A fluorescence image scattering background suppression method, system, device and storage medium

Abstract

This invention relates to the field of image processing technology, and discloses a method, system, device, and storage medium for suppressing scattering background in fluorescence images. The method includes: performing standardized preprocessing on the original fluorescence image, including full dynamic range normalization, centered square cropping, and symmetrical mirror filling; calculating the pixelated cutoff frequency based on microscope optical system parameters, and constructing a continuous Gaussian filter bank containing one low-pass, n-2 bandpasses, and one high-pass to separate sub-images of each frequency band; denoising the low-frequency band sub-image using an improved dark channel prior algorithm, and denoising the mid-to-high frequency band sub-image using Gaussian smoothing filtering; weighted superposition of the denoised low-frequency band energy and linear superposition of the mid-to-high frequency bands to obtain a preliminary background-removed image; and outputting the result after smoothing, negative value truncation, and normalization. This invention can adapt to different microscope systems and samples, significantly suppress scattering background while preserving details with high fidelity, improving image contrast and signal-to-noise ratio, and is suitable for various fluorescence microscopy imaging scenarios.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to a method, system, device, and storage medium for suppressing background scattering in fluorescence images. Background Technology

[0002] Fluorescence microscopy is a core tool for observing microstructures in life sciences, materials science, and other fields. However, its imaging quality is often severely affected by background scattering. Factors such as endogenous fluorescent impurities in the sample, non-specific staining, stray light from the optical system, and Rayleigh scattering can cause the target signal to be obscured by a blurred background, resulting in low image contrast, decreased signal-to-noise ratio, and even loss of fine structures. This greatly limits the accuracy of subsequent quantitative studies such as cell counting and protein expression analysis.

[0003] Existing scattering background suppression methods have significant limitations: frequency-domain filtering methods often use a single cutoff frequency to segment high and low frequencies, making it difficult to adapt to the background characteristics of different samples such as thick tissue slices and single-layer cells, easily causing high-frequency signal attenuation or low-frequency background residue; spatial domain local denoising methods, such as the dark channel prior algorithm, cannot adapt to the characteristics of different frequency band signals due to fixed background thresholds and suppression strengths, easily leading to blurred target edges or incomplete background suppression; physical model-based deconvolution methods rely on precise point spread function parameters, but in actual imaging, these parameters are easily affected by factors such as sample refractive index, and the calculations are complex, making it difficult to meet the needs of real-time processing of large-scale images; multi-scale fusion methods use a uniform denoising strategy for sub-images at each scale, lacking specificity and making it difficult to balance background suppression and signal fidelity. Summary of the Invention

[0004] In view of this, in order to solve the problem that existing fluorescence image scattering background suppression techniques are unable to achieve both complete suppression of scattering background and high-fidelity preservation of sample structure signals in complex imaging scenarios, this invention provides a fluorescence image scattering background suppression method, system, device and storage medium. Based on the optical diffraction limit theory, the frequency band is adaptively segmented, and different denoising strategies are adopted for the signal and noise characteristics of different frequency bands, which significantly improves the accuracy and robustness of fluorescence image scattering background removal.

[0005] In a first aspect, the present invention provides a method for suppressing background scattering in fluorescence images, comprising:

[0006] The input raw fluorescence image is subjected to normalization preprocessing, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate the image matrix to be processed;

[0007] The pixelation cutoff frequency is calculated based on the microscope optical system parameters, and a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter is constructed based on the pixelation cutoff frequency. The sub-images of each frequency band of the image matrix to be processed are separated based on the Gaussian filter group.

[0008] Different denoising algorithms are used for each frequency band sub-map according to its frequency range, including: for low frequency band sub-maps with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-maps with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising.

[0009] Energy-weighted superposition is performed on the denoised low-frequency band sub-images, and direct linear superposition is performed on the mid- and high-frequency band sub-images. All weighted low-frequency bands and all mid- and high-frequency bands are added together to obtain a preliminary descattered background image.

[0010] The preliminary descattered background image is smoothed, truncated for negative values, and normalized to output the final result.

[0011] The fluorescence image scattering background suppression method provided in this invention standardizes the image reference and eliminates boundary interference through standardized preprocessing, dynamically divides the frequency band using optical parameters to achieve precise separation of signal and background, then uses differentiated denoising to specifically process noise of different frequencies, and finally optimizes through weighted superposition and post-processing. This method solves the problems of fixed frequency band segmentation and lack of specificity in denoising in traditional methods. It can effectively suppress scattering background while preserving sample detail signals, significantly improving the contrast and signal-to-noise ratio of fluorescence images. It is adaptable to various microscopic systems and two-dimensional / three-dimensional images, enhancing the universality and practicality of the method.

[0012] In one alternative implementation, the denoising using the improved dark channel prior algorithm includes: dynamically calculating an initial background value based on the center frequency of the current low-frequency band sub-image, calculating the dark channel image, and selectively suppressing the background using an adaptive background threshold and suppression intensity.

[0013] This invention employs an improved dark channel prior algorithm tailored to the characteristics of low-frequency sub-images. By dynamically calculating initial background values ​​to adapt to different frequency band characteristics, and combining dark channel image calculation, adaptive threshold, and suppression intensity adjustment, it achieves accurate localization and selective suppression of low-frequency scattering background. This avoids background residue or signal damage caused by fixed parameters in traditional dark channel algorithms. While efficiently removing low-frequency background, it better preserves the large structural information of samples in the low-frequency band, thus improving the accuracy and adaptability of low-frequency denoising.

