An adaptive multi-scale microscope image iterative denoising method and system
Patent Information
- Application Number
- CN202610981160.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-02
- Publication Date
- 2026-08-28
AI Technical Summary
[0006]为克服上述现有技术的不足，本发明提供了一种自适应多尺度显微镜图像迭代去噪方法及系统，旨在解决现有技术中存在的去噪导致细胞轮廓模糊、对无规律随机噪声去噪能力差，对新型噪声去噪能力差且计算量大不支持交互式迭代去噪的技术问题
本发明的上述技术方案提供了一种自适应多尺度显微镜图像迭代去噪方法及系统，通过自适应尺度参数和自适应相似度阈值提高了显微镜图像去噪的泛化能力，避免了传统固定多尺度参数和固定阈值对多样化噪点显微镜图像去噪能力差的问题；通过引入非局部均值算法，有效保留显微镜图像中的细胞轮廓及结构，避免去噪图像的细胞轮廓模糊；通过引入噪点掩膜图像对显微镜图像噪点进行精准修复，提高了显微镜图像噪点修复的准确性；最后通过非锐化掩码对最终修复图像进行锐化处理，增强高对比度细胞边缘，进而提高显微镜图像去噪质量。
Smart Images
Figure CN122656908A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology for optical microscopy, and in particular to an adaptive multi-scale microscopic image iterative denoising method and system. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Transwell assays are widely used in cell biology, especially in tumor research. The acquired optical microscopic images are inevitably affected by the free channel noise of the Transwell chamber. This noise includes Poisson shot noise from the photodetector and artificial artifacts such as bubbles and dust particles. The above noise directly affects the accuracy of tasks such as cell counting, cell morphology analysis, and subcellular structure localization.
[0004] Existing methods for denoising microscope images mainly include spatial domain filtering, frequency domain filtering, and deep learning methods. Spatial domain filtering methods include mean filtering, median filtering, and Gaussian low-pass filtering. These methods severely damage edges and details, especially when the noise density is high, easily leading to blurred cell membrane contours, making them unsuitable for high-resolution microscopic image processing of subcellular structures. Frequency domain filtering methods include Wiener filtering and band-stop filtering. While frequency domain filtering is effective at suppressing periodic noise, its ability to suppress irregularly distributed random noise is limited, and its parameters depend on manual experience, resulting in insufficient adaptability. Deep learning methods include DnCNN, FFDNet, and Noise2Noise. These methods require a large amount of labeled training data, have poor denoising ability for novel noise points that the model has not learned before, and are computationally intensive, making them unsuitable for rapid interaction in laboratory settings.
[0005] Therefore, the existing microscope image denoising methods mentioned above have limitations such as limited edge protection capabilities, insufficient adaptability, and lack of support for interactive iterative denoising. Summary of the Invention
[0006] To overcome the shortcomings of the prior art, the present invention provides an adaptive multi-scale microscope image iterative denoising method and system, which aims to solve the technical problems in the prior art, such as blurred cell outlines, poor denoising ability for irregular random noise, poor denoising ability for novel noise, large computational load, and lack of support for interactive iterative denoising.
[0007] To achieve the above objectives, the technical solution provided by the present invention is as follows: In a first aspect, the present invention provides an adaptive multi-scale microscope image iterative denoising method, comprising: The original microscope image to be processed is acquired and converted into a grayscale image. The user selects a rectangular area of the target noise on the grayscale image through a graphical interface to obtain a noise template image. The scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noise template image, and multi-scale Laplacian Gaussian feature pyramid response maps are constructed for the grayscale image and the noise template image, respectively. The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The similarity threshold is determined by the maximum inter-class variance method, and the candidate noise point location set is extracted. For each candidate location in the candidate noise location set, calculate the repair radius. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. The original microscope image is denoised using a nonlocal mean algorithm to obtain the base image of the original microscope image; the original microscope image and the base image of the original microscope image are then fused using a noise mask image to obtain the restored image. Repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image; The final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
[0008] As a further technical solution, the scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noisy template image, specifically as follows: The size of the noise template image is h × w The equivalent radius is The basic scale is n scale parameters are generated at logarithmically uniform intervals. , , This is the scale interval coefficient. For the first The scale parameters corresponding to each scale level.
