A printed matter printing content online monitoring and early warning method

Abstract

The application discloses a printed matter printing content online monitoring and early warning method, and relates to the field of printed matter printing content monitoring and early warning, and comprises the following steps: the character edge contour and skeleton feature are extracted, and the character distortion degree is calculated; the recognized character sequence is compared with a printing design original draft bit by bit, the candidate deviation position character is subjected to semantic analysis by using natural language processing, printing character deviation is recognized, and the recognized character is matched with a copyright content feature library; if there is copyright content, copyright early warning is triggered; an ink color detection area is positioned according to the gradient amplitude of a printed matter image pixel point, and ink color offset color difference is calculated; the character distortion degree, printing character deviation occurrence frequency and ink color offset color difference of the same batch of printed matters are continuously collected, the content quality trend of subsequent printed matters of the batch is predicted, and batch early warning is triggered when there is quality risk. The method can realize early warning and equipment adjustment guidance before the accumulation of printing quality defects exceeds the limit.

Description

Technical Field

[0001] This invention belongs to the field of printed matter content monitoring and early warning technology, and relates to an online monitoring and early warning method for printed matter content. Background Technology

[0002] In the field of text printing, the quality of printed materials is mainly reflected in the geometric shape of the characters, the accuracy of the text content, and the stability of the ink color, which together determine the readability of the printed materials. Secondly, the legality of the content of printed materials is also related to people's protection of their own intellectual property rights.

[0003] Currently, existing online monitoring solutions for printed materials have the following shortcomings: First, geometric distortion of characters in printing production originates from fluctuations in printing pressure. However, existing technologies for detecting printed character deformation use binary area comparison, treating characters as homogeneous wholes and ignoring the internal structural orientation of characters. This makes it difficult to identify local distortions. As production continues, the risk to the readability of printed materials accumulates, resulting in the entire batch of text becoming blurry.

[0004] Secondly, current technologies for recognizing printed text largely rely on standard text matching, comparing the matching results with the design draft in isolation. This fails to identify issues such as misaligned characters that are correctly shaped but semantically incoherent, making it difficult to effectively intercept the risk of content errors in printed materials where high accuracy is required. Furthermore, if the original design contains infringing information such as unauthorized trademarks, pirated text fragments, or sensitive words, leading to the continuous production of infringing products by printing equipment, it poses legal risks to the company.

[0005] Next, for the detection of ink color in printed text, existing technologies use the color values ​​of the entire image to detect ink color deviation. The results are easily affected by changes in image content, and it is difficult to distinguish whether the character deformation is caused by fluctuations in printing pressure or the color difference shift is caused by changes in ink volume, making it impossible to locate the root cause of the problem.

[0006] Finally, actual printing quality defects often accumulate gradually as production continues, and existing technologies mostly perform single-sheet inspections, which cannot identify batch printing quality risks in advance, leading to a decline in production efficiency. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background art, the present invention provides an online monitoring and early warning method for printed content.

[0008] The objective of this invention can be achieved through the following technical solution: an online monitoring and early warning method for printed content, comprising: S1, performing character segmentation on the acquired printed image, extracting the edge contours and skeleton features of the characters, and calculating the character distortion degree.

[0009] S2. Based on the segmented characters, perform text recognition, compare the recognized text with the text in the original printed design, and use natural language processing to perform semantic analysis on the recognized text to identify deviations in the printed text.

[0010] S3. Locate the ink color detection area based on the gradient amplitude of each pixel in the printed image, obtain the colorimetric feature value of the ink color detection area, calculate the colorimetric deviation between the printed product and the original printed design, and obtain the ink color offset color difference.

[0011] S4. Continuously collect character distortion, printing text deviation, and ink color shift of the same batch of printed materials to predict the content quality trend of subsequent printed materials in the batch. When there is a quality risk in the content quality trend of the printed materials, trigger a batch warning signal.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention extracts the character edge contour and skeleton features to calculate the character distortion degree, realizes the quantification of subtle deformations such as blurred character stroke edges and broken skeletons, effectively curbs the accumulation of text readability risks and avoids the entire batch of text being unclear.

[0013] (2) The present invention compares the identified text sequence with the original printed design one by one, performs semantic analysis on the candidate deviation position text using natural language processing, identifies the deviation of printed text, and then matches the identified text with the infringing content feature library. If there is infringing content, an infringement warning is triggered, thus realizing the interception of semantic errors such as similar-looking characters confusion and text replacement, improving the content reliability of printed materials with high accuracy requirements, and blocking the risk of batch infringement caused by the violation of the original manuscript from the source.

