Image enhancement method for printing plate piece inspection of printing machine

Abstract

The invention belongs to the technical field of image enhancement, and particularly relates to an image enhancement method for printing plate inspection of a printing machine, which comprises the following steps of: S1, acquiring an original digital image of a printing plate to be inspected, and performing multi-scale decomposition, detail remapping and image reconstruction on the original digital image to obtain an enhanced image; s2, a multi-scale decomposition process: processing an input image of any scale level l from fine to rough: calculating a local gradient covariance matrix, a local information entropy graph and gradient consistency measurement of the input image; and constructing a spatial variable filtering kernel, and filtering the input image by using the spatial variable filtering kernel to obtain a baseband image. According to the method, amplification of background noise and irrelevant textures can be inhibited while the contrast of weak flaws is enhanced, and the obtained enhanced image is high in signal-to-noise ratio, outstanding in flaw feature and clear in edge contour.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image enhancement, and particularly relates to an image enhancement method for printing plate inspection of a printing machine. BACKGROUND

[0002] In the printing industry, the quality of the printing plate is a key factor determining the quality of the printed matter, and any slight defect, such as a dirty spot, a scratch, a missing or deformed screen dot, can cause a large batch of printed matter to be scrapped. Quality inspection of the printing plate is an indispensable link in the printing production process. An automatic optical inspection system based on digital image processing has gradually replaced manual visual inspection and become the mainstream inspection method. However, in the image acquisition stage, due to the complex lighting environment in the industrial field, the light reflection characteristics of the printing plate surface, and the inherent noise of the imaging equipment, the original digital image obtained often has problems such as low contrast, blurred details, and low signal-to-noise ratio, which causes the weak defect features to be submerged in the complex background texture and noise, seriously affecting the accuracy of the subsequent defect segmentation and recognition algorithms. In order to solve this problem, image enhancement technology is applied in the preprocessing stage of the printing plate inspection. Existing enhancement methods such as histogram equalization and Retinex algorithm can improve the overall contrast of the image to a certain extent, but the former is prone to over-enhancing noise and losing local details, and the latter may introduce color distortion or halation artifacts. For the printing plate inspection scene that needs to accurately retain details and suppress noise, the applicability is limited.

[0003] In order to better handle the details and structure of the image, enhancement methods based on multi-scale decomposition have attracted attention, among which local Laplacian filtering is an advanced detail enhancement technology. By constructing a Laplacian pyramid to decompose the image into different scale detail components, a nonlinear remapping is performed on each layer of detail components, and the image is reconstructed to achieve the purpose of enhancement. Compared with the sharpening algorithm, local Laplacian filtering can effectively enhance the local contrast while better suppressing halation artifacts. However, the local Laplacian filtering method still has some inherent defects. In the pyramid decomposition process, a spatially invariant isotropic filter such as a Gaussian filter is usually used to generate the baseband image. This filter does not have edge-preserving characteristics and will blur the edges and fine structures of the image during smoothing, resulting in impure detail information being decomposed and affecting the enhancement accuracy. Moreover, the detail remapping strategy is usually single, for example, only relying on local gray value or variance to design the gain function, which cannot effectively distinguish between real edges, texture details and random noise in the image. While enhancing the target features, background noise is also inevitably amplified, reducing the signal-to-noise ratio of the image, which is particularly disadvantageous for detecting small defects with low signal-to-noise ratio. SUMMARY

[0004] This invention provides an image enhancement method for printing plate inspection in printing presses, which solves the technical problems in the prior art where local Laplacian filtering blurs edges and fine structures due to the use of filters without edge preservation characteristics during pyramid decomposition, and the single detail remapping strategy easily amplifies background noise and reduces the signal-to-noise ratio.