[0009] As a further technical solution, multi-scale Gaussian Laplacian feature pyramid response maps are constructed for both the grayscale image and the noisy template image, specifically as follows: Multi-scale Gaussian Laplacian Feature Pyramid Response Map of Grayscale Images Represented as: ; ; in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of a grayscale image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a grayscale image. Represents the spatial coordinates of pixels in an image; Multi-scale Laplacian Feature Pyramid Response Map of Noisy Template Image Represented as: ; ; in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of the noisy template image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the noise template image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a noise template image. Represents the spatial coordinates of pixels in an image.
[0010] As a further technical solution, a normalized cross-correlation template matching algorithm is used to calculate the full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map. The maximum inter-class variance method is used to determine the similarity threshold, and a set of candidate noise point locations is extracted. Specifically: Multi-scale Gaussian Laplacian feature pyramid response map of grayscale images Multi-scale Laplacian feature pyramid response map for search images and noise template images As a template, The full-image similarity graph S is calculated using the normalized cross-correlation algorithm and is represented as: ; in, Represents the coordinates in the full-image similarity graph. Similarity score at each location; Represents local coordinates within the template image; This represents the response value of the search image at the corresponding local location; This represents the response value of the template at a local location; This represents the average pixel value of the search image within a local window that overlaps with the template. This represents the mean value of all pixels in the template image; The pixel value in the full image similarity map S reflects the similarity interval between the corresponding position and the template as [-1, 1]. All positive value regions in the full image similarity map S are normalized to the interval [0, 255]. The Otsu's method is used to perform optimal threshold segmentation on the normalized full-image similarity map. The threshold is then mapped back to the similarity interval, and the larger value between this threshold and a predetermined threshold is taken as the final detection threshold. All satisfied The positions constitute the candidate noise location set.
[0011] As a further technical solution, a repair radius is calculated for each candidate location in the candidate noise location set. A solid circle is drawn with the center of each noise candidate location set as the center and the repair radius as the radius. The union of these solid circles constitutes the noise mask image. Specifically: Normalize each location in the candidate noise location set and calculate the repair radius. , represented as: ; ; ; in, Indicates the first Repair radius at each location, Indicates the base radius. Indicates the radius shrinkage coefficient. Indicates the first Normalized confidence level at each position Indicates the width of the noise template image. Indicates the height of the noise template image. This indicates a round-down operation. Indicates the first Similarity score for each position This represents the minimum similarity score among the candidate noise location sets. This represents the maximum similarity score among the candidate noise location sets; Based on each noise candidate position Center of the circle The noise mask image is constructed by taking the union of solid circular regions with radius .
[0012] As a further technical solution, a nonlocal mean algorithm is used to denoise the original microscope image to obtain the base image of the original microscope image; a noise mask image is then used to fuse the original microscope image and the base image of the original microscope image to obtain the restored image, specifically: Based on the maximum repair radius in this round Calculate the denoising intensity parameters of the nonlocal means algorithm. , represented as: ; in, Indicates the noise reduction intensity. Indicates the maximum repair radius in this round; With noise reduction intensity parameter Using the filter intensity parameter and combining a preset template window and a preset search window, global denoising is performed on the original microscope image to obtain a base map of the original microscope image, which is represented as follows: ; in, Represented by coordinates The default search window is centered on the user. Represents the coordinates within the search window With current coordinates The similarity weight between them, the similarity weight is expressed as: ; in, Represents pixels within the search window With the current pixel Similarity weights between them This represents a preset template window used to extract local image patches. Indicates a local offset within the template window. This represents the weighting coefficients within the template window. Represented by pixels The pixel value at the corresponding position within the template window, centered Represented by pixels The pixel value at the corresponding position within the template window, centered on the template. Gaussian softening and normalization are applied to the noisy mask image to obtain the fusion weights. , represented as: ; ; in, Representing coordinates Softening mask at the location, The standard deviation is expressed as Gaussian kernel, Represents a noise mask image. Representing coordinates The fusion weight at the point, This represents a very small constant to prevent the denominator from being zero; The original microscope image and the base image are fused pixel by pixel according to the fusion weights to obtain the restored image. , represented as: ; in, This indicates the fused and repaired image in coordinates. Pixel value at that location, Represents the original microscope image in coordinates Pixel value at that location, Representing coordinates The fusion weight at the point, Indicates the base map in coordinates The pixel value at that location.