[0014] (3) The present invention locates the ink color detection area based on the gradient amplitude of each pixel in the printed image, calculates the ink color offset color difference in the ink color detection area, overcomes the problem that the existing technology of using the color value of the whole image is easily affected by changes in image content, realizes the accurate detection of the true ink color offset, and, combined with character distortion degree and text deviation, can accurately locate the root cause of the problem.

[0015] (4) This invention continuously collects the character distortion, number of times the printed text deviation occurs and the ink color shift color difference of the same batch of printed materials, predicts the content quality trend of subsequent printed materials in the batch, triggers batch warning when there is a quality risk, realizes the proactive prediction of the quality deterioration trend, and can intervene in advance before the defects accumulate to the limit, thereby improving production efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the implementation steps of the method of the present invention;

[0018] Figure 2 This is a flowchart of the ink color detection area positioning process of the present invention;

[0019] Figure 3 This is a flowchart of the batch warning signal triggering process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 As shown, the present invention provides an online monitoring and early warning method for printed content, including: S100, performing character segmentation on the acquired printed image and extracting the edge contours and skeleton features of the characters.

[0022] To ensure that the acquired printed images have complete coverage, eliminate the interference of ambient light changes on image acquisition, and guarantee printing clarity, the specific implementation steps for acquiring printed images are as follows: First, an industrial line scan camera is installed above the paper cutting unit of the printing press, and the scanning width of the camera covers the printed material. Next, an encoder is set on one side of the printing material conveyor roller, and the encoder is linked to the printing material conveying speed to trigger pulse signals to the camera in real time, ensuring that the camera can scan stably line by line during the high-speed movement of the printed material. At the same time, LED light sources are installed on both sides of the camera, and the illumination angle of the light sources is set to a 45-degree angle with the surface of the printed material to provide uniform illumination to the surface of the printed material.

[0023] Considering that strokes between adjacent characters may have overlapping projections or edge interference, directly extracting edge contours or skeleton features would lead to incorrect distortion judgments. Therefore, the character segmentation includes the following steps: converting the acquired color image of the printed material into a grayscale image and performing denoising processing; binarizing the grayscale image; performing horizontal projection analysis on the binary image; counting the cumulative number of character pixels in each row of pixels; generating a horizontal projection curve, where the peaks of the horizontal projection curve correspond to the positions of the text lines, and the troughs correspond to the blank areas between lines. Identifying the trough positions in the horizontal projection curve, using the troughs as line segmentation points, the binary image is divided into several independent text line images.

[0024] Vertical projection analysis is performed on each text line image to count the cumulative number of character pixels in each column, generating a vertical projection curve. The peaks of this curve correspond to the positions of character strokes, and the troughs correspond to the blank areas between characters. The troughs in the vertical projection curve are identified and used as segmentation points between characters to divide the text line image into individual images of the characters to be tested.

[0025] Each segmented character image to be tested is normalized in size, and all character images are uniformly scaled to the same height and width to eliminate comparison errors caused by differences in character size on printed materials.

[0026] Since the shape quality of printed characters is determined by both geometric accuracy and topological integrity, providing a data basis for subsequent calculation of character distortion, the extraction of character edge contours and skeleton features includes: performing binarization processing on the segmented character image to separate character pixels from background pixels.

[0027] Specifically, the segmented individual character images are converted into grayscale images; secondly, the inter-class variance value corresponding to each grayscale level is calculated, and the grayscale value corresponding to the maximum inter-class variance is used as the global segmentation threshold; finally, the grayscale image is scanned pixel by pixel. If the pixel grayscale value is greater than or equal to the global segmentation threshold, it is assigned a value of 255 and marked as a background pixel; otherwise, it is assigned a value of 0 and marked as a character pixel.

[0028] Scan each character pixel in the binarized image, identify the distribution of its eight neighboring pixels, and if there is at least one background pixel in the eight neighboring pixels, mark the pixel as an edge pixel and connect all edge pixels according to the connectivity relationship to form an edge contour line.

[0029] The position of edge pixels indicates the boundary of ink transfer. By judging the extracted edge pixels through the eight-neighborhood, the original geometric shape information of the character can be preserved: if all eight neighbors of a character pixel are character pixels, it means that the point is located inside the character and far away from the boundary between the character and the background; if at least one background pixel exists in the eight neighbors of a character pixel, it means that the point is located on the boundary between the character and the background, that is, one side of the point belongs to the character area and the other side belongs to the background area.

[0030] The binarized character image is thinned by peeling off edge pixels layer by layer until a character skeleton with a single pixel width is obtained.

