An identity card off-line intelligent identification and correction system for public security actual combat

Abstract

The application discloses an identity card offline intelligent identification and correction system for public security actual combat, relates to the technical field of image processing and artificial intelligence, and comprises an image input module, an image correction module, an OCR identification engine module, an intelligent field extraction module, a backend reasoning core, a front-end interaction interface, a standardized output module and a local storage unit, and all calculations are completed on a local terminal and are independent of a network; the overall execution process is as follows: S1, image input; S2, certificate detection and correction; S3, text detection; S4, character recognition; S5, field extraction; and S6, result output. The identity card offline intelligent identification and correction system for public security actual combat adopts a full offline localization architecture, all image acquisition, model reasoning, data identification and result output are completed on a local terminal, are independent of a network and do not upload to a cloud, the risk of public security sensitive information leakage is avoided from the root, and the data security and confidentiality are significantly improved.

Description

Technical Field

[0001] This invention relates to the fields of image processing and artificial intelligence technology, specifically to an offline intelligent identification and correction system for ID cards for public security operations. Background Technology

[0002] ID card information collection is a high-frequency basic task in grassroots public security work, widely used in scenarios such as hotel registration, internet cafe real-name management, household registration windows, case investigation, traffic enforcement, and security for large-scale events. Currently, there are three main technical solutions: First, manual input, which relies on police officers to input data manually, resulting in low efficiency and a high risk of errors; second, cloud-based OCR services, which complete recognition by calling commercial APIs (such as Baidu OCR and Alibaba OCR), requiring an internet connection and posing data security risks; and third, dedicated recognition equipment, such as ID card readers, which have high accuracy but are costly, have poor portability, are difficult to deploy on a large scale, require uploading to the cloud, pose security risks, and the recognition results are mostly plain text, lacking functions such as image correction, batch processing, and standardized file output, and have poor adaptability to complex scenarios such as tilted or distorted shooting angles.

[0003] To address the aforementioned shortcomings, Chinese Patent Publication No. N110135346A discloses a deep learning-based automatic ID card recognition method and system. The method includes: upon acquiring an input image, using a first fully convolutional neural network to perform facial and national emblem feature detection on the input ID card image, obtaining and correcting the image's rotation information; using a second fully convolutional neural network to perform text detection on the rotated image, obtaining multiple text boxes representing the actual text areas; correcting the tilt of the text boxes to adjust their angles to a horizontal direction; calculating the intersection-union ratio (CUI) between adjacent text boxes in all horizontal directions to connect multiple text boxes in the same row; and using a third fully convolutional neural network to perform character recognition on the connected text boxes to identify the characters within them. This invention's technical solution can automatically recognize ID card information and improve the accuracy of ID card recognition, exhibiting good tolerance.

[0004] Furthermore, in the existing technology, Chinese Patent Publication No. CN120563909A discloses a geometrically corrected, all-angle adaptive ID card image recognition method and system, including: acquiring an ID card image; preprocessing the ID card image; performing edge detection on the preprocessed image; enhancing corner points on the edge contour image; identifying the document language type of the enhanced image; determining the target angle based on the document language type; detecting the text line baseline angle of the enhanced image through projection analysis; constructing a perspective transformation matrix of the enhanced image based on the target angle and the text line baseline angle; and mapping the ID card image according to the perspective transformation matrix to obtain the corrected image. This invention solves the problem of correcting damaged and bent ID cards through a progressive positioning process of edge detection, corner point enhancement, and text line calibration.

[0005] The aforementioned device utilizes deep learning technology to correct and recognize ID card images during use. However, the data from this device needs to be uploaded to the cloud, posing risks to transmission and storage security. The recognition results are mostly plain text and lack functions such as image correction, batch processing, and standardized document output. Summary of the Invention

[0006] The purpose of this invention is to provide an offline intelligent identification and correction system for ID cards for practical use in public security, so as to solve the problems of data security risks and limited functionality in the existing technology for collecting ID card information mentioned in the background.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an offline intelligent ID card recognition and correction system for public security operations, comprising an image input module, an image correction module, an OCR recognition engine module, an intelligent field extraction module, a back-end inference core, a front-end interactive interface, a standardized output module, and a local storage unit. The system adopts a fully offline architecture of front-end acquisition, local inference, and standardized output, with all calculations completed on the local terminal, independent of the network. The overall execution flow is as follows: S1: Image input; S2: ID card detection and correction; S3: Text detection; S4: Character recognition; S5: Field extraction; S6: Result output. All algorithms and models of the system are deployed on the local terminal, independent of the network and cloud services, achieving fully offline operation.