[0013] As a further technical solution, unsharpened masking is used to sharpen the final repaired image, outputting the final denoised image, specifically as follows: The final restored image is converted from RGB to YUV space, the luminance channel Y is extracted, Gaussian smoothing is applied to Y, the detail layer is calculated, and a threshold mask is generated, represented as follows: ; ; in, Representing coordinates Detail layer pixel values at that location Representing coordinates The original luminance channel pixel value at that location, Representing coordinates The brightness channel pixel value after Gaussian smoothing. Representing coordinates The sharpening threshold mask value at that point, Indicates the global background texture suppression threshold; Gain superposition is performed on the mask-covered area to obtain the sharpened luminance component, represented as: ; in, Indicates the sharpened luminance component in coordinates The pixel value of the brightness channel at that location. Indicates coordinates The original luminance channel pixel value at that location, Indicates the sharpening intensity. Representing coordinates Detail layer pixel values at that location Representing coordinates The sharpening threshold mask value at that location; Sharpen the brightness component With primary chromaticity components The images are merged and converted back to the RGB color space to obtain the final denoised image, represented as follows: ; in, This represents the final denoised image. U represents the sharpened luminance component, and U and V represent the chrominance components.
[0014] Secondly, the present invention provides an adaptive multi-scale microscope image iterative denoising system, comprising: Data acquisition module: used to acquire the original microscope image to be processed and convert it into a grayscale image. The user selects the rectangular area of the target noise on the grayscale image through a graphical interactive interface to obtain the noise template image; Denoising module: used to calculate the scale parameters of the multi-scale Laplacian Gaussian operator based on the noise template image, and to construct multi-scale Laplacian Gaussian feature pyramid response maps for the grayscale image and the noise template image respectively; The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The similarity threshold is determined by the maximum inter-class variance method, and the candidate noise point location set is extracted. For each candidate location in the candidate noise location set, calculate the repair radius. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. The original microscope image is denoised using a nonlocal mean algorithm to obtain the base image of the original microscope image; the original microscope image and the base image of the original microscope image are then fused using a noise mask image to obtain the restored image. Repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image; The final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
[0015] Thirdly, the present invention provides an electronic device including a processor and a memory, wherein the memory is used to store a computer program, the computer program being loaded and executed by the processor to implement an adaptive multi-scale microscope image iterative denoising method as described in the first aspect above.
[0016] Fourthly, the present invention provides a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements an adaptive multi-scale microscope image iterative denoising method as described in the first aspect above.
[0017] As can be seen from the above technical solutions, the present invention has the following beneficial effects: The above-mentioned technical solution of the present invention provides an adaptive multi-scale iterative denoising method and system for microscope images. By using adaptive scale parameters and adaptive similarity thresholds, the method improves the generalization ability of microscope image denoising, avoiding the problem of poor denoising ability of traditional fixed multi-scale parameters and fixed thresholds for microscope images with diverse noise. By introducing a non-local mean algorithm, the method effectively preserves the cell contours and structures in the microscope image, avoiding blurring of cell contours in the denoised image. By introducing a noise mask image to accurately repair noise in the microscope image, the method improves the accuracy of noise repair in microscope images. Finally, the method uses an unsharpened mask to sharpen the final repaired image, enhancing high-contrast cell edges, thereby improving the denoising quality of the microscope image. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of an adaptive multi-scale microscope image iterative denoising method provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram showing the comparison of microscope images before and after denoising, as provided in Embodiment 1 of the present invention. Detailed Implementation
[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0021] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.
