Image feature-based cigarette packet intaglio printing surface quality evaluation method and system

Abstract

The application discloses a cigarette case intaglio printing surface quality evaluation method and system based on image features, and relates to the fields of cigarette case intaglio printing quality detection and machine vision technology; the application collects cigarette case intaglio printing images through an image acquisition unit and pre-processes the images into gray-scale images, extracts multi-scale gradient features to obtain a comprehensive feature index for preliminary defect determination, analyzes edge morphology and texture periodicity in a transition area to obtain a comprehensive defect discrimination index, obtains a defect score based on the multi-scale comprehensive feature index and the comprehensive defect discrimination index, and completes defect classification and positioning through spatial clustering and secondary verification; four quality grades are divided based on defect proportions and defect numbers, a defect labeling map, a heat map and a standardized quality evaluation report are generated, and the problem that extracted image features cannot effectively distinguish normal dots from defective dots and result in a high false detection rate is effectively solved, and the precision and efficiency of cigarette case intaglio printing surface defect detection are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of quality inspection and machine vision technology for gravure printing of cigarette boxes, specifically a method and system for evaluating the surface quality of gravure printing of cigarette boxes based on image features. Background Technology

[0002] Gravure printing, as the mainstream process for cigarette box packaging, boasts thick ink layers, saturated colors, rich tonal range, and strong anti-counterfeiting features. It perfectly presents brand logos and complex patterns, meeting the stringent requirements of high-end consumer products for both visual quality and security. During gravure printing, ink is transferred from the cells of the printing plate to the substrate to form fine dots. The microscopic accumulation morphology of these dots directly determines the gloss, color, and image clarity of the printed product. However, the production process is highly susceptible to defects such as "plate clogging" or "plate jamming" due to factors like ink characteristics, printing pressure, squeegee angle, or environmental fluctuations. These defects severely affect the consistency of product appearance and brand image. Therefore, rapid and accurate online evaluation of the printed surface quality, especially the microscopic dot morphology, is crucial for ensuring production efficiency and product quality.

[0003] Currently, machine vision-based surface defect detection technology has become an important tool in industrial quality inspection. Traditional methods typically rely on edge detection operators to extract the contour features of dots and identify defects by analyzing their shape, size, or grayscale gradient. However, in the specific scenario of gravure printing, this method faces fundamental challenges:

[0004] Due to the microscopic accumulation effect of ink, the edges of normal dots naturally exhibit a certain gradual transition characteristic in grayscale images. However, the essence of "smearing / clogging" defects is the abnormal diffusion or deposition of ink, and the grayscale gradient pattern formed by it in the image is highly similar to that of normal accumulation edges. This essential similarity at the feature level makes it difficult for algorithms based on conventional gradients or contours to establish effective distinguishing boundaries, mistakenly judging normal process fluctuations as defects, or missing minor defects. This results in a high false detection rate and poor stability of the detection system, which seriously restricts its reliable application on high-speed production lines.

[0005] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0006] The purpose of this invention is to solve the problem that the microscopic accumulation edges of ink layers in gravure printing and printing defects are highly similar in grayscale gradient features, which makes it impossible to effectively distinguish normal dots from defective dots based on conventional edge detection operators, resulting in a high false detection rate. Therefore, this invention proposes a method and system for evaluating the surface quality of gravure printing of cigarette boxes based on image features.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for evaluating the surface quality of gravure printing on cigarette boxes based on image features includes:

[0009] S1. Image acquisition of printed surface: The original color image of the gravure printing surface of the cigarette box is acquired through the image acquisition unit, and the original color image is preprocessed by grayscale conversion and noise suppression.

[0010] S2. Multi-scale gradient feature extraction: Gradient features are extracted from the preprocessed image based on windows of different scales. By analyzing the attenuation characteristics of the gradient response in multi-scale space, the edges of normal ink layer accumulation and defect edges are distinguished.

[0011] S3. Edge morphology feature analysis: Perform micro-morphological feature quantitative analysis on the extracted edge region, and distinguish normal dots from defective dots by the geometric complexity and directional consistency features of the edge contour.