[0031] Scan the eight neighborhoods of each pixel on the character skeleton. If the number of skeleton pixels in the neighborhood is 1, mark it as a skeleton endpoint. If the number is greater than or equal to 3, mark it as a skeleton intersection. Count the number of endpoints and intersections on the character skeleton as skeleton features.

[0032] By extracting the character skeleton, endpoints, and intersections, the connectivity and structural integrity of character strokes can be identified, and structural defects such as broken or stuck strokes caused by printing plate wear, printhead clogging, or abnormal ink volume can be detected. Skeleton endpoints correspond to the start and end positions of strokes; an increase in endpoints indicates broken strokes, while a decrease indicates stuck strokes. Intersections correspond to the points where strokes meet or fork; changes in their number reflect whether the stroke connection is abnormal, such as strokes that should be connected being broken, or strokes that should not be connected being stuck.

[0033] S101, Calculate character distortion degree.

[0034] By calculating character distortion, the morphological distortion of characters can be transformed into a measurable numerical index, providing a basis for predicting quality trends. The calculation of character distortion includes: calculating the variance of the Euclidean distance between each pixel on the edge contour line and the nearest point on the standard edge line of the corresponding position in the standard template image of the corresponding character in the original printed design, as an edge distortion index; realizing the quantitative measurement of the degree of character geometric distortion caused by pressure fluctuations or ink volume changes during the printing process.

[0035] The absolute values ​​of the differences between the number of endpoints and intersections obtained from the statistics and the skeleton feature values ​​of the corresponding characters in the standard character template of the original printed design are calculated. The sum of the absolute values ​​of the differences is used to obtain the topological distortion index. This index can quantify the structural deviation of characters in stroke connectivity and intersection, and transform topological defects such as broken strokes and adhesion into measurable distortion indices.

[0036] After normalizing the edge distortion index and the topological distortion index, the edge distortion index and the topological distortion index are linearly weighted and summed to obtain the character distortion degree.

[0037] It should be added that the weighting coefficients used in the linear weighted summation were pre-calibrated through regression analysis of historical printing defect samples. Specifically, sample images of the same batch or type of printed materials are collected, and the sample set should cover normal products and various defective products. The defective products include at least ghosting, blurring, and broken strokes. The edge distortion index and topological distortion index of each sample image are calculated as sample feature variables. Each sample is labeled with a quality level, with normal products marked as 0 and defective products marked as 1, as the sample category label.

[0038] Input the sample feature variables and their corresponding class labels into the logistic regression classifier, and extract the regression coefficients corresponding to the edge distortion index and the topological distortion index from the logistic regression analysis. After taking the absolute value of these two regression coefficients, normalize them so that the sum of the two is 1, and obtain the edge distortion weight α and the topological distortion weight β.

[0039] Preferably, the specific values ​​of edge distortion weight α and topological distortion weight β are determined as follows: For high-precision printed materials, such as anti-counterfeiting labels and precision instrument manuals, the geometric shape of the characters is subject to extremely high requirements, and can be set... ,For example , For printed materials where readability is paramount, such as textbooks and newspapers, the structural integrity of characters is required to be even higher, and settings can be configured accordingly. ,For example , For general printed materials, an equal weighting setting can be used, that is... .

[0040] S200: Based on the segmented characters, perform text recognition, compare the recognized text with the text in the original printed design, and use natural language processing to perform semantic analysis on the recognized text to identify deviations in the printed text.

[0041] The text recognition based on segmented characters includes: First, performing connected component analysis on the entire printed image, identifying and filtering non-text elements, selecting a set of connected components that conform to the text layout characteristics, defining the boundary coordinates of the text region according to its circumscribed rectangle, and extracting the text region image containing the text to be recognized; Next, segmenting the text region image to obtain text line images, and then performing vertical projection analysis on each text line image, dividing the text line into a sequence of individual text images to be tested according to the projection trough position; Then, matching each text image to be tested with a standard text image pixel by pixel, selecting the standard text with the highest matching degree as the recognition result, forming a recognized text sequence.

[0042] The standard text image is obtained by collecting standard characters from the first batch of sample prints and normalizing them after the printing press has been properly debugged.

[0043] Considering that image feature detection alone can only assess character morphological distortion and cannot verify whether the printed content is consistent with the original design, we ensure the integrity of printed information by identifying text deviations.

[0044] Therefore, the identification of printed text deviations includes: aligning the identified text sequence with the text sequence of the original printed design to determine the corresponding position.

