Image enhancement method for edge exposure machine

Abstract

The present application relates to the technical field of image processing, and relates to an image enhancement method for an edge exposure machine, which comprises the following steps: calculating a structure tensor of an original image, and filtering the original image to obtain a plurality of illumination components; calculating a plurality of multi-scale reflection components according to the original image and the plurality of illumination components; performing weighted fusion on the plurality of multi-scale reflection components to obtain a fused reflection component; determining the noise suppression strength of each pixel in the fused reflection component; calculating a denoised reflection component; determining the order of fractional differential processing of each pixel according to the denoised reflection component, and processing the denoised reflection component according to the order to obtain a detail enhancement layer; and fusing and reconstructing the detail enhancement layer and the denoised reflection component to obtain a final enhanced image. The present application effectively suppresses noise in a dark area without damaging details in a bright area, thereby improving the signal-to-noise ratio.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to an image enhancement method for an edge exposure machine. Background Technology

[0002] Edge exposure machines are used in the manufacturing process of printed circuit boards (PCBs) or semiconductor wafers to precisely expose the edge areas of the substrate. The image quality directly affects the yield of subsequent etching and other processes. However, in actual acquisition, due to complex factors such as light source distribution, lens optical characteristics, and diffuse reflection from the PCB surface material, the acquired edge images often suffer from uneven illumination, low overall or local contrast, blurred details, and noise interference. These problems severely affect the visual quality of the image. More importantly, they can mask or blur minute defects (such as edge burrs, cracks, or contaminants), posing a significant challenge to machine vision-based automated defect detection systems and easily leading to missed detections or misjudgments.

[0003] To improve the quality of such images, existing technologies have proposed a series of image enhancement methods. Among them, histogram equalization-based methods can globally improve image contrast, but their processing method is relatively simple, easily leading to over-enhancement of local image areas, loss of detail information, and potential amplification of background noise. Another major category of mainstream methods is based on Retinex theory, which decomposes the image into illumination and reflection components, and enhances the image by adjusting these two components. For example, the multi-scale Retinex (MSR) algorithm achieves a more balanced effect by fusing processing results from different scales. However, traditional MSR has shortcomings in illumination component estimation and multi-scale fusion strategies, easily producing halo effects at the boundary between light and dark areas, and often lacking effective means to suppress noise in the reflection component. While enhancing details, it often inevitably amplifies noise in dark areas, resulting in a decrease in the signal-to-noise ratio of the enhanced image. Summary of the Invention

[0004] To address the technical problem that enhancing image details may amplify noise in dark areas, leading to a decrease in the signal-to-noise ratio of the enhanced image, this invention provides an image enhancement method for edge exposure machines.

[0005] An image enhancement method for an edge exposure machine includes: calculating the structure tensor of an original image and filtering the original image to obtain multiple illumination components; calculating a set of multi-scale reflection components based on the original image and the multiple illumination components, wherein the set of multi-scale reflection components includes reflection components of multiple scales; performing weighted fusion on the set of multi-scale reflection components to obtain a fused reflection component; fusing the multiple illumination components to obtain a final illumination map, and determining the noise suppression intensity of each pixel in the fused reflection component based on the final illumination map, wherein the illumination value of a pixel is inversely proportional to the noise suppression intensity; performing non-local mean filtering on the fused reflection component using the noise suppression intensity to obtain a denoised reflection component; determining the order of fractional-order differential processing for each pixel based on the denoised reflection component, and processing the denoised reflection component based on the order to obtain a detail enhancement layer; fusing the detail enhancement layer and the denoised reflection component to obtain a final reflection component, and reconstructing the final reflection component and the final illumination map to obtain a final enhanced image.

[0006] Preferably, filtering the original image includes: determining the parameters of the anisotropic Gaussian kernel corresponding to each pixel based on the structure tensor, wherein the parameters of the anisotropic Gaussian kernel include the standard deviation of the anisotropic Gaussian kernel in two orthogonal directions and the principal direction of the anisotropic Gaussian kernel; and filtering the original image using the anisotropic Gaussian kernel.

[0007] Preferably, determining the parameters of the anisotropic Gaussian kernel corresponding to each pixel includes: calculating the first-order partial derivative of the original image to construct a structure tensor; performing smoothing and eigenvalue decomposition on the structure tensor to obtain the eigenvalues ​​and eigenvectors of each pixel in the original image; determining the standard deviation of the anisotropic Gaussian kernel of the pixel in two orthogonal directions based on the eigenvalues ​​of the pixel, and using the eigenvector of the pixel as the principal direction of the anisotropic Gaussian kernel of the pixel, wherein the standard deviation is inversely proportional to the eigenvalues ​​of the pixel.