[0012] S4. Defect Identification and Location: Based on a comprehensive evaluation index of multi-scale gradient features and edge morphology features, defects are identified on the printed surface, and the location and type of defects are marked.

[0013] S5. Quality assessment results generation: Statistically calculate the percentage of defects and the severity level of defects, and generate a printed surface quality assessment report.

[0014] As a further improvement of the present invention, the specific operation steps of S2 include:

[0015] The baseline scale window is set to a side length of 3 pixels. A multi-level scale window sequence is generated according to an arithmetic progression. For each pixel in the preprocessed grayscale image, the gradient components and gradient magnitude in the horizontal and vertical directions are calculated under different scale windows. The gradient response vector is constructed by integrating the gradient magnitudes under different scales for each pixel.

[0016] Based on the gradient decay rate between adjacent scales, the corresponding standard deviation and average gradient decay rate are statistically analyzed to obtain the scale response stability coefficient.

[0017] Based on the threshold comparison results of scale response stability coefficient and average gradient decay rate, the pixels are initially determined to be normal ink layer accumulation edge features or defect edge candidate features.

[0018] Based on the gradient direction angle and the difference value of gradient direction angle under each scale window, the gradient direction consistency coefficient is obtained; the scale response stability coefficient and the gradient direction consistency coefficient are weighted and fused to obtain the multi-scale comprehensive feature index, and compared with the preset interval to determine whether the pixel is a normal ink layer accumulation edge, a defect edge, or a transition area that needs to be judged a second time.

[0019] As a further improvement of the present invention, the specific operation steps of S3 include:

[0020] Based on the gradient magnitude results obtained from multi-scale gradient feature extraction, an adaptive threshold is used to extract edge points, and pixels with a first-scale gradient magnitude greater than the adaptive threshold are marked as edge points.

[0021] Adjacent edge points are aggregated into a continuous edge contour chain by using 8-neighborhood connectivity, and noisy edges with a total number of contour points less than a preset threshold are removed. The retained contour chain is divided into three-point groups, and the turning angle of each contour point is obtained based on the three-point group and the turning point ratio is statistically obtained. At the same time, the contour complexity ratio is obtained based on the ratio of the actual contour length to the theoretical shortest length.

[0022] Based on the ratio of turning point proportion to contour complexity, the edge jaggedness complexity index is obtained by combining a weighted formula. Based on the edge jaggedness complexity index, the edges are respectively determined as irregular edges caused by printing defects such as smearing or blocking, normal dot edges, and intermediate state edges.

[0023] As a further improvement of the present invention, the specific operation steps of S3 also include:

[0024] Texture periodic feature analysis is performed on the region marked as the intermediate state edge. Based on the edge contour, fixed pixels are extended inward and outward to form an edge neighborhood strip region. Gray values ​​are collected along the direction perpendicular to the edge. At each contour point, a unit normal vector perpendicular to the edge is calculated. Gray values ​​are collected in both directions along the unit normal vector with a fixed pixel step size to obtain a gray-scale profile sequence.

[0025] A one-dimensional discrete Fourier transform is performed on the grayscale profile sequence to obtain the frequency domain components and power spectrum. The dominant frequency components in the power spectrum are identified to obtain the periodic energy concentration. Based on the periodic energy concentration, the edge of normal ink layer accumulation or the edge of defect is determined. Based on the edge serration complexity index and the periodic energy concentration, a comprehensive defect discrimination index is obtained.

[0026] As a further improvement of the present invention, the specific operation steps of S4 are as follows:

[0027] The final defect score is obtained based on the multi-scale comprehensive feature index and the comprehensive defect discrimination index. Based on the defect score, the pixels are divided into three categories: deterministic defects, normal regions, and suspected defects.

[0028] Spatial clustering analysis is performed on deterministic defect pixels to aggregate them into independent defect regions and extract geometric features. Based on the geometric features, common defect types in cigarette box gravure printing are identified as smearing, blocking, and scratch defects.

[0029] For suspected defective pixels, extract a 20×20 pixel square neighborhood around them, and calculate the mean gray value and standard deviation of gray value within the neighborhood to obtain the local contrast coefficient.

