Image recognition-based printing layout detection method and system

Abstract

The application provides a printing layout detection method and system based on image recognition, and relates to the technical field of image recognition, wherein the method comprises the following steps: obtaining a printing scanning image, performing multi-dimensional interference correction, obtaining a printing layout image for feature detection, establishing a first and a second layout element feature group, based on a to-be-printed file, performing incentive driving anomaly detection on the first and the second layout element feature group according to an unbalanced detection compensation mechanism, establishing a first and a second layout anomaly detection atlas, performing global layout matching detection optimization according to the first and the second layout anomaly detection atlas, obtaining a third layout anomaly detection atlas, and performing multi-modal correction processing on the printing layout image. The technical problem that the printing layout detection in the prior art relies on single-dimensional image recognition, resulting in low detection accuracy and reliability, is solved. The technical effect of accurately and comprehensively identifying and positioning various anomalies in layout is achieved, and the detection accuracy and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and specifically to a printing layout detection method and system based on image recognition. Background Technology

[0002] In the modern printing industry, the quality of print layout directly affects the visual appeal and accuracy of information delivery in printed materials. Traditional print layout inspection largely relies on manual proofreading or simple rule-based comparisons, which suffers from low efficiency, high subjectivity, and susceptibility to missed or false detections due to human factors. As the structure of printed materials becomes increasingly diverse, multimodal elements such as text, images, and tables are often interwoven on the same page. Any minor layout error can lead to a decline in the overall quality of the printed material.

[0003] Existing automated inspection methods, while utilizing image recognition technology, mostly focus on single-dimensional inspections, such as text alignment detection or image sharpness detection in printed layouts. These methods suffer from limitations in their detection range, inability to comprehensively cover multiple elements within the layout, and difficulty in handling complex layouts. Furthermore, when faced with layouts containing mixed multimodal elements, existing image recognition technologies struggle to simultaneously and accurately extract and analyze these different types of features, thus affecting the accuracy and reliability of the inspection.

[0004] Existing technologies suffer from the problem that printing layout detection relies on single-dimensional image recognition, resulting in a limited detection range and low accuracy and reliability. Summary of the Invention

[0005] This application provides a printing layout detection method and system based on image recognition, which addresses the technical problem in the prior art where printing layout detection relies on single-dimensional image recognition, resulting in a limited detection range and low detection accuracy and reliability.

[0006] In view of the above problems, this application provides a printing layout detection method and system based on image recognition.

[0007] The first aspect of this application provides a printing layout detection method based on image recognition. The method includes: scanning a document to be printed using a scanning component of a printing press to obtain a printed scanned image; performing multidimensional interference correction on the printed scanned image to obtain a printed layout image; performing Shearlet transform feature detection on the printed layout image to establish a first layout element feature group and a second layout element feature group; performing excitation-driven anomaly detection on the first layout element feature group based on the document to be printed according to an imbalance detection compensation mechanism to establish a first layout anomaly detection map; performing excitation-driven anomaly detection on the second layout element feature group based on the document to be printed according to the imbalance detection compensation mechanism to obtain a second layout anomaly detection map; performing global layout coordination detection optimization based on the first and second layout anomaly detection maps to obtain a third layout anomaly detection map; and performing multimodal correction processing on the printed layout image based on the third layout anomaly detection map.

[0008] Optionally, the scanning environment information and scanning component status information corresponding to the printed scanned image are retrieved simultaneously; interference detection is performed on the printed scanned image based on the scanning environment information to obtain a first interference detection result; interference detection is performed on the printed scanned image based on the scanning component status information to obtain a second interference detection result; adaptive correction processing is performed on the printed scanned image based on the first interference detection result and the second interference detection result to generate the printed layout image.

[0009] Optionally, a multi-scale Shearlet transform is performed on the printed layout image to extract a first-scale image feature set and a second-scale image feature set; layout elements are classified according to the first-scale image feature set to obtain a text element feature set and a non-text element feature set; edge detail smoothing processing is performed on the text element feature set according to the second-scale image feature set to obtain the first layout element feature group; edge detail smoothing processing is performed on the non-text element feature set according to the second-scale image feature set to obtain the second layout element feature group.

[0010] Optionally, text features are captured based on the document to be printed to obtain a text baseline feature group; alignment processing is performed on the text baseline feature group and the first typesetting element feature group to construct multiple text feature comparison vectors; imbalance detection compensation learning is performed on the typesetting text anomaly detection record set according to the imbalance detection compensation mechanism to construct a typesetting text anomaly detection channel; the multiple text feature comparison vectors are traversed to extract a first text feature comparison vector, and the first text feature comparison vector is evaluated for discrepancies to determine a first text detection excitation coefficient; based on the first text detection excitation coefficient, the typesetting text anomaly detection channel is driven to perform multimodal anomaly detection on the first text feature comparison vector to obtain a first text typesetting anomaly detection result, and the first text typesetting anomaly detection result is added to the first typesetting anomaly detection map.

[0011] Optionally, the set of text anomaly detection records is classified according to multiple text defect indicators to obtain multiple text defect detection record clusters; a sample balance evaluation is performed on the multiple text defect detection record clusters to obtain a sample balance coefficient; if the sample balance coefficient does not meet the sample balance constraint, the imbalance cause tracing is performed on the multiple text defect detection record clusters to determine the imbalance tracing factor; adversarial sample injection is performed on the multiple text defect detection record clusters according to the imbalance tracing factor to obtain a text anomaly detection optimization set that meets the sample balance constraint; supervised training is performed on the text anomaly detection optimization set according to multiple local nodes to generate multiple text anomaly detection models; federated aggregation training is performed on the multiple text anomaly detection models to generate the text anomaly detection channel.

[0012] Optionally, non-text features are captured based on the document to be printed to obtain a non-text baseline feature group, and the non-text baseline feature group and the second typesetting element feature group are aligned to construct multiple non-text feature comparison vectors; the typesetting non-text anomaly detection record set is subjected to imbalance detection compensation learning through the imbalance detection compensation mechanism to build a typesetting non-text anomaly detection channel; the multiple non-text feature comparison vectors are traversed to extract the first non-text feature comparison vector, and a first non-text detection excitation coefficient is generated simultaneously; based on the first non-text detection excitation coefficient, the typesetting non-text anomaly detection channel is driven to perform multimodal anomaly detection on the first non-text feature comparison vector to obtain a first non-text typesetting anomaly detection result, and the first non-text typesetting anomaly detection result is added to the second typesetting anomaly detection map.