[0014] In one optional implementation, the dynamic initial background value setting of the improved dark channel prior algorithm is achieved by filtering the original image with an ultra-low-pass filter. The cutoff frequency of the ultra-low-pass filter is dynamically adjusted according to the current frequency band. The frequency band range is positively correlated with the cutoff frequency, and the dynamic range of the cutoff frequency is 0.01km-0.1km.

[0015] The calculation of the dark channel includes: calculating the minimum value of each pixel in the current frequency band sub-image within its local neighborhood window to obtain the dark channel image;

[0016] The adaptive background threshold is determined by any one of the following methods: using the Otsu method to automatically calculate a threshold T based on the histogram of the current frequency band sub-map, using an empirical threshold T that dynamically changes with the frequency band, or using a generalized cross-validation method to adaptively determine the background threshold T, in order to distinguish between background and potential signals.

[0017] The suppression intensity is dynamically calculated based on the statistical characteristics of the current frequency band image: , where sigma is the standard deviation of the current frequency band image;

[0018] The selective background suppression includes: performing intensity suppression only on pixels with gray values ​​below a threshold T in the dark channel image at the corresponding positions in the original frequency band sub-image.

[0019]

[0020] in, These are the pixel values ​​of the original low-frequency sub-image. These are the pixel values ​​after noise reduction. I dark These are the pixel values ​​of the dark channel image.

[0021] This invention dynamically adjusts the cutoff frequency of the ultra-low-pass filter to make the initial background value more closely match the characteristics of the current frequency band; the dark channel image calculated by the local neighborhood minimum can accurately capture the background distribution; multiple adaptive threshold methods can flexibly distinguish between the background and the signal; the dynamic suppression intensity formula adjusts the degree of intervention according to the statistical characteristics of the image, combined with a selective suppression strategy, only processing background pixels, minimizing the impact on the effective signal, solving the problem of over-suppression or under-suppression caused by fixed thresholds and intensities in traditional algorithms, and significantly improving the accuracy of low-frequency noise reduction.

[0022] In one optional implementation, the calculation of the pixelation cutoff frequency based on microscope optical system parameters includes:

[0023] Obtain the optical system parameters of the microscope, including: emitted light wavelength λ, objective lens numerical aperture NA, camera effective pixel size b, image pixel count N, and resolution correction factor used to compensate for deviations between the actual system and the ideal model. ;

[0024] The pixelation cutoff frequency is calculated using the following formula based on the acquired parameters:

[0025]

[0026]

[0027] Where R is the physical limit resolution, and the resolution correction factor α ranges from 1.5 to 2.5.

[0028] This invention calculates the pixelated cutoff frequency based on the core optical parameters of a microscope, deeply binding the frequency band segmentation with the physical characteristics of the imaging system. By compensating for actual system deviations through a resolution correction factor, the cutoff frequency km more closely matches the actual imaging resolution limit, avoiding the problem of mismatch between the traditional fixed cutoff frequency and the actual system. This provides a physically meaningful benchmark for subsequent filter bank construction, ensuring that the frequency band division is consistent with the true frequency distribution of the signal and background, and improving the scientific nature and accuracy of frequency band segmentation.

[0029] In one alternative implementation, the weighting coefficient for each low-frequency channel... The weighting coefficient is proportional to its own energy and is calculated using the following formula. :

[0030]

[0031] Where m is the number of low-frequency bands, and E i Let be the energy of the i-th low-frequency band, which is the sum of the squares of all pixel values ​​in that band.

[0032] In this embodiment of the invention, the low-frequency sub-images are superimposed using energy weighting, which gives a larger proportion of low-frequency bands with higher energy (i.e., containing more effective signals) in the reconstruction, while weakening the influence of low-frequency bands with lower energy (mainly background noise). This avoids noise accumulation caused by simple linear superposition, highlights effective low-frequency signals, solves the problem of signal dilution caused by equal weighting of each low-frequency band, improves the signal-to-noise ratio of the low-frequency reconstruction result, and makes the merged image closer to the true low-frequency characteristics of the sample.

[0033] In one optional implementation, the continuous Gaussian filter bank has 10 channels n, covering [0, 0.1 km], [0.1 km, 0.9 km], [0.9 km, ∞], and the filter standard deviation is uniformly distributed in logarithmic space, with the preset threshold being 0.2 km.

[0034] This invention specifies that the number of filter bank channels is 10 and limits the passband range to ensure complete frequency coverage and fine segmentation. The uniform standard deviation in logarithmic space makes the frequency band transition smooth and avoids signal fragmentation. The preset threshold of 0.2km accurately distinguishes high and low frequencies, providing a clear boundary for differentiated denoising. This solves the problem of insufficient targeting of denoising strategies caused by fuzzy frequency band division parameters, making low-frequency background suppression and mid-to-high frequency detail preservation more balanced, and improving the stability and effectiveness of the method.

[0035] In one alternative implementation, the Gaussian convolution kernel used with Gaussian smoothing filtering is 2×2 pixels in size, with a standard deviation σ=1.0, and the symmetric mirror padding width is at least the width of the largest Gaussian convolution kernel used in the Gaussian smoothing filtering.