[0022] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.
[0023] Example 1 like Figure 1 As shown, this embodiment provides an adaptive multi-scale microscope image iterative denoising method, including: Step 1: Obtain the original microscope image to be processed and convert it into a grayscale image. The user selects the rectangular area of the target noise on the grayscale image through the graphical interface to obtain the noise template image. Step 2: Calculate the scale parameters of the multi-scale Laplacian Gaussian operator based on the noise template image, and construct multi-scale Laplacian Gaussian feature pyramid response maps for the grayscale image and the noise template image respectively; Step 3: Calculate the full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map using the normalized cross-correlation template matching algorithm, determine the similarity threshold using the maximum inter-class variance method, and extract the candidate noise point location set; Step 4: Calculate the repair radius for each candidate location in the candidate noise location set. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. Step 5: Denoise the original microscope image using the nonlocal means algorithm to obtain the base image of the original microscope image; fuse the original microscope image and the base image of the original microscope image using a noise mask image to obtain the restored image; Step 6: Repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop and obtain the final repaired image; Step 7: Sharpen the final repaired image using an unsharpened mask to output the final denoised image.
[0024] The following content provides a more detailed description of the adaptive multi-scale microscope image iterative denoising method provided in this embodiment.
[0025] In step 1, the original image of the microscope to be processed is acquired and converted into a grayscale image. The user selects a rectangular area of the target noise on the grayscale image through a graphical interactive interface to obtain a noise template image.
[0026] In one specific implementation, the process of converting the original microscope image into a grayscale image is as follows: extract the red (R), green (G), and blue (B) color channel values of each pixel in the original microscope image, and calculate the grayscale value corresponding to that pixel using a weighted average method. The specific conversion formula is expressed as follows: ; in, This indicates the coordinates after transformation. The grayscale pixel value at that location; These represent the original microscope image in coordinates. The component values of the red, green, and blue channels at that location.
[0027] In step 2, the scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noise template image, and multi-scale Laplacian Gaussian feature pyramid response maps are constructed for the grayscale image and the noise template image, respectively.
[0028] In one specific implementation, the scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noise template image, specifically as follows: The size of the noise template image is h × w The equivalent radius is The basic scale is n scale parameters are generated at logarithmically uniform intervals. , , This is the scale interval coefficient. For the first The scale parameters corresponding to each scale level.
[0029] Multi-scale Laplacian feature pyramid response maps are constructed for both the grayscale image and the noisy template image, as follows: Multi-scale Gaussian Laplacian Feature Pyramid Response Map of Grayscale Images Represented as: ; ; in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of a grayscale image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a grayscale image. Represents the spatial coordinates of pixels in an image; Multi-scale Laplacian Feature Pyramid Response Map of Noisy Template Image Represented as: ; ; in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of the noisy template image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the noise template image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a noise template image. Represents the spatial coordinates of pixels in an image.
[0030] In step 3, the normalized cross-correlation template matching algorithm is used to calculate the full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map, the maximum inter-class variance method is used to determine the similarity threshold, and the candidate noise point location set is extracted.
[0031] In one specific implementation, a normalized cross-correlation template matching algorithm is used to calculate the full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map. The maximum inter-class variance method is then used to determine the similarity threshold, and a set of candidate noise point locations is extracted. Specifically: Multi-scale Gaussian Laplacian feature pyramid response map of grayscale images Multi-scale Laplacian feature pyramid response map for search images and noise template images As a template, The full-image similarity graph S is calculated using the normalized cross-correlation algorithm and is represented as: ; in, Represents the coordinates in the full-image similarity graph. Similarity score at each location; Represents local coordinates within the template image; This represents the response value of the search image at the corresponding local location; This represents the response value of the template at a local location; This represents the average pixel value of the search image within a local window that overlaps with the template. This represents the mean value of all pixels in the template image; The pixel value in the full image similarity map S reflects the similarity interval between the corresponding position and the template as [-1, 1]. All positive value regions in the full image similarity map S are normalized to the interval [0, 255]. The Otsu algorithm (maximum inter-class variance method) is used to perform optimal threshold segmentation on the normalized full-image similarity map. The threshold is then mapped back to the similarity interval, and the larger value between this threshold and a predetermined threshold is taken as the final detection threshold. All satisfied The positions constitute the candidate noise location set.
