Weak light sub-image denoising method and device based on bright spot protection and variance stabilization

Abstract

The present application relates to the technical field of image processing, and discloses a weak-light sub-image denoising method and device based on bright spot protection and variance stabilization, which comprises the following steps: detecting bright spots of a to-be-processed image, and generating a soft protection mask based on morphological constraints; correcting the background of the to-be-processed image by using the soft protection mask to obtain a background correction image, and performing transformation and stable field denoising on the background correction image to obtain a denoised image of the to-be-processed image; and fusing the background correction image and the denoised image based on the soft protection mask to obtain an enhanced image. The present application generates a soft protection mask by bright spot detection combined with morphological constraints, accurately distinguishes real bright spots from noise artifacts, and the background correction result is more accurate; the noise variance is balanced by Anscombe transformation, the stable field denoising can efficiently suppress noise without damaging weak details; and then the two images are adaptively fused by relying on the soft protection mask to improve the image visibility.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method and apparatus for denoising weak photon images based on bright spot protection and variance stabilization. Background Technology

[0002] Under weak photon imaging conditions, the image is simultaneously superimposed with multiple noises and distortion interferences: First, affected by Poisson statistical noise, detector dark noise, and readout noise, the image background fluctuates significantly, and weak structures are easily submerged by noise; Second, factors such as non-uniform illumination, scattering background, and long-exposure substrate drift in the imaging system introduce gradually changing backgrounds, reducing the contrast between the target and the background; Third, high bright spots, small-scale fluorescence hotspots, or local real structures in the image are not only important information for subsequent analysis, but are also easily mistaken for outliers or noise during background estimation and denoising, resulting in signal loss, boundary blunting, and decreased contrast, which seriously affects the accuracy of subsequent image quantification analysis.

[0003] Existing methods typically employ single background estimation, single-step filtering, or direct use of nonlocal means, wavelet denoising, etc. However, these methods generally suffer from the following shortcomings: First, the background estimation stage does not protect bright targets, easily incorporating real bright spots into the background model, resulting in over-subtraction of the background and loss of weak structure signals. Second, the denoising stage does not fully consider the statistical characteristics of weak photon images, making it difficult to balance strong noise suppression and weak structure preservation when filtering directly in the original grayscale domain. Third, different processing modules operate independently, lacking a joint design that addresses the relationship between bright spot structures and background noise, leading to insufficient detail preservation, increased local artifacts, or unstable overall visual quality. Summary of the Invention

[0004] This invention provides a weak photon image denoising method and apparatus based on bright spot protection and variance stabilization to solve the problems of inaccurate background estimation, insufficient bright spot protection, easy loss of weak structures, and excessive smoothing after denoising.

[0005] In a first aspect, the present invention provides a weak photon image denoising method based on bright spot protection and variance stabilization, the method comprising: Bright spot detection is performed on the image to be processed, and a soft protective mask is generated based on morphological constraints; A soft protective mask is used to perform background correction on the image to be processed, resulting in a background-corrected image. Anscombe transform is then performed on the background-corrected image to obtain a standard image. The standard image is denoised in the variance stabilization domain to obtain the stabilized domain image, and the stabilized domain image is then subjected to inverse Anscombe transform to obtain the denoised image of the image to be processed. A bright spot protection fusion is performed on the background correction image and the denoised image based on a soft protective mask to obtain an enhanced image.

[0006] The present invention provides a weak photon image denoising method based on bright spot protection and variance stabilization. It generates a soft protective mask by combining bright spot detection with morphological constraints, accurately distinguishing between real bright spots and noise artifacts, avoiding interference from bright targets in background modeling, and resulting in more accurate background correction. Anscombe transform is used to achieve noise variance equalization, adapting to the Poisson noise characteristics of weak photon images. Stabilization domain denoising effectively suppresses noise without damaging subtle details. Furthermore, the soft protective mask adaptively fuses two images, preserving the original signal in bright spot areas and focusing on noise reduction output in non-protected areas, resulting in a smooth transition without boundary artifacts. This method effectively suppresses background undulations and image noise while fully preserving minute bright structures and edge information, improving image visibility.

