Certificate copying identification method and system, and computer equipment

Abstract

The invention provides a certificate copying identification method and system, and computer equipment. The method comprises the following steps: acquiring image data related to the identity of a driver; performing standardization processing on the image data to obtain standard image data; extracting moire features in the standard image data to obtain moire feature data; and analyzing the moire feature data based on a preset confidence coefficient to obtain an identification result of whether the image data is copied or not. According to the invention, a feature extraction strategy fusing a moire recognition technology and deep learning is innovatively provided, the copied certificate image is efficiently detected, various common driver registration image types in an online car-hailing platform can be comprehensively covered, and an edge end integrated copying detection function is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of online car-hailing operation management, and in particular to a document photocopying and recognition method and system, and computer equipment. Background Art

[0002] With the popularity of smartphones in daily life, digital images have become an indispensable part. In the online car-hailing industry, drivers who do not meet the qualification requirements often use copied or other people's certificates to register. However, during the copying process, especially when the mobile phone shoots other mobile phones or computer screens, moiré fringes often appear on the monitor. This phenomenon has become an important basis for judging whether the image is copied. At the same time, with the continuous advancement of visual recognition technology, the classification of copied images and normal images has achieved a high degree of accuracy. However, cloud deployment still lacks real-time performance. Only with low latency on the mobile phone side can the corresponding guidance be directly given to the drivers who are registering to avoid using the copying method. The current main technologies for image copying detection are as follows:

[0003] 1. Manually designed feature recognition

[0004] This technology identifies whether an image has been copied by analyzing the gradient variation of the image surface or identifying features such as moiré. However, this method has limited stability and anti-interference performance. Although it performs well in dealing with situations such as copying on electronic screens, its effectiveness is greatly reduced when faced with more diverse copying forms and image types, and it is difficult to meet the audit needs in multiple scenarios and complex environments.

[0005] 2. Deep learning recognition

[0006] The core idea of ​​this method is to regard image remake recognition as an image binary classification task in the field of computer vision. By using deep learning technology, especially convolutional neural networks such as ResNet50, core features are automatically mined and extracted from massive data, and then a classifier is configured at the end of the model to make judgments. Compared with traditional models that rely on manually designed features, this method has significantly improved robustness and the end-to-end training process is simpler. However, in the specific scenario of driver registration, since multiple image types such as driving licenses, vehicle photos, and driver personal photos are involved, simply reducing the remake recognition problem to a binary classification task will encounter challenges. Specifically, there may be significant differences between remake versions of different types of images (such as driving licenses and real-life vehicle photos), while remake and non-remake versions of the same type of image may be highly similar. This "intra-class diversity" and "inter-class ambiguity" problem brings great difficulties to model optimization, making it difficult to achieve ideal convergence. Summary of the invention

[0007] To this end, the present invention provides a method and system for ID card photocopying and recognition, and a computer device, aiming to solve the technical problem that the existing technology is difficult to accurately perform ID card photocopying and recognition in multiple scenarios and complex environments when deployed at the edge.

[0008] To achieve the above objectives, the present invention adopts the following technical solutions:

[0009] According to a first aspect of the present invention, the present invention provides a method for identifying a photocopy of a certificate, the method comprising:

[0010] obtaining image data associated with the driver's identity;

[0011] Performing standardization processing on the image data to obtain standard image data;

[0012] Extracting moiré features from the standard image data to obtain moiré feature data;

[0013] The moiré feature data is analyzed based on a preset confidence level to obtain a recognition result of whether the image data is a photocopy.

[0014] Furthermore, the obtaining of image data related to the driver's identity includes:

[0015] Extracting registration data submitted by the driver when registering from the database, and screening out image data related to the driver's identity from the registration data;

[0016] The image data includes at least one of a vehicle registration certificate image, a driver's license image, an ID card image, a real-life vehicle image, and a driver's own image.

[0017] Furthermore, the step of performing standardization processing on the image data to obtain standard image data includes:

[0018] Scaling the image data according to a fixed aspect ratio so that the longest side thereof reaches a first preset pixel value, thereby obtaining first image data;

[0019] Adjusting pixel data of the first image data so that the mean of the pixel data is 0 and the variance is 1, thereby obtaining second image data;

[0020] A padding operation is performed on the second image data to fill the image size to a multiple of 32 to obtain standard image data.