[0045] The identified text sequence is compared with the text in the original printed design at the corresponding aligned positions. If there is a character inconsistency at the corresponding position, the position is determined as a candidate deviation position, including typos, text replacements, and missing text.

[0046] Extract at least one complete word or semantic unit based on punctuation marks before and after each candidate deviation position to form a candidate semantic unit. Then, convert each candidate semantic unit into a corresponding semantic vector and calculate the cosine similarity between each candidate semantic unit and the corresponding original semantic unit.

[0047] The number of characters before and after the candidate deviation position is dynamically extracted based on the punctuation marks of the sentence. The complete clause between the current punctuation mark and the next punctuation mark is taken as the semantic unit. If there is no punctuation, the beginning and end positions of the text line are used as the boundaries.

[0048] The cosine similarity of all candidate deviation positions is sorted and compared, and candidate deviation positions with cosine similarity values ​​lower than the similarity threshold are marked as printed text deviations.

[0049] Cosine similarity is calculated by taking the cosine of the angle between two vectors in the semantic space. The smaller the angle, the closer the cosine value is to 1, and the more similar the semantics are. The larger the angle, the closer the cosine value is to 0 or a negative value, and the more distant the semantics are.

[0050] Preferably, the similarity threshold determination step is as follows: With the printing press in a qualified debugging state, 100 standard sample prints are continuously collected. For each sample print, 10 non-edge locations are randomly selected, and the cosine similarity of their semantic units with the original print is calculated, resulting in 1000 similarity sample values. The mean μ and standard deviation σ of this sample set are calculated, and the similarity threshold is set to... The similarity threshold is pre-calculated and fixed for the content of the current printing batch. Since the original design content of different printing batches is different, the distribution of their semantic space may also be different. Therefore, whenever a new printing task begins, the above determination steps need to be repeated to adapt to the semantic characteristics of the current batch.

[0051] S300: Match the identified text with the infringing content feature database. If infringing content is found, trigger an infringement warning.

[0052] Considering that printed design originals may contain infringing content, the inability to identify infringement could lead to a risk of mass infringement. Specifically, the identified text sequences are combined into a text to be detected according to the reading order. This text is then segmented into words, and the word vector of each segment is extracted to generate a text feature matrix for the text to be detected.

[0053] Specifically, the maximum matching word segmentation based on the dictionary is used to perform word segmentation on the text to be detected. Then, a unique index is assigned to each word according to the size of the dictionary, and a word vector with a dimension equal to the size of the dictionary is generated, where only the position corresponding to the word is 1 and the rest are 0. Finally, the word vectors of each word are stacked in word order to form a text feature matrix.

[0054] The cosine similarity is calculated between the text feature matrix and the infringing text feature matrix in the infringing content feature database.

[0055] The process of establishing the infringement content feature database is as follows: First, various infringing content samples are collected, including but not limited to: text fragments from pirated publications, unauthorized trademarks, sensitive words, prohibited slogans, excerpts from literary works within the copyright protection period, and text data determined to be infringing in historical printing cases; then, each infringing sample is combined into a continuous infringing text string according to the reading order, and the infringing text is segmented into word sequences to ensure consistency with the feature extraction method of the text to be detected; next, using the same word vector mapping method as the text to be detected, each segment is converted into a numerical vector, and stacked in order to generate a feature matrix of the infringing text; finally, the feature matrix of each infringing text and its corresponding infringing text identifier are associated and stored to form the infringement content feature database.

[0056] If the maximum cosine similarity value is greater than or equal to the infringement threshold, it is determined that there is a risk of content infringement, and the corresponding infringing text identifier is output, triggering an infringement warning.

[0057] The infringement threshold is determined based on industry compliance requirements and historical detection data, with a preferred value of 0.75 to 0.95. Specifically, a batch of known infringing samples and normal samples are collected, their cosine similarity with the feature library is calculated, and a threshold that balances the false positive rate and the false negative rate is selected through ROC curve analysis.

[0058] Based on the infringement determination result, an early warning control instruction is generated, which can be implemented in the following ways: control the early warning indicator light to flash, prompting on-site operators to review the text.

[0059] S400: Locate the ink color detection area based on the gradient amplitude of each pixel in the printed image.

[0060] Since online monitoring needs to complete image processing and judgment while the printing press is running at high speed, by locating the ink color detection area, the color detection range is narrowed from the entire image to the ink color detection area, avoiding blindly detecting the entire surface of the printed matter.

[0061] refer to Figure 2 As shown, the positioning ink color detection area includes: calculating the grayscale difference value between each pixel in the printed image and its surrounding pixels, and generating a gradient amplitude map.