[0013] Optionally, layout coordination detection is performed based on the first layout element feature group to determine the first layout coordination detection result; layout coordination detection is performed based on the second layout element feature group to obtain the second layout coordination detection result; overall coordination detection is performed based on the first and second layout element feature groups to obtain the third layout coordination detection result; a global layout coordination detection map is generated based on the first, second, and third layout coordination detection results; and a global analysis is performed based on the global layout coordination detection map, the first layout anomaly detection map, and the second layout anomaly detection map to generate the third layout anomaly detection map.

[0014] Optionally, the file format parameters of the document to be printed are obtained; the file format parameters are subjected to compatibility testing according to the file format constraint information of the printing press to obtain the format compatibility testing result; and a file format warning signal is generated according to the format compatibility testing result.

[0015] Optionally, a format optimization reminder message is generated based on the file format warning signal; and the format optimization process is performed on the file to be printed based on the format optimization reminder message.

[0016] A second aspect of this application provides a printing layout detection system based on image recognition. The system includes: a document scanning module for scanning a document to be printed using the scanning components of a printing press to obtain a printed scanned image, and performing multidimensional interference correction on the printed scanned image to obtain a printed layout image; an image detection module for performing Shearlet transform feature detection on the printed layout image to establish a first layout element feature group and a second layout element feature group; a first anomaly map establishment module for performing excitation-driven anomaly detection on the first layout element feature group based on the document to be printed according to an imbalance detection compensation mechanism to establish a first layout anomaly detection map; a second anomaly map establishment module for performing excitation-driven anomaly detection on the second layout element feature group based on the document to be printed according to the imbalance detection compensation mechanism to obtain a second layout anomaly detection map; a map detection optimization module for performing global layout coordination detection optimization based on the first and second layout anomaly detection maps to obtain a third layout anomaly detection map; and an image correction module for performing multimodal correction processing on the printed layout image based on the third layout anomaly detection map.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application scans a document to be printed using the scanning component of a printing press to obtain a printed scanned image. Multidimensional interference correction is then applied to the scanned image to obtain a printed layout image. Shearlet transform feature detection is performed on the printed layout image to establish a first layout element feature group and a second layout element feature group. Based on the document to be printed, an excitation-driven anomaly detection is performed on the first layout element feature group according to an imbalance detection compensation mechanism to establish a first layout anomaly detection map. Based on the document to be printed, an excitation-driven anomaly detection is performed on the second layout element feature group according to the imbalance detection compensation mechanism to obtain a second layout anomaly detection map. Global layout coordination detection optimization is performed based on the first and second layout anomaly detection maps to obtain a third layout anomaly detection map. Multimodal correction processing is then performed on the printed layout image based on the third layout anomaly detection map. This achieves the technical effect of accurately and comprehensively identifying and locating various anomalies in the layout, improving the accuracy and reliability of printed layout detection. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of a printing layout detection method based on image recognition provided in this application.

[0020] Figure 2 This is a schematic diagram of the structure of a printing layout detection system based on image recognition provided in this application.

[0021] Explanation of reference numerals in the attached figures: document scanning module 11, image detection module 12, first anomaly map establishment module 13, second anomaly map establishment module 14, map detection optimization module 15, image correction module 16. Detailed Implementation

[0022] This application provides a printing layout detection method and system based on image recognition, addressing the technical problem in existing technologies where printing layout detection relies on single-dimensional image recognition, resulting in a limited detection range and low accuracy and reliability. It achieves accurate and comprehensive identification and localization of various anomalies in layout, thereby improving the accuracy and reliability of printing layout detection.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides a printing layout detection method based on image recognition, the method comprising: The printing press's scanning components scan the document to be printed to obtain a printed scan image, and then perform multidimensional interference correction on the printed scan image to obtain a printed layout image.

[0025] Furthermore, multi-dimensional interference correction is performed on the printed scanned image to obtain a printed layout image, including: synchronously retrieving scanning environment information and scanning component status information corresponding to the printed scanned image; performing interference detection on the printed scanned image based on the scanning environment information to obtain a first interference detection result; performing interference detection on the printed scanned image based on the scanning component status information to obtain a second interference detection result; and performing adaptive correction processing on the printed scanned image based on the first interference detection result and the second interference detection result to generate the printed layout image.

[0026] Specifically, the document to be printed is scanned by the scanning component of the printing press to obtain a printed scan image. The document to be printed refers to various documents or image files that need to be printed by the printing press, including text content, image content, layout format, and special elements. Special elements refer to QR codes, barcodes, watermarks, etc., used for specific identification or anti-counterfeiting functions. The scanning component of the printing press includes multiple parts such as optical sensors, light sources, scanning platforms, and drive mechanisms. The document to be printed is placed in the designated position on the scanning platform of the printing press, ensuring that the document is flat. The light source is turned on to provide illumination for the optical sensor. The scanning component is started, and the drive mechanism controls the movement of the optical sensor to achieve line-by-line scanning and recognition of the document to be printed, thereby obtaining a printed scan image. The printed scan image includes the text content, image content, layout format, special elements, and interference features such as stripes, spots, and uneven brightness introduced by scanning or environmental factors in the document to be scanned.

[0027] The scanning environment and the status of the scanning components can affect the quality of the scanned image. For example, uneven lighting can lead to inconsistent image brightness, and humidity can affect the flatness of the paper. Therefore, while scanning the document, the scanning environment information and scanning component status information corresponding to the scanning process are retrieved simultaneously. The scanning environment information includes light intensity, ambient temperature and humidity, and vibration data, which are acquired through multiple sensors, such as light sensors, temperature and humidity sensors, and vibration sensors. The scanning component status information includes the position of the optical sensors, the flatness of the scanning platform, the brightness and uniformity of the light source, and the stability of the scanning speed of the drive mechanism. The scanning component status information can be calibrated using a standard grayscale card to detect the response of the optical sensors at different grayscale values, the brightness and uniformity of the light source can be measured using a photometer or colorimeter, the scanning platform status can be detected using a laser flatness detector, and the motor speed can be measured using an encoder.