[0007] In one alternative implementation, the method further includes: The enhanced image is lightly sharpened to obtain the final image; Calculate the index parameter values ​​of the final image and evaluate the image processing effect based on the index parameter values.

[0008] The weak photon image denoising method based on bright spot protection and variance stabilization provided by this invention compensates for edge softening that may be introduced during the denoising process through lightweight sharpening, thereby improving edge clarity and overall visibility. The image processing effect is evaluated by calculating index parameters to ensure that the image processing meets the requirements, and the solution can also be improved based on the image processing effect.

[0009] In one alternative implementation, bright spot detection is performed on the image to be processed, and a soft protective mask is generated based on morphological constraints, including: If the image to be processed is a color image, then the image to be processed is converted into a grayscale image and the data format is unified before being used as the input image; The input image is Gaussian smoothed to obtain a pre-smoothed image. High-frequency components are constructed based on the pre-smoothed image, and the noise intensity is estimated using the median absolute deviation to obtain a coarse noise estimate. A segmentation threshold is constructed based on the global mean and coarse noise estimate of the pre-smoothed image, and a candidate bright spot mask is generated based on the segmentation threshold; Based on morphological constraints, candidate bright spot masks are screened for bright spots and their morphology is expanded. A soft protective mask is then generated by Gaussian smoothing.

[0010] The present invention provides a weak photon image denoising method based on bright spot protection and variance stabilization. It unifies data standards to eliminate color channel interference, extracts high-frequency bright spots through multi-scale differential extraction, accurately estimates noise through median absolute deviation, obtains candidate masks through threshold segmentation, and generates bright spot detection based on the image's own noise and brightness adaptively, making it more versatile. Furthermore, it filters out noisy false bright spots by relying on morphological constraints such as connected region area and eccentricity, and generates a gradient soft mask through dilation and Gaussian smoothing, which not only significantly reduces the false detection rate of bright spots, but also eliminates the boundary abrupt artifacts caused by hard masks.

[0011] In one optional implementation, candidate bright spot masks are screened and their shapes are dilated based on morphological constraints, and a soft protective mask is generated by Gaussian smoothing, including: Extract the morphological parameters of each position of the candidate bright spot mask, and filter the candidate bright spot regions based on morphological constraints; Based on the candidate bright spot regions, the candidate bright spot mask is subjected to morphological dilation processing to obtain a dilated protective mask, and the dilated protective mask is then smoothed to obtain a soft protective mask.

[0012] The weak photon image denoising method based on bright spot protection and variance stabilization provided by this invention extracts the morphological parameters of connected components and sets constraints to filter out false bright spots formed by noise, thereby improving the accuracy of real bright spot recognition. The filtered region is expanded to increase the protection range, and then Gaussian smoothing is used to generate a gradient soft mask to eliminate the jagged tortuous boundary of the hard mask and achieve a smooth transition of region weights.

[0013] In one optional implementation, a soft protective mask is used to perform background correction on the image to be processed to obtain a background-corrected image, including: The median filtering result is used to replace and fill the area covered by the soft protective mask to obtain the filled image; A large-scale Gaussian smoothing process is applied to the filled image to obtain a gradually changing background image. The gradually varying background image is subtracted from the image to be processed, and the negative values ​​after subtraction are truncated to zero to obtain the background-corrected image.

[0014] The weak photon image denoising method based on bright spot protection and variance stabilization provided by this invention uses the median filter value of a soft protective mask to replace the bright spot area to obtain a filled image, avoiding the participation of real bright signals in background fitting and avoiding background overestimation; the gradually changing background is accurately restored by large-scale Gaussian smoothing, and the original image is corrected after removing the background and truncating negative values, which significantly reduces the interference of bright targets on background fitting, completely preserves weak target signals, improves the accuracy of background subtraction, and obtains a background-corrected image with higher contrast and no negative pixel interference.

