A printed matter detection method and system based on pattern recognition

Abstract

The application discloses a printed matter detection method and system based on pattern recognition, comprising the following steps: collecting printed matter image data to generate standardized printed matter images; performing pattern recognition processing on the printed matter from multiple independent recognition perspectives which do not share intermediate results to generate pattern recognition results of each recognition perspective; generating a discrete decision state under each recognition perspective; collecting the decision states generated under each recognition perspective to form a decision state combination; performing consistency verification processing to determine whether the decision state combination belongs to a pre-defined co-existing decision state combination; generating an abnormal detection result when the decision state combination does not belong to the co-existing decision state combination; and generating a normal detection result when the decision state combination belongs to the co-existing decision state combination. The application adopts a multi-recognition perspective consistency verification pattern recognition method to realize printed matter structural abnormality detection, and has the advantages of no need for templates, strong interpretability and high adaptability.

Description

Technical Field

[0001] This invention relates to the field of printed matter inspection, and more particularly to a printed matter inspection method and system based on pattern recognition. Background Technology

[0002] Existing printed matter inspection technologies mainly rely on image processing, template matching, or pattern recognition methods based on training samples to detect and compare text, images, layout, or color features in printed matter. These methods typically start from a single or a few feature dimensions and output detection results through threshold judgment, rule verification, or learning models. They are suitable for the identification of obvious defects or known anomalies and have been applied to a certain extent in practical applications.

[0003] Existing technologies generally lack the ability to model the inherent relationships between various structural patterns in printed materials. They are unable to effectively determine whether there are structural contradictions between layout structure, graphic space organization, and color organization without relying on preset defect templates or a large number of training samples. When each individual feature is within a reasonable range but their combination relationship is unreasonable, existing methods often cannot identify such latent anomalies, resulting in limited accuracy and adaptability of detection results, making it difficult to meet the needs of complex printed material detection scenarios. Summary of the Invention

[0004] One objective of this invention is to propose a pattern recognition-based method and system for detecting printed materials. This invention employs a pattern recognition method with multi-view consistency verification to detect structural anomalies in printed materials, and has the advantages of being template-free, highly interpretable, and highly adaptable.

[0005] A method for detecting printed matter based on pattern recognition according to an embodiment of the present invention includes the following steps: Collect printed image data of the printed matter to be tested, and preprocess it to generate standardized printed images; Using standardized printed images as a unified input, pattern recognition processing is performed on the printed materials from multiple independent recognition perspectives that do not share intermediate results, generating pattern recognition results corresponding to each recognition perspective, including layout structure recognition perspective, text and image space organization recognition perspective, and color organization method recognition perspective. Based on the pattern recognition results corresponding to each recognition perspective, a discrete judgment state is generated under each recognition perspective; Collect the judgment states generated from the same printed material under various recognition perspectives to form the judgment state combination corresponding to that printed material; Perform consistency verification on the decision state combination to determine whether the decision state combination belongs to a predefined coexisting decision state combination. When the judgment state combination does not belong to the coexisting judgment state combination, the printed product is judged to have a judgment conflict anomaly, and an anomaly detection result is generated. When the judgment state combination belongs to the coexisting judgment state combination, the printed matter is judged to be in a normal state, and a normal detection result is generated.

[0006] Optionally, the preprocessing specifically includes noise reduction processing, brightness normalization processing, contrast normalization processing, geometric correction processing, page positioning processing, and effective page area cropping processing.

[0007] Optionally, the generation of the pattern recognition result specifically includes: From the perspective of layout structure recognition, based on standardized printed images, page partitioning and image block localization processes are performed to analyze the spatial distribution of each page area and the arrangement of image blocks in the printed material, and generate layout structure pattern recognition results. From the perspective of image and text spatial organization recognition, character region extraction and character row and column structure parsing are performed on standardized printed images to analyze the spatial organization state between character regions and the relative positional relationship between images and text, and generate image and text spatial organization pattern recognition results. From the perspective of color organization pattern recognition, color channel analysis and regional color distribution analysis are performed on standardized printed images to analyze the color hierarchy distribution and regional color correspondence of each region in the printed material, and generate color organization pattern recognition results. In the process of pattern recognition processing from the perspectives of layout structure recognition, text-image space organization recognition, and color organization recognition, an independent processing flow is established for each recognition perspective. Each processing flow is based solely on standardized printed images and does not accept intermediate results generated by other recognition perspectives. In the independent processing flow corresponding to each recognition perspective, the results of layout structure pattern recognition, text and image space organization pattern recognition, and color organization pattern recognition are generated respectively. After completing the pattern recognition processing for each recognition perspective, the results of layout structure pattern recognition, text-image space organization pattern recognition, and color organization pattern recognition are output respectively.