[0008] Furthermore, the image correction module uses a ResNet18 network to detect the ID card area, locates the four corner points of the document, and corrects the tilted and perspective-distorted image into a standard rectangle based on a perspective transformation algorithm, outputting the corrected image for subsequent OCR processing.

[0009] Furthermore, the OCR recognition engine module is implemented based on the domestic PaddleOCRv5 framework, and adopts a two-stage architecture of detection and recognition. In the detection stage, the DBNet algorithm is used to locate the text region, and in the recognition stage, the CRNN+CTC algorithm is used to complete the text recognition. The model performs inference locally through ONNXRuntime, supporting CPU and GPU acceleration.

[0010] Furthermore, the intelligent field extraction module extracts six key pieces of information—name, gender, ethnicity, date of birth, address, and ID number—from the recognition results in a structured manner based on regular expressions, keyword matching, and field location information, and adds manual verification prompts to fields with low confidence.

[0011] Furthermore, the backend inference core is developed using the Rust language and performs model inference through ONNXRuntime. The frontend interactive interface is built using the Tauri framework and TypeScript, supporting Windows and Linux operating systems. The frontend and backend communicate securely through TauriIPC.

[0012] Furthermore, the standardized output module includes a batch multi-image parallel recognition unit, an Excel export unit, a PDF copy generation unit, and an intelligent renaming unit.

[0013] Furthermore, the ResNet18 in the image correction module can be replaced with a lightweight network such as MobileNet or ShuffleNet to reduce the consumption of computing resources.

[0014] Furthermore, the OCR recognition engine module can be replaced with Tesseract or a self-developed offline OCR model, and ONNXRuntime can be replaced with TensorRT or OpenVINO inference framework.

[0015] Furthermore, the backend inference core can be replaced with a C++ or Go language implementation, and the frontend interactive interface can be replaced with an Electron framework implementation.

[0016] Furthermore, the process includes the following steps: S1: Obtain the ID card image locally through the image input module; S2: Detect the corner points of the ID card and perform perspective transformation correction through the image correction module; S3: Perform text detection and character recognition locally through the OCR recognition engine module; S4: Automatically extract six pieces of identity information through the intelligent field extraction module; S5: Batch output Excel, PDF and standardized named files through the standardized output module. The entire process is completed on the local terminal without connecting to the internet or uploading to the cloud.

[0017] Compared with the prior art, the beneficial effects of the present invention are: (1) By adopting a fully offline localized architecture, all image acquisition, model inference, data recognition and result output are completed on the local terminal without relying on the network or uploading to the cloud, thus avoiding the risk of leakage of sensitive public security information from the source and significantly improving data security and confidentiality. (2) The automatic correction method of ResNet18 corner detection and perspective transformation can be used to correct the tilted, deflected and perspective deformed ID card images with high precision, convert non-standard images into standard ID card images, and greatly improve the recognition success rate and accuracy under complex shooting conditions. (3) It has standardized output capabilities such as batch recognition, Excel export with images, automatic generation of PDF copies, and intelligent renaming, which can meet the practical needs of public security file collection, ledger production, and copy archiving in one stop, greatly reducing the repetitive workload of police officers. (4) Based on the domestic PaddleOCR offline engine and ONNXRuntime inference framework, it realizes local high-speed, low-latency and high-accuracy text recognition. The single image recognition speed is fast and the resource consumption is low. It can still run stably in the offline environment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall main interface of the present invention; Figure 2 This is a schematic diagram of the execution flow structure of the present invention; Figure 3 This is a schematic diagram of the operating module structure of the present invention; Figure 4 This is a schematic diagram of the execution steps of the present invention; Figure 5 This is a schematic diagram of the batch ID card recognition structure of the present invention.

[0019] In the diagram: 1. Image input module; 2. Image correction module; 3. OCR recognition engine module; 4. Intelligent field extraction module; 5. Backend inference core; 6. Frontend interactive interface; 7. Standardized output module; 8. Local storage unit. Detailed Implementation

[0020] 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 embodiments of the present invention, and not all embodiments. 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.

[0021] Example 1: Please refer to Figures 1-2This invention provides the following technical solution: an offline intelligent ID card recognition and correction system for public security operations, comprising an image input module 1, an image correction module 2, an OCR recognition engine module 3, an intelligent field extraction module 4, a backend inference core 5, a frontend interactive interface 6, a standardized output module 7, and a local storage unit 8. The system adopts a fully offline architecture of frontend acquisition, local inference, and standardized output, and all calculations are completed on the local terminal without relying on the network. The overall execution flow is as follows: S1: Image input; S2: ID card detection and correction; S3: Text detection; S4: Character recognition; S5: Field extraction; S6: Result output. All algorithms and models of the system are deployed on the local terminal, without relying on the network or cloud services, achieving fully offline operation.