[0032] In step 4, the repair radius is calculated for each candidate position in the candidate noise position set. A solid circle is drawn with the center of each noise candidate position set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image.
[0033] In one specific implementation, a repair radius is calculated for each candidate location in the candidate noise location set. A solid circle is drawn with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. Specifically: Normalize each location in the candidate noise location set and calculate the repair radius. , represented as: ; ; ; in, Indicates the first Repair radius at each location, Indicates the base radius. Indicates the radius shrinkage coefficient. Indicates the first Normalized confidence level at each position Indicates the width of the noise template image. Indicates the height of the noise template image. This indicates a round-down operation. Indicates the first Similarity score for each position This represents the minimum similarity score among the candidate noise location sets. This represents the maximum similarity score among the candidate noise location sets; Based on each noise candidate position Center of the circle The noise mask image M is formed by taking the union of solid circular regions with radius M.
[0034] In step 5, the nonlocal mean algorithm is used to denoise the original microscope image to obtain the base image of the original microscope image; the noise mask image is used to fuse the original microscope image and the base image of the original microscope image to obtain the repaired image.
[0035] In one specific implementation, a nonlocal mean algorithm is used to denoise the original microscope image to obtain a background image of the original microscope image; a noise mask image is then used to fuse the original microscope image and the background image of the original microscope image to obtain a restored image, specifically: Based on the maximum repair radius in this round Calculate the denoising intensity parameters of the nonlocal means algorithm. , represented as: ; in, Indicates the noise reduction intensity. Indicates the maximum repair radius in this round; With noise reduction intensity parameter Using the filter intensity parameter and combining a preset template window and a preset search window, global denoising is performed on the original microscope image to obtain a base map of the original microscope image, which is represented as follows: ; in, Represented by coordinates The default search window is centered on the user. Represents the coordinates within the search window With current coordinates The similarity weight between them, the similarity weight is expressed as: ; in, Represents pixels within the search window With the current pixel Similarity weights between them This represents a preset template window used to extract local image patches. Indicates a local offset within the template window. This represents the weighting coefficients within the template window. Represented by pixels The pixel value at the corresponding position within the template window, centered Represented by pixels The value is the pixel value centered within the template window. Therefore, in the nonlocal mean algorithm, the template window explicitly participates in the similarity weight calculation through local image patch differences, while the search window is used to limit the range of candidate similar pixels, thereby generating a base map using redundant information between similar structures in the image. While suppressing noise, it preserves cell outlines and local structures in microscope images; Gaussian softening and normalization are applied to the noisy mask image to obtain the fusion weights. , Specifically, to avoid abrupt changes at the boundaries of the repaired area, the noise mask image is subjected to Gaussian softening to obtain a softened mask, represented as follows: ; in, Representing coordinates Softening mask at the location, The standard deviation is expressed as Gaussian kernel, Represents a noise mask image; The blending weights are obtained by normalizing the softening mask, and are expressed as follows: ; in, Representing coordinates The fusion weight at the point, Representing coordinates Softening mask at the location, This represents a very small constant to prevent the denominator from being zero. ; The original microscope image and the base image are fused pixel by pixel according to the fusion weights to obtain the restored image. , represented as: ; in, This indicates the fused and repaired image in coordinates. Pixel value at that location, Indicates the original microscope image in Pixel value at that location, Representing coordinates The fusion weight at the point, Indicates the base map in coordinates The pixel value at that location.
[0036] In step 6, repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image.
[0037] Specifically, after completing the current round of repair and obtaining the current repaired image, the user re-selects the rectangular area of the target noise based on the current repaired image to update the noise template image; then repeats steps 2 to 5 above until the user confirms that all target noise has been eliminated, exits the iteration loop, and obtains the final repaired image.