[0005] In a first aspect, the present invention provides an image enhancement method for inspecting printing plates for a printing press, comprising the following steps: S1. Obtain the original digital image of the printing plate to be inspected, and perform multi-scale decomposition, detail remapping and image reconstruction on the original digital image to obtain the enhanced image; S2, the multi-scale decomposition process is as follows: for any scale level from fine to coarse... l Input image Processing: Calculate the input image Local gradient covariance matrix Local information entropy diagram And gradient consistency metric Construct a spatially variable filter kernel and use it to filter the input image. Filtering is performed to obtain the baseband image. Input image With baseband image The difference is used as the scale level. l Laplace detail ; for baseband images Downsampling is performed, and the resulting image is used as the input for the next scale level. ; S3, the detail remapping process is as follows: for each scale level, calculate the input image... Local phase consistency map Based on the local phase consistency diagram Phase consistency value pc and gradient consistency metric , details of Laplace Remapping is performed to obtain the remapping detail components. If the phase consistency value pc and the gradient consistency metric If all values ​​are greater than their respective high thresholds, then an S-shaped gain function is used for enhancement; if the phase consistency value pc is greater than the high threshold of the phase consistency value pc, then gradient consistency measurement... Not greater than gradient consistency metric If the low threshold is not greater than the high threshold of phase consistency value pc, then a logarithmic gain function is used for enhancement; if the phase consistency value pc is not greater than the high threshold of phase consistency value pc, then a soft threshold function is used for coefficient shrinkage.

[0006] Furthermore, in S2, anisotropy is determined by the local gradient covariance matrix. The eigenvectors determine the spatial support range, which is determined by the local information entropy map. The entropy value is adjusted in the opposite direction.

[0007] Furthermore, the image reconstruction process involves taking the input image at the coarsest scale level obtained from multi-scale decomposition and remapping it with the detail components after remapping at all scale levels. The image is enhanced by upsampling and accumulating the data step by step, from coarse to fine.

[0008] Furthermore, a spatially variable filter kernel is constructed, including: For the input image For each pixel, calculate the local gradient covariance matrix within a 5×5 neighborhood window of that pixel. ; For the input image Perform eigenvalue decomposition to obtain eigenvectors. and ,Will and The anisotropic direction serves as the spatially variable filter kernel; Calculate the local information entropy map within a 5×5 neighborhood window. The spatial support range of the spatially variable filter kernel is set according to its entropy value, and the calculation formula is as follows: ,in It is a constant. represents the scale parameter of the spatially variable filter kernel in the principal direction.

[0009] Furthermore, if the phase consistency value pc and the gradient consistency metric... If all values ​​exceed their respective high thresholds, then an S-shaped gain function is used for enhancement, including: When phase consistency value Greater than the first preset threshold and gradient consistency metric When the value is greater than the second preset threshold, the S-shaped gain function is used to adjust the Laplace detail components. Enhancement is performed to obtain remapped detail components. , The gain function The calculation method is as follows .

[0010] Furthermore, if the phase consistency value pc is greater than the high threshold of the phase consistency value pc, then the gradient consistency measure... Not greater than gradient consistency metric For low threshold values, a logarithmic gain function is used for enhancement, including: When phase consistency value Greater than the first preset threshold and gradient consistency metric When the value is not greater than the second preset threshold, a logarithmic gain function is used to evaluate the Laplace detail components. Enhancement is performed to obtain remapped detail components. ,in It is a symbolic function.

[0011] Furthermore, if the phase consistency value pc is not greater than the high threshold of the phase consistency value pc, then a soft thresholding function is used for coefficient shrinkage, including: When phase consistency value When the value is not greater than the first preset threshold, a soft thresholding function is used to evaluate the Laplace detail components. By performing coefficient shrinkage, the remapped detail components are obtained. ,in The shrinkage threshold is determined based on the local standard deviation of the image at the current scale level. To remap detail components, For Laplace detail components.

[0012] Furthermore, the input image at the coarsest scale level obtained from multi-scale decomposition is compared with the detail components after remapping across all scale levels. Upsampling and accumulation are performed step by step from coarse to fine to obtain the enhanced image, including: Assume multi-scale decomposition This process involves several levels to obtain the input image at the coarsest scale level. And detail components after remapping all levels ; Let the initial layer of the reconstructed image ; According to Upsampling and accumulation are performed sequentially up to 0, and the calculation method is as follows: ,in For bilinear interpolation upsampling operation, Enlarged to the size of the remapped detail component Same size, In the process of image reconstruction, the first l Reconstructed images at multiple levels; The final image obtained This refers to image enhancement.

[0013] Furthermore, in S2, for the input image For each pixel in the algorithm, the horizontal gradient gx and vertical gradient gy of the pixel are calculated using the Sobel operator. A 5×5 window is taken around the pixel, and the local gradient covariance matrix is ​​calculated using the gradient values ​​of all pixels within the window. .