[0030] The pixel upgrade or downgrade process is completed based on the local contrast coefficient; based on the final results of defect classification and secondary verification, a defect annotation map is generated with different colors and shapes marking the defect location range, and a structured defect location information list is generated at the same time.

[0031] As a further improvement of the present invention, the specific operation steps of S5 are as follows:

[0032] Traverse the defect regions corresponding to all deterministic defect pixels and calculate the proportion of defect regions to the total pixel area of ​​the grayscale image.

[0033] Based on the defect ratio and the number of defects, and combined with the quality standards of the cigarette box printing industry, the surface quality of gravure printing of cigarette boxes is divided into four levels: excellent, good, qualified, and unqualified. The determination of each level is completed through dual thresholds.

[0034] The printed surface image is evenly divided into 4×4 grid areas. The number of defects in each grid is counted to obtain the uniformity index. When the uniformity index is less than a preset threshold, the corresponding grid is marked as having an abnormal concentration of defects, and the administrator is notified that there is a local fault in the printing equipment.

[0035] For printed surfaces that are deemed unqualified, a defect heat map is generated, which uses pseudo-color to map the defect density and visually display the location of concentrated defects, generating a standardized quality assessment report.

[0036] A second aspect of the present invention provides a surface quality assessment system for gravure printing of cigarette boxes based on image features, comprising:

[0037] Printed Surface Image Acquisition Module: Acquires images of the printed area of ​​the cigarette box through the image acquisition unit, and performs preprocessing operations such as grayscale conversion and median filtering for noise reduction;

[0038] Multi-scale gradient feature extraction module: Extracts multi-scale gradient features of image pixels and preliminarily determines the edge type of the pixels;

[0039] Edge morphology feature analysis module: Performs contour extraction, jaggedness complexity calculation and texture periodicity analysis on the extracted edge regions, and outputs a comprehensive defect discrimination index;

[0040] Defect identification and localization module: Based on multi-scale gradient features and edge morphology features, it obtains defect scores, completes defect classification, secondary verification of suspected defects, and defect location labeling and information statistics;

[0041] Quality assessment result generation module: Based on the defect ratio, it divides the quality into four levels, analyzes the uniformity of defect distribution, and generates a defect heat map and a standardized quality assessment report.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This invention acquires printed images of cigarette boxes through an image acquisition unit and performs grayscale conversion and noise reduction preprocessing. It extracts multi-scale gradient features to initially determine edge types, and performs edge contour extraction, jaggedness complexity calculation, and texture periodicity analysis on transition areas. By integrating multi-scale comprehensive feature indices and comprehensive defect discrimination indices, it completes defect discrimination, classification, and secondary verification of suspected defects. Finally, it calculates the defect proportion to classify four quality levels and generates defect annotation maps, heatmaps, and standardized evaluation reports. This significantly improves the automation and accuracy of surface defect detection in gravure printing of cigarette boxes, enabling precise defect location, classification, and quantitative evaluation. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0045] 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.

[0046] Example:

[0047] like Figure 1 As shown, the surface quality assessment method for gravure printing of cigarette boxes based on image features includes image acquisition of the printed surface, multi-scale gradient feature extraction, edge morphology feature analysis, defect identification and localization, and generation of quality assessment results.

[0048] S1. Image acquisition of printed surface:

[0049] The image acquisition unit consists of an industrial color CCD camera, a ring-shaped LED shadowless light source, an image acquisition card, and a fixed bracket. The printed cigarette box to be inspected is placed flat on the stage, the camera axis is adjusted to be perpendicular to the printing surface, and the camera is triggered to acquire a color original image of a specific area of ​​the cigarette box through the image acquisition card.

[0050] The acquired original color image is converted into a grayscale image using a weighted average method, and the grayscale image is then subjected to noise suppression using a median filtering method to eliminate random noise and interference introduced during the imaging process, while preserving printing edges and texture details, thus obtaining a preprocessed grayscale image.