[0038] In step 7, the final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
[0039] If at least one noise point is removed in this processing, then unsharpened masking sharpening is applied to the repaired image.
[0040] In one specific implementation, the final repaired image is sharpened using an unsharpened mask to output the final denoised image, specifically as follows: The final restored image is converted from RGB to YUV space, and the luminance channel Y is extracted; the RGB to YUV conversion relationship can be expressed as: ; ; ; Where Y represents the luminance component, and U and V represent the chrominance components. This method performs threshold-controlled unsharpened masking only on the luminance channel Y, while keeping the chrominance channels U and V unchanged, in order to reduce the risk of color shift caused by the sharpening process; Gaussian smoothing is applied to Y, the detail layer is calculated, and a threshold mask is generated, as shown below: ; ; in, Representing coordinates Detail layer pixel values at that location Representing coordinates The original luminance channel pixel value at that location, Representing coordinates The brightness channel pixel value after Gaussian smoothing. Representing coordinates The sharpening threshold mask value at that point, Indicates the global background texture suppression threshold; Gain superposition is performed on the mask-covered area to obtain the sharpened luminance component, represented as: ; in, Indicates the sharpened luminance component in coordinates The pixel value of the brightness channel at that location. Indicates coordinates The original luminance channel pixel value at that location, Indicates the sharpening intensity. Representing coordinates Detail layer pixel values at that location Representing coordinates The sharpening threshold mask value at that location.
[0041] Sharpen the brightness component With primary chromaticity components The images are merged and converted back to the RGB color space to obtain the final denoised image, represented as follows: ; in, This represents the final denoised image. U represents the sharpened luminance component, and U and V represent the chrominance components.
[0042] Therefore, the sharpening process only applies to the high-contrast brightness structure area of the final repaired image and does not participate in the aforementioned multiple rounds of noise detection and repair iterations, thereby avoiding excessive edge enhancement, background texture amplification, or artifact accumulation caused by multiple sharpening processes.
[0043] To visually demonstrate the changes in an image before and after denoising, this application provides, for example... Figure 2 The denoising comparison diagram shown below. Figure 2 The image in the middle left is the image before denoising during microscopic examination. Figure 2 The image on the right in the middle is the denoised image after microscopic image processing, which visually demonstrates the denoising effect of the technical solution of this application.
[0044] Example 2 This embodiment provides an adaptive multi-scale microscope image iterative denoising system, including: Data acquisition module: used to acquire the original microscope image to be processed and convert it into a grayscale image. The user selects the rectangular area of the target noise on the grayscale image through a graphical interactive interface to obtain the noise template image; Denoising module: used to calculate the scale parameters of the multi-scale Laplacian Gaussian operator based on the noise template image, and to construct multi-scale Laplacian Gaussian feature pyramid response maps for the grayscale image and the noise template image respectively; The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The similarity threshold is determined by the maximum inter-class variance method, and the candidate noise point location set is extracted. For each candidate location in the candidate noise location set, calculate the repair radius. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. The original microscope image is denoised using a nonlocal mean algorithm to obtain the base image of the original microscope image; the original microscope image and the base image of the original microscope image are then fused using a noise mask image to obtain the restored image. Repeat the above process until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image; The final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
[0045] Example 3 This embodiment provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement an adaptive multi-scale microscope image iterative denoising method as described in Embodiment 1.
[0046] Example 4 This embodiment provides a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements an adaptive multi-scale microscope image iterative denoising method as described in Embodiment 1.