[0008] Optionally, the generation of the discrete decision states specifically includes: Read the layout structure pattern recognition results, image and text space organization pattern recognition results, and color organization pattern recognition results as input data for generating judgment states from each recognition perspective; Based on the layout structure pattern recognition results, extract the layout partition structure information, graphic block arrangement information and layout consistency information, combine and compare them to generate the judgment status of the layout structure recognition perspective. Based on the results of graphic and text spatial organization pattern recognition, character arrangement information, row and column structure information, and graphic and text spatial organization relationships are extracted, correlation analysis is performed, and a judgment state from the perspective of graphic and text spatial organization recognition is generated. The color hierarchy distribution information, regional color correspondence information, and color consistency information are extracted from the color organization pattern recognition results, and a comprehensive evaluation is performed to generate a judgment status from the perspective of color organization pattern recognition. The judgment states generated by each recognition perspective are mapped to judgment states in discrete value form, with each recognition perspective generating only one discrete judgment state.

[0009] Optionally, the generation of the determination state combination specifically includes: The determination status of the layout structure recognition perspective, the determination status of the graphic space organization recognition perspective, and the determination status of the color organization method recognition perspective are obtained as the basic data for forming the combination of determination status. Fixed perspective identifiers are assigned to the layout structure recognition perspective, the graphic space organization recognition perspective, and the color organization method recognition perspective, respectively, and the collection order of judgment states is determined based on the perspective identifiers. According to the order of the judgment states, the judgment states of the layout structure recognition perspective, the judgment states of the graphic space organization recognition perspective, and the judgment states of the color organization method recognition perspective are written into the combination carrier in sequence to generate a judgment state sequence. Based on the decision state sequence, a combination encapsulation process is performed on the decision state sequence to form a decision state combination that characterizes the decision results of the same printed material under various recognition perspectives. The judgment state combination is bound to the corresponding printed matter identifier to generate the judgment state combination corresponding to the printed matter.

[0010] Optionally, the consistency verification process specifically includes: The overall structure of the decision state combination is analyzed to generate a combination structure representation; Based on the combination structure representation, a unique combination identifier is generated for the combination of judgment states; Based on the value range of the judgment state from each recognition perspective, a set of coexisting judgment state combinations is constructed. Using the combined identifier as the matching object, perform combination matching processing in the set of coexisting decision state combinations to generate matching results; Based on the matching results, a consistency verification result is generated to determine whether the combination of judgment states belongs to the coexisting judgment state combinations.

[0011] Optionally, the generation of the anomaly detection result specifically includes: The consistency verification results are parsed to determine the attribution, and the attribution results of the determination state combination are generated. When the attribution determination result indicates that the combination of determination states does not belong to the combination of coexisting determination states, the combination of determination states is used as the input for anomaly analysis to generate an anomaly determination input object. Based on the anomaly detection input object, the detection state corresponding to each detection perspective in the detection state combination is structurally expanded to generate a set of detection perspective detection states, including the detection state of layout structure detection perspective, the detection state of graphic space organization detection perspective, and the detection state of color organization method detection perspective. Based on the recognition perspective, the set of judgment states is determined, and the combination of judgment state values ​​that causes the combination of judgment states to not belong to the combination of coexisting judgment states is located, and evidence of judgment conflict is generated. Based on the evidence of conflict of judgment, the anomaly type of the printed matter is identified as a conflict of judgment anomaly, and anomaly detection results are generated, including the printed matter identifier, the anomaly type, and the evidence of conflict of judgment.

[0012] Optionally, the generation of the normal detection result specifically includes: The consistency verification results are analyzed to determine the attribution of the judgment state combination to the coexisting judgment state combination. When the attribution result characterizes the combination of judgment states as a combination of coexisting judgment states, the combination of judgment states is used as the normal judgment input to generate a normal judgment input object. Based on the normal judgment input object, the judgment state combination is structurally expanded under each recognition perspective to generate a set of judgment states for recognition perspectives. Based on the set of states determined by the recognition perspective, a coexistence state record representing the consistency of the state combination under each recognition perspective is generated. Based on the coexistence status record, the status of the printed matter is determined to be normal, and a normal inspection result is generated, including the printed matter identification and the normal status identification.

[0013] A pattern recognition-based printed matter inspection system according to an embodiment of the present invention includes: The image acquisition module is used to acquire printed image data of the printed matter to be tested. The preprocessing module is used to preprocess printed image data to generate standardized printed images; The multi-view pattern recognition module is used to perform pattern recognition processing from multiple viewpoints in parallel based on standardized printed images, and generate pattern recognition results corresponding to each viewpoint. The determination state generation module is used to generate a discrete determination state for each recognition view based on the pattern recognition results corresponding to each recognition view. The determination state combination module is used to collect the determination states generated by the same printed material under various recognition perspectives to form the determination state combination corresponding to the printed material. The consistency verification module is used to perform consistency verification on the combination of judgment states to determine whether the combination of judgment states is a coexisting combination of judgment states. The detection result generation module is used to generate a conflict-type abnormal detection result when the judgment state combination does not belong to the coexisting judgment state combination, and to generate a normal detection result when the judgment state combination belongs to the coexisting judgment state combination.

[0014] The beneficial effects of this invention are: This invention, by constructing a pattern recognition-based printed matter detection method, breaks through the dependence of existing printed matter detection technologies on defect templates and training samples at the overall technical level. It standardizes and preprocesses printed matter images, and then introduces multiple independent recognition perspectives that do not share intermediate results. Parallel pattern recognition is performed on the same printed matter from multiple dimensions, including layout structure, text-image spatial organization, and color organization. This invention can comprehensively characterize various structural patterns of printed matter without pre-setting specific defect types. Furthermore, this invention uniformly maps the pattern recognition results of each recognition perspective into discrete judgment states, and abstractly expresses the overall structural state of the printed matter through combinations of judgment states. This provides a stable, unified, and scalable structured foundation for subsequent consistency verification and anomaly detection.