[0062] For each pixel in the printed image, the Sobel operator is used to calculate its first derivative in the horizontal and vertical directions, and then the gradient magnitude of that pixel is calculated to generate a gradient magnitude map.

[0063] The gradient magnitude map searches for connected regions composed of adjacent pixels and calculates the average gradient magnitude of all pixels within each connected region. At the same time, connected regions with an area smaller than a preset threshold are filtered out to avoid selecting meaningless tiny regions, such as isolated white dots. The preset threshold is obtained by collecting the gradient magnitude map of the current batch of printed materials, statistically analyzing the area distribution histogram of all connected regions, and using the area value corresponding to the first trough in the area distribution histogram as the preset threshold.

[0064] The connected region with the smallest average gradient magnitude is determined as the ink detection region.

[0065] The mean gradient magnitude represents the average level of grayscale variation among all pixels within a connected region. A smaller mean indicates less grayscale difference between pixels within the region, and a flatter overall region; a larger mean indicates more texture or noise within the region. Ink color detection requires sampling in flat regions with gentle grayscale changes; therefore, the connected region with the smallest mean gradient magnitude is the flattest region in the image for ink color sampling.

[0066] S401. Obtain the chromaticity characteristic value of the ink color detection area, calculate the chromaticity deviation between the printed matter and the original printing design, and obtain the ink color offset color difference.

[0067] The process of obtaining the chromaticity feature value of the ink detection area includes: obtaining the RGB values ​​of all pixels in each ink detection area, converting the RGB values ​​of each pixel into chromaticity values ​​in a uniform color space, and obtaining the chromaticity value of each pixel.

[0068] The steps for converting the RGB values ​​of each pixel to chromaticity values ​​in a uniform color space are as follows: Divide the RGB value of the pixel by 255, normalize it to the [0, 1] interval, and output the normalized RGB value; convert the normalized RGB value to a linear RGB value according to the gamma correction formula; convert the linear RGB value to CIE XYZ tristimulus values ​​according to the CIE standard; convert the XYZ values ​​to CIE XYZ tristimulus values ​​according to the conversion formula in the CIE standard color difference formula, using the reference white point of the standard illuminant D65 as the benchmark. Colorimetric value.

[0069] Calculate the arithmetic mean of the chromaticity values ​​of all pixels within each ink color detection area, and use it as the current chromaticity feature value of that area.

[0070] Printed surfaces may contain minute ink droplets, exposed paper fibers, or tiny impurities, causing random fluctuations in the chromaticity value of individual pixels. Selecting only the chromaticity value of a single pixel can easily lead to measurement distortion due to accidental factors. By calculating the arithmetic mean of all pixels, these high-frequency noises can be smoothed out, ensuring that the detection results are not affected by microscopic defects.

[0071] Since excessive ink volume results in a darker color, while insufficient ink volume results in a lighter color, uneven ink distribution leads to color differences or ink spots in localized areas, thus affecting the visual effect of printed materials and, in severe cases, rendering the entire batch of products unusable. Therefore, obtaining the ink color offset color difference includes: acquiring the standard area chromaticity values ​​of the corresponding positions in the original printing design artwork for each ink color detection area.

[0072] According to the CIE color difference formula, the color difference between the current colorimetric feature value of each ink color detection area and the colorimetric feature value of the corresponding standard area is calculated to obtain the color difference values ​​of several areas on the current printed matter.

[0073] The color difference values ​​of several regions are recorded according to their regional locations to form the regional color difference distribution set of the current printed material.

[0074] The arithmetic mean of the color difference values ​​of each region in the regional color difference distribution set is used as the ink color offset color difference of the current printed material.

[0075] The color difference value of a single ink color detection area on a printed material can only reflect the ink color state of that local location. By averaging the color difference values ​​of multiple areas distributed in different locations on the entire printed material surface, local abnormal interference can be eliminated.

[0076] S500 continuously collects character distortion, printing text deviation, and ink color shift of the same batch of printed materials to predict the content quality trend of subsequent printed materials in the same batch.

[0077] Because the printing industry continuously produces large quantities of printed materials, batch-level defects may occur due to undetected quality problems, leading to a continuous accumulation of substandard products. Predicting the subsequent quality trend of a batch allows for addressing the root cause before the problem escalates. Therefore, predicting the content quality trend of subsequent printed materials in a batch includes: extracting the character distortion degree of printed materials from the same batch, performing linear regression fitting according to the production timeline to obtain the slope of the character distortion degree change, and determining that the character distortion degree is increasing when the slope of the character distortion degree change is positive.