[0036] The embodiments of this invention employ a 2×2 pixel Gaussian convolution kernel with σ=1.0 to suppress mid-to-high frequency random noise while preserving detailed edges to the maximum extent. The symmetrical mirror fill width matches the maximum convolution kernel, which can completely eliminate edge artifacts during filtering and avoid signal distortion at the boundaries. This solves the problem that mid-to-high frequency filtering easily introduces edge noise or blurring of details, ensuring that the mid-to-high frequency sub-images are clean and undistorted after denoising, providing high-quality data for subsequent overlay and reconstruction, and improving the detail clarity of the final image.

[0037] In a second aspect, the present invention provides a fluorescence image scattering background suppression system, the system comprising:

[0038] The image preprocessing module is used to perform normalization preprocessing on the input raw fluorescence image, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate the image matrix to be processed.

[0039] The frequency band sub-division module is used to calculate the pixelation cutoff frequency based on the microscope optical system parameters, and construct a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter based on the pixelation cutoff frequency, and separate each frequency band sub-map from the image matrix to be processed based on the Gaussian filter group.

[0040] The differential denoising module is used to perform differential denoising on each frequency band sub-map according to its frequency range, including: for low frequency band sub-maps with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-maps with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising.

[0041] The initial descattering background image generation module is used to perform energy-weighted superposition on the denoised low-frequency band sub-images and direct linear superposition on the mid- and high-frequency band sub-images. All weighted low-frequency bands and all mid- and high-frequency bands are added together to obtain the initial descattering background image.

[0042] The final image output module is used to perform smoothing, negative value truncation, and normalization on the preliminary descattered background image and output the final result.

[0043] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the fluorescence image scattering background suppression method of the first aspect or any corresponding embodiment described above.

[0044] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the fluorescence image scattering background suppression method of the first aspect or any corresponding embodiment thereof.

[0045] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the fluorescence image scattering background suppression method of the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0046] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0047] Figure 1 This is a schematic flowchart of a fluorescence image scattering background suppression method according to an embodiment of the present invention;

[0048] Figure 2 This is a flowchart of low-frequency dark channel prior denoising according to an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of frequency band segmentation and reconstruction according to an embodiment of the present invention;

[0050] Figure 4 These are comparison images of the dark channel prior and multi-channel differential background removal effects in wide-field fluorescence microscopy images according to embodiments of the present invention.

[0051] Figure 5 These are comparison images of the dark channel prior and multi-channel differential background removal effects of confocal fluorescence microscopy images according to embodiments of the present invention.

[0052] Figure 6This is a structural block diagram of a fluorescence image scattering background suppression system according to an embodiment of the present invention;

[0053] Figure 7 A schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0056] According to an embodiment of the present invention, a method for suppressing background scattering in a fluorescence image is provided, which can be applied to the energy management system described above. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here. Figure 1 This is a flowchart of a fluorescence image scattering background suppression method according to an embodiment of the present invention, such as... Figure 1 As shown, the specific steps are as follows:

[0057] Step S1 involves standardizing the input raw fluorescence image, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate the image matrix to be processed.

[0058] Specifically, in this embodiment of the invention, the original fluorescence image is preprocessed using standardization. First, full dynamic range normalization is performed to uniformly map the pixel values ​​of the fluorescence image under different imaging conditions (such as differences in exposure time and light source intensity) to the range of [0, 255], eliminating the grayscale scale differences caused by fluctuations in device parameters. This provides a unified numerical benchmark for subsequent frequency band segmentation and denoising algorithms, improving the adaptability of the method to different imaging systems. Then, a centered square crop is performed, specifically cropping the largest inscribed square region based on the image center to eliminate the influence of asymmetric imaging field of view. Finally, symmetrical filling is performed, using mirror filling to extend the square image boundary to a suitable width (the filling width can be selected according to the image size and Gaussian convolution kernel size; the default filling width of 50 is large enough) to completely eliminate the edge effects in subsequent frequency domain filtering. By expanding the symmetrical distribution of pixels at the image edge, the frequency band characteristics of the boundary region during the filtering process are closer to the real sample region, ensuring the accuracy of frequency band sub-image segmentation and providing a more reliable image matrix to be processed for subsequent differential denoising and frequency band reconstruction.

[0059] The synergistic effect of these three elements can standardize the input data to enhance the robustness of the method, optimize the effective area of ​​the image, and eliminate boundary interference, laying the foundation for improving the accuracy of the entire scattering background suppression process.

[0060] Step S2: Calculate the pixelation cutoff frequency based on the microscope optical system parameters, and construct a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter based on the pixelation cutoff frequency. Then, separate the frequency band sub-images of the image matrix to be processed based on the Gaussian filter group.

