a computer device, such as... Figure 5 The diagram shown is a schematic representation of the structure of a computer device provided in an embodiment of this application, including: a processor 51, a memory 52, and a network interface 53. The memory 52 stores machine-readable instructions executable by the processor 51 (e.g., ...). Figure 4In the system, the image detection module 401, the seal analysis module 402, the feature analysis module 403, and the voucher authentication module 404 (and their corresponding execution instructions, etc.) are executed. When the computer device is running, the processor 51 and the memory 52 communicate through the network interface 53. When the machine-readable instructions are executed by the processor 51, the steps of the voucher document validity authentication method in the above embodiment are executed.

[0099] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it performs the steps of the authentication method for the validity of the credential document described in the above embodiments.

[0100] The above-described embodiments are preferred embodiments of this application and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape, structure, and method of this application are within the protection scope of this application.

Claims

1. A method for verifying the validity of a voucher document, characterized in that: Includes the following steps: Perform image processing and text recognition on business voucher images to build a voucher image library; Based on the official seal area location information of the reference voucher images in the voucher image library, an extended voucher seal image library is constructed; Based on the point feature information of the benchmark voucher image, a voucher image point feature extension library is constructed. In this process, the handwritten text region in the benchmark voucher image is identified, and the feature points and their corresponding feature descriptors in the handwritten text region are removed to obtain target feature points representing the fixed format of the voucher. The voucher image point feature extension library is then constructed based on the target feature points. The voucher image to be identified is analyzed based on the voucher imprint extension library and the voucher image point feature extension library to obtain the identification result. The steps of analyzing the voucher image to be identified include: Obtain the official seal area image, official seal unit information, and feature points of the image to be identified; based on the voucher image point feature expansion library, find target feature points that are the same as the official seal unit information of the voucher image to be identified; Calculate the point matching results between the feature points of the image to be identified and the target feature points, obtain the difference in the relative position of the official seal in the image of the official seal area, and determine that the image to be identified is a suspected counterfeit image based on the fact that the difference in the relative position of the official seal is less than a preset threshold. The suspected counterfeit document image is subjected to print recognition to obtain the counterfeit document identification result; Furthermore, the authenticity of the seal impressions on the image of the voucher to be authenticated is verified, and the verification result is obtained. Specifically: based on the voucher seal impression extension library, a set of voucher image seal impressions from the same unit is constructed by filtering voucher image seal impressions from the same unit; a model for recognizing identical official seal impressions is constructed based on the set of voucher image seal impressions from the same unit; a seal impression identical to the official seal information of the voucher image to be authenticated is searched in the voucher seal impression extension library, and this seal impression is merged with the seal impression of the voucher image to be authenticated as input to the model for recognizing identical official seal impressions, to determine whether the seal impression of the voucher image to be authenticated and the seal impression of the reference voucher image belong to the same official seal, and the verification result is output. When the identification result of the forged document is a forged document image, and the identification result of the seal authenticity is a genuine seal, the document image to be identified is determined to be a high-risk forged document.

2. The method for verifying the validity of a voucher document according to claim 1, characterized in that: The process of image processing and text recognition of business voucher images to construct a voucher image library includes: The table borders in the business voucher image are extracted based on the edge detection algorithm and the adaptive threshold binarization algorithm, and the corner coordinates and attribute information of the largest rectangle are calculated. The corner coordinates are corrected and transformed based on the rectangular frame attribute information in order to correct the business voucher image; Text detection and recognition are performed on the corrected business voucher images to extract the official seal and unit information; Based on the official seal unit information, historical valid voucher images are indexed to construct the voucher image library containing several benchmark voucher images.

3. The method for verifying the validity of a voucher document according to claim 2, characterized in that: The step of constructing an extended library of document seals based on the official seal area location information of the reference document images in the document image library includes: The location of the official seal in the business voucher image is marked to obtain the official seal location detection dataset, and an official seal detection model is constructed. The official seal area location information of each of the benchmark voucher images is detected based on the official seal detection model; Based on the location information of the official seal area, the corresponding image of the official seal area is extracted. After corner correction of the image of the official seal area, the voucher seal image extension library is constructed by combining it with the official seal unit information.