[0014] Further, the method for obtaining the original digital image of the printing plate to be inspected comprises: acquiring the surface image of the printing plate by an industrial camera, and uniformly converting the acquired color or gray image into a gray image as the original digital image.

[0015] The beneficial effect is that: in the multi-scale decomposition stage, a spatially variable filter kernel determined by the local gradient covariance matrix and the local information entropy is constructed, a filtering process with edge preservation characteristics is realized, the anisotropy and spatial support range of the spatially variable filter kernel are controlled by the local structure characteristics of the image, and the blurring of important structures such as the edge of the printing plate defect and the screen dot during the smoothing of the baseband image can be avoided. In the detail remapping stage, the local phase consistency and gradient consistency are jointly used for measurement, the image content is distinguished and processed, the S-shaped strong gain, the logarithmic moderate gain and the soft threshold shrinkage are respectively applied to the high structural defect features, the ordinary texture and the noise area, the contrast of the weak defect can be enhanced, the amplification of the background noise and the irrelevant texture can be suppressed, and the obtained enhanced image has high signal-to-noise ratio, prominent defect features and clear edge outline. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the image enhancement method for printing plate inspection of a printing machine. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0018] Embodiments of the image enhancement method for printing plate inspection of a printing machine provided by the present application: As shown in the drawings, Figure 1 The image enhancement method for printing plate inspection of a printing machine comprises the following steps: S1, obtaining the original digital image of the printing plate to be inspected, and performing multi-scale decomposition, detail remapping and image reconstruction on the original digital image to obtain an enhanced image.

[0019] The surface image of the printing plate is acquired by an industrial camera, and the acquired color or gray image is uniformly converted into a gray image as the original digital image. The total number of scale levels of multi-scale decomposition is obtained, for example, 5 layers, and the original digital image is taken as the input image of the 0th layer , and the decomposition is started layer by layer.

[0020] S2, the multi-scale decomposition process is: for any scale level from fine to coarsel Input image Processing: Calculate the input image Local gradient covariance matrix Local information entropy diagram And gradient consistency metric Construct a spatially variable filter kernel, with anisotropy determined by the local gradient covariance matrix. The eigenvectors determine the spatial support range, which is determined by the local information entropy map. The entropy value is adjusted inversely; the input image is checked using a spatially variable filter. Filtering is performed to obtain the baseband image. Input image With baseband image The difference is used as the scale level. l Laplace detail ; for baseband images Downsampling is performed, and the resulting image is used as the input for the next scale level. .

[0021] For the input image For each pixel, the horizontal gradient gx and vertical gradient gy of that pixel are calculated using operators such as Sobel. A local window, such as a 5×5 window, is taken around the pixel, and the local gradient covariance matrix is ​​calculated using the gradient values ​​of all pixels within the window. For local information entropy graphs Take a local window, such as a 7x7 window, around each pixel, and calculate the gray-level histogram of the pixels within the window. Calculate the information entropy of the local window using the Shannon entropy formula. The information entropy values ​​of all pixels constitute the information entropy map. For gradient consistency metrics Similarly, within a local window surrounding each pixel, the magnitude of the sum of the gradient vectors of all pixels within the window is calculated, and then divided by the sum of the magnitudes of the gradient vectors of all pixels within the window. The ratio is the gradient consistency measure of the pixel, with a value range between 0 and 1.

[0022] At each pixel, the local gradient covariance matrix of that pixel... Eigenvalue decomposition yields two eigenvalues ​​and two corresponding eigenvectors. These two eigenvectors indicate the principal and secondary directions of the local image structure, which are then used to determine the direction of an anisotropic Gaussian kernel. Simultaneously, based on the pixel's position in the information entropy map... corresponding value to obtain the size of the Gaussian kernel. The higher the entropy value, the more complex the local region, and a smaller size of the Gaussian kernel is obtained to preserve details; the lower the entropy value, the more flat the local region, and a larger size of the Gaussian kernel is obtained to perform stronger smoothing. In this way, each pixel point in the image corresponds to a spatially variable filter kernel with different shapes and sizes.

[0023] When filtering the input image , the output value of each pixel point thereon is obtained by placing the spatially variable filter kernel corresponding to the pixel point thereon, and performing weighted summation on the pixel values in the corresponding neighborhood in the input image . After traversing all the pixel points, the generated image is the baseband image . Subtracting the input image from the baseband image pixel by pixel, the difference image obtained is the hierarchical Laplacian detail component .