[0015] In one optional implementation, the standard image is denoised in the variance stabilization domain to obtain a stabilized domain image, and the stabilized domain image is subjected to an inverse Anscombe transform to obtain a denoised image of the image to be processed, including: Denoising is performed in the variance stabilization domain using nonlocal means to obtain a stabilized domain image. Smoothing intensity parameters are estimated based on the local high-frequency differences in the stabilized domain image for adaptive denoising. The denoised image of the image to be processed is obtained by performing an inverse Anscombe transform on the stabilized domain image.

[0016] The weak photon image denoising method based on bright spot protection and variance stabilization provided by this invention adopts Anscombe variance stabilization before denoising, which adapts to the Poisson noise characteristics in weak photon images. It is superior to performing uniform filtering directly in the original intensity domain, is compatible with different denoising modules, and can also adjust the bright spot detection threshold, background smoothing scale and fusion weight according to different imaging systems. It has good versatility and scalability.

[0017] In one optional implementation, a bright spot protection fusion is performed on the background correction image and the denoised image based on a soft protective mask to obtain an enhanced image, including: Based on the original background correction information retention coefficients of bright spot protected areas and non-bright spot protected areas, the bright spot fidelity path results and non-bright spot denoising path results are calculated using the background correction image and the denoised image. The enhanced image is obtained by fusing bright spot protection based on the soft protective mask, the bright spot fidelity path results, and the non-bright spot region denoising path results.

[0018] The weak photon image denoising method based on bright spot protection and variance stabilization provided by this invention can effectively preserve real hot spots, small-scale fluorescence structures and local edge information through geometric screening, protective masking and protective fusion, reducing the risk of signal being accidentally erased. Through soft protective fusion and a small amount of original information re-injection, it can effectively avoid the texture being too smooth and edge distortion caused by single denoising output.

[0019] Secondly, the present invention provides a weak photon image denoising device based on bright spot protection and variance stabilization, the device comprising: The soft protective mask generation module is used to detect bright spots in the image to be processed and generate a soft protective mask based on morphological constraints. The background correction module is used to perform background correction on the image to be processed using a soft protective mask to obtain a background-corrected image, and then perform Anscombe transform on the background-corrected image to obtain a standard image. The stabilization domain denoising module is used to denoise the standard image in the variance stabilization domain to obtain the stabilization domain image, and then perform an inverse Anscombe transform on the stabilization domain image to obtain the denoised image of the image to be processed. The image fusion module is used to perform bright spot protection fusion on the background correction image and the denoised image based on a soft protective mask to obtain an enhanced image.

[0020] Thirdly, the present invention provides an electronic 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 method described in the first aspect or any corresponding embodiment thereof.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0022] 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.

[0023] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the second process of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the bright spot protection fusion mechanism using a soft protective mask in the weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention. Figure 5 This is a complete flowchart of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the effect of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention; Figure 7This is a structural block diagram of a weak photon image denoising device based on bright spot protection and variance stabilization according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] 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.

[0025] 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.

[0026] As an optional application scenario of this invention, such as Figure 1 As shown, the weak photon image denoising system may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0027] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0028] This invention provides a weak photon image denoising method based on bright spot protection and variance stabilization. It corrects the background image by protecting the bright spot and combines this with variance stabilization for denoising. During background estimation and noise suppression, it separates the real signal from the background noise, effectively removing both background and noise while preserving as much of the real bright spot, weak structure, and spatial resolution information as possible. This improves background suppression, noise control, and structural fidelity under low signal-to-noise ratio conditions, enhancing image visibility and the reliability of subsequent quantitative analysis. It is applicable to sparse structure image preprocessing and enhancement analysis in scenarios such as fluorescence microscopy, biomedical weak signal imaging, astronomical low-light detection, and low-illuminance industrial inspection.

[0029] According to an embodiment of the present invention, a weak photon image denoising method based on bright spot protection and variance stabilization is provided. 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.