[0051] S2, Multi-scale gradient feature extraction:

[0052] The baseline scale window is set to a side length of 3 pixels. A scale window sequence is generated based on an arithmetic progression, and then processed using the formula... Calculations are performed to obtain the first The side length of the scaled window, where, Generate the corresponding scale window side length sequence based on the scale window side length;

[0053] For each pixel in the preprocessed grayscale image, the gradient components and gradient magnitudes in the horizontal and vertical directions are calculated sequentially at different scale windows using the following formulas:

[0054] Horizontal gradient components: ;

[0055] Vertical gradient components: ;

[0056] Gradient magnitude: ;

[0057] in, This indicates rounding down. They represent the first Under the scale window, pixels The horizontal and vertical gradient components at point (). Indicates the horizontal pixel offset. Indicates the first Gradient magnitude under a scale window Represents pixels The grayscale value at ()

[0058] For each pixel, the gradient magnitudes of the pixel at different scales are integrated to construct a gradient response vector;

[0059] Through formula The gradient decay rate between adjacent scales is calculated, where, Indicates the first Gradient magnitude under a scale window This is a very small constant used to prevent the denominator from being zero;

[0060] Calculate the gradient decay rate between all adjacent scales to obtain the standard deviation and average gradient decay rate.

[0061] The scale response stability coefficient is obtained by using the ratio of the standard deviation of the gradient decay rate to the average gradient decay rate. The scale response stability coefficient represents the stability of the gradient magnitude in a multi-scale space. The larger the scale response stability coefficient, the more stable the edge feature response is in a multi-scale space.

[0062] When both the scale response stability coefficient and the average gradient decay rate are greater than the preset threshold, the current pixel is determined to be a normal ink layer accumulation edge feature.

[0063] When the scale response stability coefficient is less than a preset threshold or the absolute value of the average gradient decay rate is less than a preset threshold, the current pixel is determined to be a candidate feature for defect edge.

[0064] Through formula Calculations are performed to obtain the first Gradient direction angle under a scale window;

[0065] Based on the gradient direction angle, using the formula The gradient direction angle difference value is obtained by calculation; where, Describes the minimum value function. Pi is a mathematical constant that represents the total radian value of a circular angle and corresponds to the upper limit of the period of the gradient direction angle.

[0066] Based on the gradient direction angle difference value, using the formula The gradient direction consistency coefficient is calculated, where, This represents the total number of windows of different sizes.

[0067] Based on the scale response stability coefficient and the gradient direction consistency coefficient, a multi-scale comprehensive feature index is obtained by combining a weighted formula.

[0068] When the multi-scale comprehensive characteristic index is greater than the maximum value of the preset range, it is determined to be a normal ink layer accumulation edge, and no further edge morphology feature analysis is required.

[0069] When the multi-scale comprehensive feature index is less than the minimum value of the preset interval, it is judged as a defect edge and directly included in the subsequent defect discrimination;

[0070] When the multi-scale comprehensive feature index is within the preset range, it is marked as a transitional region that needs further edge morphology feature analysis and requires secondary judgment through edge micromorphology and texture features.

[0071] S3. Edge Morphology Feature Analysis: Quantitative analysis of the edge morphology features of the transition regions marked in S2 is performed. The specific implementation method is as follows:

[0072] S301. Edge point and contour extraction:

[0073] Based on the gradient magnitude calculation results of S2, an adaptive threshold is used to extract edge points; the adaptive threshold calculation formula is as follows:

[0074]

[0075] in, These represent the mean and standard deviation of the gradient magnitude across the entire grayscale image, respectively.

[0076] When the magnitude of the first-scale gradient of a pixel is greater than the adaptive threshold, the current pixel is marked as an edge point;

[0077] Neighborhood connectivity analysis is performed on all extracted edge points, and adjacent edge points are aggregated to form a continuous edge contour chain. The neighborhood connectivity determination rule is: if there are other edge points among the 8 neighboring pixels of any edge point, then the current edge point is determined to be connected to any one of its 8 neighboring pixels and is included in the same contour chain.

[0078] For each edge contour chain, the sequence of pixel coordinate points is recorded sequentially along the contour direction. When the total number of contour points in the pixel coordinate point sequence is less than a preset threshold, it is determined to be a noise edge and is directly removed.