[0015] This invention utilizes a consistency verification mechanism to determine whether a combination of judgment states is a coexisting combination, effectively enabling the detection of latent structural anomalies in printed materials. When the judgment results from each individual identification perspective are within a reasonable range but their combination relationships are contradictory, this invention can accurately identify such conflicting anomalies and further generate traceable evidence of judgment conflicts, thereby significantly improving the interpretability and reliability of the detection results. When a combination of judgment states meets the coexistence condition, this invention also verifies the normal state through structural unfolding and coexistence state recording, avoiding the simplistic view of normal results as abnormal omissions. Thus, this invention not only improves the accuracy and adaptability of printed material detection but also significantly enhances the detection capability for complex layouts, diverse designs, and latent anomaly scenarios. It is applicable to various printed material detection application environments and has good engineering practice value and promotional significance. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a pattern recognition-based printed matter detection method and system proposed in this invention; Figure 2This is a schematic diagram of the multi-view parallel pattern recognition processing structure of a pattern recognition-based printed matter detection method and system proposed in this invention. Figure 3 This is a schematic diagram illustrating the generation of discrete decision states and the combination of decision states in a pattern recognition-based printed matter detection method and system proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A pattern recognition-based method for detecting printed materials includes the following steps: Collect printed image data of the printed matter to be tested, and preprocess it to generate standardized printed images; Using standardized printed images as a unified input, pattern recognition processing is performed on the printed materials from multiple independent recognition perspectives that do not share intermediate results, generating pattern recognition results corresponding to each recognition perspective, including layout structure recognition perspective, text and image space organization recognition perspective, and color organization method recognition perspective. Based on the pattern recognition results corresponding to each recognition perspective, a discrete judgment state is generated under each recognition perspective; Collect the judgment states generated from the same printed material under various recognition perspectives to form the judgment state combination corresponding to that printed material; Perform consistency verification on the decision state combination to determine whether the decision state combination belongs to a predefined coexisting decision state combination. When the judgment state combination does not belong to the coexisting judgment state combination, the printed product is judged to have a judgment conflict anomaly, and an anomaly detection result is generated. When the judgment state combination belongs to the coexisting judgment state combination, the printed matter is judged to be in a normal state, and a normal detection result is generated.

[0019] In this embodiment, the preprocessing specifically includes noise reduction processing, brightness normalization processing, contrast normalization processing, geometric correction processing, page positioning processing, and effective page area cropping processing.

[0020] In this embodiment, the generation of the pattern recognition result specifically includes: From the perspective of layout structure recognition, based on standardized printed images, page partitioning and image block localization processes are performed to analyze the spatial distribution of each page area and the arrangement of image blocks in the printed material, and generate layout structure pattern recognition results. From the perspective of image and text spatial organization recognition, character region extraction and character row and column structure parsing are performed on standardized printed images to analyze the spatial organization state between character regions and the relative positional relationship between images and text, and generate image and text spatial organization pattern recognition results. From the perspective of color organization pattern recognition, color channel analysis and regional color distribution analysis are performed on standardized printed images to analyze the color hierarchy distribution and regional color correspondence of each region in the printed material, and generate color organization pattern recognition results. In the process of pattern recognition processing from the perspectives of layout structure recognition, text-image space organization recognition, and color organization recognition, an independent processing flow is established for each recognition perspective. Each processing flow is based solely on standardized printed images and does not accept intermediate results generated by other recognition perspectives. The establishment of an independent processing flow specifically includes: A separate image processing module is established for the layout structure recognition perspective. This module uses only standardized printed images to partition and locate the layout areas within the printed material, generating layout partitioning structure and text / image block arrangement information. This ensures that information extraction from this perspective is unaffected by other perspectives. Similarly, a separate image processing module is established for the text / image spatial organization recognition perspective. This module uses only standardized printed images to extract character regions, analyzes the row and column structure of these regions, and generates character arrangement information and text / image spatial relationship information. This ensures that processing from this perspective is unaffected by other perspectives. A separate image processing module is also established for the color organization recognition perspective. This module uses only standardized printed images to analyze the color hierarchy distribution and regional color distribution within the printed material, generating color hierarchy distribution information and regional color correspondences. This ensures that information extraction from this perspective is unaffected by other perspectives. Within each independent processing flow corresponding to a specific perspective, after each processing flow completes its independent feature extraction operation, the generated intermediate results are only used by subsequent steps within that perspective and are not shared or interacted with by intermediate results from other perspectives, guaranteeing the independence and accuracy of the information. In the independent processing flow corresponding to each recognition perspective, the results of layout structure pattern recognition, text and image space organization pattern recognition, and color organization pattern recognition are generated respectively. After completing the pattern recognition processing for each recognition perspective, the results of layout structure pattern recognition, text-image space organization pattern recognition, and color organization pattern recognition are output respectively.