[0030] This embodiment provides a weak photon image denoising method based on bright spot protection and variance stabilization, which can be used in the aforementioned computer system. Figure 2 This is a flowchart of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention, as follows: Figure 2 As shown, the process includes the following steps: Step S201: Perform bright spot detection on the image to be processed and generate a soft protective mask based on morphological constraints.

[0031] Specifically, the image to be processed is pre-smoothed, and bright spot detection is achieved by combining noise coarse estimation. Candidate signal regions with high intensity in the image are identified and used as protection regions. When performing background estimation, the protection regions are constrained to avoid real signals from participating in background modeling, thereby obtaining more accurate background distribution estimation results. The final real bright spot region is slightly enlarged, and a Gaussian blur is used to create a gradual soft protective mask. The center of the bright spot has the highest weight, which gradually decreases outward to avoid abrupt boundary breaks.

[0032] Step S202: Use a soft protective mask to perform background correction on the image to be processed to obtain a background-corrected image, and then perform Anscombe transform on the background-corrected image to obtain a standard image.

[0033] Specifically, a background estimation neutron image is constructed, and the background estimation seed image is corrected based on a soft protective mask to obtain a slowly varying background image. Then, the slowly varying background image is subtracted from the original image, and the negative values ​​after subtraction are truncated to zero to obtain the background corrected image.

[0034] Anscombe variance stabilization transformation is performed on the corrected image to obtain a standard image, making the Poisson dominant noise closer to the variance stable distribution.

[0035] In low-photon imaging, noise is primarily Poisson noise (variance equal to mean). Bright areas have high signal strength and high noise, while dark areas have low signal strength and low noise. Most denoising algorithms assume a fixed Gaussian noise variance. Mapping the Poisson-dominated signal-dependent noise in the original low-photon image to an approximately variance-stable transform domain essentially transforms the high noise in bright areas and low noise in dark areas into a more uniform noise across the entire image. The Anscombe transform is a mathematical method that converts Poisson noise into uniform Gaussian noise, primarily used to process noise in images or signals. The noise level in the original signal varies with brightness (brighter areas have higher noise), but after the transform, the noise fluctuations become more consistent, facilitating unified processing. It converts the irregular Poisson data into a near-standard Gaussian distribution, which can then be denoised using mature denoising tools.

[0036] Step S203: Denoise the standard image in the variance stabilization domain to obtain the stabilized domain image, and perform inverse Anscombe transform on the stabilized domain image to obtain the denoised image of the image to be processed.

[0037] Specifically, the converted equalized acoustic image is denoised in the stabilization domain, removing noise from both bright and dark areas simultaneously without excessively damaging subtle details, thus improving noise suppression. After denoising, the image is restored to the original data domain using an inverse Anscombe transform.

[0038] Step S204: Based on the soft protective mask, the background correction image and the denoised image are fused with bright spot protection to obtain the enhanced image.

[0039] Specifically, based on the soft protective mask, the denoised image and the background correction image are adaptively fused. In bright areas, the real strong signal is retained first, and the blurred image after noise reduction is used less to avoid the loss of details caused by the smooth transition. In the non-protected area (ordinary background area), the noise reduction effect is enhanced, mainly using the clean image after noise reduction, and only retaining a small amount of original image information, so as to balance noise reduction and signal fidelity.

[0040] The weak photon image denoising method provided in this embodiment, based on bright spot protection and variance stabilization, generates a soft protective mask by combining bright spot detection with morphological constraints. This accurately distinguishes between real bright spots and noise artifacts, avoiding interference from bright targets in background modeling, resulting in more accurate background correction. Anscombe transform is used to achieve noise variance equalization, adapting to the Poisson noise characteristics of weak photon images. Stabilization domain denoising can efficiently suppress noise without damaging subtle details. Furthermore, the soft protective mask adaptively fuses two images, preserving the original signal in bright spot areas and focusing on noise reduction output in non-protected areas. The transition is smooth with no boundary artifacts, effectively suppressing background undulations and image noise while fully preserving minute bright structures and edge information, thus improving image visibility.

[0041] In some alternative implementations, the method further includes: Step S205: Perform a light sharpening process on the enhanced image to obtain the final image.