[0022] Through a fully offline architecture design, the system ensures that data is not connected to the internet or uploaded to the cloud during the ID card information collection process, effectively avoiding the risk of data leakage and meeting the high information security requirements of public security operations. The image correction module 2 uses a ResNet18 network that can accurately locate the corner points of the document. Combined with the perspective transformation algorithm, it can correct various tilted and perspective-distorted ID card images into standard rectangles, providing high-quality image input for subsequent OCR processing and improving the accuracy of text recognition. The OCR recognition engine module 3 is based on the domestic PaddleOCRv5 framework. Through a two-stage detection-recognition architecture, the DBNet algorithm first locates the text region, and then the CRNN+CTC algorithm is used to complete the text recognition. The model performs inference locally through ONNXRuntime and supports CPU and GPU acceleration, ensuring the efficiency and real-time performance of recognition. The intelligent field extraction module 4 comprehensively uses regular expressions, keyword matching, and field position information to extract six key pieces of information such as name and gender from the recognition results in a structured manner. It also marks low-confidence fields with manual review prompts, which not only improves the efficiency of information extraction but also ensures the accuracy of information.

[0023] Example 2: Based on Example 1, the problem of poor security in the prior art is solved. Please refer to Example 2. Figures 2-5The OCR recognition engine module 3 was also disclosed, with the following specific structure: Image correction module 2 uses ResNet18 network to detect the ID card area, locate the four corner points of the document, and correct the tilted and perspective-distorted image into a standard rectangle based on the perspective transformation algorithm, outputting the corrected image for subsequent OCR processing. OCR recognition engine module 3 is implemented based on the domestic PaddleOCRv5 framework, adopting a two-stage architecture of detection and recognition. In the detection stage, the DBNet algorithm is used to locate the text area, and in the recognition stage, the CRNN+CTC algorithm is used to complete the text recognition. The model performs inference locally through ONNXRuntime, supporting CPU and GPU acceleration. Intelligent field extraction module 4 extracts six key information items, name, gender, ethnicity, date of birth, address, and ID card number, in a structured manner from the recognition results based on regular expressions, keyword matching, and field position information, and marks low-confidence fields with manual review prompts. Backend inference core 5 is developed using Rust language and performs model inference through ONNXRuntime. Frontend interactive interface 6 is built using Tauri framework + TypeScript, supporting Windows and Linux operating systems. The frontend and backend achieve secure communication through TauriIPC.

[0024] The tilted, skewed, and perspective-distorted ID card photos are input into the system through image input module 1. A ResNet18 convolutional neural network is used for forward inference of the image, outputting the coordinates of the four corner points of the ID card (top left, top right, bottom left, and bottom right). Based on the coordinates of the four corner points, a perspective transformation algorithm is called to map the irregular quadrilateral into a standard 85.6mm×54mm rectangular image, outputting a clear, upright, and distortion-free corrected image, which is then fed into the subsequent OCR recognition engine module 3. At the same time, ResNet18 can be replaced with a lightweight network such as MobileNet or ShuffleNet, which can still achieve real-time correction on low-end terminals. The DBNet algorithm is used to locate the text region of the corrected ID card image, selecting text lines such as name, gender, ethnicity, address, and ID number. A CRNN network combined with the CTC loss function is used to perform end-to-end character recognition of the text region. All models are stored in ONNX format and inference is performed locally through ONNXRuntime, supporting CPU single-precision acceleration. The time to recognize a single ID card is ≤300ms.

[0025] Example 3: Based on Example 1, the problem of limited functionality in the prior art is solved. Please refer to Example 3. Figures 3-5The image correction module 2 was also disclosed, with the following specific structure: the normalized output module 7 includes a batch multi-image parallel recognition unit, an Excel export unit, a PDF copy generation unit, and an intelligent renaming unit. The ResNet18 in the image correction module 2 can be replaced with a lightweight network such as MobileNet or ShuffleNet to reduce the consumption of computing resources. The OCR recognition engine module 3 can be replaced with Tesseract or a self-developed offline OCR model. The ONNXRuntime can be replaced with TensorRT or OpenVINO inference framework. The backend inference core 5 can be replaced with C++ or Go language implementation. The frontend interactive interface 6 can be replaced with Electron framework implementation. The process includes the following steps: S1: Obtain the ID card image locally through the image input module 1; S2: Detect the corner points of the ID card and perform perspective transformation correction through the image correction module 2; S3: Perform text detection and character recognition locally through the OCR recognition engine module 3; S4: Automatically extract six identity information items through the intelligent field extraction module 4; S5: Batch output Excel, PDF, and standardized named files through the normalized output module 7. The entire process is completed on the local terminal without connecting to the internet or uploading to the cloud.