[0079] S302, Calculation of the complexity of microscopic serrations in the profile:

[0080] For each retained edge contour chain, the edge contour is divided into a continuous group of three points. ,in, , indicating the first Each contour point has pixel coordinates;

[0081] The turning angle of each contour point is calculated using the following formula:

[0082] Adjacent line segment vectors: , ;

[0083] Steering angle: ;

[0084] in, Indicates the first The turning angle of each contour point;

[0085] The number of contour points with a turning angle greater than a preset threshold on the entire edge contour chain is counted and compared with the total number of contour points to obtain the turning point ratio.

[0086] Through formula The actual length of the contour is calculated.

[0087] Through formula The theoretical shortest length of the profile is obtained through calculation, where, This represents the total number of contour points in the edge contour chain. This represents the pixel coordinates of the first contour point, i.e., the coordinates of the starting pixel point of the edge contour chain; This represents the pixel coordinates of the last contour point, i.e., the coordinates of the terminating pixel point of the edge contour chain.

[0088] The contour complexity ratio is obtained based on the ratio of the actual length of the contour to the theoretical shortest length.

[0089] The edge jaggedness complexity index is obtained by combining the ratio of turning point proportion to contour complexity with a weighted formula.

[0090] When the edge jaggedness complexity index is greater than the maximum value of the preset range, it is determined to be an irregular edge caused by a clogging or blocking defect.

[0091] When the edge jaggedness complexity index is less than the minimum value of the preset range, it is judged as a normal dot edge;

[0092] When the edge jaggedness complexity index is within a preset range, it is marked as an intermediate state edge.

[0093] S303, Texture Periodicity Feature Analysis:

[0094] For the region marked as the intermediate state edge, based on the edge contour, a fixed number of pixels are extended to the inside and outside of the contour to form an edge neighborhood strip region.

[0095] For pixels in the edge neighborhood strip region, grayscale values ​​are sampled along a direction perpendicular to the edge; at the contour points... At that point, through the formula Calculate the unit normal vector perpendicular to the edge. ,in, , indicating the next contour point Compared to the previous contour point The difference in horizontal pixel coordinates, , indicating the next contour point Compared to the previous contour point The vertical pixel coordinate difference;

[0096] Along the unit normal vector direction, sample from the edge points to both sides with a fixed pixel step size to obtain a grayscale profile sequence;

[0097] A one-dimensional discrete Fourier transform is performed on the grayscale profile sequence to obtain the frequency domain components and the power spectrum; where the power spectrum represents the energy proportion of each frequency component.

[0098] Identify the dominant frequency components in the power spectrum, eliminate the DC component, and use the formula The periodic energy concentration is calculated, where, Represents the power spectrum. Indicates the power-dominant frequency The corresponding power spectral density value, Indicates frequency index, Indicates the number of pixels extending inwards and outwards from the outline;

[0099] When the periodic energy concentration is greater than the preset threshold, it indicates that there is a regular periodic texture of dots in the edge neighborhood, which is judged as a normal ink layer accumulation edge.

[0100] When the periodic energy concentration is less than the preset threshold, it indicates that the texture is irregular and is judged as a smear or blockage defect edge.

[0101] Based on the edge zigzag complexity index and periodic energy concentration, through the formula... The comprehensive defect discrimination index is calculated, where, This represents the complexity index of the jagged edges. Indicates a non-periodic index of texture. These represent the weighting coefficients of the edge jaggedness complexity index and the texture aperiodicity index, respectively.

[0102] The higher the comprehensive defect discrimination index value, the higher the probability that the current pixel is a defect edge.

[0103] S4. Defect Identification and Location:

[0104] For each pixel, the final defect score is obtained by combining the multi-scale comprehensive feature index and the comprehensive defect discrimination index with a weighted fusion formula.

[0105] When the defect score is greater than the maximum value of the preset range, the current pixel is determined to be a deterministic defect pixel.

[0106] When the defect score is less than the minimum value of the preset range, the current pixel is determined to be a normal pixel.

[0107] When the defect score is within a preset range, the current pixel is determined to be a suspected defective pixel.

[0108] Spatial clustering analysis is performed on pixels identified as having deterministic defects to aggregate adjacent defective pixels into independent defective regions. For each defective region, the corresponding geometric features are extracted, including the defect area, center coordinates, and aspect ratio of the bounding rectangle.