[0042] Specifically, the fused enhanced image can also undergo light sharpening to moderately enhance image edges and details, compensating for edge softening that may be introduced during the denoising process, improving edge sharpness and overall visibility, and outputting the final enhanced image. Light sharpening is used to slightly compensate for edge details and improve the sharpness of target contours, while strictly controlling the sharpening intensity to avoid amplifying background noise or producing edge overshoot artifacts.

[0043] Lightweight sharpening can be achieved using a low-intensity unsharp mask (USM). The enhanced image is then subjected to a small-scale Gaussian blur. The blurred image is subtracted from the original image to obtain edge detail components, which are then superimposed back onto the enhanced image with very low weight. The sharpening parameters can be adaptively adjusted, only slightly brightening fluorescent points and small target edges, without significantly amplifying noise in dark background areas. The final output image after sharpening is one with no significant blurring, sharp edges, and no amplified noise.

[0044] Step S206: Calculate the index parameter values ​​of the final image and evaluate the image processing effect based on the index parameter values.

[0045] Specifically, further evaluation parameters such as the standard deviation of the background region, the overall contrast index, and the proportion of the protected area can be calculated to quantify the enhancement effect. The calculation of the standard deviation of the background region involves selecting a blank background area without a target and statistically analyzing the standard deviation of pixel grayscale values; a smaller value indicates better background noise suppression. The grayscale difference / ratio between the bright target area and the background area is calculated to measure the distinguishability of the target; a higher value indicates better detail preservation and background subtraction. The proportion of pixels covered by the soft protective mask to the total number of pixels in the entire image is statistically analyzed to reflect the number of effective small bright targets within the image, assisting in judging the completeness of bright spot recognition.

[0046] The calculated parameter values ​​are compared with preset reference thresholds and the original unprocessed image parameters. A decrease in background standard deviation, an improvement in overall contrast, and no significant loss in the protected area ratio indicate that the denoising, background correction, and bright spot protection processing have met the standards. If the parameters are abnormal, parameters such as the bright spot detection threshold, background smoothing scale, and fusion weight can be adjusted in reverse to optimize the process. This provides an objective and quantitative evaluation basis, overcoming the limitations of relying solely on subjective visual judgment of image quality, and meeting the needs of application scenarios such as biological imaging and astronomical exploration that require quantitative data analysis.

[0047] The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment compensates for edge softening that may be introduced during the denoising process through lightweight sharpening, thereby improving edge clarity and overall visibility. The image processing effect is evaluated by calculating index parameters to ensure that the image processing meets the requirements. The solution can also be improved based on the image processing effect.

[0048] This embodiment provides a weak photon image denoising method based on bright spot protection and variance stabilization, which can be used in the aforementioned computer system. Figure 3 This is a flowchart of a weak photon image denoising method based on bright spot protection and variance stabilization according to an embodiment of the present invention, as follows: Figure 3 As shown, the process includes the following steps: Step S301: Perform bright spot detection on the image to be processed and generate a soft protective mask based on morphological constraints.

[0049] Specifically, step S301 includes: Step S3011: If the image to be processed is a color image, then the image to be processed is converted into a grayscale image and the data format is unified before being used as the input image.

[0050] Specifically, the image to be processed is read. If the image to be processed is a color image, it is first converted into a grayscale image. Then, the pixel data is normalized to a double-precision floating-point data format, and invalid negative grayscale pixels are truncated to zero. This completes the standardization preprocessing and yields an input image in a uniform format.

[0051] Step S3012: Gaussian smoothing is applied to the input image to obtain a pre-smoothed image. High-frequency components are constructed based on the pre-smoothed image, and the noise intensity is estimated using median absolute deviation to obtain a coarse noise estimate.

[0052] Specifically, the input image is first smoothed by Gaussian to obtain a pre-smoothed image; then a Gaussian kernel with a larger kernel is used to blur the pre-smoothed image again, and the difference between the two smoothed images is calculated to extract the high-frequency detail components (bright spots, edges, and noise).