[0078] The slope of the character distortion rate indicates the average change in character distortion rate as the number of printed sheets increases. When the slope is positive, it means that the character distortion rate is gradually increasing as printing progresses, i.e., the print quality is gradually deteriorating, thus indicating an upward trend.

[0079] The printed materials in the same batch are divided into the front and back sections according to the production sequence. The number of occurrences of printing text deviations in the front and back sections are counted respectively. The ratio of the number of occurrences in the back section to the number of occurrences in the front section is calculated. If the ratio is greater than 1, it is determined that the frequency of printing text deviations is increasing.

[0080] It's important to note that the ratio of the number of occurrences in the later stage to the number of occurrences in the earlier stage represents the multiple of change in the later stage relative to the earlier stage. If the ratio equals 1, it indicates that the frequency of deviations in the later and earlier stages is the same; if the ratio is greater than 1, it indicates that the frequency of deviations in the later stage is higher than that in the earlier stage, suggesting that the frequency of deviations increases as production progresses; if the ratio is less than 1, it indicates that the frequency of deviations in the later stage is lower than that in the earlier stage, meaning the frequency decreases. Therefore, a ratio greater than 1 corresponds to more deviations in the later stage than in the earlier stage, showing an increasing trend in a temporal sense.

[0081] Calculate the change in ink color shift between two adjacent printed sheets from the same batch, and take the arithmetic mean of all adjacent changes to obtain the average rate of change of ink color shift.

[0082] If the character distortion shows an upward trend, or the frequency of printed text deviations continues to increase, or the average rate of change of ink color shift is consistently positive, then subsequent printed products in that batch are deemed to have a quality risk. To avoid misjudgment due to a single fluctuation, the judgments of a continuous upward trend, a continuous increase, and a consistently positive value are based on the assumption that data from five consecutive printed samples meet the corresponding conditions before a final determination of quality risk is made.

[0083] An increasing trend in character distortion indicates that the mechanical condition of the printing press is continuously deteriorating, and the geometric quality of characters in subsequent printed materials will inevitably decline, thus indicating a quality risk.

[0084] As the number of printed sheets increases, wear on the oleophilic layer of the printing plate leads to the loss of fine dots, or ink sedimentation in the printhead causes blockages in individual nozzles, resulting in missing text and misprints becoming more frequent. Printing plate wear and nozzle blockages cannot repair themselves; once the frequency of deviations increases, it will continue to worsen until the printing plate is unusable. Therefore, a continuous increase in the frequency of printed text deviations indicates an escalation of the fault sources, suggesting a quality risk.

[0085] The consistently positive average rate of change in ink color shift indicates that the ink color shift between adjacent printed materials increases with each sheet, meaning the ink color continues to deviate from the original design. Furthermore, the ink volume deviation and temperature will only increase and will not automatically revert to their previous state; if this trend continues, subsequent printed materials will face quality risks.

[0086] S501. When there is a quality risk in the trend of printed content quality, a batch warning signal is triggered.

[0087] refer to Figure 3 As shown, the batch warning signal is triggered if only one of the following has a quality risk: character distortion, printing text deviation, and ink color shift.

[0088] When only one indicator is at risk, it indicates that the source of the fault is relatively singular and locatable. Operators can check the printing unit corresponding to that indicator without having to shut down the entire machine for inspection. Therefore, the lowest level yellow warning signal is used.

[0089] If any two of the following three conditions—character distortion, printed text deviation, and ink color shift—exhibit a quality risk, an orange warning signal will be triggered.

[0090] When two quality indicators are at risk simultaneously, it indicates that at least two independent dimensions of the printing system are deteriorating. This dual-dimensional risk is often not an isolated event, but rather a situation where pressure fluctuations may simultaneously lead to increased character distortion and abnormal ink delivery; plate wear may simultaneously lead to increased text deviation and ink color shift caused by dot gain. Utilizing orange alerts allows time for potentially immediate, comprehensive troubleshooting.

[0091] If there are quality risks in all three aspects—character distortion, printed text deviation, and ink color offset—a red warning signal will be triggered.

[0092] The simultaneous occurrence of all three risks often indicates a systemic failure, rather than a localized problem. For example, abnormal overall pressure in the printing press can lead to character distortion; abnormal pressure can also affect ink delivery, causing ink color shift; and mechanical vibration can cause printhead misalignment, resulting in missing text. Such systemic failures cannot be resolved through localized adjustments, thus triggering a red warning to prompt a shutdown for comprehensive overhaul.