[0109] Defect types are classified based on geometric features, adapting to common defect types in intaglio printing on cigarette boxes. The specific rules are as follows:

[0110] When the defect area is greater than the preset threshold and the aspect ratio of the bounding rectangle is less than the preset threshold, it is determined to be a smearing defect, that is, a planar defect formed by excessive ink accumulation.

[0111] When the defect area is less than the preset threshold and the defect area is distributed in a clumping pattern, it is determined to be a plate blockage defect, that is, a dot-shaped defect caused by the blockage of the printing plate dots.

[0112] When the defect is distributed in a linear pattern and the aspect ratio is greater than the preset threshold, it is judged as a scratch defect, that is, a linear defect generated during the printing or transportation process.

[0113] For pixels identified as potentially defective, a secondary verification process is performed:

[0114] Extract a 20×20 pixel square neighborhood around the suspected defect area; count the gray values ​​of the pixels within the square neighborhood to obtain the mean and standard deviation of the grayscale image within the square neighborhood;

[0115] The local contrast coefficient is obtained by comparing the gray mean and gray standard deviation of the gray image within a square neighborhood.

[0116] When the local contrast coefficient is greater than the preset threshold, it indicates that the current region has typical characteristics of a real defect. The pixels in the current region are upgraded to deterministic defect pixels, and spatial clustering analysis and defect type classification are performed.

[0117] When the local contrast coefficient is less than the preset threshold, the pixels in the current region are downgraded to normal region pixels;

[0118] Based on the results of the defect type classification and secondary verification process, a defect annotation map is generated, which covers the marked defect locations and ranges on the original color image with different colors and shapes; for example, for a smear defect, a red solid line rectangle is used to annotate the corresponding outer rectangle.

[0119] At the same time, a structured list of defect location information is generated, recording the unique ID, defect type, center coordinates, defect area, and circumscribed rectangle parameters of each defect region.

[0120] S5. Generation of quality assessment results:

[0121] Iterate through the defect regions of all pixels identified as having deterministic defects, using the formula... The defect percentage was calculated, where, Indicates the first The pixel area of ​​each defective region This represents the total pixel area of ​​a grayscale image. This indicates the total number of deterministic defect regions detected.

[0122] Based on the defect percentage and the number of defects, the surface quality of gravure printing on cigarette boxes is divided into four levels. The level classification rules are adapted to the industry quality standards for cigarette box printing, as follows:

[0123] Premium grade: And the number of defects ;

[0124] Good grade: and ;

[0125] Passing grade: and ;

[0126] Unacceptable level: or ;

[0127] The printed surface image is uniformly divided into 4×4 grid areas. The number of defects in each grid is counted, and the standard deviation of the defect number is obtained. Through formula The uniformity index is calculated, where, , representing the average number of defects per grid cell;

[0128] When the uniformity index is less than the preset threshold, the corresponding grid area is marked as having an abnormal concentration of defects, and the administrator is notified that there is a local malfunction in the printing equipment.

[0129] For printed surfaces that are judged to be substandard, a defect heat map is generated, and the defect density is mapped to pseudo-color. Areas with dense defects are displayed as red high-temperature areas, areas with sparse defects are displayed as yellow medium-temperature areas, and normal areas are displayed as blue low-temperature areas, which intuitively shows the concentrated distribution of defects.

[0130] Generate a standardized quality assessment report containing core information about the surface quality of cigarette box printing. The report includes: inspection image number, inspection timestamp, quality grade, total number of defects, details of the number of each type of defect, defect percentage, distribution uniformity index, list of defect location coordinates, and defect-annotated images.

[0131] A surface quality assessment system for gravure printing of cigarette boxes based on image features includes:

[0132] Printed Surface Image Acquisition Module: Configures an image acquisition unit to acquire a color original image of a specific area of ​​the cigarette box, completes image grayscale conversion through weighted averaging, uses median filtering to suppress noise and retain printing edge and texture details, and outputs a pre-processed grayscale image.

[0133] Multi-scale gradient feature extraction module: generates multi-scale window sequences, calculates gradient components and gradient magnitude parameters at each scale of the pixel, obtains scale response stability coefficient and gradient direction consistency coefficient, weighted fusion to obtain multi-scale comprehensive feature index, and initially determines whether the pixel is a normal ink layer accumulation edge, defect edge or transition region, and marks the transition region to be analyzed.