[0021] In this embodiment, the generation of the discrete determination states specifically includes: Read the layout structure pattern recognition results, image and text space organization pattern recognition results, and color organization pattern recognition results as input data for generating judgment states from each recognition perspective; Based on the layout structure pattern recognition results, extract the layout partition structure information, graphic block arrangement information and layout consistency information, combine and compare them to generate the judgment status of the layout structure recognition perspective. Based on the results of graphic and text spatial organization pattern recognition, character arrangement information, row and column structure information, and graphic and text spatial organization relationships are extracted, correlation analysis is performed, and a judgment state from the perspective of graphic and text spatial organization recognition is generated. The color hierarchy distribution information, regional color correspondence information, and color consistency information are extracted from the color organization pattern recognition results, and a comprehensive evaluation is performed to generate a judgment status from the perspective of color organization pattern recognition. The judgment states generated by each recognition perspective are mapped to judgment states in discrete value form, with each recognition perspective generating only one discrete judgment state. The generation of the discrete decision states specifically includes: Based on the pattern recognition results under the corresponding recognition perspective, the state value domain for judgment under that recognition perspective is determined; the judgment-related information extracted from the pattern recognition results is mapped to the candidate state set in the state value domain; consistency constraint verification is performed on each candidate state in the candidate state set to exclude candidate states that do not meet the conditions for being valid under that recognition perspective; among the candidate states that meet the consistency constraint verification, the unique valid state value is determined as the judgment result of that recognition perspective; the unique valid state value is encoded into a discrete value form to generate the discrete judgment state corresponding to that recognition perspective; discretization processing is performed on each recognition perspective to ensure that only one discrete judgment state is generated for each recognition perspective.

[0022] In this embodiment, the generation of the determination state combination specifically includes: The determination status of the layout structure recognition perspective, the determination status of the graphic space organization recognition perspective, and the determination status of the color organization method recognition perspective are obtained as the basic data for forming the combination of determination status. Fixed perspective identifiers are assigned to the layout structure recognition perspective, the graphic space organization recognition perspective, and the color organization method recognition perspective, respectively, and the collection order of judgment states is determined based on the perspective identifiers. According to the order of the judgment states, the judgment states of the layout structure recognition perspective, the judgment states of the graphic space organization recognition perspective, and the judgment states of the color organization method recognition perspective are written into the combination carrier in sequence to generate a judgment state sequence. Based on the decision state sequence, a combination encapsulation process is performed on the decision state sequence to form a decision state combination that characterizes the decision results of the same printed material under various recognition perspectives. The generation of the determination state combination specifically includes: The decision state sequence is used as the input object for combination encapsulation, keeping the order of the recognition perspectives corresponding to each decision state in the decision state sequence unchanged; the decision state sequence is subjected to structured mapping processing, which maps the decision state sequence to a combination representation with a fixed field structure; in the combination representation, a perspective identifier association relationship is established for the decision state corresponding to each recognition perspective, forming a state combination structure with perspective semantics; the state combination structure with perspective semantics is subjected to overall encapsulation processing to generate a decision state combination for uniformly representing the decision results of multiple recognition perspectives; The judgment state combination is bound to the corresponding printed matter identifier to generate the judgment state combination corresponding to the printed matter.

[0023] In this embodiment, the consistency verification process specifically includes: The overall structure of the decision state combination is analyzed to generate a combination structure representation; Based on the combination structure representation, a unique combination identifier is generated for the combination of judgment states; Based on the value range of the judgment state from each recognition perspective, a set of coexisting judgment state combinations is constructed. The generation of the set of coexisting decision states specifically includes: Based on the value ranges of judgment states from the perspectives of layout structure recognition, text-image spatial organization recognition, and color organization recognition, the discrete value ranges of judgment states under each recognition perspective are determined, generating a set of judgment state value ranges. Using this set as a constraint, the judgment states from these perspectives are combined and enumerated to generate a set of candidate judgment state combinations. For each judgment state combination in the candidate set, based on the structural consistency and coexistence constraints between the judgment states from each recognition perspective, combinations that can be simultaneously valid are selected, generating a set of coexisting candidate combinations. Each judgment state combination in the coexisting candidate combination set is then standardized to unify the order and structural form of each judgment state in the combination, generating a standardized set of judgment state combinations. This standardized set of judgment state combinations is then determined as the set of coexisting judgment state combinations. Using the combined identifier as the matching object, perform combination matching processing in the set of coexisting decision state combinations to generate matching results; The generation of the matching results specifically includes: Read the combination identifier corresponding to the judgment state combination as the matching input identifier for the combination matching process; sequentially read the set combination identifier corresponding to each coexisting judgment state combination from the set of coexisting judgment state combinations to generate a sequence of matchable combination identifiers; perform identifier consistency comparison between the matching input identifier and each set combination identifier in the sequence of matchable combination identifiers to generate an identifier comparison result; when the matching input identifier matches any set combination identifier, record the corresponding coexisting judgment state combination and generate a matching success flag; after completing all comparisons of the sequence of matchable combination identifiers, output the matching result based on whether a matching success flag has been generated. Based on the matching results, a consistency verification result is generated to determine whether the combination of judgment states belongs to the coexisting judgment state combinations.