[0028] Interference detection is performed on the scanned image based on scanning environment information and scanning component status information to identify the interference effects of multiple information sources on the printed scanned image. Environmental interference detection can be achieved by: analyzing the brightness and color distribution of the printed scanned image to detect areas of abnormal brightness or color fluctuations; performing frequency domain analysis to identify periodic stripes or noise peaks; comparing the texture consistency of the printed scanned image to detect vibration or blur interference, and obtaining a first interference detection result. The first interference detection result reflects the type, area, and intensity of interference caused by the scanning environment information on the printed scanned image. Component interference detection can be achieved by: performing geometric distortion analysis on straight lines or text edges in the scanned printed image to detect bending or tilting caused by scanning head or optical lens offset; performing defect detection on image pixels to identify sensor dead pixels or fixed bright / dark spots caused by light source aging; comparing the scanning head motion control signal with the image line synchronization information to identify stretching or compression caused by uneven scanning speed or synchronization deviation; calculating the overall signal-to-noise ratio of the image; and identifying the sensor noise level. The second interference detection result is obtained based on the detection. The second interference detection result is used to reflect the type, area and intensity of interference generated by the scanning component's own operating state on the printed scanned image.

[0029] Based on the first interference detection result, an environmental interference correction model is established. For brightness deviations caused by uneven illumination, histogram equalization or adaptive gamma correction is used to optimize the brightness distribution in different regions. For periodic stripe interference, frequency domain filtering, such as band-stop filters, is used to suppress interference peaks in the spectrum. For vibration blur, motion blur and inverse filters are used for compensation. Based on the second interference detection result, a component interference correction model is established. For geometric distortion, the component interference correction model recalibrates the image edges through perspective transformation matrices or polynomial geometric correction. For sensor dead pixels, pixel interpolation repair methods are used to perform weighted compensation for the neighborhood of the dead pixels. For stretching or compression caused by unstable scanning speed, a dynamic resampling algorithm based on the scanning synchronization signal is used for proportional scaling correction. For a decrease in the overall signal-to-noise ratio, adaptive filtering, such as Wiener filtering, can be used to improve image clarity. A weighted adaptive mechanism is used to fuse the environmental interference correction model and the component interference correction model. Based on the interference intensity parameters in the interference detection results, the weights of different correction models are dynamically allocated to obtain a multi-dimensional correction strategy. For example, when illumination interference is detected as the main influencing factor, the weight of brightness optimization in the environmental interferometry correction model is increased. Adaptive correction processing is applied to the printed scanned image according to a multi-dimensional correction strategy to improve brightness consistency, geometric integrity, and detail clarity, generating a printed layout image that accurately reproduces the layout features of the document to be printed. By obtaining the printed scanned image through scanning and performing multi-dimensional interferometry correction, interference factors that may be introduced during the scanning process are eliminated, improving the reliability and accuracy of the image.

[0030] Furthermore, scanning the document to be printed using the scanning component of the printing press includes: obtaining the file format parameters of the document to be printed; performing compatibility testing on the file format parameters based on the file format constraint information of the printing press to obtain a format compatibility test result; and generating a file format warning signal based on the format compatibility test result.

[0031] Furthermore, generating a file format warning signal includes: generating a format optimization reminder message based on the file format warning signal; and performing format optimization processing on the file to be printed based on the format optimization reminder message.

[0032] Specifically, the process involves obtaining the file format parameters of the document to be printed, including file type, color mode, resolution, page size, font embedding, and layer information. Based on the printing press's operating manual, the process also involves obtaining the printing press's file format constraints, including supported file types, maximum resolution, recognizable color modes, font requirements, and processable layer structures. By comparing each file format parameter and constraint with the specifications, a compatibility check is performed to determine if the document to be printed has incompatibility issues, such as resolution lower than printing requirements, color mode mismatch, or missing fonts.

[0033] Based on compatibility testing, format compatibility test results are obtained, indicating whether the format parameters of the document to be printed meet the printing press requirements. A file format warning signal is generated based on the format compatibility test results. For example, if the color mode of the document to be printed is RGB while the printing press requires CMYK, it indicates incompatibility, and a file format warning signal is generated. Based on the file format warning signal, further format optimization reminder information is generated, clearly indicating the specific parameters that need optimization and the optimization methods, such as converting RGB to CMYK, adjusting the resolution to 300 dpi, embedding missing fonts, or merging unnecessary layers. The document to be printed is then format optimized according to the format optimization reminder information to ensure that the document structure meets the printing press's processing requirements. By judging the file format parameters of the document to be printed, the format of the document to be printed is effectively ensured to be compatible with the printing press, avoiding scanning failures or printing errors due to format mismatch.

[0034] Shearlet transform feature detection is performed on the printed layout image to establish a first layout element feature group and a second layout element feature group.

[0035] Furthermore, based on the printed layout image, Shearlet transform feature detection is performed to establish a first layout element feature group and a second layout element feature group, including: performing multi-scale Shearlet transform on the printed layout image to extract a first-scale image feature set and a second-scale image feature set; classifying layout elements based on the first-scale image feature set to obtain a text element feature set and a non-text element feature set; performing edge detail smoothing processing on the text element feature set based on the second-scale image feature set to obtain the first layout element feature group; and performing edge detail smoothing processing on the non-text element feature set based on the second-scale image feature set to obtain the second layout element feature group.

[0036] Specifically, the Shearlet transform is a multi-scale geometric analysis tool that effectively captures the geometric structure and texture features in an image by combining scaling and shearing operations. Using the Shearlet transform, based on the characteristics of printed layout images, two scales are selected for transformation. Large-scale image features and small-scale image features are extracted from the printed layout image. Large-scale image features refer to larger features in the printed layout image, such as the overall text area, the overall image area, the overall table area, and the outline of large graphics. Small-scale image features refer to relatively smaller details in the printed layout image, such as text edges, image details, line boundaries, and small graphic texture information. The extracted features are integrated to form a first-scale image feature set and a second-scale image feature set. Morphological processing, connected component analysis, and region statistical feature analysis methods are used to classify the layout elements of the first-scale image feature set, dividing it into a text element feature set and a non-text element feature set. The text element feature set includes font type, font size, and text layout format, such as line spacing, character spacing, alignment, and paragraph format. The non-text element feature set includes feature information related to non-text content such as images, tables, and graphics. The text element feature set helps detect problems in text layout, such as inconsistent fonts and uneven line spacing; while the non-text element feature set helps detect layout problems of elements such as images, tables, and graphics, such as image distortion and table misalignment.

