Paper paper overprinting quality detection method and system based on image recognition

Abstract

The invention relates to the technical field of image data processing, in particular to a paperboard overprinting quality detection method and system based on image recognition, and the method comprises the steps: obtaining a to-be-detected paperboard overprinting image, and dividing the to-be-detected paperboard overprinting image into a plurality of pixel blocks which are consistent in size and have overlapped pixels between adjacent pixel blocks; performing two-dimensional fast Fourier transform on each pixel block to obtain a corresponding local spectrogram; carrying out filtering processing on each pixel block by utilizing improved frequency domain filtering, and carrying out image reconstruction on the pixel blocks after the filtering processing so as to obtain a corrected paper jam overprinting image; and carrying out edge detection on the corrected paperboard overprinting image to obtain a plurality of edge contours, and if the number of the detected edge contours is greater than a set threshold, judging that the to-be-detected paperboard overprinting image has quality defects. The method solves the problem that an existing algorithm is not high in quality detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing. More particularly, the present application relates to a paper sheet overprint quality detection method and system based on image recognition. BACKGROUND

[0002] With the rapid development of modern packaging and printing industry, the quality of paper sheet overprint directly affects the appearance consistency and added value of printed products. Overprint refers to the process of superimposing multi-color patterns on the same printing substrate through multiple printing, and its precision and stability depend on the strict alignment of different color plates in space position. However, in high-speed production environment, due to factors such as mechanical vibration, paper expansion, ink layer thickness variation and environmental temperature and humidity fluctuation, color misregistration, edge ghosting, blur diffusion and other quality problems are prone to occur in the overprint process. Such defects are often difficult to identify in real time by naked eye, and traditional manual sampling inspection method has low detection efficiency, strong subjectivity and difficulty in realizing full-process online monitoring.

[0003] In recent years, with the wide application of machine vision and image recognition technology, overprint quality detection methods based on image analysis have gradually become the main research direction of the printing industry. Such methods usually obtain high-resolution images of the printed surface, extract color, texture and edge features, and realize automatic identification of printing offset, blur and defects. Among them, frequency domain filtering technology is widely used in image denoising, feature enhancement and defect extraction, because it can effectively distinguish low-frequency structural information and high-frequency texture details in the image. By mapping the image to the frequency domain through Fourier transform, filter functions can be designed for different frequency components to suppress noise interference and enhance texture features, thereby improving the accuracy of subsequent feature analysis.

[0004] However, the existing frequency domain filtering method still has obvious technical limitations. First, the traditional filter function mostly uses fixed parameters or static templates, which cannot adaptively adjust to the texture complexity and structural features of different image regions, resulting in overblurring in texture smooth areas and retaining too much noise in high-frequency detail areas, which affects the accuracy and stability of the overall detection, thereby causing the problem of low quality detection accuracy of the existing algorithm. SUMMARY

[0005] To solve the problem of low quality detection accuracy of the existing algorithm in the background art, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for detecting the quality of overprinted cardboard based on image recognition, comprising: acquiring an image of overprinted cardboard to be detected, and dividing the image into multiple pixel blocks of the same size with pixel overlap between adjacent pixel blocks; performing a two-dimensional fast Fourier transform on each pixel block to obtain a corresponding local spectrum map; filtering each pixel block using an improved frequency domain filter, and reconstructing the filtered pixel blocks to obtain a corrected image of overprinted cardboard; performing edge detection on the corrected image of overprinted cardboard to obtain multiple edge contours, and determining that the image of overprinted cardboard to be detected has a quality defect if the number of detected edge contours is greater than a set threshold; wherein, the improved frequency domain filter includes a filter function, the filter function being inversely correlated with the structural determinism index and corresponding to the pixel block with the highest spectral amplitude excluding the DC component. The structural determinism index is positively correlated with the Euclidean distance from each pixel to the center of the spectrum; it is also positively correlated with the information complexity of the pixel block and negatively correlated with the degree of structure of the pixel block; the information complexity is related to the highest spectral amplitude (excluding the DC component) in the pixel block. The spectral amplitude of an individual pixel and its Euclidean distance to the center of the spectrum are positively correlated, while the sum of the spectral amplitudes of all pixels in the pixel block is negatively correlated.