[0024] In this embodiment, the generation of the anomaly detection result specifically includes: The consistency verification results are parsed to determine the attribution, and the attribution results of the determination state combination are generated. When the attribution determination result indicates that the combination of determination states does not belong to the combination of coexisting determination states, the combination of determination states is used as the input for anomaly analysis to generate an anomaly determination input object. Based on the anomaly detection input object, the detection state corresponding to each detection perspective in the detection state combination is structurally expanded to generate a set of detection perspective detection states, including the detection state of layout structure detection perspective, the detection state of graphic space organization detection perspective, and the detection state of color organization method detection perspective. The generation of the recognition perspective determination state set specifically includes: Using the combination of judgment states contained in the anomaly judgment input object as the structure expansion object, the perspective affiliation of each judgment state in the judgment state combination is analyzed to generate a judgment state-recognition perspective correspondence. Based on the judgment state-recognition perspective correspondence, the judgment states in the judgment state combination are split according to the recognition perspective, and the judgment states of layout structure recognition perspective, graphic space organization recognition perspective, and color organization method recognition perspective are extracted respectively. The judgment states corresponding to each recognition perspective obtained by splitting are processed to be independent within the perspective, eliminating the constraint influence of the judgment state combination structure on the judgment state of a single recognition perspective, and generating perspective-independent judgment states. The perspective-independent judgment states corresponding to each recognition perspective are collected to form a recognition perspective judgment state set used to represent the judgment results of each recognition perspective under anomaly conditions. Based on the recognition perspective, the set of judgment states is determined, and the combination of judgment state values ​​that causes the combination of judgment states to not belong to the combination of coexisting judgment states is located, and evidence of judgment conflict is generated. The generation of conflicting evidence specifically includes: Taking the set of judgment states based on recognition perspectives as the analysis object, this paper extracts the judgment states from the perspectives of layout structure recognition, text-image space organization recognition, and color organization recognition, generating a judgment state analysis input set. Based on the judgment state analysis input set, the judgment states under different recognition perspectives are combined and expanded to generate judgment state value combinations composed of multiple recognition perspective judgment states. Each judgment state value combination is compared with the set of coexisting judgment state combinations to identify judgment state value combinations that do not satisfy the coexistence relationship, generating a conflict value combination set. Conflict attribution labeling is performed on each judgment state value combination in the conflict value combination set to determine the recognition perspectives involved in the conflict and the corresponding judgment state values, generating conflict location information. Based on the conflict location information, a structured conflict description is constructed to characterize the incompatible judgment relationships in the judgment state combinations, generating judgment conflict evidence. Based on the evidence of conflict of judgment, the anomaly type of the printed matter is identified as a conflict of judgment anomaly, and anomaly detection results are generated, including the printed matter identifier, the anomaly type, and the evidence of conflict of judgment.

[0025] In this embodiment, the generation of the normal detection result specifically includes: The consistency verification results are analyzed to determine the attribution of the judgment state combination to the coexisting judgment state combination. When the attribution result characterizes the combination of judgment states as a combination of coexisting judgment states, the combination of judgment states is used as the normal judgment input to generate a normal judgment input object. Based on the normal judgment input object, the judgment state combination is structurally expanded under each recognition perspective to generate a set of judgment states for recognition perspectives. Based on the set of states determined by the recognition perspective, a coexistence state record representing the consistency of the state combination under each recognition perspective is generated. The generation of the coexistence state record specifically includes: Taking the set of recognition perspective judgment states as input, the judgment states of layout structure recognition perspective, text-image space organization recognition perspective, and color organization method recognition perspective are aggregated in parallel to generate a multi-recognition perspective judgment state parallel set. Based on the multi-recognition perspective judgment state parallel set, the judgment states under each recognition perspective are compared for consistency to determine the combination of judgment states that do not conflict with each recognition perspective judgment state. The coexistence relationship confirmation process is performed on the judgment states of each recognition perspective in the judgment state combination to generate a coexistence relationship description that represents the simultaneous validity of the judgment states of each recognition perspective. The coexistence relationship description is associated with the corresponding judgment state combination to form a coexistence state record used to represent the consistent state of the judgment state combination under multiple recognition perspectives. Based on the coexistence status record, the status of the printed matter is determined to be normal, and a normal inspection result is generated, including the printed matter identification and the normal status identification.

[0026] A pattern recognition-based printed matter inspection system includes: The image acquisition module is used to acquire printed image data of the printed matter to be tested. The preprocessing module is used to preprocess printed image data to generate standardized printed images; The multi-view pattern recognition module is used to perform pattern recognition processing from multiple viewpoints in parallel based on standardized printed images, and generate pattern recognition results corresponding to each viewpoint. The determination state generation module is used to generate a discrete determination state for each recognition view based on the pattern recognition results corresponding to each recognition view. The determination state combination module is used to collect the determination states generated by the same printed material under various recognition perspectives to form the determination state combination corresponding to the printed material. The consistency verification module is used to perform consistency verification on the combination of judgment states to determine whether the combination of judgment states is a coexisting combination of judgment states. The detection result generation module is used to generate a conflict-type abnormal detection result when the judgment state combination does not belong to the coexisting judgment state combination, and to generate a normal detection result when the judgment state combination belongs to the coexisting judgment state combination.