[0037] Edge detection algorithms, such as Canny edge detection or the Sobel operator, are used to extract edge information from text elements in the second-scale image feature set. This edge information reflects the contours and structural details of the text. Smoothing filters, such as Gaussian filters or bilateral filters, are applied to the text edges to reduce edge noise and make the transitions smoother. Morphological operations, including dilation and erosion, are then used to further optimize the smoothed edges, filling in any small holes and eliminating small protrusions, forming a first set of typographic element features. This first set of typographic element features fully reflects the shape, contour, and structural details of the text region. Based on the second-scale image feature set, edge detail smoothing is applied to the non-text element feature set. Curve fitting techniques, such as Bezier curves, are used to smooth the contours of graphics, making the boundaries smoother. Furthermore, techniques such as contrast enhancement and sharpening are used to enhance the details of the non-text element feature set, resulting in a second set of typographic element features. This second set of typographic element features includes the contours and edge features of images, tables, and graphics.

[0038] By using multi-scale Shearlet transform, features at different scales in printed layout images can be effectively extracted, and layout elements can be accurately classified, thereby improving the targeting and accuracy of layout anomaly detection and optimization.

[0039] Based on the document to be printed, the first typesetting element feature group is stimulated to perform anomaly detection according to the imbalance detection compensation mechanism, and a first typesetting anomaly detection map is established.

[0040] Furthermore, based on the document to be printed, an excitation-driven anomaly detection is performed on the first typesetting element feature group according to the imbalance detection compensation mechanism to establish a first typesetting anomaly detection map, including: capturing text features based on the document to be printed to obtain a text baseline feature group; aligning the text baseline feature group and the first typesetting element feature group to construct multiple text feature comparison vectors; performing imbalance detection compensation learning on the typesetting text anomaly detection record set according to the imbalance detection compensation mechanism to construct a typesetting text anomaly detection channel; traversing the multiple text feature comparison vectors, extracting a first text feature comparison vector, and performing a separation evaluation on the first text feature comparison vector to determine a first text detection excitation coefficient; driving the typesetting text anomaly detection channel to perform multimodal anomaly detection on the first text feature comparison vector according to the first text detection excitation coefficient, obtaining a first text typesetting anomaly detection result, and adding the first text typesetting anomaly detection result to the first typesetting anomaly detection map.

[0041] Specifically, Optical Character Recognition (OCR) technology is used to capture text features from the document to be printed, obtaining a set of text reference features. These reference features include the content, font, font size, line spacing, and layout position of the text in the document. Through spatial location mapping and morphological feature matching, the text reference feature set and the first layout element feature set are aligned. Spatial location mapping involves translating, scaling, or rotating the text features extracted from the printed layout image based on the row and column coordinates of the reference text on the page, making them coincide as closely as possible with the reference text in the two-dimensional coordinate system. The reference text is the feature information within the text reference feature set. Morphological feature matching involves calculating a similarity score using information such as text outline, edge direction, stroke width, and grayscale distribution, finding the most matching text region for each reference text in the image.

[0042] Through alignment, each reference character and its corresponding image text features are combined to form a text feature comparison vector. This vector contains information such as the feature attributes of the reference character, the bounding box position of the corresponding image text, morphological features, grayscale statistics, similarity score, and positional deviation. Multiple text feature comparison vectors are obtained by traversing and aligning the entire set of reference text features and the first set of layout element features. An imbalanced detection compensation mechanism is used to learn from the historical layout text anomaly detection record set, constructing a layout text anomaly detection channel. This mechanism addresses the data imbalance problem in the historical layout text anomaly detection record set by employing methods such as sample balance evaluation, adversarial sample injection, and federated aggregation training to improve the detection capability of the layout text anomaly detection channel for different anomaly categories, thereby ensuring the comprehensiveness and accuracy of anomaly detection.

[0043] The first text feature comparison vector is sequentially extracted from multiple text feature comparison vectors, and a discrepancy evaluation is performed on the first text feature comparison vector to assess the degree of deviation or anomaly between the first text feature comparison vector and the baseline features. This discrepancy evaluation includes positional deviation analysis, morphological deviation analysis, stroke integrity assessment, and texture feature comparison. For example, positional deviation analysis is achieved by comparing the center coordinate differences between the baseline text bounding box and the image text bounding box, calculating the horizontal and vertical offsets. Morphological deviation analysis is achieved by comparing contour shape differences using text contour matching algorithms, such as Hausdorff distance or shape context matching, quantifying the degree of text deformation. Stroke integrity assessment is achieved by using edge detection, such as the Canny edge operator and skeletonization algorithm, to analyze whether text strokes are missing or broken. Texture feature comparison is achieved by comparing the difference between the baseline text texture distribution and the image text to detect uneven ink distribution or localized ink loss.

[0044] Through anomaly assessment analysis, anomaly assessment results are obtained, including multiple deviation data points. Each deviation data point corresponds to the degree of anomaly of the text in dimensions such as position, shape, stroke count, and grayscale. Based on the anomaly assessment results, each deviation data point is normalized, and weights are assigned according to the severity of the anomaly. For example, the weight increases when the positional offset exceeds a threshold or the percentage of missing strokes exceeds a threshold. A weighted summation is used to calculate the first text detection activation coefficient, which is between 0 and 1. A higher value indicates a greater degree of anomaly, and the corresponding detection channel trigger intensity is also higher.

[0045] Based on the first text detection excitation coefficient, the text anomaly detection channel is driven to perform multimodal anomaly detection on the first text feature comparison vector, including multi-dimensional detection of morphological anomalies, contour offsets, missing strokes, and positional misalignments. After detection, the first text layout anomaly detection result is obtained. Multiple text feature comparison vectors are traversed, and the extraction, anomaly evaluation, and multimodal anomaly detection steps are repeated to obtain multiple text layout anomaly detection results. The obtained first text layout anomaly detection results are added to the first layout anomaly detection map, providing a visual description of the anomalies in the entire text layout area, facilitating subsequent analysis and optimization, and improving the efficiency and accuracy of print layout detection.

[0046] Furthermore, based on the imbalance detection compensation mechanism, imbalance detection compensation learning is performed on the text anomaly detection record set to construct a text anomaly detection channel. This includes: classifying the text anomaly detection record set according to multiple text defect indicators to obtain multiple text defect detection record clusters; evaluating sample balance based on the multiple text defect detection record clusters to obtain a sample balance coefficient; if the sample balance coefficient does not meet the sample balance constraint, tracing the imbalance causes of the multiple text defect detection record clusters to determine the imbalance tracing factor; injecting adversarial examples into the multiple text defect detection record clusters based on the imbalance tracing factor to obtain a text anomaly detection optimization set that meets the sample balance constraint; supervising training of the text anomaly detection optimization set based on multiple local nodes to generate multiple text anomaly detection models; and performing federated aggregation training based on the multiple text anomaly detection models to generate the text anomaly detection channel.