[0021] Further, the extracting the moiré features in the standard image data to obtain the moiré feature data includes:

[0022] For the standard image data, multi-scale features are extracted using bottleneck blocks with different receptive fields, and the multi-scale features are aggregated to obtain moiré aggregate features; and,

[0023] For the standard image data, a channel target edge predictor is used to detect moiré edge features; and moiré feature data including moiré aggregation features and the moiré edge features are obtained.

[0024] Furthermore, aggregating the multi-scale features includes:

[0025] The multi-scale features are fused through the connection and compression excitation SE blocks, and the function is expressed as follows:

[0026] F m =SE(Cat[NU1(F1),NU2(F2),...,NU n (F n )])

[0027] Among them, SE() represents the operation of SE block; Cat represents connection; F i Represents different feature maps; NU i It is a nonlinear upsampling used to map features to the same spatial size.

[0028] Further, the detecting the moiré edge feature by using the channel target edge predictor comprises:

[0029] Sobel kernel convolution is used to enhance the individual edge maps of each color channel (Esr, Esg, Esb) corresponding to the source image. The formula is as follows:

[0030] E p =Cat(Epr,Epg,Epb)=ge(Is,Esr,Esg,Esb)

[0031] Among them, E p represents the target edge map; ge() represents the channel target edge predictor; Cat represents connection; Epr, Epg, Epb represent the predicted edge maps of R, G, and B channels respectively; Esr, Esg, and Esb represent the separate edge maps of R, G, and B channels respectively.

[0032] Furthermore, the method further comprises:

[0033] When extracting moiré features from the standard image data, a quantization tool is used to quantize the feature layer into an onnx format and deployed on the mobile phone;

[0034] The Softmax function of the last layer is encapsulated into a classification layer written based on CUDA to shorten the detection time of moiré features.

[0035] Furthermore, before analyzing the moiré feature data based on a preset confidence level, the method further includes:

[0036] Pre-collecting a plurality of image data samples, and configuring corresponding confidence thresholds for the image data samples as a test set for testing, so as to generate precision-recall (PR) curves under different confidence thresholds;

[0037] The confidence threshold corresponding to the best balance of the precision-recall (PR) curve is selected as the preset confidence.

[0038] According to a second aspect of the present invention, the present invention provides a document photocopying recognition system, the system comprising:

[0039] A data acquisition module, used to acquire image data related to the driver's identity;

[0040] A data processing module, used for performing standardization processing on the image data to obtain standard image data;

[0041] A feature extraction module, used to extract moiré features from the standard image data to obtain moiré feature data;

[0042] The copy recognition module is used to analyze the moiré feature data based on a preset confidence level to obtain a recognition result of whether the image data is a copy.

[0043] According to the third aspect of the present invention, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for identifying a photocopied certificate as described in any one of the first aspect of the present invention is implemented.

[0044] The present invention adopts the above technical solution and has at least the following beneficial effects:

[0045] Through the scheme of the present invention, image data related to the driver's identity is obtained; the image data is standardized to obtain standard image data; the moiré features in the standard image data are extracted to obtain moiré feature data; the moiré feature data is analyzed based on a preset confidence level to obtain an identification result of whether the image data is a duplicate. Thus, an innovative feature extraction strategy integrating moiré recognition technology and deep learning is proposed to efficiently detect duplicated document images, and can fully cover a variety of driver registration image types commonly seen in online car-hailing platforms, realizing an integrated duplicate detection function at the edge.

[0046] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A schematic diagram showing the flow of a method for identifying a photocopy of a certificate provided by an embodiment of the present invention is shown;

[0049] Figure 2 A schematic diagram showing the structure of a system for identifying a certificate photocopy provided by an embodiment of the present invention is shown;

[0050] Figure 3 A schematic diagram showing the structure of a system for identifying a certificate photocopy provided by another embodiment of the present invention is shown;

[0051] Figure 4 A schematic diagram of the physical structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0052] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0054] The embodiment of the present invention provides a method for identifying a photocopy of a certificate. Figure 1 As shown, at least the following steps S101 to S104 may be included:

[0055] Step S101, acquiring image data related to the driver's identity.

[0056] First, the registration data submitted by the driver when registering is extracted from the user database, and the image data related to the driver's identity is screened out from the registration data. The image data includes the vehicle registration certificate image, driver's license image, ID card image, vehicle real scene image, and driver's own image.