[0061] Specifically, this embodiment of the invention calculates the inherent spatial frequency response range of the microscope based on its optical system parameters. These optical system parameters include: emitted light wavelength λ (in nm), objective lens numerical aperture NA, effective camera pixel size b (in nm), image pixel count N, and a resolution correction factor used to compensate for deviations between the actual system and the ideal model. (Typically, the value is between 1.5 and 2.5). Further, calculate the physical limit resolution R and pixelation cutoff frequency of the system. :

[0062] (1)

[0063] (2)

[0064] in, This represents the highest effective spatial frequency that the system can express in a digital image after pixel sampling. By dynamically calculating the cutoff frequency through system parameters, the filter bank can adapt to the imaging characteristics of different types of microscopes (such as wide-field, confocal, and super-resolution systems); flexible adjustment of the n value can adapt to different sample complexities (such as single-layer cells and thick tissues), ensuring both segmentation accuracy and computational efficiency. After obtaining the pixelated cutoff frequency... After that, with A continuous Gaussian filter bank with n channels (default n=10) is constructed as a baseline. The frequency threshold range of each filter is empirically determined based on the spectral distribution of the image. This filter bank includes:

[0065] One low-pass filter (LPF): Allows frequencies below 0.1 km to pass without attenuation, with a passband coverage of [0, 0.1 km]. n-2 band-pass filters (BPF): These filters are continuously and logarithmically uniformly distributed in the frequency range of 0.1 km to 0.9 km. For example, when n=10, eight band-pass filters are constructed, with their center frequencies uniformly distributed between 0.1 km and 0.9 km on a logarithmic scale. The standard deviation of each Gaussian filter is set logarithmically uniformly within the passband to ensure relatively balanced energy distribution across the bands after decomposition. One high-pass filter (HPF): Allows frequencies above 0.9 km to pass, with a passband coverage of [0.9 km, ∞].

[0066] This invention utilizes a continuous Gaussian filter bank comprising one low-pass filter, n-2 band-pass filters, and one high-pass filter to perform continuous and non-overlapping layered segmentation of image signals within the frequency range [0,∞]. Compared to single high-low frequency segmentation, this design can capture the specific characteristics of different frequency bands (such as low-frequency slowly varying backgrounds, mid-frequency target edges, and high-frequency detail signals). It is particularly suitable for complex scenes in fluorescence images where the scattering background and target signal spectrum partially overlap. The standard deviation of the filter bank is uniformly distributed in logarithmic space, and the passband coverage is complete and continuous. This avoids signal distortion caused by frequency band fragmentation and ensures the progressiveness of frequency characteristics of each frequency band sub-image, creating conditions for the targeted application of subsequent differentiated denoising algorithms (such as low-frequency strong background suppression and mid-to-high-frequency weak signal protection), while also improving signal integrity during frequency band reconstruction.

[0067] Step S3: Differentiated denoising is performed on each frequency band sub-map according to its frequency range, including: for low-frequency band sub-maps with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-maps with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising.

[0068] Specifically, the low-frequency components in fluorescence images mainly consist of a uniform background and slowly changing signals. Noise here manifests as large-area, low-intensity background fluctuations. The dark channel prior algorithm is based on a natural statistical law: in most non-background local regions, there are always some pixels with a very low channel value (close to black). By calculating the "dark channel" of the image, the distribution of background noise can be accurately estimated.

[0069] This invention employs an improved dark channel prior denoising algorithm. Its core lies in introducing a frequency band adaptive initial background value and noise suppression threshold, replacing the fixed extremely low-frequency components and thresholds used in traditional methods. Figure 2 As shown, the specific steps include:

[0070] 1. Dynamic Initial Background Value Setting: A dynamic initial background value is independently calculated for each frequency band, replacing the traditional approach of using a fixed extremely low-frequency component. Specifically, an extremely low-pass filter is used to filter the original image, and its cutoff frequency is dynamically adjusted according to the current frequency band. The frequency band range is positively correlated with the cutoff frequency, following the principle that "the higher the frequency band, the larger the cutoff frequency" (the dynamic range of the cutoff frequency is 0.01 km ~ 0.1 km). This method better reflects the physical reality that the scattered background is non-uniformly distributed across different frequency components.

[0071] 2. Calculate the dark channel: Calculate the current frequency band sub-map Dark channel image For each pixel, the outer rectangle of its local neighborhood window Ω(x) (the system point spread function) is used as the dark channel size, or fixed as... Find the minimum value within a pixel:

[0072] (3)

[0073] 3. Adaptive threshold calculation: The Otsu method is adopted, based on... This invention provides three methods to distinguish between background and potential signals: automatically calculating a threshold T using a histogram, using a dynamically changing empirical threshold T based on the frequency band, and adaptively determining a background threshold T using a generalized cross-validation method. The Otsu method automatically segments the background and signal using a histogram, making it suitable for scenarios with significant differences in grayscale distribution. The dynamic empirical threshold adjusts with frequency band characteristics to adapt to the intensity differences of backgrounds at different frequencies. Generalized cross-validation requires no prior knowledge, making it suitable for complex noise distributions and solving the problem of insufficient adaptability of single threshold methods. All three methods calculate the threshold based on the characteristics of the frequency band sub-map itself, avoiding "oversegmentation" or "undersegmentation" caused by a fixed threshold.

[0074] 4. Dynamic Suppression Intensity: The suppression intensity dep is not a fixed value, but is dynamically calculated based on the statistical characteristics of the current frequency band image. sigma is the standard deviation of the current frequency band image. This ensures stronger suppression of backgrounds with large fluctuations.