[0007] The above technical solution applies an improved frequency domain filter to each pixel block, enabling the filtering intensity to be dynamically adjusted according to local structural features. This preserves more detailed information in areas with complex textures and obvious directionality, while achieving smooth suppression in noisy or abnormal areas. This significantly improves the adaptability and robustness of the filtering process, enabling the detection to maintain high resolution, high sensitivity, and low false detection rate in complex printing backgrounds.

[0008] Furthermore, pixels filter function for: , To find the maximum value function, For pixels The structural determinism index of the pixel block it belongs to. , Each pixel The x and y coordinates of its pixel block, , The first Each pixel has its x and y coordinates within its pixel block. The set Gaussian kernel width, For the natural constant An exponential function with base 0. The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels.

[0009] The above technical solution achieves adaptive control of local frequency domain characteristics by introducing structural determinism and Gaussian weighting mechanism into the filter function, thus avoiding edge breakage or spectrum jump problems that may occur in traditional filtering.

[0010] Furthermore, pixels Structural determinism index of the pixel block for: , For pixels The information complexity of the pixel block, For pixels The degree of structuring of the pixel block.

[0011] The aforementioned technical solution can accurately distinguish between textured and noise-perturbed regions. When the information complexity is high and the degree of structure is low, the structural determinism index increases, indicating that the region has strong uncertainty and variability, making it suitable for enhanced filtering to smooth anomalous frequency components. Conversely, when the structural direction is obvious and the spectrum is concentrated, the structural determinism index decreases, and the filtering intensity automatically decreases to protect texture details. Through this coupling relationship, the method achieves adaptive matching between filtering and regional features, making frequency domain processing more selective and targeted. This effectively balances noise reduction and detail preservation in complex printing backgrounds, improving the accuracy and stability of subsequent defect detection.

[0012] Furthermore, pixels Information complexity of the pixel block for: , For the first The spectral amplitude of each pixel The pixel block with the highest spectral amplitude excluding the DC component. The first pixel The Euclidean distance from each pixel to the center of the spectrum The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels, This represents the total number of pixels in the pixel block.

[0013] The above technical solution, through a quantitative description method based on spectral energy distribution, can effectively distinguish between smooth regions and complex texture regions, providing a precise basis for the adaptive adjustment of subsequent filtering intensity.

[0014] Furthermore, pixels The degree of structuring of the pixel block for: , For the first The spectral amplitude of each pixel For pixels The average spectral amplitude of all pixels within the pixel block. For pixels The standard deviation of the spectral amplitude of all pixels within the pixel block. These are the preset hyperparameters.

[0015] The above technical solution calculates the deviation of the spectral amplitude of each pixel within a pixel block from its overall distribution, and normalizes the local spectral features by combining the mean and standard deviation. This effectively reflects the concentration and directional consistency of spectral energy within a pixel block, thereby quantifying the degree of structure of local texture in the image.

[0016] Furthermore, it also includes normalizing the structural determinism index.

[0017] Furthermore, a CCD camera is used to acquire an overprinted image of the cardstock to be inspected.

[0018] Furthermore, the edge detection is Canny edge detection.

[0019] Furthermore, the image reconstruction specifically involves: for regions with overlapping pixels, extracting the pixel values ​​of each overlapping pixel block, and performing a weighted average fusion according to a preset weight to obtain the final pixel value.

[0020] In a second aspect, the present invention provides an image recognition-based paperboard overprint quality inspection system, including a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the image recognition-based paperboard overprint quality inspection method described above.

[0021] The beneficial effects of this invention are as follows: This invention adaptively analyzes and filters local features of an image in the frequency domain, enabling differentiated processing based on the texture complexity and structure of different regions. This effectively suppresses false feature responses caused by noise and printing texture interference while preserving the main structural information in the cardboard overprint image. The improved filter function dynamically adjusts the filtering intensity according to the local spectral energy distribution, keeping high-structure-consistency regions clear while smoothing and optimizing low-structure-deterministic regions, significantly improving the overall spectral balance and texture representation accuracy of the image. The edge features of the filtered and reconstructed image are more continuous and the contours are more stable, helping edge detection algorithms accurately extract overprint boundaries, thereby improving the sensitivity and reliability of quality defect detection. This invention achieves closed-loop optimization from frequency domain feature analysis and dynamic filtering control to structured image reconstruction, significantly improving the accuracy and robustness of cardboard overprint quality detection. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an image recognition-based method for detecting overprint quality according to an embodiment of the present invention; Figure 2 This is a schematic block diagram illustrating the structure of a cardboard overprint quality inspection system based on image recognition according to an embodiment of the present invention. Detailed Implementation

[0023] An example of an image recognition-based paper overprint quality detection method.