[0008] Preferably, the weighted fusion of the set of multi-scale reflectance components includes: calculating the contrast feature map of each scale reflectance component; constructing the spatial saliency map and the local frequency entropy map of the original image; multiplying the contrast feature map of each scale reflectance component, the spatial saliency map, and the local frequency entropy map to obtain an initial weight map of each scale reflectance component, wherein the initial weight map includes the initial weight value of each pixel in the original image; normalizing the initial weight value of each pixel across scales to obtain a final weight map; and performing a weighted summation of the set of multi-scale reflectance components based on the final weight map to obtain the fused reflectance component.

[0009] Preferably, fusing multiple illumination components to obtain a final illumination map includes: performing an arithmetic average calculation on the multiple illumination components to obtain the final illumination map.

[0010] Preferably, determining the noise suppression intensity of each pixel in the fused reflection component based on the final illumination map includes: fusing multiple illumination components to obtain the final illumination map; and mapping the normalized illumination values ​​to noise suppression intensity using a preset monotonically decreasing function based on the noise suppression intensity determined from the pixel values ​​of the final illumination map.

[0011] Preferably, determining the order of fractional-order differential processing for each pixel based on the denoised reflection component includes: calculating the local gradient norm of the denoised reflection component, and determining the order of the fractional-order differential processing based on the local gradient norm of the denoised reflection component, wherein the local gradient norm of the pixel region is inversely proportional to the differential order.

[0012] Preferably, the formula for determining the order of the fractional derivative processing is:

[0013] .

[0014] in, v max , k , c The control parameters are preset sizes. G The local gradient norm of the denoised reflection component.

[0015] Preferably, fusing the detail enhancement layer and the denoised reflection component to obtain the final reflection component includes: linearly combining the detail enhancement layer and the denoised reflection component to obtain the final reflection component.

[0016] Preferably, reconstructing the final reflection component and the final illumination map includes: multiplying the final reflection component and the final illumination map pixel by pixel to obtain the final enhanced image.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention provides an image enhancement method for edge exposure machines. By utilizing the structure tensor to determine the anisotropic Gaussian kernel, it achieves illumination component estimation, suppressing the edge halo effect commonly found in traditional Retinex methods and obtaining more realistic reflection components. This invention uses spatial saliency and local frequency entropy maps for weighting, ensuring that the fusion result of the reflection components focuses on preserving and highlighting information-rich and visually important areas in the image. This invention determines the noise suppression intensity of each pixel based on the illumination map, applying stronger denoising processing to areas with darker illumination and more readily visible noise. This effectively suppresses noise in dark areas without compromising details in bright areas, thereby improving the signal-to-noise ratio. Attached Figure Description

[0019] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein:

[0020] Figure 1 This is a schematic flowchart illustrating the steps of an image enhancement method for an edge exposure machine according to an embodiment of the present invention. Detailed Implementation

[0021] 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, not all, of the embodiments of the present invention. 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.

[0022] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Figure 1 This is a schematic flowchart illustrating the steps of an image enhancement method for an edge exposure machine according to an embodiment of the present invention.

[0024] like Figure 1 As shown, an image enhancement method for an edge exposure machine includes steps S1 to S7.

[0025] Step S1: Calculate the structure tensor of the original image and filter the original image to obtain multiple illumination components.

[0026] In one embodiment, filtering the original image includes: determining the parameters of the anisotropic Gaussian kernel corresponding to each pixel based on the structure tensor, wherein the parameters of the anisotropic Gaussian kernel include the standard deviation of the anisotropic Gaussian kernel in two orthogonal directions and the principal direction of the anisotropic Gaussian kernel; and filtering the original image using the anisotropic Gaussian kernel. It should be noted that the anisotropic Gaussian kernel parameters are used for multi-scale filtering.

[0027] The parameters for determining the anisotropic Gaussian kernel corresponding to each pixel include: calculating the first-order partial derivative of the original image to construct a structure tensor; performing smoothing and eigenvalue decomposition on the structure tensor to obtain the eigenvalues ​​and eigenvectors of each pixel in the original image; determining the standard deviation of the anisotropic Gaussian kernel of the pixel in two orthogonal directions based on the eigenvalues ​​of the pixel, and using the eigenvector of the pixel as the principal direction of the anisotropic Gaussian kernel of the pixel, wherein the standard deviation is inversely proportional to the eigenvalues ​​of the pixel.