4. The method for verifying the validity of a voucher document according to claim 2, characterized in that: The step of constructing an extended library of point features for voucher images based on the point feature information of the benchmark voucher image includes: Based on text detection and recognition of the business voucher image, several text regions are labeled with text style categories to obtain text region image samples. A text classification model is constructed based on the image samples of the text region. Based on the text classification model, the handwritten text region in the benchmark voucher image is identified, and the feature points and their corresponding feature descriptors in the handwritten text region are removed to obtain the target feature points and feature descriptor set of the benchmark voucher image. The extended feature library of the voucher image points is constructed based on the target feature points and their feature descriptor subsets of all the benchmark voucher images.

5. The method for verifying the validity of a voucher document according to claim 4, characterized in that: The step of performing print recognition on the suspected counterfeit document image to obtain the counterfeit document identification result includes: Based on the aforementioned voucher seal extension library, multiple copies of the same voucher seal are combined as positive samples and marked as positive samples, while different voucher seals are combined and marked as negative samples. A signature consistency recognition model is obtained by training the Siamese network model using the positive and negative samples as input parameters. The text of the suspected counterfeit document image is fused with the text of the document image to be identified as input to the text consistency recognition model to determine whether the suspected counterfeit document image is a counterfeit document image, and the counterfeit document identification result is output.

6. A system for verifying the validity of a voucher document, characterized in that: A method for verifying the validity of a credential document as described in any one of claims 1-5, the system comprising: The image detection module is used to perform image processing and text recognition on business voucher images to build a voucher image library; The seal analysis module is used to construct a voucher seal extension library based on the official seal area location information of the benchmark voucher image in the voucher image library. Specifically, it identifies the handwritten text area in the benchmark voucher image, removes the feature points and their corresponding feature descriptors in the handwritten text area to obtain target feature points that represent the fixed format of the voucher, and constructs the voucher image point feature extension library based on the target feature points. The feature analysis module is used to construct an extended library of point features for the voucher image based on the point feature information of the benchmark voucher image. The voucher authentication module is used to analyze the voucher image to be authenticated based on the voucher imprint extension library and the voucher image point feature extension library to obtain authentication results. The steps for analyzing the voucher image to be authenticated include: The process involves: acquiring the official seal area image, official seal unit information, and feature points of the image to be identified; searching for target feature points that match the official seal unit information of the image to be identified based on the extended feature library of the voucher image; calculating the point matching results between the feature points of the image to be identified and the target feature points; obtaining the difference in the relative position of the official seal in the official seal area image; and determining that the image to be identified is a suspected counterfeit voucher image based on the fact that the difference in the relative position of the official seal is less than a preset threshold. The suspected counterfeit document image is subjected to print recognition to obtain the counterfeit document identification result; Furthermore, the authenticity of the seal impressions on the image of the voucher to be authenticated is verified, and the verification result is obtained. Specifically: based on the voucher seal impression extension library, a set of voucher image seal impressions from the same unit is constructed by filtering voucher image seal impressions from the same unit; a model for recognizing identical official seal impressions is constructed based on the set of voucher image seal impressions from the same unit; a seal impression identical to the official seal information of the voucher image to be authenticated is searched in the voucher seal impression extension library, and this seal impression is merged with the seal impression of the voucher image to be authenticated as input to the model for recognizing identical official seal impressions, to determine whether the seal impression of the voucher image to be authenticated and the seal impression of the reference voucher image belong to the same official seal, and the verification result is output. When the identification result of the forged document is a forged document image, and the identification result of the seal authenticity is a genuine seal, the document image to be identified is determined to be a high-risk forged document.

7. A computer device, characterized in that: include: The computer device includes a processor, a memory, and a network interface, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the network interface, and the processor executes the machine-readable instructions to perform the steps of the authentication method for the validity of a credential document as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the authentication method for the validity of a credential document as described in any one of claims 1-5.

Images