[0053] The median absolute deviation algorithm is used to statistically calculate the high-frequency components, thus obtaining a coarse estimate of the noise in the entire image. To avoid distortion in noise estimation caused by interference from bright spots in the mean and standard deviation.

[0054] A Gaussian filter is used to pre-smooth the input image; then preliminary bright spot detection is performed, the high-frequency residual is calculated by the difference between the pre-smoothed image and its further blurred version, and the noise intensity is estimated by the median absolute deviation method.

[0055] Step S3013: Construct a segmentation threshold based on the global mean and coarse noise estimate of the pre-smoothed image, and generate a candidate bright spot mask based on the segmentation threshold.

[0056] Specifically, a detection threshold is generated based on a linear combination of the image mean and a coarse noise estimate. This threshold is then used to segment potential bright spot regions in the image, obtaining candidate point masks. Let the original image be I(x,y), where (x,y) represents the pixel coordinates. The global grayscale mean of the pre-smoothed image is extracted. Combined with the noise coarse estimate Construct a segmentation threshold , where k represents the threshold coefficient.

[0057] An initial background approximation is obtained through local background estimation or multi-scale smoothing. And construct a highlight response map:

[0058] in, This represents the enhanced response of the original image relative to the local background. Initial bright spot candidate regions are constructed based on the response intensity:

[0059] in, This is the initial candidate mask for bright spots, i.e., the candidate bright spot mask.

[0060] Step S3014: Based on the morphological constraints, candidate bright spot masks are screened for bright spots and their morphological expansion is performed, and a soft protective mask is generated by Gaussian smoothing.

[0061] Specifically, morphological screening is performed on candidate bright spot masks. Through connected component analysis, the area, eccentricity, major axis length, and minor axis length of each connected component are extracted. Regions with excessively small or large areas, abnormal aspect ratios, or excessively high eccentricities are removed, thus obtaining more reliable candidate bright spot regions. The screened bright spot regions are then expanded using structuring elements to increase the protection range, and Gaussian smoothing is used to transform them into spatially continuous soft protective masks.

[0062] Parameters such as area, eccentricity, major axis length, and minor axis length are extracted from the connected components of candidate bright spots. Abnormal regions are then eliminated by area range constraints, aspect ratio constraints, and eccentricity constraints, retaining only candidate bright spot regions that meet the geometric features, thereby reducing false detections due to noise and non-target structures. Subsequently, the filtered bright spot regions are subjected to morphological dilation, and a soft protective mask is generated by Gaussian smoothing.

[0063] The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment eliminates color channel interference by unifying data standards, extracts high-frequency bright spots through multi-scale differential extraction, accurately estimates noise by median absolute deviation, obtains candidate masks through threshold segmentation, and generates bright spot detection based on the image's own noise and brightness adaptively, making it more versatile. Furthermore, it filters out noisy false bright spots by relying on morphological constraints such as connected region area and eccentricity, and generates a gradient soft mask through dilation and Gaussian smoothing, which not only significantly reduces the false detection rate of bright spots, but also eliminates the boundary abrupt artifacts caused by hard masks.

[0064] In some optional implementations, step S3014 above includes: Step a1: Extract the morphological parameters of each position of the candidate bright spot mask, and filter the candidate bright spot regions based on the morphological constraints.

[0065] Specifically, it does not merely distinguish bright spots using grayscale thresholds, but rather introduces connected component geometric constraints based on the initial detection results, extracting each connected component of the candidate bright spot mask. The morphological parameters, including but not limited to: area eccentricity Major axis length and minor axis length The morphological constraints are as follows: ， ，

[0066] in, This represents the set of truly valuable highlights that have been preserved. and Let represent the lower and upper bounds of the area of ​​the connected region, respectively. Indicates the maximum permissible eccentricity. This indicates the upper limit of the major-minor axis ratio. This geometric screening effectively eliminates isolated noise points, elongated artifacts, and anomalous bright spots, improving the reliability of the bright spot protection area.