[0026] The intelligent field extraction module 4 performs structured parsing on the raw OCR results, reading the recognized text and corresponding location information. Through keyword matching, regular expressions, and location priors, it extracts six key fields: name, gender, ethnicity, date of birth, address, and ID number. The confidence level of the recognition results is assessed, and low-confidence fields are marked for manual review. Structured JSON data is generated and sent to the standardized output module 7. The standardized output module 7 then performs batch processing: it supports importing 1 to 100 ID card images at once, performing correction, recognition, and extraction in parallel. It exports to Excel: generating a table of the six fields and embedding the corrected ID card image into the cells for easy archiving. Then, it generates a PDF copy: automatically stitching the front and back of the ID card into a standard A4 copy format and outputting a printable PDF. Finally, it intelligently renames the original image and exported file according to name and ID number rules for easy file retrieval.

[0027] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An offline intelligent ID card recognition and correction system for public security operations, comprising an image input module (1), characterized in that: It also includes an image correction module (2), an OCR recognition engine module (3), an intelligent field extraction module (4), a back-end inference core (5), a front-end interactive interface (6), a normalized output module (7), and a local storage unit (8). The system adopts a fully offline architecture of front-end acquisition, local inference, and normalized output, and all calculations are completed on the local terminal without relying on the network. The overall execution flow is as follows: S1: Image input; S2: Document detection and correction; S3: Text Detection; S4: Text recognition; S5: Field extraction; S6: Result output; All algorithms and models of the system are deployed on local terminals, without relying on networks and cloud services, achieving fully offline operation.

2. The offline intelligent identification and correction system for ID cards for public security operations as described in claim 1, characterized in that: The image correction module (2) uses a ResNet18 network to detect the ID card area, locate the four corner points of the document, and correct the tilted and perspective-distorted image into a standard rectangle based on the perspective transformation algorithm, and outputs the corrected image for subsequent OCR processing.

3. The offline intelligent identification and correction system for ID cards for practical public security operations as described in claim 1, characterized in that: The OCR recognition engine module (3) is based on the domestic PaddleOCRv5 framework and adopts a two-stage architecture of detection and recognition. In the detection stage, the DBNet algorithm is used to locate the text region, and in the recognition stage, the CRNN+CTC algorithm is used to complete the text recognition. The model performs inference locally through ONNXRuntime and supports CPU and GPU acceleration.

4. The offline intelligent identification and correction system for ID cards for practical public security operations as described in claim 1, characterized in that: The intelligent field extraction module (4) extracts six key information items—name, gender, ethnicity, date of birth, address, and ID number—from the recognition results based on regular expressions, keyword matching, and field location information, and marks low-confidence fields with manual review prompts.

5. The offline intelligent identification and correction system for ID cards for practical public security operations as described in claim 1, characterized in that: The backend inference core (5) is developed using Rust language and performs model inference through ONNXRuntime. The frontend interactive interface (6) is built using Tauri framework + TypeScript and supports Windows and Linux operating systems. The front and back ends communicate securely through TauriIPC.

6. The offline intelligent identification and correction system for ID cards for practical public security operations as described in claim 1, characterized in that: The standardized output module (7) includes a batch multi-image parallel recognition unit, an Excel export unit, a PDF copy generation unit, and an intelligent renaming unit.

7. The offline intelligent identification and correction system for ID cards for public security operations as described in claim 1, characterized in that: The ResNet18 in the image correction module (2) can be replaced with a lightweight network such as MobileNet or ShuffleNet to reduce the consumption of computing resources.

8. The offline intelligent identification and correction system for ID cards for practical public security operations as described in claim 1, characterized in that: The OCR recognition engine module (3) can be replaced with Tesseract or a self-developed offline OCR model, and ONNXRuntime can be replaced with TensorRT or OpenVINO inference framework.

9. The offline intelligent identification and correction system for ID cards for public security operations as described in claim 1, characterized in that: The backend inference core (5) can be replaced with C++ or Go language, and the frontend interactive interface (6) can be replaced with Electron framework.

10. The identification and correction method of the offline intelligent identification and correction system for public security operations as described in claim 1, characterized in that: Includes the following steps: S1: Obtain the ID card image locally through the image input module (1); S2: Detect the corner points of the ID card and perform perspective transformation correction through the image correction module (2); S3: The OCR recognition engine module (3) performs text detection and character recognition locally; S4: The intelligent field extraction module (4) automatically extracts six identity information items; S5: The standardized output module (7) outputs Excel, PDF and standardized named files in batches. The whole process is completed on the local terminal without connecting to the Internet or uploading to the cloud.

Images