[0047] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. An adaptive multi-scale microscope image iterative denoising method, characterized in that, include: The original microscope image to be processed is acquired and converted into a grayscale image. The user selects a rectangular area of the target noise on the grayscale image through a graphical interface to obtain a noise template image. The scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noise template image, and multi-scale Laplacian Gaussian feature pyramid response maps are constructed for the grayscale image and the noise template image, respectively. The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The similarity threshold is determined by the maximum inter-class variance method, and the candidate noise point location set is extracted. For each candidate location in the candidate noise location set, calculate the repair radius. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. The original microscope image is denoised using a nonlocal mean algorithm to obtain the base image of the original microscope image; the original microscope image and the base image of the original microscope image are then fused using a noise mask image to obtain the restored image. Repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image; The final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
2. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, The scale parameters of the multi-scale Laplacian Gaussian operator are calculated based on the noisy template image, specifically as follows: The size of the noise template image is h × w The equivalent radius is The basic scale is n scale parameters are generated at logarithmically uniform intervals. , , This is the scale interval coefficient. For the first The scale parameters corresponding to each scale level.
3. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, Multi-scale Laplacian feature pyramid response maps are constructed for both the grayscale image and the noisy template image, as follows: Multi-scale Gaussian Laplacian Feature Pyramid Response Map of Grayscale Images Represented as: ； ； in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of a grayscale image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a grayscale image. Represents the spatial coordinates of pixels in an image; Multi-scale Laplacian Feature Pyramid Response Map of Noisy Template Image Represented as: ； ； in, Coordinates in the multi-scale Gaussian Laplacian feature pyramid response map of the noisy template image The response value at that location, Indicates in The pixel-wise maximum value of the normalized Gaussian Laplacian response at different scale levels. Indicates the noise template image in coordinates The absolute value of the scale-normalized Gaussian Laplace response. Indicates the first The scale parameters corresponding to each scale level Represents the Laplace operator. The scale is represented as The corresponding Gaussian kernel, Represents the convolution operator. Represents a noise template image. Represents the spatial coordinates of pixels in an image.
4. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The maximum inter-class variance method is used to determine the similarity threshold, and a set of candidate noise point locations is extracted. Specifically: Multi-scale Gaussian Laplacian feature pyramid response map of grayscale images Multi-scale Laplacian feature pyramid response map for search images and noise template images As a template, The full-image similarity graph S is calculated using the normalized cross-correlation algorithm and is represented as: ； in, Represents the coordinates in the full-image similarity graph. Similarity score at each location; Represents local coordinates within the template image; This represents the response value of the search image at the corresponding local location; This represents the response value of the template at a local location; This represents the average pixel value of the search image within a local window that overlaps with the template. This represents the mean value of all pixels in the template image; The pixel value in the full image similarity map S reflects the similarity interval between the corresponding position and the template as [-1, 1]. All positive value regions in the full image similarity map S are normalized to the interval [0, 255]. The Otsu's method is used to perform optimal threshold segmentation on the normalized full-image similarity map. The threshold is then mapped back to the similarity interval, and the larger value between this threshold and a predetermined threshold is taken as the final detection threshold. All satisfied The positions constitute the candidate noise location set.
5. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, For each candidate location in the candidate noise location set, calculate the restoration radius. Draw a solid circle with the center of each noise candidate location set as the center and the restoration radius as the radius. The union of these solid circles constitutes the noise mask image. Specifically: Normalize each location in the candidate noise location set and calculate the repair radius. , represented as: ； ； ； in, Indicates the first Repair radius at each location, Indicates the base radius. Indicates the radius shrinkage coefficient. Indicates the first Normalized confidence level at each position Indicates the width of the noise template image. Indicates the height of the noise template image. This indicates a round-down operation. Indicates the first Similarity score for each position This represents the minimum similarity score among the candidate noise location sets. This represents the maximum similarity score among the candidate noise location sets; Based on each noise candidate position Center of the circle The noise mask image is constructed by taking the union of solid circular regions with radius .
6. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, The original microscope image is denoised using a nonlocal means algorithm to obtain the base image of the original microscope image. A noise mask image is then used to fuse the original microscope image and the base image of the original microscope image to obtain the restored image. Specifically: Based on the maximum repair radius in this round Calculate the denoising intensity parameters of the nonlocal means algorithm. , represented as: ； in, Indicates the noise reduction intensity. Indicates the maximum repair radius in this round; With noise reduction intensity parameter Using the filter intensity parameter and combining a preset template window and a preset search window, global denoising is performed on the original microscope image to obtain a base map of the original microscope image, which is represented as follows: ； in, Represented by coordinates The default search window is centered on the user. Represents the coordinates within the search window With current coordinates The similarity weight between them, the similarity weight is expressed as: ； in, Represents pixels within the search window With the current pixel Similarity weights between them This represents a preset template window used to extract local image patches. Indicates a local offset within the template window. This represents the weighting coefficients within the template window. Represented by pixels The pixel value at the corresponding position within the template window, centered Represented by pixels The pixel value at the corresponding position within the template window, centered on the template. Gaussian softening and normalization are applied to the noisy mask image to obtain the fusion weights. , represented as: ； ； in, Representing coordinates Softening mask at the location, The standard deviation is expressed as Gaussian kernel, Represents a noise mask image. Representing coordinates The fusion weight at the point, This represents a very small constant to prevent the denominator from being zero; The original microscope image and the base image are fused pixel by pixel according to the fusion weights to obtain the restored image. , represented as: ； in, This indicates the fused and repaired image in coordinates. Pixel value at that location, Represents the original microscope image in coordinates Pixel value at that location, Representing coordinates The fusion weight at the point, Indicates the base map in coordinates The pixel value at that location.
7. The adaptive multi-scale microscope image iterative denoising method as described in claim 1, characterized in that, The final repaired image is then sharpened using an unsharpened mask to output the final denoised image, as follows: The final restored image is converted from RGB to YUV space, the luminance channel Y is extracted, Gaussian smoothing is applied to Y, the detail layer is calculated, and a threshold mask is generated, represented as follows: ； ； in, Representing coordinates Detail layer pixel values at that location Representing coordinates The original luminance channel pixel value at that location, Representing coordinates The brightness channel pixel value after Gaussian smoothing. Representing coordinates The sharpening threshold mask value at that point, Indicates the global background texture suppression threshold; Gain superposition is performed on the mask-covered area to obtain the sharpened luminance component, represented as: ； in, Indicates the sharpened luminance component in coordinates The pixel value of the brightness channel at that location. Indicates coordinates The original luminance channel pixel value at that location, Indicates the sharpening intensity. Representing coordinates Detail layer pixel values at that location Representing coordinates The sharpening threshold mask value at that location; Sharpen the brightness component With primary chromaticity components The images are merged and converted back to the RGB color space to obtain the final denoised image, represented as follows: ； in, This represents the final denoised image. U represents the sharpened luminance component, and U and V represent the chrominance components.
8. An adaptive multi-scale microscope image iterative denoising system, characterized in that, include: Data acquisition module: used to acquire the original microscope image to be processed and convert it into a grayscale image. The user selects the rectangular area of the target noise on the grayscale image through a graphical interactive interface to obtain the noise template image; Denoising module: used to calculate the scale parameters of the multi-scale Laplacian Gaussian operator based on the noise template image, and to construct multi-scale Laplacian Gaussian feature pyramid response maps for the grayscale image and the noise template image respectively; The full-image similarity map of the multi-scale Gaussian Laplacian feature pyramid response map is calculated using the normalized cross-correlation template matching algorithm. The similarity threshold is determined by the maximum inter-class variance method, and the candidate noise point location set is extracted. For each candidate location in the candidate noise location set, calculate the repair radius. Draw a solid circle with the center of each noise candidate location set as the center and the repair radius as the radius. The union of the solid circles constitutes the noise mask image. The original microscope image is denoised using a nonlocal mean algorithm to obtain the base image of the original microscope image; the original microscope image and the base image of the original microscope image are then fused using a noise mask image to obtain the restored image. Repeat the above steps until the user confirms that all target noise has been eliminated, then exit the iteration loop to obtain the final repaired image; The final repaired image is sharpened using an unsharpened mask, and the final denoised image is output.
9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement an adaptive multi-scale microscope image iterative denoising method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement an adaptive multi-scale microscope image iterative denoising method as described in any one of claims 1-7.