[0047] Specifically, multiple text layout defect indicators include character offset, stroke missing degree, outline deformation degree, font errors, and positional errors. The historical text layout detection record set data is clustered or classified using these multiple text layout defect indicators. For example, the K-means clustering algorithm is used to group records with similar abnormal characteristics into the same cluster, forming multiple text defect detection record clusters. Sample balance evaluation is performed on these multiple text defect detection record clusters. This is done by counting the number of samples in each text defect detection record cluster and calculating the ratio of the number of samples in each category to the total sample data, thus calculating a sample balance coefficient to measure the degree of distribution balance of samples of different defect types in the record set.

[0048] Set sample balance constraints to define the minimum proportion or distribution range of each text defect category in the training data. For example, set the sample balance of each defect category to account for no less than 5% of the total sample size. If the sample balance coefficient does not meet the sample balance constraint, perform imbalance tracing on the multiple text defect detection record clusters corresponding to the sample balance coefficients that do not meet the constraint. Statistically analyze the number and detection difficulty distribution of each type of text defect sample, analyze the frequency of minority class samples and historical detection success rate, and combine the clustering results to identify the differences in the number of samples between clusters. Determine the factors causing sample imbalance, i.e., imbalance tracing factors. For example, the number of a certain type of text defect in the sample set is too small, the frequency of text samples of a specific font or size is low, the detection difficulty of text samples in complex layout positions is high, and the recognition success rate is low due to complex stroke structure. Based on the imbalanced tracing factor, adversarial examples are injected into the text defect detection record cluster through data augmentation methods. For example, affine transformation, random noise superposition, or stroke missing simulation are performed on a few defect samples in the text defect detection record cluster to form adversarial examples. The generated adversarial examples are then added to the text defect detection record cluster to increase the number and diversity of minority class samples, thereby obtaining an optimized set of text anomaly detection that satisfies the sample balance constraint.

[0049] Supervised training is performed on a text anomaly detection optimization set using multiple local nodes. These local nodes are distributed servers or computing units. Each node independently acquires a portion of the optimized text anomaly detection set and undergoes supervised training to reduce the risk of overfitting from single-node data. Supervised training specifically uses labeled data to iteratively update the model. Labels for each text sample include text offset, missing strokes, and contour anomalies. The model type can employ a Convolutional Neural Network (CNN) or Transformer architecture. The input is the optimized text anomaly detection set and its corresponding labels. Spatial and structural features are extracted through multi-layer convolution or self-attention mechanisms, and the text anomaly prediction results are output through fully connected layers. During training, cross-entropy loss or mean squared error loss functions are used, and model parameters are updated using gradient descent, such as the Adam optimizer, until convergence, generating multiple local models, i.e., multiple text anomaly detection models.

[0050] Multiple text anomaly detection models are trained through federated aggregation. Each local node uploads its model parameters, such as convolutional kernel weights and fully connected layer parameters, to the central aggregation node. The central aggregation node performs a weighted average based on the sample size or training quality of each model to obtain globally updated model parameters. The aggregated parameters are then distributed back to each local node for the next training iteration. This process is repeated several times until the global model converges, ultimately generating a high-precision text anomaly detection channel. This channel can perform multimodal detection of various text layout anomalies while remaining sensitive to a few anomaly categories and adapting to different data distributions, thus improving the reliability and accuracy of print layout detection.

[0051] Based on the document to be printed, the second typesetting element feature group is subjected to excitation-driven anomaly detection according to the imbalance detection compensation mechanism to obtain the second typesetting anomaly detection map.

[0052] Furthermore, based on the document to be printed, the second typesetting element feature group is subjected to excitation-driven anomaly detection according to the imbalance detection compensation mechanism to obtain a second typesetting anomaly detection map, including: capturing non-text features based on the document to be printed to obtain a non-text baseline feature group, and aligning the non-text baseline feature group and the second typesetting element feature group to construct multiple non-text feature comparison vectors; performing imbalance detection compensation learning on the typesetting non-text anomaly detection record set through the imbalance detection compensation mechanism to build a typesetting non-text anomaly detection channel; traversing the multiple non-text feature comparison vectors, extracting a non-text feature first comparison vector, and simultaneously generating a non-text detection first excitation coefficient; based on the non-text detection first excitation coefficient, driving the typesetting non-text anomaly detection channel to perform multimodal anomaly detection on the non-text feature first comparison vector to obtain a non-text typesetting anomaly first detection result, and adding the non-text typesetting anomaly first detection result to the second typesetting anomaly detection map.

[0053] Specifically, similar to the method for obtaining the first typesetting anomaly detection map, the non-textual features of the document to be printed are first captured to obtain a non-textual baseline feature group. This non-textual baseline feature group includes the position, size, shape, boundary information, and texture features of images, tables, graphics, lines, and other graphic elements within the document. The non-textual baseline feature group is then aligned with the second typesetting element feature group to generate multiple non-textual feature comparison vectors. The alignment process includes spatial transformations such as translation, scaling, and rotation to ensure that the non-textual element features extracted from the scanned printed image are as consistent as possible with the non-textual baseline features of the document to be printed in two-dimensional coordinates and shape. Simultaneously, similarity matching is performed by combining contour features, texture features, and boundary shapes.

[0054] Based on an imbalanced detection compensation mechanism, an imbalanced detection compensation learning process is performed on the non-textual anomaly detection record set for typography, establishing a non-textual anomaly detection channel. This mechanism analyzes the quantity distribution and detection difficulty of different types of non-textual defect samples, employing adversarial example generation or data augmentation methods to obtain an optimized non-textual anomaly detection set, ensuring that a minority of non-textual defect classes can be sufficiently trained. Supervised training is then performed on the optimized non-textual anomaly detection set using multiple local nodes, generating multiple non-textual anomaly detection models. Finally, federated aggregation training is performed on these multiple models to generate the typography non-textual anomaly detection channel.