[0027] Example 1: To verify the feasibility of the present invention in practice, it was applied to a printing production environment in a coastal city that specializes in commercial printing and publication printing. This production environment has long been responsible for producing promotional brochures, product manuals, and multi-format combined printed materials. The printed materials often contain multiple layouts, text arrangements of different densities, and diverse color organization methods in the same batch. Due to the high requirements of customers for printing quality, in addition to obvious omissions and misprints, there are also strict requirements for hidden problems such as inconsistent layout structure, abnormal spatial relationship between text and images, and color organization conflicts. These problems are often difficult to reliably identify through traditional detection methods in actual production.

[0028] In this application scenario, the printed materials are first photographed by an image acquisition device to obtain complete image data. The acquired image data is then sent to the detection system of this invention for processing. During system operation, a uniform preprocessing operation is first performed on the printed images to correct for differences in lighting, noise interference, and shooting angle deviations introduced during the shooting process. This ensures that the image data used in subsequent processing maintains consistency in brightness, contrast, and geometric structure, thereby guaranteeing the comparability of printed images acquired from different batches and at different times.

[0029] After preprocessing, standardized printed images are simultaneously fed into multiple independent recognition perspectives for pattern recognition processing. From the layout structure recognition perspective, the system analyzes the overall layout of the printed material, identifying the division of different layout areas and the arrangement of text and image blocks, thus obtaining pattern recognition results reflecting the overall layout structure characteristics. From the image space organization recognition perspective, the system focuses on the spatial distribution relationship between text and graphic areas, analyzing character arrangement, row and column structure, and the relative positions of text and images to generate pattern recognition results reflecting the image space organization characteristics. From the color organization recognition perspective, the system analyzes the color distribution of each area in the printed material, identifying the distribution state between different color levels and the color correspondence between areas, forming pattern recognition results reflecting the color organization method. These multiple recognition perspectives do not share intermediate results during processing, ensuring that their respective judgments are not interfered with by other perspectives.

[0030] After obtaining the pattern recognition results corresponding to each recognition perspective, the system further generates a discrete judgment state under each recognition perspective to abstractly express the structural state of the printed matter under that perspective. Subsequently, the system aggregates the judgment states generated for the same printed matter under multiple recognition perspectives to form a unified judgment state combination, which is used to characterize the overall state of the printed matter at the multi-dimensional structural level. Through this judgment state combination, the system no longer relies on a single feature to judge whether the printed matter is abnormal, but instead conducts a comprehensive analysis of the printed matter from the perspective of multi-view structural relationships.

[0031] In practical applications, the system performs consistency verification on the combination of judgment states to determine whether the combination belongs to a coexisting judgment state combination. When some printed materials appear normal from a single perspective, but there is an unreasonable combination relationship between their layout structure, graphic space organization, and color organization, the system can identify such structural conflicts through consistency verification, thereby determining that the printed materials have judgment conflict anomalies. In the actual operation of this production environment, the system successfully identified a variety of hidden anomalies that were previously difficult to detect stably by manual quality inspection and traditional detection methods. For example, printed materials where the text layout is inconsistent with the page layout logic but do not show obvious errors when viewed alone, or where the color organization is not coordinated with the graphic structure but does not show obvious color distortion.

[0032] When the system determines that the printed material has a conflict-type anomaly, it will further expand the structure of the judgment state combination to locate the specific identification perspective combination that caused the conflict and generate corresponding judgment conflict evidence. This evidence is presented in a structured form, which can clearly reflect the source of the anomaly and provide an intuitive basis for production personnel to adjust the layout design or printing process. When the judgment state combination meets the coexistence condition, the system will also expand and verify the judgment state under multiple identification perspectives and generate coexistence state records to confirm the consistency of the structural relationship of the printed material under each identification perspective, thereby determining that the printed material is in a normal state.

[0033] After running continuously in this application scenario for a period of time, relevant production personnel reported that by introducing the detection method of this invention, the overall stability and reliability of printed material quality inspection have been significantly improved. The system can effectively detect printed materials with complex layouts and diverse designs without relying on specific defect templates and a large number of historical samples, reducing reliance on human experience. At the same time, the structured presentation of the causes of anomalies during the detection process helps to shorten the problem location and processing cycle and reduce the probability of repeated rework. It can be seen that this invention has good feasibility in actual printing production environment, can effectively solve the problem that existing technologies are difficult to identify hidden structural anomalies in printed materials, and fully demonstrates the beneficial effects achieved by this invention.

[0034] Table 1. Comparison of the overall performance of different printing inspection methods

[0035] As can be seen from the comparison results shown in Table 1, the present invention demonstrates a stable, reasonable, and interpretable improvement trend in overall detection performance compared to traditional methods. In terms of comprehensive detection accuracy, the present invention achieves 94.6%, which is about 2.5 percentage points higher than template matching-based methods and about 1.6 percentage points higher than single feature learning-based methods. This improvement is within the range commonly seen in engineering practice, demonstrating the effectiveness of the present invention in improving the stability of detection results through a structured judgment method without introducing overly complex models.