[0028] Specifically, for the original imageI Calculate its in x and y gradient of direction I x and I y Construct the structure tensor for each pixel. J The structure tensor J It is a matrix. Next, for the structure tensor... J Eigenvalues ​​and eigenvectors are obtained by performing eigenvalue decomposition, and the eigenvectors indicate the principal and secondary directions of the local structure of the image.

[0029] Step S2: Calculate a set of multi-scale reflection components based on the original image and multiple illumination components.

[0030] Among them, a set of multi-scale reflection components includes reflection components of multiple scales.

[0031] In one embodiment, weighted fusion of the set of multi-scale reflectance components includes: calculating the contrast feature map of each scale reflectance component; constructing the spatial saliency map and local frequency entropy map of the original image; multiplying the contrast feature map, the spatial saliency map, and the local frequency entropy map of each scale reflectance component to obtain an initial weight map of each scale reflectance component, wherein the initial weight map includes the initial weight value of each pixel in the original image; normalizing the initial weight value of each pixel across scales to obtain a final weight map; and weighted summing of the set of multi-scale reflectance components based on the final weight map to obtain the fused reflectance component.

[0032] Specifically, anisotropic Gaussian kernels are constructed for three different scales, such as large, medium, and small. The orientation of each kernel is determined by the eigenvectors of the structure tensor, and its shape, i.e., the ratio of its major axis to its minor axis, is determined by the ratio of its eigenvalues. This ensures that the filtering is isotropic in flat regions and along the edge direction in edge regions, thus preserving the edges. These three scales of anisotropic Gaussian kernels are then used to filter the original image. I Convolution filtering is performed to obtain three illumination components. L 1. L 2. L 3. According to Retinex theory, in the logarithmic domain, by subtracting the logarithmic values ​​of each illumination component from the logarithmic value of the original image, i.e. , R k Indicates that xx is in the 1st month. k The reflection components at each scale are the illumination splits at the corresponding scales, and the reflection components at the three scales are obtained accordingly. R 1. R 2. R 3.

[0033] Step S3: Weighted fusion of the set of multi-scale reflection components to obtain fused reflection components.

[0034] In one embodiment, weighted fusion of the set of multi-scale reflectance components includes: calculating the contrast feature map of each scale reflectance component; constructing the spatial saliency map and local frequency entropy map of the original image; multiplying the contrast feature map, the spatial saliency map, and the local frequency entropy map of each scale reflectance component to obtain an initial weight map of each scale reflectance component, wherein the initial weight map includes the initial weight value of each pixel in the original image; normalizing the initial weight value of each pixel across scales to obtain a final weight map; and weighted summing of the set of multi-scale reflectance components based on the final weight map to obtain the fused reflectance component.

[0035] For example, using a frequency-tuned saliency model, the image is converted to the CIELAB color space, and the Euclidean distance between the color value of each pixel and the average color value of the image after Gaussian blurring is calculated to obtain a spatial saliency map. W The local frequency entropy map is obtained by performing a two-dimensional discrete Fourier transform on the spectral amplitude of each pixel within a local neighborhood window, such as a 9x9 window. W e Subsequently, W s and W e After normalization, the results are multiplied to obtain the final fusion weight map. W Using the aforementioned fusion weight graph W The multi-scale reflection components obtained in the previous step R 1. R 2. R 3. Perform a weighted summation, that is... The fused reflection component is obtained and denoted as the fused reflection component. R f .

[0036] Specifically, the purpose of this fusion strategy is to combine the advantages of different scales while retaining the most informative image features. For the reflection component at each scale, its contrast map is calculated using methods such as the Laplacian operator. A high contrast value indicates that the scale component contains rich edge or detail information at that pixel. Simultaneously, a pre-computed global spatial saliency map is used to identify regions of greater visual interest to the human eye, such as the main object in the image. Furthermore, a local frequency entropy map is calculated to measure texture complexity; a higher entropy value indicates richer texture information. These three maps are multiplied pixel-by-pixel to obtain the initial weight map for each scale component. A pixel will only have a high initial weight at a given scale if it simultaneously possesses high contrast, is located in a salient region, and has complex texture.