[0055] The multiple non-text feature comparison vectors are traversed, and a single comparison vector is extracted as the first non-text feature comparison vector. The first non-text feature comparison vector is then evaluated for discrepancies, and a first non-text detection excitation coefficient is calculated. This first excitation coefficient dynamically adjusts the sensitivity of the non-text anomaly detection channel, making it more responsive to elements with larger deviations. Based on the first non-text detection excitation coefficient, the layout non-text anomaly detection channel is driven to perform multimodal anomaly detection on the first non-text feature comparison vector, including morphological anomalies such as image deformation, table misalignment, boundary offset, size anomalies, and texture anomalies. After detection, a first non-text layout anomaly detection result is generated and added to a second layout anomaly detection map, forming a comprehensive description of anomalies in the entire non-text layout area, improving the comprehensiveness and accuracy of print layout detection.

[0056] Based on the first and second layout anomaly detection maps, global layout coordination detection and optimization are performed to obtain the third layout anomaly detection map.

[0057] Furthermore, a global layout coordination detection and optimization is performed based on the first and second layout anomaly detection maps to obtain a third layout anomaly detection map. This includes: performing layout coordination detection based on the first layout element feature group to determine the first layout coordination detection result; performing layout coordination detection based on the second layout element feature group to obtain the second layout coordination detection result; performing overall coordination detection based on the first and second layout element feature groups to obtain the third layout coordination detection result; generating a global layout coordination detection map based on the first, second, and third layout coordination detection results; and performing a global review based on the global layout coordination detection map, the first layout anomaly detection map, and the second layout anomaly detection map to generate the third layout anomaly detection map.

[0058] Specifically, by calculating the line spacing vector, column baseline, and paragraph boundary matrix, and comparing the first set of layout element feature groups with the baseline layout parameters of the document to be printed, layout matching detection is performed on the first set of layout element feature groups. This checks whether the text conforms to specifications in terms of line spacing, column alignment, and paragraph boundaries, generating the first set of layout matching detection results. Layout matching detection is then performed based on the second set of layout element feature groups. By performing morphological analysis on non-text elements and combining positional constraints and size ratio rules, the system detects whether tables have row and column misalignment, whether images have geometric deformation or scaling distortion, and whether lines have offset or breakage, obtaining the second set of layout matching detection results. The first and second sets of layout element feature groups are used for overall coordination detection. Spatial constraint analysis is used to check whether the positional relationship between text and non-text elements in the page layout conforms to the set spatial boundaries. Adjacent area relationship analysis is used to determine whether the spacing, wrapping, or embedding relationships between text and images / tables are reasonable. Boundary consistency comparison is used to detect whether there are overlapping, white space, or occlusion anomalies in element boundaries, resulting in a third set of layout coordination detection results. These results reflect whether there are layout anomalies such as misalignment, occlusion, or improper white space between text and non-text elements in the overall page. Visualization techniques, such as image heatmaps or vector-annotated anomaly distribution maps, are used to map the first, second, and third set of layout coordination detection results onto the page space, generating a global layout coordination detection map.

[0059] Finally, the global layout coordination detection map, the first layout anomaly detection map, and the second layout anomaly detection map are comprehensively reviewed. By cross-comparison, anomalies with duplicate annotations in different maps are eliminated. Conflict resolution algorithms are used to resolve contradictions between detection results. For example, when the text detection map and the non-text detection map give different conclusions in the same area, the final result is determined by weight fusion. The global layout coordination detection map, the first layout anomaly detection map, and the second layout anomaly detection map are fused in a multimodal manner to generate a third layout anomaly detection map. The third layout anomaly detection map not only shows the local anomaly detection results, but also fully reflects the coordination and consistency of the local and overall layout in terms of spatial layout.

[0060] By conducting multi-level joint detection of text and non-text elements, not only can single-element anomalies in printed layout images be detected, but also coordination problems in the overall layout structure can be reflected, thereby significantly improving the completeness and accuracy of printed layout anomaly detection and enhancing the reliability and accuracy of the detection results.

[0061] The printed layout image is subjected to multimodal correction processing based on the third layout anomaly detection map.

[0062] Specifically, the abnormal information in the third typesetting anomaly detection map is decomposed into categories, including text-related anomalies, non-text-related anomalies, and overall coordination anomalies. Text-related anomalies include, but are not limited to, text offset, uneven line spacing, and missing strokes. Non-text-related anomalies include, but are not limited to, image distortion, table misalignment, and line offset. Overall coordination anomalies include, but are not limited to, layer misalignment, occlusion, and improper white space. A corresponding correction strategy is selected based on the anomaly type, and multimodal correction processing is performed on the printed typesetting image. Text-related anomalies are corrected based on a geometric alignment algorithm using baseline features, i.e., character bounding box matching and morphological constraints are used to restore normal typesetting. Non-text-related anomalies are optimized through image resampling, geometric correction, and boundary smoothing. For example, affine transformations are performed on distorted images to restore the original proportions, and row and column alignment is performed on misaligned tables. Overall coordination anomalies are corrected through multi-element joint correction, i.e., spatial constraints and neighborhood relationship optimization algorithms are used to globally rearrange the relative positional relationships between text and non-text elements. By performing multimodal correction processing on the printed layout image, the overall layout coordination can be maintained while ensuring the accuracy of local details, thereby improving the high quality and accuracy of the printed layout image.

[0063] Example 2 is based on the same inventive concept as the image recognition-based printing layout detection method in the foregoing examples, such as... Figure 2 As shown, this application provides a printing layout detection system based on image recognition, wherein the system includes: Document scanning module 11 is used to scan the document to be printed using the scanning components of the printing press to obtain a printed scanned image, and to perform multi-dimensional interference correction on the printed scanned image to obtain a printed layout image; image detection module 12 is used to perform Shearlet transform feature detection on the printed layout image to establish a first layout element feature group and a second layout element feature group; first anomaly map establishment module 13 is used to perform excitation-driven anomaly detection on the first layout element feature group based on the document to be printed according to an imbalance detection compensation mechanism to establish a first layout anomaly detection map; second anomaly map establishment module 14 is used to perform excitation-driven anomaly detection on the second layout element feature group based on the document to be printed according to the imbalance detection compensation mechanism to obtain a second layout anomaly detection map; map detection optimization module 15 is used to perform global layout coordination detection optimization based on the first layout anomaly detection map and the second layout anomaly detection map to obtain a third layout anomaly detection map; image correction module 16 is used to perform multimodal correction processing on the printed layout image based on the third layout anomaly detection map.