[0057] Step S102, performing standardization processing on the image data to obtain standard image data.

[0058] Specifically, the collected image data is scaled according to a fixed aspect ratio so that its longest side reaches a first preset pixel value to obtain first image data; the pixel data of the first image data is adjusted so that the mean of the pixel data is 0 and the variance is 1 to obtain second image data; a filling operation is performed on the second image data to fill the image size to a multiple of 32 to obtain standard image data.

[0059] Preferably, the first preset pixel value can be 512, that is, the longest side reaches 512 pixels while keeping the aspect ratio of the image unchanged. Then, the image is normalized to adjust the pixel data so that its mean is 0 and its variance is 1. Finally, the image is filled to fill the image size to a multiple of 32 to ensure that it meets the model step size requirements.

[0060] Step S103, extracting moiré features from the standard image data to obtain moiré feature data.

[0061] The extraction of moiré features in the embodiment of the present invention includes two aspects: moiré aggregation features and moiré edge features. In actual operation, multi-scale features can be extracted for standard image data using bottleneck blocks with different receptive fields, and the multi-scale features can be aggregated to obtain moiré aggregation features; and, for standard image data, moiré edge features can be detected using a channel target edge predictor; and moiré feature data including moiré aggregation features and moiré edge features can be obtained.

[0062] Specifically, the embodiment of the present invention uses feature aggregation to study the complex distribution of broadband moiré fringes. The performance of moiré fringes in the frequency domain is quite complex. They not only occupy a wide frequency range, but also are affected by noise. The frequency of moiré fringes is more concentrated in the low-frequency region, which makes them visually difficult to distinguish from ordinary moiré-free images. In addition, the spectrum of moiré fringes varies in various directions, and there are multiple response peaks in the frequency domain, which further increases its complexity. Therefore, in-depth study of these complex characteristics of moiré fringes in the frequency domain is crucial for effectively removing moiré fringes. To meet this challenge, the embodiment of the present invention proposes a method for aggregating multi-scale features in a network to fully consider the frequency characteristics of moiré fringes. In actual operation, bottleneck blocks with different receptive fields are used to extract multi-scale features. The multi-scale features are then fused through connection and compression excitation (SE) blocks. The SE block enhances the features learned by the SE block by calculating the normalized weights of each channel and re-weighting the feature map. The function for fusing multi-scale features through connection and compression excitation SE blocks is expressed as follows:

[0063] F m =SE(Cat[NU1(F1),NU2(F2),...,NU n (F n )])

[0064] Among them, SE() represents the operation of SE block; Cat represents connection; F i Represents different feature maps; NU i It is a nonlinear upsampling method used to map features to the same spatial size, thereby achieving the detection of moiré aggregate features.

[0065] Furthermore, the embodiment of the present invention uses a target edge predictor to estimate the edge of a moiré-free image. Moiré patterns often appear as significant curves and stripes, and their edges change dramatically. When dealing with such edge-sensitive tasks, edge cues, as a common means, have been widely verified to be effective in various layer separation tasks. Compared with natural images, the edge intensity of moiré in the R and B channels is more significant. Therefore, when the edge amplitude difference between color channels is larger, the edge of the moiré will be easier to separate. The embodiment of the present invention uses a channel target edge predictor to analyze the edge map of the target content and the moiré. The input of the target edge predictor includes the image contaminated by the moiré and the edge information of each channel thereof. The channel edge map E is predicted by a neural network. p, thereby obtaining the target image after removing the moiré fringes. Considering the imbalance of the RGB channel edges, the embodiment of the present invention does not use a grayscale edge map, but uses Sobel kernel convolution to enhance the individual edge maps of each color channel (Esr, Esg, Esb) of the source image. The target edge predictor maps such an enhanced input target edge map Ep, and the formula is as follows:

[0066] E p =Cat(Epr,Epg,Epb)=ge(Is,Esr,Esg,Esb)

[0067] Among them, E p represents the target edge map; ge() represents the channel target edge predictor; Cat represents the connection; Epr, Epg, Epb represent the predicted edge maps of the R, G, and B channels respectively; Esr, Esg, Esb represent the separate edge maps of the R, G, and B channels respectively. Thus, the detection of moiré edge features is realized.