[0037] Specifically, suppose we process an image containing a foreground figure and a blurred background, obtaining reflection components at three scales: small, medium, and large. For the skin texture region of the foreground figure's face, the contrast is high in the small-scale reflection component, and this region also has a high value in the spatial saliency map and a high local frequency entropy. Therefore, the initial weight value of the small-scale reflection component in this region will be very large. Conversely, for the background region, its contrast, saliency, and entropy values ​​are low at all scales, and its weight is correspondingly low. After calculating the initial weights for each pixel at all scales, normalization is performed; for example, for a given pixel, its initial weight value at the small scale is divided by the sum of the initial weight values ​​at the three scales to obtain its final weight. These final weights are used to perform a weighted summation of the reflection components at each scale, so that the final fused reflection component preserves the fine texture of the foreground figure while smoothing the background region.

[0038] Step S4: Fuse multiple illumination components to obtain a final illumination map, and determine the noise suppression intensity of each pixel in the fused reflection component based on the final illumination map.

[0039] The illumination value of a pixel is inversely proportional to the noise suppression strength.

[0040] In one embodiment, fusing multiple illumination components to obtain a final illumination map includes: performing an arithmetic average calculation on the multiple illumination components to obtain the final illumination map.

[0041] In one embodiment, determining the noise suppression intensity of each pixel in the fused reflection component based on the final illumination map includes: fusing multiple illumination components to obtain the final illumination map; and mapping the normalized illumination values ​​to noise suppression intensity using a preset monotonically decreasing function based on the noise suppression intensity determined from the pixel values ​​of the final illumination map.

[0042] Specifically, it leverages the characteristics of the human visual system: it is insensitive to noise in bright areas of an image, but highly sensitive to noise in dark areas. To achieve content-based adaptive denoising, a final illumination map representing the overall illumination distribution is needed. This can be achieved by merging multiple illumination components obtained from multi-scale decomposition. After obtaining the final illumination map, its pixel values ​​are normalized to the range of zero to one, where zero represents the darkest and one represents the brightest.

[0043] Specifically, a pre-defined monotonically decreasing function maps the normalized illumination value of each pixel to a noise suppression intensity. For example, a simple linear function can be defined where the noise suppression intensity equals one minus the denormalized illumination value. For a dark area with a pixel value of 0.1, its noise suppression intensity will be mapped to 0.9, indicating the need for strong denoising. Conversely, for a bright area with a pixel value of 0.9, its noise suppression intensity will be mapped to 0.1, indicating only slight denoising or no processing, thus avoiding the problem of excessive smoothing in bright areas leading to detail loss. This adaptive mechanism makes noise suppression processing more intelligent, effectively removing noise in dark areas while perfectly preserving image details in bright areas.

[0044] Step S5: Perform nonlocal mean filtering on the fused reflection component using the noise suppression intensity to obtain the denoised reflection component.

[0045] Specifically, the aforementioned multiple illumination components L 1. L 2. L 3. Perform arithmetic mean calculation to obtain a smooth final illumination map with well-preserved structure. Determine the filtering parameters in the nonlocal mean filtering algorithm based on the final illumination map. h Filter parameters h This is the noise suppression strength. h The value of follows an exponential decay function, for example... ,in h 0 and c A constant of a preset size. L f This is the illumination value for a pixel. Based on this, make... L f Lower dark areas, h The value is high, and the noise reduction intensity is strong; in L f Higher bright areas, h A lower value results in less noise reduction, thus preserving more detail. This spatially varying filter parameter... h Applied to nonlocal mean filtering algorithms, for fused reflection components R f The denoised reflection component is obtained through processing.R d .

[0046] Step S6: Determine the order of fractional derivative processing for each pixel based on the denoised reflection component, and process the denoised reflection component according to the order to obtain the detail enhancement layer.

[0047] In one embodiment, determining the order of fractional-order differentiation processing for each pixel based on the denoised reflection component includes: calculating the local gradient norm of the denoised reflection component, and determining the order of fractional-order differentiation processing based on the local gradient norm of the denoised reflection component, wherein the local gradient norm of the pixel region is inversely proportional to the differentiation order.