[0064] Furthermore, the document scanning module 11 in the image recognition-based printing layout detection system is also used to: synchronously retrieve the scanning environment information and scanning component status information corresponding to the printed scanned image; perform interference detection on the printed scanned image according to the scanning environment information to obtain a first interference detection result; perform interference detection on the printed scanned image according to the scanning component status information to obtain a second interference detection result; and perform adaptive correction processing on the printed scanned image according to the first interference detection result and the second interference detection result to generate the printed layout image.

[0065] Furthermore, the image detection module 12 in the image recognition-based printing layout detection system is also used to: perform multi-scale Shearlet transform on the printed layout image to extract a first-scale image feature set and a second-scale image feature set; classify layout elements according to the first-scale image feature set to obtain a text element feature set and a non-text element feature set; perform edge detail smoothing processing on the text element feature set according to the second-scale image feature set to obtain the first layout element feature group; and perform edge detail smoothing processing on the non-text element feature set according to the second-scale image feature set to obtain the second layout element feature group.

[0066] Furthermore, the first anomaly map establishment module 13 in the image recognition-based printing layout detection system is also used for: capturing text features based on the document to be printed to obtain a text baseline feature group; performing alignment processing on the text baseline feature group and the first layout element feature group to construct multiple text feature comparison vectors; performing imbalance detection compensation learning on the layout text anomaly detection record set according to the imbalance detection compensation mechanism to construct a layout text anomaly detection channel; traversing the multiple text feature comparison vectors, extracting a first text feature comparison vector, and performing a separation evaluation on the first text feature comparison vector to determine a first text detection excitation coefficient; driving the layout text anomaly detection channel to perform multimodal anomaly detection on the first text feature comparison vector according to the first text detection excitation coefficient, obtaining a first text layout anomaly detection result, and adding the first text layout anomaly detection result to the first layout anomaly detection map.

[0067] Furthermore, the first anomaly map establishment module 13 in the image recognition-based printing layout detection system is also used for: classifying the layout text anomaly detection record set according to multiple layout text defect indicators to obtain multiple text defect detection record clusters; performing sample balance evaluation on the multiple text defect detection record clusters to obtain a sample balance coefficient; if the sample balance coefficient does not meet the sample balance constraint, tracing the imbalance cause of the multiple text defect detection record clusters to determine the imbalance tracing factor; injecting adversarial examples into the multiple text defect detection record clusters according to the imbalance tracing factor to obtain a text anomaly detection optimization set that meets the sample balance constraint; performing supervised training on the text anomaly detection optimization set according to multiple local nodes to generate multiple text anomaly detection models; and performing federated aggregation training on the multiple text anomaly detection models to generate the layout text anomaly detection channel.

[0068] Furthermore, the second anomaly map establishment module 14 in the image recognition-based printing layout detection system is also used for: capturing non-text features based on the document to be printed, obtaining a non-text baseline feature group, aligning the non-text baseline feature group and the second layout element feature group to construct multiple non-text feature comparison vectors; performing imbalanced detection compensation learning on the layout non-text anomaly detection record set through the imbalanced detection compensation mechanism to build a layout non-text anomaly detection channel; traversing the multiple non-text feature comparison vectors, extracting a non-text feature first comparison vector, and synchronously generating a non-text detection first excitation coefficient; based on the non-text detection first excitation coefficient, driving the layout non-text anomaly detection channel to perform multimodal anomaly detection on the non-text feature first comparison vector, obtaining a non-text layout anomaly first detection result, and adding the non-text layout anomaly first detection result to the second layout anomaly detection map.

[0069] Furthermore, the image recognition-based printing layout detection system's image detection optimization module 15 is also used for: performing layout coordination detection based on the first layout element feature group to determine the first layout coordination detection result; performing layout coordination detection based on the second layout element feature group to obtain the second layout coordination detection result; performing overall coordination detection based on the first and second layout element feature groups to obtain the third layout coordination detection result; generating a global layout coordination detection image based on the first, second, and third layout coordination detection results; and performing a global analysis based on the global layout coordination detection image, the first layout anomaly detection image, and the second layout anomaly detection image to generate the third layout anomaly detection image.

[0070] Furthermore, the document scanning module 11 in the image recognition-based printing layout detection system is also used to: obtain the file format parameters of the document to be printed; perform compatibility detection on the file format parameters according to the file format constraint information of the printing press, and obtain the format compatibility detection result; and generate a file format warning signal according to the format compatibility detection result.

[0071] Furthermore, the document scanning module 11 in the image recognition-based printing layout detection system is also used to: generate format optimization reminder information based on the document format warning signal; and perform format optimization processing on the document to be printed based on the format optimization reminder information.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The image recognition-based printing layout detection method and specific examples in Embodiment 1 are also applicable to the image recognition-based printing layout detection system in this embodiment. Through the foregoing detailed description of the image recognition-based printing layout detection method, those skilled in the art can clearly understand the image recognition-based printing layout detection system in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A printing layout detection method based on image recognition, characterized in that, The method includes: The printing press scans the document to be printed using its scanning components to obtain a printed scan image. Then, multidimensional interference correction is performed on the printed scan image to obtain a printed layout image. Shearlet transform feature detection is performed on the printed layout image to establish a first layout element feature group and a second layout element feature group. Based on the document to be printed, the first typesetting element feature group is stimulated to drive anomaly detection according to the imbalance detection compensation mechanism, and a first typesetting anomaly detection map is established. The imbalance detection compensation mechanism refers to a detection model training mechanism that uses historical anomaly detection record sets for sample balance evaluation, imbalance cause tracing, adversarial sample injection, and federated aggregation training. Based on the document to be printed, the second typesetting element feature group is excited and driven to detect anomalies according to the imbalance detection compensation mechanism to obtain the second typesetting anomaly detection map. Based on the first and second typesetting anomaly detection maps, global typesetting standardization detection and optimization are performed to obtain the third typesetting anomaly detection map. The printed layout image is subjected to multimodal correction processing based on the third layout anomaly detection map; The multimodal correction processing refers to a comprehensive correction process that employs corresponding correction strategies for text-related anomalies, non-text-related anomalies, and overall coordination anomalies. Based on the document to be printed, the first typesetting element feature group is subjected to excitation-driven anomaly detection according to the imbalance detection compensation mechanism to establish a first typesetting anomaly detection map, including: Based on the document to be printed, text features are captured to obtain a set of text baseline features; Alignment processing is performed based on the text reference feature group and the first typesetting element feature group to construct multiple text feature comparison vectors; Based on the aforementioned imbalance detection compensation mechanism, imbalance detection compensation learning is performed on the set of abnormal text detection records to construct an abnormal text detection channel. Traverse the multiple text feature comparison vectors, extract the first text feature comparison vector, perform a separation evaluation on the first text feature comparison vector, and determine the first text detection excitation coefficient; The evaluation of the discrepancy includes positional deviation analysis, morphological deviation analysis, stroke integrity assessment, and texture feature comparison; Based on the first text detection excitation coefficient, the text layout anomaly detection channel is driven to perform multimodal anomaly detection on the first text feature comparison vector to obtain the first text layout anomaly detection result, and the first text layout anomaly detection result is added to the first layout anomaly detection map. The multimodal anomaly detection includes multi-dimensional detection of morphological anomalies, boundary offsets, size anomalies, and texture anomalies.