[0068] It should be noted that, when extracting moiré features from standard image data, the embodiment of the present invention uses a quantization tool to quantize the feature layer into an onnx format and deploy it on the mobile phone; the Softmax function of the last layer is encapsulated in a classification layer written based on cuda to shorten the detection time of the moiré features. That is to say, in practical applications, a quantization tool is generally used to quantize the feature layer into an onnx format and deploy it on the mobile phone, but the traditional pytorch code can only convert the previous layer of classification into an acceleration format callable by CUDA (mobile phone GPU). The Softmax function of the last layer calls the mobile phone CPU, which causes the edge prediction to increase from 10ms to 100ms for a single picture. Therefore, the embodiment of the present invention optimizes this and encapsulates the Softmax function in a classification layer written by CUDA, so that the final prediction time is restored to 12ms, which significantly improves the operating efficiency on the edge device and reduces the prediction time.

[0069] Step S104: analyzing the moiré feature data based on a preset reliability to obtain a recognition result of whether the image data is a copy.

[0070] Before analyzing the moiré feature data based on the preset confidence, the setting of the preset confidence needs to be solved first. Specifically, multiple image data samples can be collected in advance, and corresponding confidence thresholds can be configured for the image data samples as a test set for testing to generate precision-recall (PR) curves under different confidence thresholds; the confidence threshold corresponding to the best balance of the precision-recall (PR) curve is selected as the preset confidence.

[0071] That is to say, after the image data samples are standardized and extracted through multi-feature moiré, different confidence thresholds for the image data samples can be set, so that the precision-recall rate (PR) curves under different confidence thresholds can be generated according to the known information of these image data samples (i.e., whether they are reprinted images). According to actual business needs, a suitable confidence threshold can be selected to achieve the best balance between precision and recall rate, and the confidence threshold can be used as a preset confidence to analyze the moiré feature data of unknown image data to accurately determine whether the unknown image data is a reprint.

[0072] An embodiment of the present invention provides a method for identifying duplicated documents, by acquiring image data related to the driver's identity; standardizing the image data to obtain standard image data; extracting moiré features in the standard image data to obtain moiré feature data; and analyzing the moiré feature data based on a preset confidence level to obtain an identification result of whether the image data is duplicated. Through the present invention, in response to the deficiencies of the prior art, an innovative feature extraction strategy that integrates moiré recognition technology and deep learning is proposed, aiming to efficiently detect duplicated document images. The present invention can comprehensively cover a variety of driver registration image types commonly seen in online ride-hailing platforms, including but not limited to identity cards, vehicle registration certificates, driver's licenses, personal photos of drivers, etc., and realizes an integrated duplicate detection function at the edge. In addition, the present invention has at least the following advantages:

[0073] (1) Enhanced feature recognition capabilities:

[0074] Multi-scale feature aggregation and edge prediction enhancement: By aggregating multi-scale features and using edge predictors, this method not only improves the recognition accuracy of moiré fringes, but also enhances the ability to detect edges. This enables the technology to more effectively handle a variety of complex image features, not just gradient variation or moiré recognition. The diversity and complexity processing capabilities of this method make it more stable and resistant to interference when faced with a variety of remake forms and image types.

[0075] (2) Optimized edge deployment:

[0076] Through quantization and CUDA optimization, this technology significantly improves the running efficiency on edge devices and reduces the prediction time. This optimization makes the technology more suitable for real-time or near-real-time application scenarios, such as image processing on mobile devices or other edge computing devices. In contrast, relying solely on basic image surface analysis may not be able to effectively achieve efficient operation on edge devices, especially in resource-constrained environments.

[0077] Further, as Figure 1 The specific implementation of the present invention provides a document copy recognition system, such as Figure 2 As shown, the device may include: a data acquisition module 210, a data processing module 220, a feature extraction module 230 and a copy recognition module 240.

[0078] The data acquisition module 210 may be used to acquire image data related to the driver's identity;

[0079] The data processing module 220 can be used to perform standardization processing on the image data to obtain standard image data;

[0080] The feature extraction module 230 can be used to extract moiré features in the standard image data to obtain moiré feature data;

[0081] The copy recognition module 240 can be used to analyze the moiré feature data based on a preset confidence level to obtain a recognition result of whether the image data is a copy.