[0067] Step a2: Based on the candidate bright spot region, perform morphological dilation processing on the candidate bright spot mask to obtain a dilated protective mask, and then perform smoothing processing on the dilated protective mask to obtain a soft protective mask.

[0068] Specifically, the filtered bright spot candidate regions are a binary mask. To perform morphological expansion:

[0069] in, This indicates the expanded protective area (i.e., the expanded protective mask). This indicates a morphological dilation operation. Represents a structural element.

[0070] The expansion mask is then smoothed to obtain spatially continuously varying soft protection weights:

[0071] in, It is a soft protective mask, with values ​​ranging from [0,1]. The standard deviation is Gaussian kernel, This indicates a convolution operation. Compared to traditional hard masks, soft protective masks allow for a smooth weight transition between bright areas and background areas, reducing abrupt boundary changes and fusion artifacts.

[0072] The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment extracts the morphological parameters of connected components and sets constraints to filter out false bright spots formed by noise, thereby improving the accuracy of real bright spot recognition. The filtered region is expanded to increase the protection range, and then Gaussian smoothing is used to generate a gradient soft mask to eliminate the jagged tortuosity of the hard mask boundary and achieve a smooth transition of region weights.

[0073] Step S302: Use a soft protective mask to perform background correction on the image to be processed to obtain a background-corrected image, and then perform Anscombe transform on the background-corrected image to obtain a standard image.

[0074] Specifically, step S302 includes: S3021, the median filtering result is used to replace and fill the area covered by the soft protective mask to obtain the filled image.

[0075] Specifically, the background seed image, also known as the fill image, is the base image used to fit the real background. Using a soft protective mask, all pixels identified as real bright spots in the original image are replaced with the median filter output value from their neighborhood, obtaining a grayscale reference more consistent with the surrounding background and eliminating interference from bright signals. Non-bright areas retain the original pixels of the original image, resulting in an image free from bright target interference and with grayscale values ​​continuously consistent with the surrounding background. Eliminating the interference of bright spots that inflate the background grayscale serves as the input base for generating a gradually changing background image, avoiding background overestimation.

[0076] Based on a soft protective mask, bright spots are replaced and filled to obtain a filled image for background estimation:

[0077] in, Indicates filling in the image, This indicates fill operations such as neighborhood interpolation, median replacement, morphological reconstruction, or local low-order fitting.

[0078] S3022, perform large-scale Gaussian smoothing on the filled image to obtain a gradually changing background image.

[0079] Specifically, by using large-scale Gaussian smoothing, all high-frequency details such as stars, fluorescent points, and noise in the image are filtered out, and only low-frequency slow gray-scale fluctuations caused by light drift and scattering are retained. The entire image is blurred over a large area to obtain a real light and dark background color that is not affected by bright spots, which serves as a gradually changing background image.

[0080] Based on filling image Perform large-scale Gaussian smoothing to estimate the background:

[0081] in, This is for the final estimation of the background (i.e., the slowly varying background map). This represents a large-scale smoothing, morphological opening, or low-frequency background modeling operator with a scale parameter of r. Real bright spots do not participate in the background modeling process, avoiding the problem of bright spots inflating the background estimate in conventional "direct filtering background estimation" methods. This reduces background overestimation and target oversubtraction, and better reflects the structural characteristics of locally strong targets and globally slowly changing backgrounds in weak photon images.

[0082] S3023, subtract the gradually changing background image from the image to be processed, and truncate the negative value after subtraction to zero to obtain the background-corrected image.

[0083] Specifically, the gradually changing background image is subtracted from the original image to be processed, and negative values ​​are truncated to zero to obtain a background-corrected image with the substrate gradient interference removed.

[0084] Background correction is performed on the original image using the results of protective background estimation.