2. The printing layout detection method based on image recognition as described in claim 1, characterized in that, Perform multidimensional interference correction on the printed scanned image to obtain a printed layout image, including: Simultaneously retrieve the scanning environment information and scanning component status information corresponding to the printed scanned image; Interference detection is performed on the printed scanned image based on the scanning environment information to obtain a first interference detection result; Based on the scanning component status information, interference detection is performed on the printed scanned image to obtain a second interference detection result; The printed scan image is adaptively corrected based on the first interference detection result and the second interference detection result to generate the printed layout image.

3. The printing layout detection method based on image recognition as described in claim 1, characterized in that, Based on the printed layout image, Shearlet transform feature detection is performed to establish a first layout element feature group and a second layout element feature group, including: Perform multi-scale Shearlet transform on the printed layout image to extract the first-scale image feature set and the second-scale image feature set; Based on the first scale image feature set, the layout elements are classified to obtain the text element feature set and the non-text element feature set; The text element feature set is smoothed for edge details based on the second-scale image feature set to obtain the first typesetting element feature group; The edge detail smoothing process is performed on the non-text element feature set based on the second scale image feature set to obtain the second typesetting element feature group.

4. The printing layout detection method based on image recognition as described in claim 1, characterized in that, Based on the aforementioned imbalanced detection compensation mechanism, an imbalanced detection compensation learning process is performed on the set of text anomaly detection records to construct a text anomaly detection channel, including: The set of text anomaly detection records is classified according to multiple text defect indicators to obtain multiple text defect detection record clusters; Based on the multiple text defect detection record clusters, a sample balance evaluation is performed to obtain the sample balance coefficient; If the sample balance coefficient does not meet the sample balance constraint, trace the imbalance cause of the multiple text defect detection record clusters to determine the imbalance tracing factor. Adversarial sample injection is performed on the multiple text defect detection record clusters according to the imbalance tracing factor to obtain an optimized set of text anomaly detection that satisfies the sample balance constraint; Multiple text anomaly detection models are generated by supervising the text anomaly detection optimization set using multiple local nodes respectively. The layout text anomaly detection channel is generated by performing federated aggregation training based on the multiple text anomaly detection models.

5. The printing layout detection method based on image recognition as described in claim 1, characterized in that, Based on the document to be printed, the second typesetting element feature group is subjected to excitation-driven anomaly detection according to the imbalance detection compensation mechanism to obtain a second typesetting anomaly detection map, including: Non-text features are captured based on the document to be printed to obtain a non-text baseline feature group. The non-text baseline feature group and the second typesetting element feature group are aligned to construct multiple non-text feature comparison vectors. The imbalanced detection compensation mechanism is used to learn the imbalanced detection compensation record set of non-textual anomalies in typesetting, thereby building a non-textual anomaly detection channel in typesetting. Traverse the multiple non-text feature comparison vectors, extract the first non-text feature comparison vector, and simultaneously generate the first non-text detection activation coefficient; Based on the non-text detection first excitation coefficient, the non-text anomaly detection channel is driven to perform multimodal anomaly detection on the non-text feature first comparison vector to obtain the non-text layout anomaly first detection result, and the non-text layout anomaly first detection result is added to the second layout anomaly detection map.

6. The printing layout detection method based on image recognition as described in claim 1, characterized in that, Based on the first and second layout anomaly detection maps, a global layout standardization detection and optimization is performed to obtain a third layout anomaly detection map, including: The layout standardization is checked based on the first layout element feature group, and the result of the first layout standardization check is determined. The typesetting standardization is checked based on the second typesetting element feature group to obtain the second typesetting standardization check result; The first layout element feature group and the second layout element feature group are used for overall coordination detection to obtain the third layout standardization detection result; Based on the first typesetting standardization test result, the second typesetting standardization test result, and the third typesetting standardization test result, a global typesetting standardization test map is generated; The third typesetting anomaly detection map is generated by performing a global review based on the global typesetting standardization detection map, the first typesetting anomaly detection map, and the second typesetting anomaly detection map.

7. The printing layout detection method based on image recognition as described in claim 1, characterized in that, The document to be printed is scanned according to the scanning components of the printing press, including: Obtain the file format parameters of the document to be printed; Based on the file format constraint information of the printing press, a compatibility test is performed on the file format parameters to obtain the format compatibility test result; Based on the format compatibility test results, a file format warning signal is generated.

8. The printing layout detection method based on image recognition as described in claim 7, characterized in that, Generate file format warning signals, including: Based on the file format warning signal, generate a format optimization reminder message; The format of the document to be printed is optimized according to the format optimization reminder information.

9. A printing layout detection system based on image recognition, characterized in that, The steps for implementing the image recognition-based print layout detection method according to any one of claims 1 to 8 include: The document scanning module is used to scan the document to be printed according to the scanning component of the printing press, obtain the printed scan image, and perform multidimensional interference correction on the printed scan image to obtain the printed layout image. The image detection module is used to perform Shearlet transform feature detection based on the printed layout image and establish a first layout element feature group and a second layout element feature group. The first anomaly map establishment module is used to establish a first typesetting anomaly detection map based on the file to be printed, according to the imbalance detection compensation mechanism, to stimulate and drive anomaly detection of the first typesetting element feature group. The second anomaly map establishment module is used to perform excitation-driven anomaly detection on the second typesetting element feature group based on the file to be printed and according to the imbalance detection compensation mechanism to obtain the second typesetting anomaly detection map. The graph detection and optimization module is used to perform global layout standardization detection and optimization based on the first layout anomaly detection graph and the second layout anomaly detection graph to obtain a third layout anomaly detection graph. The image correction module is used to perform multimodal correction processing on the printed layout image based on the third layout anomaly detection map.

Images