[0036] In terms of the detection rate of latent structural anomalies, the advantages of this invention are more obvious. The data in the table show that the traditional methods achieve 71.2% and 74.5% respectively in this indicator, while the invention improves it to 82.3%. This result shows that when the anomaly is not manifested as a single feature anomaly, but rather as a combination of layout structure, text-image spatial organization, and color organization, the recognition ability of the traditional method is significantly limited. This invention, through the combination and consistency verification mechanism of discrete judgment states under multiple recognition perspectives, can effectively identify structural conflicts that are difficult to detect under a single perspective, thereby improving the detection ability of latent anomalies.

[0037] In terms of the consistency and stability of the judgment results, the present invention also shows a relatively stable improvement trend. Since traditional methods are more sensitive to changes in thresholds or model parameters, the detection results are prone to fluctuations under different printing batches or layout changes. The present invention maps the pattern recognition results of each recognition perspective into discrete judgment states and performs overall verification based on the combination of judgment states, thereby reducing the sensitivity of the detection results to changes in local features. Its stability index reaches 0.92, which has better continuity performance than the comparison method.

[0038] The positive effects of the present invention can be further verified by the abnormal false alarm rate index. The abnormal false alarm rate of the present invention is controlled at about 5.0%, which is significantly lower than that of traditional methods. This shows that the present invention improves the ability to detect anomalies without introducing additional risk of misjudgment. The reason is that the consistency verification process focuses on the coexistence relationship between judgment states, rather than amplifying the judgment of a single feature, thereby avoiding the misjudgment of reasonable design changes as anomalies.

[0039] In terms of detection efficiency, the average detection time per sheet of printed matter of this invention is basically close to that of the method based on single feature learning, and slightly higher than the template matching method, but the overall difference is small. This indicates that the parallel processing of multiple recognition perspectives and state combination analysis did not have a significant impact on the real-time performance of the system. In terms of adaptability to layout changes and the ability to identify abnormal color and structural combinations, this invention shows a relatively stable improvement, demonstrating its adaptability advantage in complex and diverse printed matter detection scenarios.

[0040] Overall, the revised comparative data shows that, while maintaining a basically stable detection efficiency, the present invention has achieved a robust improvement in the detection capability and stability of hidden structural anomalies in printed materials through a combination mechanism of multi-identification perspective consistency verification and judgment status, fully demonstrating the beneficial effects of the present invention in practical applications.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting printed matter based on pattern recognition, characterized in that, Includes the following steps: Collect printed image data of the printed matter to be tested, and preprocess it to generate standardized printed images; Using standardized printed images as a unified input, pattern recognition processing is performed on the printed materials from multiple independent recognition perspectives that do not share intermediate results, generating pattern recognition results corresponding to each recognition perspective, including layout structure recognition perspective, text and image space organization recognition perspective, and color organization method recognition perspective. Based on the pattern recognition results corresponding to each recognition perspective, a discrete judgment state is generated under each recognition perspective; Collect the judgment states generated from the same printed material under various recognition perspectives to form the judgment state combination corresponding to that printed material; Perform consistency verification on the decision state combination to determine whether the decision state combination belongs to a predefined coexisting decision state combination. When the judgment state combination does not belong to the coexisting judgment state combination, the printed product is judged to have a judgment conflict anomaly, and an anomaly detection result is generated. When the judgment state combination belongs to the coexisting judgment state combination, the printed matter is judged to be in a normal state, and a normal detection result is generated.

2. The method for detecting printed matter based on pattern recognition according to claim 1, characterized in that, The preprocessing specifically includes noise reduction, brightness normalization, contrast normalization, geometric correction, layout positioning, and effective layout area cropping.

3. The printed matter detection method based on pattern recognition according to claim 1, characterized in that, The generation of the pattern recognition result specifically includes: From the perspective of layout structure recognition, based on standardized printed images, page partitioning and image block localization processes are performed to analyze the spatial distribution of each page area and the arrangement of image blocks in the printed material, and generate layout structure pattern recognition results. From the perspective of image and text spatial organization recognition, character region extraction and character row and column structure parsing are performed on standardized printed images to analyze the spatial organization state between character regions and the relative positional relationship between images and text, and generate image and text spatial organization pattern recognition results. From the perspective of color organization pattern recognition, color channel analysis and regional color distribution analysis are performed on standardized printed images to analyze the color hierarchy distribution and regional color correspondence of each region in the printed material, and generate color organization pattern recognition results. In the process of pattern recognition processing from the perspectives of layout structure recognition, text-image space organization recognition, and color organization recognition, an independent processing flow is established for each recognition perspective. Each processing flow is based solely on standardized printed images and does not accept intermediate results generated by other recognition perspectives. In the independent processing flow corresponding to each recognition perspective, the results of layout structure pattern recognition, text and image space organization pattern recognition, and color organization pattern recognition are generated respectively. After completing the pattern recognition processing for each recognition perspective, the results of layout structure pattern recognition, text-image space organization pattern recognition, and color organization pattern recognition are output respectively.