[0075] 5. Selective background suppression: only for... Pixels with gray values ​​below the threshold T (i.e., "potential background" pixels) in the original image Intensity suppression is applied to the corresponding positions:

[0076] (4)

[0077] in, These are the pixel values ​​of the original low-frequency sub-image. These are the pixel values ​​after denoising. Non-background pixels are retained. Intensity suppression is applied only to "potential background" pixels with grayscale values ​​below the threshold T in the dark channel image, while non-background pixels (i.e., effective signal pixels carrying sample structural information) are completely preserved. This fundamentally solves the problem that traditional global denoising algorithms easily lead to the mis-suppression of effective signals (such as weak fluorescence structures and fine textures), achieving a precise balance between background removal and signal fidelity.

[0078] Subsequent steps are similar to those of traditional dark channel prior algorithms, including calculating the image's transmittance matrix based on the suppressed low-frequency sub-image, refining the transmittance matrix, combining the optimized transmittance matrix with the initial background value, and recovering the original signal. These steps are consistent with the workflow of traditional dark channel prior algorithms, ensuring that the signal recovery process of the low-frequency sub-image conforms to mature image restoration logic within the differential denoising framework. Simultaneously, the selective background suppression in the early stages compensates for the shortcomings of traditional algorithms in adapting to frequency band characteristics, ultimately achieving accurate suppression of low-frequency background and high-fidelity preservation of the effective signal.

[0079] The mid-to-high frequency components in fluorescence images contain detailed information such as image edges and textures, but noise is also mainly concentrated here, exhibiting rapid and random variations. Gaussian smoothing filtering is used for processing: a small Gaussian convolution kernel (recommended size 2×2 pixels, standard deviation σ=1.0) is used to convolve the mid-to-high frequency sub-image. This operation aims to smooth and suppress random high-frequency noise, while, due to the favorable properties of the Gaussian kernel, it can preserve the sharpness of details to the maximum extent. This process only performs smoothing filtering and does not involve any subtraction operations, fundamentally avoiding the introduction of negative values ​​or attenuation of the true signal.

[0080] Step S4: Perform energy-weighted superposition on the denoised low-frequency band sub-images, and directly linearly superimpose the mid-to-high frequency band sub-images. Add all the weighted low-frequency bands and all mid-to-high frequency bands together to obtain a preliminary descattered background image.

[0081] This step involves fusing the sub-images of each frequency band after differential denoising into a complete denoised image. For the low-frequency components, this embodiment of the invention employs a weighted superposition strategy, with weight coefficients for each low-frequency channel. It is proportional to its own energy (i.e., the sum of the squares of all pixel values), that is: , where m is the number of low-frequency bands. This allows higher-energy bands (which typically contain more valid information) to have a larger proportion in the final result. For the mid-to-high frequency components, there is usually more noise (based on experience with the noise distribution of fluorescence images, low frequencies can be identified as background noise, while mid-to-high frequencies contain various types of noise, including the real signal and shot noise). For example, shot noise might be amplified by energy weighting; therefore, linear superposition is performed directly. Finally, all weighted low-frequency bands are added to all mid-to-high frequency bands to obtain a preliminary descattered background image.

[0082] The differentiated strategies of low-frequency weighted superposition and mid-to-high-frequency linear superposition are adapted to the signal and noise characteristics of the two frequency bands respectively. This approach eliminates low-frequency background interference and highlights large structures, while protecting mid-to-high-frequency details and suppressing random noise. The preliminary background-removed image formed by adding the two strategies achieves balanced integration of signals across the entire frequency band, avoiding the "low-frequency blurring" or "high-frequency noise" caused by a single superposition method. This significantly improves the image's contrast, signal-to-noise ratio, and detail fidelity, laying a high-quality foundation for subsequent post-processing. Figure 3 The diagram shown illustrates frequency band segmentation and multi-channel differential denoising.

[0083] Step S5: Perform smoothing, negative value truncation, and normalization on the preliminary descattered background image, and output the final result.

[0084] This step involves the final optimization of the reconstructed image, specifically:

[0085] 1. Smoothing: A very small Gaussian kernel (size 2×2 pixels, σ≈0.5) is used for global slight smoothing to eliminate the slight discontinuity effect that may be caused by frequency band reconstruction.

[0086] 2. Negative value truncation: Truncates all negative pixel values ​​in the image to 0, eliminating physically meaningless negative values ​​that may be generated during the noise reduction process.

[0087] 3. Normalization: The pixel values ​​of the final image are linearly transformed back to the standard range of 16-bit integers [0, 65535] for easy display, storage and subsequent quantitative analysis.

[0088] The method provided in this invention is applicable to two-dimensional or three-dimensional images acquired by wide-field fluorescence microscopy, confocal microscopy, light field microscopy, light sheet microscopy, and super-resolution fluorescence microscopy systems. When processing a three-dimensional image stack, steps S1-S5 are executed sequentially for each frame of two-dimensional image in the stack, and parallel computing is used to accelerate the processing. In a wide-field fluorescence image processing embodiment, the method of this invention is applied to perform the following process:

[0089] 1. Input image: Wide-field fluorescence image of animal tissue section, size 1800×1800 pixels, 16-bit depth, stored in TIFF format.

[0090] 2. Obtain optical system parameters: excitation wavelength (λ) = 550 nm; objective lens numerical aperture (NA) = 0.6; pixel size (b) = 0.095 µm; number of image pixels (N) = 1800; resolution correction factor (α) = 2.0.

[0091] 3. Cutoff frequency calculation: The pixelation cutoff frequency km ≈ 611 pixels is calculated according to formula (2).