[0024] To prevent aliasing effects caused by downsampling, the baseband image may be first subjected to a slight Gaussian smoothing, and the pixel points are extracted row by row and column by column, so that the width and height are both halved, and the processed image is the input image l plus 1 of the next hierarchical level . The process is repeated until the preset total number of decomposition levels is reached.

[0025] In an optional embodiment, the spatially variable filter kernel is constructed, including: For each pixel point in the input image , a local gradient covariance matrix is calculated within a 5x5 neighborhood window of the pixel point; The input image is subjected to eigenvalue decomposition to obtain eigenvectors and , and and are taken as the anisotropic directions of the spatially variable filter kernel; A local information entropy map within the 5x5 neighborhood window is calculated, and the spatial support range of the spatially variable filter kernel is set according to the entropy value, and the calculation formula is: wherein is a constant, is a scale parameter of the spatially variable filter kernel in the principal direction.

[0026] For a certain pixel point located on a vertical edge in the image, the gradients within the 5x5 neighborhood of the pixel point are mainly concentrated in the horizontal direction. At this time, the local gradient covariance matrix will exhibit a clear anisotropy, for example, the matrix elements are approximately [150, 2; 2, 5]. After eigen-decomposition of the matrix elements, the principal eigenvector is approximately directed in the vertical direction, i.e., along the edge, while the secondary eigenvector is approximately directed in the horizontal direction, i.e., perpendicular to the edge. The two mutually orthogonal vectors and define the anisotropic directions for the subsequent filtering operation, so that the filtering can be performed along the edge rather than across the edge, preserving the sharpness of the edge.

[0027] Further, the scale or size of the spatially variable filter kernel is adaptively adjusted by the local information entropy map . In the above edge region, the pixel values vary sharply, and the computed local information entropy map will be relatively high, for example, the value of the local information entropy map is 4.5. Assuming the constant is set to 15, then the computed scale parameter is approximately 2.73. In a flat region of the image, the pixel values vary little within the 5x5 neighborhood, and the computed local information entropy map will be relatively low, for example, the value of the local information entropy map is 0.2. The computed scale parameter is 12.5. Thus, a small scale spatially variable filter kernel is used in the edge and texture regions to preserve the details, and a large scale spatially variable filter kernel is used in the flat regions to effectively smooth the noise.

[0028] S3, the detail remapping procedure is as follows: for each scale level, a local phase coherence map of the input image is computed; based on the phase coherence value pc and the gradient coherence measure in the local phase coherence map , the Laplacian detail component is remapped to obtain a remapped detail component : if both the phase coherence value pc and the gradient coherence measure are greater than their respective high thresholds, an S-shaped gain function is used for enhancement; if the phase coherence value pc is greater than the high threshold of the phase coherence value pc but the gradient coherence measure is not greater than the low threshold of the gradient coherence measure , a logarithmic gain function is used for enhancement; if the phase coherence value pc is not greater than the high threshold of the phase coherence value pc, a soft threshold function is used for coefficient shrinkage.

[0029] For each scale level of the input image , a set of multi-scale and multi-directional log-Gabor wavelet filters are used to convolve the input image , and the local energy and local amplitude at each pixel point are calculated. The ratio of the two is the phase consistency value pc of the pixel point. All phase consistency values pc form a phase consistency map .

[0030] The high threshold value of phase consistency is 0.6, and the high threshold value of gradient consistency is 0.8. If the phase consistency value pc of a pixel point is greater than 0.6 and the gradient consistency measure value is greater than 0.8, it indicates that it is a strong edge with clear structure, corresponding to the coefficient value of the corresponding position in the Laplacian detail component , an S-shaped function is used for mapping. The S-shaped function weakly enhances the smaller coefficient value, significantly amplifies the intermediate coefficient value, and weakens the amplification degree of the very large coefficient value, so as to enhance the flaw contrast while avoiding over-enhancement.

[0031] If the phase consistency value pc of a pixel point is greater than 0.6 and the gradient consistency measure value is less than 0.2, it indicates that it is a region with significant structure but inconsistent gradient direction, such as fine texture, and the corresponding position coefficient in the Laplacian detail component is mapped using a logarithmic function. The logarithmic function provides a moderate degree of nonlinear gain, which moderately enhances the texture details.