[0048] In one embodiment, the Sobel operator is used to calculate the denoised reflection component. R d The gradient of the gradient is calculated, and its local gradient norm is obtained. G Next, the local gradient norm is established. G With fractional order of differential order v The mapping relationship between them, for example, using the arctangent function ,in v max , k , c These are control parameters. This mapping makes the local gradient norm... G A smaller order v in a flat region indicates a stronger enhancing effect; while the local gradient norm... G Larger, i.e., stronger edge regions have smaller order v, resulting in weaker enhancement effects, thus avoiding over-enhancement.

[0049] It's important to note that fractional derivatives are powerful detail enhancement tools, and their order determines the enhancement characteristics. Lower orders tend to enhance texture information over large areas, while higher orders are closer to first-order derivatives and can sharpen edges. This method adaptively selects the derivative order to apply the most suitable enhancement effect to different types of regions. The gradient of the denoised reflection component is calculated, and its norm is determined. This local gradient norm effectively measures whether a pixel's neighborhood is a flat region, a textured region, or an edge region. Regions with large local gradient norms typically correspond to well-defined edges in the image.

[0050] Specifically, this invention establishes an inverse relationship, mapping the local gradient norm to the order of the fractional derivative. For example, the derivative order is set to range from 0.1 to 0.9. For an edge pixel in an image with a large local gradient norm, a smaller order close to 0.1 is assigned. Using a small-order derivative to process strong edges avoids distortions such as oversharpening and ringing. Conversely, for a pixel in a flat or weakly textured region with a small local gradient norm, a larger order close to 0.9 is assigned. Using a large-order derivative effectively enhances subtle texture details in these regions, making them more clearly visible. In this way, the overall visual quality of the image is improved, enhancing detail while preserving edges.

[0051] It should be noted that by determining the order of the fractional derivative based on the local gradient norm, subtle edges and texture details are finely enhanced, avoiding over-sharpening of strong edges. The final reconstructed image has uniform illumination, clear details, and low noise levels, which greatly improves image quality and provides a high-quality image foundation for subsequent defect detection.

[0052] Step S7: Fuse the detail enhancement layer with the denoised reflection component to obtain the final reflection component, and reconstruct the final reflection component with the final illumination map to obtain the final enhanced image.

[0053] In one embodiment, fusing the detail enhancement layer with the denoised reflection component to obtain the final reflection component includes: linearly combining the detail enhancement layer with the denoised reflection component to obtain the final reflection component.

[0054] In one embodiment, reconstructing the final reflection component and the final illumination map includes: multiplying the final reflection component and the final illumination map pixel by pixel to obtain the final enhanced image.

[0055] Specifically, this step (step S7) is the final synthesis stage of image enhancement, where the detail enhancement layer obtained in the previous step through adaptive fractional derivative processing is fused with the denoised reflection component. The fusion is typically a linear combination, most commonly involving multiplying the detail enhancement layer by a weighting coefficient and adding it to the denoised reflection component. The weighting coefficient controls the intensity of the detail enhancement, which can be adjusted by the user as needed. For example, if the weighting coefficient is set to 0.8, the final reflection component equals the denoised reflection component plus 0.8 times the detail enhancement layer, thus adding the extracted texture and edge information back to the structural parts of the image in a controllable manner.

[0056] After obtaining the final reflection component, which contains rich details, it is recombined with the final lighting map representing the scene's illumination to restore the image's natural appearance. This reconstruction process follows the classic Retinex theory, where an image is obtained by multiplying the illumination and reflection components. The final reflection component and the final lighting map are multiplied pixel by pixel. During the calculation, the pixel values ​​of both components are typically normalized to the range of zero to one. For example, if the final reflection component of a pixel has a normalized value of 0.6 and the final lighting map has a normalized value of 0.8, then the final image obtained after multiplication will have a normalized value of 0.48 for that pixel. Multiplying this result by 255 yields the final pixel value of 122 for the eight-bit image. This process ensures that the enhanced details blend harmoniously into the original lighting environment, generating a final enhanced image that is both sharp and natural.

[0057] This invention adopts the Grunwald-Letnikov fractional derivative definition, utilizing the calculated order corresponding to each pixel. v For the denoised reflection component R d Processing is performed to generate a detail enhancement layer D. This detail enhancement layer D is then combined with the denoised reflection component. R d Perform linear addition, that is The final reflection component is obtained. R final The final reflected component R final With the final lighting diagram L f When adding in the logarithmic field or multiplying in the real number field, that is... The reconstructed image yields a final enhanced image with uniform illumination and clear details.