[0092] 4. Filter bank construction: Based on this, a 10-channel Gaussian filter bank is constructed, consisting of one low-pass filter (0-0.1km), eight band-pass filters (0.1km to 0.9km), and one high-pass filter (>0.9km). The standard deviation of the filters is uniformly distributed in logarithmic space.

[0093] 5. Differentiated noise reduction processing:

[0094] (1) Low-frequency band (bands 1-2): An improved dark channel prior algorithm is adopted. A 23×23 pixel local window is used to calculate the dark channel, and the Otsu method is used to adaptively determine the background threshold T. The suppression intensity dep is calculated according to the formula... The calculation is dynamic, and the intensity suppression is only applied to pixels that are determined to be background.

[0095] (2) Mid-to-high frequency band (band 3-10): Gaussian low-pass filtering is used for smoothing. The filter kernel size is 2×2 pixels and the standard deviation σ=1.0, so as to suppress noise while preserving tissue texture structure.

[0096] (3) Frequency band reconstruction: The low frequency band after noise reduction is weighted and superimposed according to its energy (the sum of squares of the pixel values ​​of each frequency band image), while the mid-to-high frequency bands are directly superimposed linearly.

[0097] 6. Post-processing: A slight Gaussian smoothing with a filter kernel size of 2×2 pixels and a standard deviation of σ=0.5 is used to eliminate block artifacts, all negative pixel values ​​are truncated to 0, and the final image intensity is linearly transformed to the range of 16-bit integer values ​​[0, 65535].

[0098] Implementation Environment and Results: Completed on the MATLAB R2019a platform, running on a computer equipped with an Intel i7-10700K processor and 64GB of memory. Processing time was approximately 12 seconds. Figure 4 As shown, the background fluorescence of the processed image is effectively suppressed, and the tissue structure is more clearly visible. Compared with the dark channel prior background removal algorithm (the algorithm's direct output is used for comparison in the figure without contrast stretching post-processing), this method removes background noise more thoroughly and the tissue structure is clearer.

[0099] In a co-focused fluorescence image processing embodiment, the following process is performed using the method of the present invention:

[0100] 1. Input image: confocal fluorescence image, 800×800 pixels, 16-bit depth.

[0101] 2. Obtain optical system parameters: emitted light wavelength (λ) = 595 nm; objective lens numerical aperture (NA) = 0.95; pixel size (b) = 0.095 µm; resolution correction factor (α) = 2.0.

[0102] 3. Cutoff frequency calculation: The calculated value is approximately 895 pixels (km).

[0103] 4. Filter bank construction: Construct a 10-channel filter bank containing 1 low-pass filter, 8 band-pass filters, and 1 high-pass filter.

[0104] 5. Differentiated noise reduction processing:

[0105] (1) Low frequency band: The local window for dark channel prior denoising is adjusted to 16×16 pixels to adapt to the characteristics of confocal images with low noise and more uniform background.

[0106] (2) Mid-to-high frequency band: The Gaussian smoothing kernel size is reduced to 2×2 pixels, σ=0.5, to more finely protect the high-resolution details unique to confocal images.

[0107] (3) Frequency band reconstruction: The low frequency band after noise reduction is weighted and superimposed according to its energy (the sum of squares of the pixel values ​​of each frequency band image), while the mid-to-high frequency bands are directly superimposed linearly.

[0108] 6. Post-processing: A slight Gaussian smoothing with a filter kernel size of 2×2 pixels and a standard deviation of σ=0.5 is used to eliminate block artifacts, all negative pixel values ​​are truncated to 0, and the final image intensity is linearly transformed to the range of 16-bit integer values ​​[0, 65535].

[0109] Implementation environment and results: Implemented in MATLAB environment as well. Processing time is approximately 4 seconds. Figure 5 As shown, the processed image effectively suppresses the background while perfectly preserving the cell structure, with no obvious loss of detail or excessive smoothing. Compared with the dark channel prior background removal algorithm (the figure uses the direct output result of the algorithm as a comparison, without contrast stretching post-processing), the method provided by the embodiment of the present invention removes background noise more thoroughly and the cell structure is clearer.

[0110] This embodiment also provides a fluorescence image scattering background suppression system, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0111] This embodiment provides a fluorescence image scattering background suppression system, such as Figure 6 As shown, it includes:

[0112] Image preprocessing module 61 is used to perform normalization preprocessing on the input raw fluorescence image, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate an image matrix to be processed.

[0113] The frequency band sub-division module 62 is used to calculate the pixelation cutoff frequency based on the microscope optical system parameters, and construct a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter based on the pixelation cutoff frequency, and separate each frequency band sub-map of the image matrix to be processed based on the Gaussian filter group.

[0114] The differential denoising module 63 is used to perform differential denoising for each frequency band sub-map according to its frequency range, including: for low frequency band sub-maps with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-maps with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising.

[0115] The preliminary descattering background image generation module 64 is used to perform energy-weighted superposition on the denoised low-frequency band sub-images, and to directly linearly superimpose the mid- and high-frequency band sub-images. All weighted low-frequency bands and all mid- and high-frequency bands are added together to obtain the preliminary descattering background image.

[0116] The final image output module 65 is used to perform smoothing, negative value truncation and normalization on the preliminary descattered background image and output the final result.