[0082] Alternatively, if Figure 3 As shown, another embodiment of the present invention provides a system for identifying a certificate by photocopying, further comprising: a threshold configuration module 250.

[0083] The threshold configuration module 250 can be used to collect multiple image data samples in advance before analyzing the moiré feature data based on the preset confidence, and configure the corresponding confidence threshold for the image data samples as a test set for testing to generate a precision-recall rate (PR) curve under different confidence thresholds; select the confidence threshold corresponding to the best balance of the precision-recall rate (PR) curve as the preset confidence.

[0084] Optionally, the data acquisition module 210 may also be used to extract the registration data submitted by the driver when registering from the database, and filter out the image data related to the driver's identity in the registration data;

[0085] The image data includes at least one of a vehicle registration certificate image, a driver's license image, an ID card image, a real-life vehicle image, and a driver's own image.

[0086] Optionally, the data processing module 220 may also be used to scale the image data according to a fixed aspect ratio so that the longest side thereof reaches a first preset pixel value, thereby obtaining first image data;

[0087] Adjusting pixel data of the first image data so that the mean of the pixel data is 0 and the variance is 1, thereby obtaining second image data;

[0088] A padding operation is performed on the second image data to fill the image size to a multiple of 32 to obtain standard image data.

[0089] Optionally, the feature extraction module 230 may also be used to extract multi-scale features for standard image data using bottleneck blocks with different receptive fields, and aggregate the multi-scale features to obtain moiré aggregate features; and

[0090] For standard image data, a channel target edge predictor is used to detect moiré edge features; and moiré feature data including moiré aggregation features and moiré edge features are obtained.

[0091] Optionally, the feature extraction module 230 can also be used to fuse multi-scale features through connection and compression-excitation SE blocks, and the function is expressed as follows:

[0092] F m =SE(Cat[NU1(F1),NU2(F2),...,NU n (F n )])

[0093] Among them, SE() represents the operation of SE block; Cat represents connection; F i Represents different feature maps; NU i It is a nonlinear upsampling used to map features to the same spatial size.

[0094] Optionally, the feature extraction module 230 may also be used to enhance the individual edge maps of each color channel (Esr, Esg, Esb) corresponding to the source image using Sobel kernel convolution, which is expressed as follows:

[0095] E p =Cat(Epr,Epg,Epb)=ge(Is,Esr,Esg,Esb)

[0096] Among them, E p represents the target edge map; ge() represents the channel target edge predictor; Cat represents connection; Epr, Epg, Epb represent the predicted edge maps of R, G, and B channels respectively; Esr, Esg, and Esb represent the separate edge maps of R, G, and B channels respectively.

[0097] Optionally, the feature extraction module 230 may also be used to quantize the feature layer into an onnx format using a quantization tool when extracting moiré features from standard image data, and deploy it on a mobile phone;

[0098] The Softmax function of the last layer is encapsulated into a classification layer written based on CUDA to shorten the detection time of moiré features.

[0099] It should be noted that for other corresponding descriptions of the functional modules involved in the document photocopying recognition system provided in the embodiment of the present invention, reference can be made to Figure 1The corresponding description of the method shown will not be repeated here.

[0100] Based on the above Figure 1 The method shown, accordingly, an embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the document photocopying and recognition method of any of the above embodiments are implemented.

[0101] Based on the above Figure 1 The method shown and Figure 3 The embodiment of the system shown in the figure, the embodiment of the present invention also provides a physical structure diagram of a computer device, such as Figure 4 As shown, the computer device may include a communication bus, a processor, a memory and a communication interface, and may also include an input / output interface and a display device, wherein each functional unit may communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the steps of the document photocopying and recognition method of the above embodiment.

[0102] Those skilled in the art can clearly understand that the specific working processes of the systems, devices, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and for the sake of brevity, they are not further described here.

[0103] In addition, the functional units in various embodiments of the present invention may be physically independent of each other, or two or more functional units may be integrated together, or all functional units may be integrated into one processing unit. The above integrated functional units may be implemented in the form of hardware, or in the form of software or firmware.

[0104] Those skilled in the art can understand that if the integrated functional unit is implemented in the form of software and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can essentially or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, which includes a number of instructions to enable a computing device (such as a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention when running the instructions. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes.