[0134] Edge morphology feature analysis module: Extracts edge points and contours in the transition area and removes noisy edges, calculates the contour micro-jagged complexity index, performs texture periodic feature analysis on intermediate state edges, obtains periodic energy concentration, and obtains a weighted comprehensive defect discrimination index;

[0135] Defect identification and localization module: It integrates multi-scale comprehensive feature index and comprehensive defect identification index to obtain defect score, classifies pixel type, clusters deterministic defect pixels and extracts geometric features to achieve accurate classification of defects such as smearing, blockage and scratches, performs secondary verification on suspected defects, and generates defect annotation map and structured defect location information list.

[0136] The quality assessment result generation module calculates the defect percentage and combines it with the number of defects. It classifies the printed surface quality into four levels according to industry standards, calculates the defect distribution uniformity index and provides early warning of local equipment failures, generates a defect heat map for unqualified printed products, and finally outputs a standardized quality assessment report containing core inspection information.

[0137] 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.

Claims

1. A method for evaluating the surface quality of a cigarette packet intaglio printing based on image features, characterized in that, include: S1. Image acquisition of printed surface: The original color image of the gravure printing surface of the cigarette box is acquired through the image acquisition unit, and the original color image is preprocessed by grayscale conversion and noise suppression. S2. Multi-scale gradient feature extraction: Gradient features are extracted from the preprocessed image based on windows of different scales. By analyzing the attenuation characteristics of the gradient response in multi-scale space, the edges of normal ink layer accumulation and defect edges are distinguished. S3. Edge morphology feature analysis: Perform micro-morphological feature quantitative analysis on the extracted edge region, and distinguish normal dots from defective dots by the geometric complexity and directional consistency features of the edge contour. S4. Defect Identification and Location: Based on a comprehensive evaluation index of multi-scale gradient features and edge morphology features, defects are identified on the printed surface, and the location and type of defects are marked. S5. Quality assessment results generation: Statistically calculate the percentage of defects and the severity level of defects, and generate a printed surface quality assessment report.

2. The image feature-based cigarette pack intaglio printing surface quality evaluation method according to claim 1, characterized by, The specific operation steps of S2 include: The baseline scale window is set to a side length of 3 pixels. A multi-level scale window sequence is generated according to an arithmetic progression. For each pixel in the preprocessed grayscale image, the gradient components and gradient magnitude in the horizontal and vertical directions are calculated under different scale windows. The gradient response vector is constructed by integrating the gradient magnitudes under different scales for each pixel. Based on the gradient decay rate between adjacent scales, the corresponding standard deviation and average gradient decay rate are statistically analyzed to obtain the scale response stability coefficient. Based on the threshold comparison results of scale response stability coefficient and average gradient decay rate, the pixels are initially determined to be normal ink layer accumulation edge features or defect edge candidate features. Based on the gradient direction angle and the difference value of gradient direction angle under each scale window, the gradient direction consistency coefficient is obtained; the scale response stability coefficient and the gradient direction consistency coefficient are weighted and fused to obtain the multi-scale comprehensive feature index, and compared with the preset interval to determine whether the pixel is a normal ink layer accumulation edge, a defect edge, or a transition area that needs to be judged a second time.

3. The image feature-based cigarette pack intaglio printing surface quality evaluation method according to claim 1, characterized by, The specific operation steps of S3 include: Based on the gradient magnitude results obtained from multi-scale gradient feature extraction, an adaptive threshold is used to extract edge points, and pixels with a first-scale gradient magnitude greater than the adaptive threshold are marked as edge points. Adjacent edge points are aggregated into a continuous edge contour chain by using 8-neighborhood connectivity, and noisy edges with a total number of contour points less than a preset threshold are removed. The retained contour chain is divided into three-point groups, and the turning angle of each contour point is obtained based on the three-point group and the turning point ratio is statistically obtained. At the same time, the contour complexity ratio is obtained based on the ratio of the actual contour length to the theoretical shortest length. Based on the ratio of turning point proportion to contour complexity, the edge jaggedness complexity index is obtained by combining a weighted formula. Based on the edge jaggedness complexity index, the edges are respectively determined as irregular edges caused by printing defects such as smearing or blocking, normal dot edges, and intermediate state edges.