[0117] In some optional implementations, the differential denoising module 63 employs an improved dark channel prior algorithm for denoising, which includes: dynamically calculating an initial background value based on the center frequency of the current low-frequency band sub-image, calculating the dark channel image, and selectively suppressing the background using an adaptive background threshold and suppression intensity.

[0118] In some optional implementations, the dynamic initial background value setting of the improved dark channel prior algorithm is achieved by filtering the original image with an ultra-low-pass filter. The cutoff frequency of the ultra-low-pass filter is dynamically adjusted according to the current frequency band. The frequency band range is positively correlated with the cutoff frequency, and the dynamic range of the cutoff frequency is 0.01km-0.1km.

[0119] The calculation of the dark channel includes: calculating the minimum value of each pixel in the current frequency band sub-image within its local neighborhood window to obtain the dark channel image;

[0120] The adaptive background threshold is determined by any one of the following methods: using the Otsu method to automatically calculate a threshold T based on the histogram of the current frequency band sub-map, using an empirical threshold T that dynamically changes with the frequency band, or using a generalized cross-validation method to adaptively determine the background threshold T, in order to distinguish between background and potential signals.

[0121] The suppression intensity is dynamically calculated based on the statistical characteristics of the current frequency band image: , where sigma is the standard deviation of the current frequency band image;

[0122] The selective background suppression includes: performing intensity suppression only on pixels with gray values ​​below a threshold T in the dark channel image at the corresponding positions in the original frequency band sub-image.

[0123]

[0124] in, These are the pixel values ​​of the original low-frequency sub-image. These are the pixel values ​​after noise reduction. I dark These are the pixel values ​​of the dark channel image.

[0125] In some optional implementations, the frequency band sub-division module 62 calculates the pixelation cutoff frequency based on microscope optical system parameters, including:

[0126] Obtain the optical system parameters of the microscope, including: emitted light wavelength λ, objective lens numerical aperture NA, camera effective pixel size b, image pixel count N, and resolution correction factor used to compensate for deviations between the actual system and the ideal model. ;

[0127] The pixelation cutoff frequency is calculated using the following formula based on the acquired parameters:

[0128]

[0129]

[0130] Where R is the physical limit resolution, and the resolution correction factor α ranges from 1.5 to 2.5.

[0131] In some alternative implementations, the weighting coefficient for each low-frequency channel... The weighting coefficient is proportional to its own energy and is calculated using the following formula. :

[0132]

[0133] Where m is the number of low-frequency bands, and E i Let be the energy of the i-th low-frequency band, which is the sum of the squares of all pixel values ​​in that band.

[0134] In one optional implementation, the continuous Gaussian filter bank has 10 channels n, covering [0, 0.1 km], [0.1 km, 0.9 km], [0.9 km, ∞], and the filter standard deviation is uniformly distributed in logarithmic space, with the preset threshold being 0.2 km.

[0135] In one alternative implementation, the Gaussian convolution kernel used with Gaussian smoothing filtering is 2×2 pixels in size, with a standard deviation σ=1.0, and the symmetric mirror padding width is at least the width of the largest Gaussian convolution kernel used in the Gaussian smoothing filtering.

[0136] The fluorescence image scattering background suppression system provided in this embodiment of the invention can execute the fluorescence image scattering background suppression method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0137] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0138] The following is a detailed reference. Figure 7This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0139] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0140] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the nuclear security network design data operation and maintenance delivery and automatic configuration method of the embodiments of the present invention.

[0141] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0142] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the nuclear security network design data operation and maintenance delivery and automatic configuration method shown in the above embodiments is implemented.

[0143] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0144] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for suppressing background scattering in fluorescence images, characterized in that, include: The input raw fluorescence image is subjected to normalization preprocessing, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate the image matrix to be processed; The pixelation cutoff frequency is calculated based on the microscope optical system parameters, and a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter is constructed based on the pixelation cutoff frequency. The sub-images of each frequency band of the image matrix to be processed are separated based on the Gaussian filter group. The calculation of the pixelation cutoff frequency based on microscope optical system parameters includes: Obtain the optical system parameters of the microscope, including: emitted light wavelength λ, objective lens numerical aperture NA, camera effective pixel size b, image pixel count N, and resolution correction factor used to compensate for deviations between the actual system and the ideal model. ; The pixelation cutoff frequency is calculated using the following formula based on the acquired parameters: Where R is the physical limit resolution, and the resolution correction factor α ranges from 1.5 to 2.5; Different denoising algorithms are used for each frequency band sub-image based on its frequency range, including: for low-frequency band sub-images with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-images with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising. The improved dark channel prior algorithm for denoising includes: dynamically calculating the initial background value based on the center frequency of the current low-frequency band sub-image, calculating the dark channel image, and selectively suppressing the background using an adaptive background threshold and suppression intensity. The dynamic initial background value setting of the improved dark channel prior algorithm is achieved by filtering the original image with an ultra-low pass filter. The cutoff frequency of the ultra-low pass filter is dynamically adjusted according to the current frequency band. The frequency band range is positively correlated with the cutoff frequency, and the dynamic range of the cutoff frequency is 0.01km-0.1km. The calculation of the dark channel includes: calculating the minimum value of each pixel in the current frequency band sub-image within its local neighborhood window to obtain the dark channel image; The adaptive background threshold is determined by any one of the following methods: using the Otsu method to automatically calculate a threshold T based on the histogram of the current frequency band sub-map, using an empirical threshold T that dynamically changes with the frequency band, or using a generalized cross-validation method to adaptively determine the background threshold T, in order to distinguish between background and potential signals. The suppression intensity is dynamically calculated based on the statistical characteristics of the current frequency band image: , where sigma is the standard deviation of the current frequency band image; The selective background suppression includes: performing intensity suppression only on pixels with gray values ​​below a threshold T in the dark channel image at the corresponding positions in the original frequency band sub-image. in, These are the pixel values ​​of the original low-frequency sub-image. These are the pixel values ​​after noise reduction. I dark These are the pixel values ​​of the dark channel image; Energy-weighted superposition is performed on the denoised low-frequency band sub-images, and direct linear superposition is performed on the mid- and high-frequency band sub-images. All weighted low-frequency bands and all mid- and high-frequency bands are added together to obtain a preliminary descattered background image. The preliminary descattered background image is smoothed, truncated for negative values, and normalized to output the final result.