[0058] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0059] 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. An image enhancement method for a border exposure machine, characterized by, include: Calculate the structure tensor of the original image and filter the original image to obtain multiple illumination components; A set of multi-scale reflection components is calculated based on the original image and multiple illumination components, wherein the set of multi-scale reflection components includes reflection components of multiple scales; The set of multi-scale reflection components are weighted and fused to obtain the fused reflection components; Multiple illumination components are fused to obtain a final illumination map, and the noise suppression intensity of each pixel in the fused reflection component is determined based on the final illumination map, wherein the illumination value of a pixel is inversely proportional to the noise suppression intensity. The fused reflection component is subjected to nonlocal mean filtering using the noise suppression intensity to obtain a denoised reflection component. The order of fractional derivative processing for each pixel is determined based on the denoised reflection component, and the denoised reflection component is processed according to the order to obtain the detail enhancement layer. The detail enhancement layer is fused with the denoised reflection component to obtain the final reflection component, and the final reflection component is reconstructed with the final illumination map to obtain the final enhanced image; The weighted fusion of the set of multi-scale reflectance components includes: calculating the contrast feature map of each scale reflectance component; constructing the spatial saliency map and local frequency entropy map of the original image; multiplying the contrast feature map, the spatial saliency map, and the local frequency entropy map of each scale reflectance component to obtain an initial weight map of each scale reflectance component, wherein the initial weight map includes the initial weight value of each pixel in the original image; normalizing the initial weight value of each pixel across scales to obtain a final weight map; and weighting and summing the set of multi-scale reflectance components according to the final weight map to obtain the fused reflectance component.

2. The image enhancement method for edge exposure machine according to claim 1, wherein, Filtering the original image includes: The parameters of the anisotropic Gaussian kernel corresponding to each pixel are determined based on the structure tensor, wherein the parameters of the anisotropic Gaussian kernel include the standard deviation of the anisotropic Gaussian kernel in two orthogonal directions and the principal direction of the anisotropic Gaussian kernel. The original image is filtered using the anisotropic Gaussian kernel.

3. The image enhancement method for an edge exposure machine according to claim 2, characterized in that, The parameters for determining the anisotropic Gaussian kernel for each pixel include: Calculate the first-order partial derivatives of the original image to construct the structure tensor; The structural tensor is smoothed and decomposed into eigenvalues ​​to obtain the eigenvalues ​​and eigenvectors of each pixel in the original image. The standard deviation of the anisotropic Gaussian kernel of the pixel in two orthogonal directions is determined based on the feature value of the pixel, and the feature vector of the pixel is used as the principal direction of the anisotropic Gaussian kernel of the pixel, wherein the standard deviation is inversely proportional to the feature value of the pixel.

4. The image enhancement method for an edge exposure machine according to claim 1, characterized in that, The process of fusing multiple illumination components to obtain the final illumination map includes: performing an arithmetic average calculation on the multiple illumination components to obtain the final illumination map.

5. The image enhancement method for an edge exposure machine according to claim 1, characterized in that, Determining the noise suppression intensity of each pixel in the fused reflection component based on the final illumination map includes: The final illumination map is obtained by fusing multiple illumination components; The noise suppression intensity is determined based on the pixel values ​​of the final illumination map, and the normalized illumination values ​​are mapped to the noise suppression intensity using a preset monotonically decreasing function.

6. The image enhancement method for an edge exposure machine according to claim 1, characterized in that, Determining the order of fractional derivative processing for each pixel based on the denoised reflection component includes: calculating the local gradient norm of the denoised reflection component, and determining the order of fractional derivative processing based on the local gradient norm of the denoised reflection component, wherein the local gradient norm of the denoised reflection component is inversely proportional to the derivative order.

7. The image enhancement method for an edge exposure machine according to claim 6, characterized in that, The formula for determining the order of the fractional derivative process is: ,in, v max , k , c The control parameters are preset sizes. G The local gradient norm of the denoised reflection component.

8. The image enhancement method for an edge exposure machine according to claim 1, characterized in that, The step of fusing the detail enhancement layer and the denoised reflection component to obtain the final reflection component includes: linearly combining the detail enhancement layer and the denoised reflection component to obtain the final reflection component.

9. The image enhancement method for an edge exposure machine according to claim 8, characterized in that, Reconstructing the final reflection component and the final illumination map includes multiplying the final reflection component and the final illumination map pixel by pixel to obtain the final enhanced image.

Images