4. The image feature-based cigarette pack intaglio printing surface quality evaluation method according to claim 3, characterized in that, The specific operation steps of S3 also include: Texture periodic feature analysis is performed on the region marked as the intermediate state edge. Based on the edge contour, fixed pixels are extended inward and outward to form an edge neighborhood strip region. Gray values ​​are collected along the direction perpendicular to the edge. At each contour point, a unit normal vector perpendicular to the edge is calculated. Gray values ​​are collected in both directions along the unit normal vector with a fixed pixel step size to obtain a gray-scale profile sequence. A one-dimensional discrete Fourier transform is performed on the grayscale profile sequence to obtain the frequency domain components and power spectrum. The dominant frequency components in the power spectrum are identified to obtain the periodic energy concentration. Based on the periodic energy concentration, the edge of normal ink layer accumulation or the edge of defect is determined. Based on the edge serration complexity index and the periodic energy concentration, a comprehensive defect discrimination index is obtained.

5. The image feature-based cigarette pack intaglio printing surface quality evaluation method according to claim 1, characterized by, The specific operation steps of S4 are as follows: The final defect score is obtained based on the multi-scale comprehensive feature index and the comprehensive defect discrimination index. Based on the defect score, the pixels are divided into three categories: deterministic defects, normal regions, and suspected defects. Spatial clustering analysis is performed on deterministic defect pixels to aggregate them into independent defect regions and extract geometric features. Based on the geometric features, common defect types in cigarette box gravure printing are identified as smearing, blocking, and scratch defects. For suspected defective pixels, extract a 20×20 pixel square neighborhood around them, and calculate the mean gray value and standard deviation of gray value within the neighborhood to obtain the local contrast coefficient. Pixel upgrade or downgrade processing is performed based on local contrast coefficients; Based on the final results of defect classification and secondary verification, a defect annotation map is generated, which marks the defect location range with different colors and shapes. At the same time, a structured defect location information list is generated.

6. The image feature-based cigarette pack intaglio printing surface quality evaluation method according to claim 1, characterized by, The specific operation steps of S5 are as follows: Traverse the defect regions corresponding to all deterministic defect pixels and calculate the proportion of defect regions to the total pixel area of ​​the grayscale image. Based on the defect ratio and the number of defects, and combined with the quality standards of the cigarette box printing industry, the surface quality of gravure printing of cigarette boxes is divided into four levels: excellent, good, qualified, and unqualified. The determination of each level is completed through dual thresholds. The printed surface image is evenly divided into 4×4 grid areas. The number of defects in each grid is counted to obtain the uniformity index. When the uniformity index is less than a preset threshold, the corresponding grid is marked as having an abnormal concentration of defects, and the administrator is notified that there is a local fault in the printing equipment. For printed surfaces that are deemed unqualified, a defect heat map is generated, which uses pseudo-color to map the defect density and visually display the location of concentrated defects, generating a standardized quality assessment report.

7. A system applied to the image feature-based surface quality assessment method for gravure printing of cigarette boxes according to any one of claims 1-6, comprising: Printed Surface Image Acquisition Module: Acquires images of the printed area of ​​the cigarette box through the image acquisition unit, and performs preprocessing operations such as grayscale conversion and median filtering for noise reduction; Multi-scale gradient feature extraction module: Extracts multi-scale gradient features of image pixels and preliminarily determines the edge type of the pixels; Edge morphology feature analysis module: Performs contour extraction, jaggedness complexity calculation and texture periodicity analysis on the extracted edge regions, and outputs a comprehensive defect discrimination index; Defect identification and localization module: Based on multi-scale gradient features and edge morphology features, it obtains defect scores, completes defect classification, secondary verification of suspected defects, and defect location labeling and information statistics; Quality assessment result generation module: Based on the defect ratio, it divides the quality into four levels, analyzes the uniformity of defect distribution, and generates a defect heat map and a standardized quality assessment report.

Images