2. The method according to claim 1, characterized in that, Weighting coefficient for each low-frequency channel The weighting coefficient is proportional to its own energy and is calculated using the following formula. : Where m is the number of low-frequency bands, E i Let be the energy of the i-th low-frequency band, which is the sum of the squares of all pixel values ​​in that band.

3. The method according to claim 1, characterized in that, The continuous Gaussian filter bank has 10 channels, covering [0, 0.1 km], [0.1 km, 0.9 km], [0.9 km, ∞], and the filter standard deviation is uniformly distributed in logarithmic space. The preset threshold is 0.2 km.

4. The method according to claim 1, characterized in that, When using Gaussian smoothing filtering, the Gaussian convolution kernel is 2×2 pixels in size, with a standard deviation σ=1.0, and the symmetric mirror padding width is at least the width of the largest Gaussian convolution kernel used in Gaussian smoothing filtering.

5. A fluorescence image scattering background suppression system, characterized in that, include: The image preprocessing module is used to perform normalization preprocessing on the input raw fluorescence image, including full dynamic range normalization, square cropping based on the image center, and symmetrical mirror filling, to generate the image matrix to be processed. The frequency band sub-division module is used to calculate the pixelation cutoff frequency based on the microscope optical system parameters, and construct a continuous Gaussian filter group including one low-pass filter, n-2 band-pass filters and one high-pass filter based on the pixelation cutoff frequency, and separate each frequency band sub-map from the image matrix to be processed based on the Gaussian filter group. The calculation of the pixelation cutoff frequency based on microscope optical system parameters includes: Obtain the optical system parameters of the microscope, including: emitted light wavelength λ, objective lens numerical aperture NA, camera effective pixel size b, image pixel count N, and resolution correction factor used to compensate for deviations between the actual system and the ideal model. ; The pixelation cutoff frequency is calculated using the following formula based on the acquired parameters: Where R is the physical limit resolution, and the resolution correction factor α ranges from 1.5 to 2.5; The differential denoising module is used to perform differential denoising on each frequency band sub-image according to its frequency range, including: for low-frequency band sub-images with frequencies less than a preset threshold, an improved dark channel prior algorithm is used for denoising; for mid-to-high frequency band sub-images with frequencies not less than a preset threshold, Gaussian smoothing filtering is used for denoising; the improved dark channel prior algorithm for denoising includes: dynamically calculating the initial background value based on the center frequency of the current low-frequency band sub-image, calculating the dark channel image, and selectively suppressing the background using an adaptive background threshold and suppression intensity; The dynamic initial background value setting of the improved dark channel prior algorithm is achieved by filtering the original image with an ultra-low pass filter. The cutoff frequency of the ultra-low pass filter is dynamically adjusted according to the current frequency band. The frequency band range is positively correlated with the cutoff frequency, and the dynamic range of the cutoff frequency is 0.01km-0.1km. The calculation of the dark channel includes: calculating the minimum value of each pixel in the current frequency band sub-image within its local neighborhood window to obtain the dark channel image; The adaptive background threshold is determined by any one of the following methods: using the Otsu method to automatically calculate a threshold T based on the histogram of the current frequency band sub-map, using an empirical threshold T that dynamically changes with the frequency band, or using a generalized cross-validation method to adaptively determine the background threshold T, in order to distinguish between background and potential signals. The suppression intensity is dynamically calculated based on the statistical characteristics of the current frequency band image: , where sigma is the standard deviation of the current frequency band image; The selective background suppression includes: performing intensity suppression only on pixels with gray values ​​below a threshold T in the dark channel image at the corresponding positions in the original frequency band sub-image. in, These are the pixel values ​​of the original low-frequency sub-image. These are the pixel values ​​after noise reduction. I dark These are the pixel values ​​of the dark channel image; The initial descattering background image generation module is used to perform energy-weighted superposition on the denoised low-frequency band sub-images and direct linear superposition on the mid- and high-frequency band sub-images. All weighted low-frequency bands and all mid- and high-frequency bands are added together to obtain the initial descattering background image. The final image output module is used to perform smoothing, negative value truncation, and normalization on the preliminary descattered background image and output the final result.

6. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the fluorescence image scattering background suppression method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the fluorescence image scattering background suppression method according to any one of claims 1 to 4.

Images