[0105] Alternatively, all or part of the steps of implementing the aforementioned method embodiments may be accomplished by hardware associated with program instructions (such as a computing device such as a personal computer, a server, or a network device), and the program instructions may be stored in a computer-readable storage medium. When the program instructions are executed by a processor of a computing device, the computing device executes all or part of the steps of the methods described in the embodiments of the present invention.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that within the spirit and principles of the present invention, the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate from the protection scope of the present invention.

Claims

1. A method for identifying a photocopy of a certificate, characterized in that: The method comprises: Obtaining image data associated with the driver's identity; Performing standardization processing on the image data to obtain standard image data; Extracting moiré features from the standard image data to obtain moiré feature data; The moiré feature data is analyzed based on a preset confidence level to obtain a recognition result of whether the image data is a photocopy.

2. The method according to claim 1, characterized in that The step of obtaining image data related to the driver's identity includes: Extracting registration data submitted by the driver when registering from the database, and screening out image data related to the driver's identity from the registration data; The image data includes at least one of a vehicle registration certificate image, a driver's license image, an ID card image, a real-life vehicle image, and a driver's own image.

3. The method according to claim 1, characterized in that The step of performing standardization processing on the image data to obtain standard image data includes: Scaling the image data according to a fixed aspect ratio so that the longest side thereof reaches a first preset pixel value, thereby obtaining first image data; Adjusting pixel data of the first image data so that the mean of the pixel data is 0 and the variance is 1, thereby obtaining second image data; A padding operation is performed on the second image data to fill the image size to a multiple of 32 to obtain standard image data.

4. The method according to claim 1, characterized in that The step of extracting the moiré features from the standard image data to obtain the moiré feature data includes: For the standard image data, multi-scale features are extracted using bottleneck blocks with different receptive fields, and the multi-scale features are aggregated to obtain moiré aggregate features; and, For the standard image data, a channel target edge predictor is used to detect moiré edge features; and moiré feature data including moiré aggregation features and the moiré edge features are obtained.

5. The method according to claim 4, characterized in that The aggregating the multi-scale features includes: The multi-scale features are fused through the connection and compression excitation SE blocks, and the function is expressed as follows: <h2 style=";text-align:left;direction:ltr">F<h2 style=";text-align:left;direction:ltr"> m <h2 style=";text-align:left;direction:ltr"> =SE(Cat[NU1(F1),NU2(F2),...,NU<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> (F<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> )]) Among them, SE() represents the operation of SE block; Cat represents connection; F i Represents different feature maps; NU i It is a nonlinear upsampling used to map features to the same spatial size.

6. The method according to claim 4, characterized in that The method of detecting moiré edge features by using a channel target edge predictor comprises: Sobel kernel convolution is used to enhance the individual edge maps of each color channel (Esr, Esg, Esb) corresponding to the source image. The formula is as follows: E p =Cat(Epr,Epg,Epb)=ge(Is,Esr,Esg,Esb) Among them, E p represents the target edge map; ge() represents the channel target edge predictor; Cat represents connection; Epr, Epg, Epb represent the predicted edge maps of R, G, and B channels respectively; Esr, Esg, and Esb represent the separate edge maps of R, G, and B channels respectively.

7. The method according to claim 6, characterized in that The method further comprises: When extracting moiré features from the standard image data, a quantization tool is used to quantize the feature layer into an onnx format and deployed on the mobile phone; The Softmax function of the last layer is encapsulated into a classification layer written based on CUDA to shorten the detection time of moiré features.

8. The method according to claim 1, characterized in that Before analyzing the moiré feature data based on a preset confidence level, the method further includes: Pre-collecting a plurality of image data samples, and configuring corresponding confidence thresholds for the image data samples as a test set for testing, so as to generate precision-recall (PR) curves under different confidence thresholds; The confidence threshold corresponding to the best balance of the precision-recall (PR) curve is selected as the preset confidence.

9. A system for identifying a photocopy of a certificate, characterized in that: The system comprises: A data acquisition module, used to acquire image data related to the driver's identity; A data processing module, used for performing standardization processing on the image data to obtain standard image data; A feature extraction module, used to extract moiré features from the standard image data to obtain moiré feature data; The copy recognition module is used to analyze the moiré feature data based on a preset confidence level to obtain a recognition result of whether the image data is a copy.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the document photocopying and recognition method according to any one of claims 1 to 8 is implemented.

Images