[0093] If the risk is character distortion, output the printing pressure adjustment direction; if the risk is ink color deviation, output the ink volume adjustment direction; if the risk is printing text deviation, output the printing plate or printhead inspection prompt.

[0094] After triggering an alert, in order to pinpoint the root cause of the problem and guide operators to quickly repair it, this method outputs specific adjustment directions, as follows: When a character distortion risk is triggered, if the edge distortion index is high while the topological distortion index is normal, it indicates that the blurry character edges and diffusion are caused by excessive printing pressure, and the adjustment direction to reduce printing pressure is output; if the topological distortion index is high while the edge distortion index is normal, it indicates that the broken strokes and missing characters are caused by insufficient printing pressure, and the adjustment direction to increase printing pressure is output; if both are high, a prompt to check the stability of the pressure system is provided.

[0095] When a color difference risk due to ink offset is triggered, the direction of ink volume adjustment is determined based on the color difference components in the CIE L*a*b* color space: If the lightness difference is positive, it indicates that the printed color is too light and the ink volume is too small, so the direction of increasing ink volume is output; if the lightness difference is negative, it indicates that the color is too dark and the ink volume is too large, so the direction of decreasing ink volume is output; if the red-green difference is positive, it indicates that the printed color is too red and the magenta ink volume needs to be reduced; if the red-green difference is negative, it indicates that the color is too green and the magenta ink volume needs to be increased; if the yellow-blue difference is positive, it indicates that the printed color is too yellow and the yellow ink volume needs to be reduced; if the yellow-blue difference is negative, it indicates that the color is too blue and the yellow ink volume needs to be increased.

[0096] When a printing text deviation risk is triggered, the fault source is located based on the spatial coordinates of the deviation location and corresponding inspection prompts are output: For offset printing, prompts are output based on the corresponding area of ​​the deviation text on the printing plate to check for local wear or smearing of the printing plate; for digital printing, prompts are output based on the position coordinates of the deviation text in the printing width direction to check for blockage of the corresponding printhead or nozzle; when the deviation shows a regular distribution, prompts are output to check the stability of the encoder, grating ruler or media transmission system.

[0097] When a printing text deviation risk is triggered, for offset printing, a prompt to check for local wear or smearing of the printing plate is output based on the corresponding area of ​​the deviation text on the printing plate; for digital printing, a prompt to check for blockage of the corresponding printhead or nozzle is output based on the position coordinates of the deviation text in the printing width direction.

[0098] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0099] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0102] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online monitoring and early warning of printed content, characterized in that: include: The acquired printed images are segmented into characters, and the edge contours and skeleton features of the characters are extracted. The character distortion is then calculated. The extraction of character edge contours and skeleton features includes: identifying edge pixels and connecting all edge pixels according to connectivity to form edge contour lines; thinning the binarized character image by peeling off edge pixels layer by layer until a character skeleton of single-pixel width is obtained; scanning the eight-neighborhood of each pixel on the character skeleton, marking it as a skeleton endpoint if the number of skeleton pixels in the neighborhood is 1, and marking it as a skeleton intersection if the number is greater than or equal to 3, and counting the number of endpoints and intersections on the character skeleton as skeleton features; the calculation of character distortion includes: calculating the variance of the Euclidean distance between each pixel on the edge contour line and the nearest point on the standard edge line of the corresponding position in the standard template image of the corresponding character in the original print design, as an edge distortion index; calculating the absolute value of the difference between the count of endpoints and the count of intersections and the skeleton feature value of the corresponding character in the original print design, and summing the absolute values ​​of the differences to obtain the topological distortion index; and linearly weighting the edge distortion index and the topological distortion index to obtain the character distortion degree; Based on the segmented characters, text recognition is performed. The recognized text is compared with the text in the original printed design, and semantic analysis is performed on the recognized text using natural language processing to identify printing text deviations. The identification of printing text deviations includes: aligning the recognized text sequence with the text sequence in the original printed design to determine the corresponding positions; comparing the recognized text sequence at the aligned corresponding positions with the text in the original printed design, and identifying candidate deviation positions; identifying candidate semantic units based on the candidate deviation positions, and then using a cosine similarity algorithm to mark the printing text deviations. The identified text is matched with the infringing content feature database. If infringing content is found, an infringement warning is triggered. The ink color detection area is located based on the gradient amplitude of each pixel in the printed image. The color feature value of the ink color detection area is obtained, and the color deviation between the printed product and the original printed design is calculated to obtain the ink color offset color difference. Continuously collect data on character distortion, printing deviation, and ink color shift of the same batch of printed materials to predict the content quality trend of subsequent printed materials in the batch. When the content quality trend of the printed materials indicates a quality risk, a batch warning signal is triggered. The prediction of the content quality trend of subsequent printed materials in the batch includes: The character distortion degree of the same batch of printed materials is extracted, and the slope of the change in character distortion degree is obtained by linear regression fitting according to the production time. When the slope of the change in character distortion degree is positive, it is judged that the character distortion degree is on an upward trend. The printed materials in the same batch are divided into the front and back sections according to the production time sequence. The number of occurrences of printing text deviations in the front and back sections are counted respectively. The ratio of the number of occurrences in the back section to the number of occurrences in the front section is calculated. If the ratio is greater than 1, it is determined that the frequency of printing text deviations is increasing. Calculate the change in ink color shift between two adjacent printed sheets in the same batch, and take the arithmetic mean of all adjacent changes to obtain the average rate of change of ink color shift. If the character distortion shows an upward trend, or the frequency of printed text deviation continues to increase, or the average rate of change of ink color shift is consistently positive, then the subsequent printed products in this batch are deemed to have quality risks.

2. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The extracted character edge contour and skeleton features also include: The segmented character image is binarized to separate character pixels from background pixels; Scan each character pixel in the binarized image, identify the distribution of its eight neighboring pixels, and if there is at least one background pixel in the eight neighboring pixels, mark that pixel as an edge pixel.

3. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The method for identifying printed text deviations also includes: The identified text sequence is compared with the text in the original printed design at the corresponding aligned positions. If there is a character inconsistency at the corresponding position, the position is determined as a candidate deviation position, including typos, text replacements and missing text. Extract at least one complete word or semantic unit divided according to punctuation marks before and after each candidate deviation position to form a candidate semantic unit. Then, convert each candidate semantic unit into a corresponding semantic vector and calculate the cosine similarity between each candidate semantic unit and the corresponding original semantic unit. The cosine similarity of all candidate deviation positions is sorted and compared, and candidate deviation positions with cosine similarity values ​​lower than the similarity threshold are marked as printed text deviations.

4. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The triggering of an infringement warning if infringing content is found includes: The identified text sequences are combined into the text to be detected according to the reading order. The text to be detected is then segmented into words, and the word vectors of each segment are extracted to generate the text feature matrix of the text to be detected. Calculate the cosine similarity between the text feature matrix and the infringing text feature matrix in the infringing content feature database; If the maximum cosine similarity value is greater than or equal to the infringement threshold, it is determined that there is a risk of content infringement, and the corresponding infringing text identifier is output, triggering an infringement warning.

5. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The positioning ink color detection area includes: Calculate the grayscale difference between each pixel in the printed image and its surrounding pixels to generate a gradient magnitude map; Search for connected regions consisting of adjacent pixels in the gradient magnitude map, and calculate the average gradient magnitude of all pixels in each connected region; The connected region with the smallest average gradient magnitude is determined as the ink detection region.

6. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The acquisition of chromaticity feature values ​​of the ink color detection area includes: Obtain the RGB values ​​of all pixels within each ink color detection area, convert the RGB values ​​of each pixel into chromaticity values ​​in a uniform color space, and obtain the chromaticity values ​​of each pixel. Calculate the arithmetic mean of the chromaticity values ​​of all pixels within each ink color detection area, and use it as the current chromaticity feature value of that area.

7. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The obtained ink color shift color difference includes: Obtain the standard area chromaticity values ​​of the corresponding positions in the original print design artwork that correspond to each ink color detection area; According to the CIE color difference formula, the color difference between the current colorimetric feature value of each ink color detection area and the colorimetric feature value of the corresponding standard area is calculated to obtain the color difference values ​​of several areas on the current printed matter. Record the color difference values ​​of several regions according to their location to form the regional color difference distribution set of the current printed material; The arithmetic mean of the color difference values ​​of each region in the regional color difference distribution set is used as the ink color offset color difference of the current printed material.

8. The online monitoring and early warning method for printed content according to claim 1, characterized in that: The trigger batch warning signals include: If any one of the following—character distortion, printed text deviation, or ink color shift—poses a quality risk, a yellow warning signal will be triggered. If any two of the following are quality risks: character distortion, printed text deviation, and ink color shift, an orange warning signal will be triggered. If there are quality risks in all three aspects—character distortion, printed text deviation, and ink color shift—a red warning signal will be triggered. If the risk is character distortion, output the printing pressure adjustment direction; if the risk is ink color deviation, output the ink volume adjustment direction; if the risk is printing text deviation, output the printing plate or printhead inspection prompt.

Images