4. The printed matter detection method based on pattern recognition according to claim 1, characterized in that, The generation of the discrete decision states specifically includes: Read the layout structure pattern recognition results, image and text space organization pattern recognition results, and color organization pattern recognition results as input data for generating judgment states from each recognition perspective; Based on the layout structure pattern recognition results, extract the layout partition structure information, graphic block arrangement information and layout consistency information, combine and compare them to generate the judgment status of the layout structure recognition perspective. Based on the results of graphic and text spatial organization pattern recognition, character arrangement information, row and column structure information, and graphic and text spatial organization relationships are extracted, correlation analysis is performed, and a judgment state from the perspective of graphic and text spatial organization recognition is generated. The color hierarchy distribution information, regional color correspondence information, and color consistency information are extracted from the color organization pattern recognition results, and a comprehensive evaluation is performed to generate a judgment status from the perspective of color organization pattern recognition. The judgment states generated from each recognition perspective are mapped to judgment states in discrete value form, with each recognition perspective generating only one discrete judgment state.

5. The printed matter detection method based on pattern recognition according to claim 1, characterized in that, The generation of the determination state combination specifically includes: The determination status of the layout structure recognition perspective, the determination status of the graphic space organization recognition perspective, and the determination status of the color organization method recognition perspective are obtained as the basic data for forming the combination of determination status. Fixed perspective identifiers are assigned to the layout structure recognition perspective, the graphic space organization recognition perspective, and the color organization method recognition perspective, respectively, and the collection order of judgment states is determined based on the perspective identifiers. According to the order of the judgment states, the judgment states of the layout structure recognition perspective, the judgment states of the graphic space organization recognition perspective, and the judgment states of the color organization method recognition perspective are written into the combination carrier in sequence to generate a judgment state sequence. Based on the decision state sequence, a combination encapsulation process is performed on the decision state sequence to form a decision state combination that characterizes the decision results of the same printed material under various recognition perspectives. The judgment state combination is bound to the corresponding printed matter identifier to generate the judgment state combination corresponding to the printed matter.

6. The printed matter detection method based on pattern recognition according to claim 1, characterized in that, The consistency verification process specifically includes: The overall structure of the decision state combination is analyzed to generate a combination structure representation; Based on the combination structure representation, a unique combination identifier is generated for the combination of judgment states; Based on the value range of the judgment state from each recognition perspective, a set of coexisting judgment state combinations is constructed. Using the combined identifier as the matching object, perform combination matching processing in the set of coexisting decision state combinations to generate matching results; Based on the matching results, a consistency verification result is generated to determine whether the combination of judgment states belongs to the coexisting judgment state combinations.

7. The method for detecting printed matter based on pattern recognition according to claim 1, characterized in that, The generation of the anomaly detection results specifically includes: The consistency verification results are parsed to determine the attribution, and the attribution results of the determination state combination are generated. When the attribution determination result indicates that the combination of determination states does not belong to the combination of coexisting determination states, the combination of determination states is used as the input for anomaly analysis to generate an anomaly determination input object. Based on the anomaly detection input object, the detection state corresponding to each detection perspective in the detection state combination is structurally expanded to generate a set of detection perspective detection states, including the detection state of layout structure detection perspective, the detection state of graphic space organization detection perspective, and the detection state of color organization method detection perspective. Based on the recognition perspective, the set of judgment states is determined, and the combination of judgment state values ​​that causes the combination of judgment states to not belong to the combination of coexisting judgment states is located, and evidence of judgment conflict is generated. Based on the evidence of conflict of judgment, the anomaly type of the printed matter is identified as a conflict of judgment anomaly, and anomaly detection results are generated, including the printed matter identifier, the anomaly type, and the evidence of conflict of judgment.

8. The printed matter detection method based on pattern recognition according to claim 1, characterized in that, The generation of the normal test results specifically includes: The consistency verification results are analyzed to determine the attribution of the judgment state combination to the coexisting judgment state combination. When the attribution result characterizes the combination of judgment states as a combination of coexisting judgment states, the combination of judgment states is used as the normal judgment input to generate a normal judgment input object. Based on the normal judgment input object, the judgment state combination is structurally expanded under each recognition perspective to generate a set of judgment states for recognition perspectives. Based on the set of states determined by the recognition perspective, a coexistence state record representing the consistency of the state combination under each recognition perspective is generated. Based on the coexistence status record, the status of the printed matter is determined to be normal, and a normal inspection result is generated, including the printed matter identification and the normal status identification.

9. A pattern recognition-based printed matter inspection system, comprising executing the pattern recognition-based printed matter inspection method according to any one of claims 1 to 8, characterized in that, include: The image acquisition module is used to acquire printed image data of the printed matter to be tested. The preprocessing module is used to preprocess printed image data to generate standardized printed images; The multi-view pattern recognition module is used to perform pattern recognition processing from multiple viewpoints in parallel based on standardized printed images, and generate pattern recognition results corresponding to each viewpoint. The determination state generation module is used to generate a discrete determination state for each recognition view based on the pattern recognition results corresponding to each recognition view. The determination state combination module is used to collect the determination states generated by the same printed material under various recognition perspectives to form the determination state combination corresponding to the printed material. The consistency verification module is used to perform consistency verification on the combination of judgment states to determine whether the combination of judgment states is a coexisting combination of judgment states. The detection result generation module is used to generate a conflict-type abnormal detection result when the judgment state combination does not belong to the coexisting judgment state combination, and to generate a normal detection result when the judgment state combination belongs to the coexisting judgment state combination.

Images