[0024] like Figure 1 As shown, the flowchart of the image recognition-based cardboard overprint quality detection method according to an embodiment of the present invention includes the following steps: S1: Obtain the overprinted image of the card to be tested, and divide the overprinted image of the card to be tested into multiple pixel blocks of the same size with pixel overlap between adjacent pixel blocks.

[0025] In a preferred embodiment, a high-resolution CCD industrial camera deployed on the production line acquires image data of the overprinted surface of the cardstock under inspection in real time. The CCD camera features high sensitivity and low noise, enabling it to accurately capture minute color differences, edge shifts, and local texture variations in the overprinted area under high-speed production conditions, thereby ensuring high signal-to-noise ratio and detail fidelity in the image data. To further improve the accuracy of subsequent frequency domain analysis, the acquired complete cardstock overprinted image is processed according to a set pixel size. The process is divided into blocks, and there are pre-defined overlapping areas between each pixel block.

[0026] This overlapping block method effectively avoids spectral truncation or feature loss at block edges, ensuring that each pixel block fully preserves the local spatial continuity and frequency consistency of the image during 2D Fast Fourier Transform. Simultaneously, it enhances the ability to distinguish between high-frequency texture regions and low-frequency background regions, making subsequent filtering and image reconstruction processes smoother and more natural, and reducing artifacts or edge distortion caused by discontinuities at block boundaries.

[0027] S2: Filter each pixel block using an improved frequency domain filter, and reconstruct the filtered pixel block into a corrected cardboard overprint image.

[0028] In a preferred embodiment, the improved frequency domain filtering includes a filter function, pixel points filter function for: , To find the maximum value function, For pixels The structural determinism index of the pixel block it belongs to. , Each pixel The x and y coordinates of its pixel block, , The first Each pixel has its x and y coordinates within its pixel block. The set Gaussian kernel width, For the natural constant An exponential function with base 0. The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels. Preferably, The minimum number of pixels whose cumulative spectral energy accounts for the proportion of the total energy of the pixel block to a set threshold ρ is determined by sorting the spectral amplitudes (excluding the DC component) from high to low. The set threshold ρ ranges from [0.85, 0.95].

[0029] By introducing an adaptive filter function related to the structural determinism index in the frequency domain, differentiated smoothing and enhancement processing for different texture regions is achieved. The filter design incorporates local structural features and spatial distribution information, automatically reducing the filter intensity in regions with stable texture structures and consistent orientations, effectively preserving key high-frequency details and edge information; while in regions with uncertain structures or strong noise interference, the filter response is enhanced to suppress stray frequency components. Through a Gaussian weighting mechanism, the filter exhibits spatial continuity and smoothness, avoiding artifacts caused by spectral abrupt changes. By adaptively adjusting the filter intensity, a better balance between noise reduction and detail preservation is achieved, significantly improving the visual quality after frequency domain filtering and the accuracy of subsequent defect detection.

[0030] pixel Structural determinism index of the pixel block for: , For pixels The information complexity of the pixel block, For pixels The degree of structuring of the pixel block.

[0031] By jointly modeling information complexity and structuring degree, a structural determinism index is constructed that comprehensively reflects the organizational state of image texture. It considers both the richness and regularity of texture information; when a region has complex texture but unstable structure, the index value is high, highlighting potential abnormal or heterogeneous regions; conversely, when a region has clear texture structure and consistent orientation, the index value is low, indicating strong structural consistency. This coupled measurement method effectively distinguishes between regular and abnormal textures, achieving high sensitivity detection of local structural differences in complex images, and providing more accurate discrimination criteria for subsequent frequency domain filtering optimization and defect identification.

[0032] pixel Information complexity of the pixel block for: , For the first The spectral amplitude of each pixel The pixel block with the highest spectral amplitude excluding the DC component. The first pixel The Euclidean distance from each pixel to the center of the spectrum The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels, This represents the total number of pixels in the pixel block.