[0032] If the phase consistency value pc of a pixel point is not greater than 0.6, it indicates that it is not important in perception and is likely to be noise. The corresponding position coefficient in the Laplacian detail component is subjected to soft threshold processing, that is, the absolute value of the coefficient is first reduced by a preset threshold. If the result is negative, it is set to zero, otherwise the result and the original sign are retained. This can effectively suppress the noise coefficient.

[0033] In an optional embodiment, if the phase consistency value pc and the gradient consistency measure are both greater than the respective high threshold values, an S-shaped gain function is used for enhancement, including: When the phase consistency value is greater than a first preset threshold and the gradient consistency measure is greater than a second preset threshold, the S-shaped gain function is used to enhance the Laplacian detail component to obtain a remapped detail component , , wherein the gain function is .

[0034] In an image of a printed piece, the pixel points on the contour line, the phase consistency value pc of the pixel points Generally high, such as 0.86, while the gradient direction in the neighborhood of the pixel point is highly consistent, the gradient consistency measure is also high, such as 0.93. The two conditions are met simultaneously, indicating that the pixel point is located on a significant structural feature. Assuming that the Laplacian detail component of the pixel point is 18, representing a relatively strong edge signal.

[0035] At this time, an S-shaped gain function is used to calculate the enhancement gain. The calculation process of the gain function g is to take the absolute value of the Laplacian detail component , which is about 0.018. The gain function g is about 1.96. The remapped detail component is equal to the gain function g multiplied by the original Laplacian detail component , which is about 35.28. In this way, the details of such strong structural edges are significantly and nonlinearly enhanced, making them more prominent and clear in the image. At the same time, the function tends to saturation for excessively strong detail signals, which can avoid over-enhancement.

[0036] In an optional embodiment, if the phase consistency value pc is greater than a high threshold value of the phase consistency value pc and the gradient consistency measure is not greater than a low threshold value of the gradient consistency measure , a logarithmic gain function is used for enhancement, including: When the phase consistency value is greater than a first preset threshold value and the gradient consistency measure is not greater than a second preset threshold value, a logarithmic gain function is used to enhance the Laplacian detail component to obtain a remapped detail component , where is a sign function.

[0037] The image has rich details in a local area, so the phase consistency value is relatively high, such as 0.83, but due to the complexity of the texture, the gradient direction in the neighborhood varies greatly and does not have high consistency, resulting in a low gradient consistency measure , such as 0.65. This indicates that the area is rich in details but not simple straight or curved edges. Assuming that the Laplacian detail component of a pixel point is -6, representing a medium-intensity texture detail.

[0038] According to the judgment condition, a logarithmic gain function is selected for processing. The remapped detail component is about -12.8. It can be seen that the original Laplacian detail component of -6 It has been enhanced to -12.8. The logarithmic gain function has a greater boosting effect on details with smaller values, while the boosting effect on details with larger values ​​is weakened. It is very suitable for enhancing the visibility of textures, so that the texture's sense of layering and richness are improved, without producing an overly sharp effect as when dealing with strong edges.

[0039] In an optional embodiment, if the phase consistency value pc is not greater than the high threshold of the phase consistency value pc, then a soft thresholding function is used for coefficient shrinkage, including: When phase consistency value When the value is not greater than the first preset threshold, a soft thresholding function is used to evaluate the Laplace detail components. By performing coefficient shrinkage, the remapped detail components are obtained. ,in The shrinkage threshold is determined based on the local standard deviation of the image at the current scale level.

[0040] In cases such as out-of-focus situations, there is no clear structural information, therefore the phase consistency value... The value will be lower than the first preset threshold of 0.8, for example, a measured value of 0.4. This indicates that the signal in the region is likely dominated by noise or has very weak texture. A shrinkage threshold is determined based on the local standard deviation of the region in the image. For example, if the local standard deviation of the image is 3.5, a shrinkage threshold can be set. It is 3.5.