[0033] By comprehensively considering the spectral amplitude of high-energy spectral components in pixel blocks and their distribution in the spectral space, a quantitative expression of the complexity of local texture information is achieved. The core idea lies in using the spatial relationship between energy-concentrated regions and the spectral center to reflect the directionality and dispersion of texture structure: when high-energy components are widely distributed and far from the spectral center, it indicates that the region contains richer texture details and more complex structures; conversely, it indicates that the texture tends to be smooth or simple. Through weighted analysis of high-energy frequency components, regions with significant structural changes in the image can be effectively captured, thereby enhancing the sensitivity to differences in texture complexity and providing a more accurate basis for subsequent frequency domain filtering and anomaly detection.

[0034] pixel The degree of structuring of the pixel block for: , For the first The spectral amplitude of each pixel For pixels The average spectral amplitude of all pixels within the pixel block. For pixels The standard deviation of the spectral amplitude of all pixels within the pixel block. The parameters are preset hyperparameters. The method also includes normalizing the structural determinism index, where the normalization is a Sigmoid normalization process.

[0035] By statistically analyzing the overall distribution of pixel spectral amplitudes within each pixel block, and comprehensively considering the deviation and stability of the spectral amplitudes, the degree of structure in a local region can be measured. By calculating the normalized deviation of the pixel spectrum relative to the average value and dispersion of its block, the concentration and regularity of texture information in that region can be effectively reflected. This results in regions with clear textures, strong directionality, or obvious periodicity exhibiting higher response values ​​in terms of structure. Furthermore, this enhances the sensitivity to local texture complexity and spatial organization features, facilitating differentiated processing of different texture regions in subsequent frequency domain filtering or defect detection, thereby improving the overall detection accuracy and robustness.

[0036] Image reconstruction specifically involves extracting the pixel values ​​of each overlapping pixel block for regions with pixel overlap, and then performing a weighted average fusion according to preset weights to obtain the final pixel value. By performing weighted average fusion on multiple pixel blocks within overlapping regions, smooth transitions and continuous image reconstruction are effectively achieved. The weight allocation mechanism balances the pixel contributions of each overlapping region, avoiding boundary abrupt changes, inconsistent brightness, or texture breaks that may result from direct stitching. This provides a more stable and artifact-free base image input for subsequent defect identification, thereby improving the accuracy and reliability of the detection results.

[0037] S3: Perform edge detection on the corrected cardboard overprint image. If the number of edge contours exceeds the set threshold, it is determined that there is a quality defect.

[0038] In a preferred embodiment, in a normal cardboard overprint image, the positions of each printing layer are accurate, the colors are uniformly superimposed, and the edges are smooth and continuous. Therefore, the number of effective contours obtained after edge detection is relatively small, mainly concentrated at the design texture and pattern boundaries. However, when the image has quality defects such as overprinting misalignment, printing blur, ink breaks, ghosting, or smudges, the local grayscale gradient distribution of the image will change abnormally, resulting in a significant increase in the number of detected edges.

[0039] Edge detection processing is performed on the corrected cardboard overprint image to obtain edge information reflecting image details and contour changes. Specifically, an edge detection algorithm with strong noise suppression and gradient response characteristics is used to perform pixel-level gradient calculation and edge extraction on the corrected image. This effectively eliminates false edge responses caused by uneven lighting, printing textures, or weak noise while ensuring detection sensitivity. Subsequently, the number of detected edge contours is compared with a set threshold. When the number of detected edge contours exceeds the preset threshold, it can be determined that there are potential quality defects in the cardboard overprint image, such as printing misalignment, color overprinting errors, or structural damage. The edge detection method used is Canny edge detection.

[0040] This invention analyzes and adaptively filters local pixel blocks of overprinted images in the frequency domain. This fully leverages the differences in spectral distribution, energy concentration, and structural features across different regions to achieve precise differentiation between image details and noise. By introducing a structural determinism index, information complexity and structuring are organically combined, allowing the filtering process to dynamically adjust the filtering intensity based on local texture characteristics. This effectively suppresses false features caused by background texture, uneven illumination, or printing noise, while preserving high-frequency details of genuine overprinting defects. After reconstruction and edge detection, the filtered image significantly improves the clarity and continuity of overprinting edges, enabling highly sensitive identification of minor misalignments, blurring, or ghosting defects. Overall, this solution improves the frequency domain representation accuracy and structural perception capability of overprinted images, enhancing the stability and accuracy of the quality inspection system in complex printing scenarios.