[0041] Laplacian detail component for each pixel in the region Apply a soft thresholding function. Assume a Laplacian detail component for a pixel. A value of 2.5 is likely a noise signal. The absolute value of 2.5 is less than the contraction threshold. That is, 3.5, therefore minus The result is negative. Take the maximum value between the result and 0, which is 0. Remap the detail components. A value of 0 indicates that noise has been completely filtered out. Assume the Laplacian detail component of another pixel. A value of -5 might indicate a minor detail that needs to be preserved. An absolute value of 5 is greater than the shrinkage threshold of 3.5. minus The result is 1.5. Taking the largest value between 1.5 and 0, which is 1.5, and then multiplying by the Laplace detail component... The sign is -1, which yields the remapped detail components. The value is -1.5. This indicates that details are preserved, but the amplitude is reduced. This helps to suppress noise in smooth areas while retaining true details beyond the noise level.

[0042] S4, the image reconstruction process is as follows: the input image at the coarsest scale level obtained from multi-scale decomposition is remapped with the detail components after remapping at all scale levels. The image is enhanced by upsampling and accumulating the data step by step, from coarse to fine.

[0043] The last baseband image obtained from the decomposition, which is the input image at the coarsest scale level, is used as the initial image for reconstruction. The detail components after the coarsest scale remapping are then used. The remapped detail components Add the result to the initial image. Upsample the result so that its size is aligned with the detail components after remapping to the next higher level (the second coarser level). With the same dimensions, upsampling can be achieved by inserting zero-value rows and columns and performing interpolation filtering. The upsampling result is then remapped to the detail components of the second-coarsening level. Add. Repeat the above sampling and accumulation process, progressively from coarse to fine, until all levels of remapping detail components are completed. Once all additions are complete, the resulting image is the enhanced image of the printing plate.

[0044] In an optional embodiment, the input image at the coarsest scale level obtained from multi-scale decomposition is remapped to the detail components after remapping across all scale levels. Upsampling and accumulation are performed step by step from coarse to fine to obtain the enhanced image, including: Suppose multi-scale decomposition This process involves several levels to obtain the input image at the coarsest scale level. And detail components after remapping all levels ; Let the initial layer of the reconstructed image ; According to Upsampling and accumulation are performed sequentially up to 0, and the calculation method is as follows: ,in For bilinear interpolation upsampling operation, Enlarged to the size of the remapped detail component Same size, In the process of image reconstruction, the first l Reconstructed images at multiple levels; The final image obtained This refers to image enhancement.

[0045] The reconstruction process is a coarse-to-fine image synthesis workflow. Assume a 1024×1024 image is decomposed into three levels, where L equals 3. This results in a coarsest input image of size 128×128. and three detail component layers with detail remapping with size 1024x1024, with size 512x512 and with size 256x256. The first step of reconstruction is to set the initial layer image of the input image as the coarsest layer At this time, the initial layer image is a 128x128 image, containing the most basic outline and color information of the image.

[0046] Next, the intermediate layer image is calculated by stepwise reconstruction starting from equal to 1. The initial layer image of the reconstructed image, i.e. the 128x128 image, is up-sampled by bilinear interpolation to the same size 512x512. The up-sampled image is added pixel by pixel with the 512x512 detail component to obtain the reconstructed intermediate layer image . The intermediate layer image fuses the enhanced details of medium scale on the basis of the initial layer image . The final image is calculated by up-sampling the intermediate layer image , i.e. the 512x512 image, to 1024x1024, and adding it with the finest detail component with size 1024x1024. The obtained final image is the enhanced image of 1024x1024, fusing all scale levels of the details information selectively enhanced or suppressed.

[0047] The above are the preferred embodiments of the present application, which are not intended to limit the protection scope of the present application. Any equivalent changes made on the basis of the structure, shape and principle of the present application should be covered within the protection scope of the present application.​

Claims

1. An image enhancement method for plate inspection of a printing press, characterized in that, The method comprises the following steps: S1, obtaining an original digital image of a printing plate to be inspected, performing multi-scale decomposition, detail remapping and image reconstruction on the original digital image to obtain an enhanced image; S2, the multi-scale decomposition process is as follows: for any scale level from fine to coarse... l Input image Processing: Calculate the input image Local gradient covariance matrix Local information entropy diagram And gradient consistency metric Construct a spatially variable filter kernel and use it to filter the input image. Filtering is performed to obtain the baseband image. Input image With baseband image The difference is used as the scale level. l Laplace detail ; for baseband images Downsampling is performed, and the resulting image is used as the input for the next scale level. ; S3, the detail remapping process is as follows: for each scale level, calculate the input image... Local phase consistency map Based on the local phase consistency diagram Phase consistency value pc and gradient consistency metric , details of Laplace Remapping is performed to obtain the remapping detail components. If the phase consistency value pc and the gradient consistency metric If all values ​​are greater than their respective high thresholds, then an S-shaped gain function is used for enhancement; if the phase consistency value pc is greater than the high threshold of the phase consistency value pc, then gradient consistency measurement... Not greater than gradient consistency metric If the threshold is low, a logarithmic gain function is used for enhancement; If the phase consistency value pc is not greater than the high threshold value of the phase consistency value pc, a soft threshold function is used for coefficient shrinkage.

2. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, In S2, the anisotropy is determined by the eigenvector of the local gradient covariance matrix and the spatial support range is inversely adjusted by the entropy value of the local information entropy map .

3. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, The image reconstruction process is: the input image of the coarsest scale level obtained by multi-scale decomposition is up-sampled and accumulated in order from coarse to fine with all scale levels remapped detail components , to obtain an enhanced image.

4. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, The spatial variable filter kernel is constructed, comprising: For each pixel in the input image a local gradient covariance matrix is computed within a 5x5 neighborhood window of the pixel ; performing feature decomposition on the input image to obtain a feature vector performing feature decomposition on the input image to obtain a feature vector and performing feature decomposition on the input image to obtain a feature vector and performing feature decomposition on the input image to obtain a feature vector calculating a local information entropy map within a 5x5 neighborhood window , according to which the spatial support range of the spatially variable filter kernel is set, the calculation formula being: , wherein is a constant, is a scale parameter of the spatially variable filter kernel in the main direction.

5. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, If the phase consistency value pc and the gradient consistency measure are both greater than their respective high thresholds, then enhancement is applied using an S-shaped gain function, including: When the phase consistency value is greater than a first preset threshold and the gradient consistency measure is greater than a second preset threshold, an S-shaped gain function is adopted to enhance the Laplacian detail component to obtain a remapped detail component , wherein the gain function is calculated as .

6. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, If the phase consistency value pc is greater than a high threshold value of the phase consistency value pc and the gradient consistency measure is not greater than a low threshold value of the gradient consistency measure , enhancement is performed using a logarithmic gain function, comprising: When the phase consistency value is greater than a first preset threshold and the gradient consistency measure is not greater than a second preset threshold, a logarithmic gain function is adopted to enhance the Laplacian detail component to obtain a remapped detail component , wherein is a sign function.

7. The image enhancement method for printing plate inspection of a printing press according to claim 1, characterized in that, If the phase consistency value pc is not greater than the high threshold value of the phase consistency value pc, a soft threshold function is used for coefficient shrinkage, comprising: When the phase consistency value is not greater than a first preset threshold value, a soft threshold function is adopted to perform coefficient shrinkage on the Laplacian detail component to obtain a remapped detail component , wherein is a shrinkage threshold value determined according to a local standard deviation of the image at a current scale level, is the remapped detail component, is the Laplacian detail component.

8. The image enhancement method for printing plate inspection of a printing press according to claim 3, characterized in that, re-mapping the input image of the coarsest scale level obtained by the multi-scale decomposition with all scale levels of the re-mapped detail components up-sampling and accumulating in order from coarse to fine, to obtain an enhanced image, comprising: Set the input image to the coarsest scale level and all levels remapped detail components and all levels remapped detail components ; Let the initial layer of the reconstructed image ; According to the order from to 0, up-sampling and accumulation are performed step by step, and the calculation is as follows: wherein is a bilinear interpolation up-sampling operation that upsizes the dimensions of to the same dimensions as the remapped detail component represents the reconstructed image at the l th level in the image reconstruction process;​ The resulting final image i.e. the enhanced image.

9. The image enhancement method for printing plate inspection of a printing press according to any of claims 1 to 8, characterized in that, In S2, for each pixel point in the input image , the horizontal gradient gx and the vertical gradient gy of the pixel point are calculated using the Sobel operator, a 5x5 window is taken around the pixel point, and the local gradient covariance matrix is calculated using the gradient values of all the pixel points in the window.

10. The image enhancement method for printing plate inspection of a printing press according to claim 9, characterized in that, The method for obtaining the original digital image of the printing plate to be inspected comprises: collecting a surface image of the printing plate by an industrial camera, and uniformly converting the obtained color or gray-scale image into a gray-scale image as the original digital image.

Images