[0041] Example of an image recognition-based paper overprint quality inspection system: like Figure 2 As shown in the figure, the structural block diagram of the image recognition-based cardboard overprint quality inspection system of the present invention includes a processor and a memory.

[0042] This invention also provides an image recognition-based paperboard overprint quality inspection system. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the image recognition-based cardboard overprint quality detection method according to the present invention.

[0043] The image recognition-based cardboard overprint quality inspection system also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0044] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0045] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0046] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for detecting overprint quality on cardboard based on image recognition, characterized in that, include: Acquire an overprint image of the paper to be tested, and divide the overprint image of the paper to be tested into multiple pixel blocks of the same size with pixel overlap between adjacent pixel blocks; Perform a two-dimensional fast Fourier transform on each pixel block to obtain the corresponding local spectrogram; The improved frequency domain filtering is used to filter each pixel block, and the filtered pixel block is reconstructed to obtain the corrected cardboard overprint image. Edge detection is performed on the corrected cardboard overprint image to obtain multiple edge contours. If the number of detected edge contours is greater than a set threshold, it is determined that the cardboard overprint image to be detected has quality defects. The improved frequency domain filtering includes a filter function that is inversely correlated with the structural determinism index and corresponds to the pixel block with the highest spectral amplitude excluding the DC component. The Euclidean distance from each pixel to the center of the spectrum is positively correlated; The structural determinism index is positively correlated with the information complexity of the pixel block and inversely correlated with the degree of structure of the pixel block; the information complexity is related to the highest spectral amplitude (excluding DC component) in the pixel block. The spectral amplitude of an individual pixel and its Euclidean distance to the center of the spectrum are positively correlated, while the sum of the spectral amplitudes of all pixels in the pixel block is negatively correlated.

2. The method for detecting overprint quality on cardboard based on image recognition according to claim 1, characterized in that, pixel filter function for: , To find the maximum value function, For pixels The structural determinism index of the pixel block it belongs to. , Each pixel The x and y coordinates of its pixel block, , The first Each pixel has its x and y coordinates within its pixel block. The set Gaussian kernel width, For the natural constant An exponential function with base 0. The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels.

3. The method for detecting overprint quality of cardboard based on image recognition according to claim 1, characterized in that, pixel Structural determinism index of the pixel block for: , For pixels The information complexity of the pixel block in question. For pixels The degree of structuring of the pixel block.

4. The method for detecting overprint quality of cardboard based on image recognition according to claim 1, characterized in that, pixel Information complexity of the pixel block for: , For the first The spectral amplitude of each pixel The pixel block with the highest spectral amplitude excluding the DC component. The first pixel The Euclidean distance from each pixel to the center of the spectrum The pixel block with the highest spectral amplitude excluding the DC component. The total number of pixels, This represents the total number of pixels in the pixel block.

5. The method for detecting overprint quality of cardboard based on image recognition according to claim 1, characterized in that, pixel The degree of structuring of the pixel block for: , For the first The spectral amplitude of each pixel For pixels The average spectral amplitude of all pixels within the pixel block. For pixels The standard deviation of the spectral amplitude of all pixels within the pixel block. These are the preset hyperparameters.

6. The method for detecting overprint quality of cardboard based on image recognition according to claim 1, characterized in that, It also includes normalizing the structural determinism index.

7. The method for detecting overprint quality of cardboard based on image recognition according to claim 1, characterized in that, A CCD camera is used to acquire an overprinted image of the paper card to be inspected.

8. The method for detecting overprint quality on cardboard based on image recognition according to claim 1, characterized in that, The edge detection method is Canny edge detection.

9. The method for detecting overprint quality on cardboard based on image recognition according to claim 1, characterized in that, The image reconstruction specifically involves: for regions with overlapping pixels, extracting the pixel values ​​of each overlapping pixel block, and performing a weighted average fusion according to a preset weight to obtain the final pixel value.

10. A paper overprinting quality inspection system based on image recognition, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the image recognition-based paper overprint quality detection method according to any one of claims 1 to 9 is implemented.

Images