[0085] In the formula, This represents the background-corrected image. To estimate the background, Background deduction factor, This is used to avoid negative values ​​after deduction. The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment uses the median filter value of a soft protective mask to replace the bright spot area to obtain a filled image, avoiding the participation of real bright signals in background fitting and avoiding background overestimation; the gradually changing background is accurately restored by large-scale Gaussian smoothing, and the original image is corrected after removing the background and truncating negative values, which significantly reduces the interference of bright targets on background fitting, completely preserves weak target signals, improves the accuracy of background subtraction, and obtains a background-corrected image with higher contrast and no negative pixel interference.

[0086] Step S303: Denoise the standard image in the variance stabilization domain to obtain the stabilized domain image, and perform inverse Anscombe transform on the stabilized domain image to obtain the denoised image of the image to be processed.

[0087] Specifically, step S303 includes: Step S3031: Denoising is performed in the variance stabilization domain using nonlocal means to obtain a stabilized domain image. Smoothing intensity parameters are estimated based on the local high-frequency differences in the stabilized domain image, and adaptive denoising is performed.

[0088] Specifically, for weak photon images, noise typically exhibits pronounced Poisson statistical characteristics, meaning the noise variance varies with signal intensity. To ensure subsequent denoising processing is performed within a more stable statistical domain, the background correction image undergoes an Anscombe variance stabilization transform:

[0089] in, This represents the image after Anscombe variance stabilization.

[0090] Then, denoising is performed in the stabilization transform domain:

[0091] in, This represents the denoised transform domain image. This represents nonlocal means, wavelet domain Bayesian soft thresholding denoising, BM3D, anisotropic diffusion, or other statistical filtering operators. If a nonlocal means algorithm is used, the smoothing intensity parameter is estimated based on the difference between the transform domain image and its smoothed map to improve denoising adaptability.

[0092] Step S3032: Perform inverse Anscombe transform on the stabilized domain image to obtain the denoised image of the image to be processed.

[0093] Specifically, by using the inverse Anscombe transform, the denoised stabilized domain image is restored to the original intensity expression domain, and the negative values ​​after restoration are truncated to zero, resulting in the denoised image of the image to be processed.

[0094] in, This represents the image after statistical denoising. Instead of simply denoising the original image directly, it first suppresses non-stationary backgrounds, then processes Poisson-dominated noise in the variance stabilization domain, making subsequent filtering more consistent with the noise statistical characteristics of weak photon images.

[0095] The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment adopts Anscombe variance stabilization before denoising, which adapts to the Poisson noise characteristics in weak photon images. It is superior to performing uniform filtering directly in the original intensity domain, is compatible with different denoising modules, and can also adjust the bright spot detection threshold, background smoothing scale and fusion weight according to different imaging systems. It has good versatility and scalability.

[0096] Step S304: Based on the soft protective mask, the background correction image and the denoised image are fused with bright spot protection to obtain the enhanced image.

[0097] Specifically, step S304 includes: Step S3041: Based on the original background correction information retention coefficients of the bright spot protection area and the non-bright spot protection area, calculate the bright spot fidelity path result and the non-bright spot denoising path result using the background correction image and the denoised image.

[0098] Specifically, within the protected area, a higher proportion of the original bright spot response in the background correction image is retained to avoid weakening of the real bright signal during denoising; in the unprotected area, the denoising result is mainly used, while a small amount of the original background correction information is retained to reduce oversmoothing problems and maintain natural texture transitions, preventing structural voids.

[0099] The highlight fidelity path is represented as follows:

[0100] The denoising path for non-protected regions is represented as follows:

[0101] in, and These are the original background correction information retention coefficients for the protected and unprotected areas, respectively, and they typically satisfy the following: This means that a higher proportion of the original background correction signal is retained in the bright spot area to avoid the bright spot intensity being reduced due to denoising; while in the non-protected area, the denoising result is mainly used to enhance the background smoothness and reduce noise.

[0102] Step S3042: Based on the soft protective mask, the bright spot fidelity path result and the non-bright spot region denoising path result, bright spot protection fusion is performed to obtain the enhanced image.

[0103] Specifically, using soft protective masks An adaptive fusion weight is constructed, and a bright spot preservation fusion is performed on the bright spot fidelity path result and the non-bright spot region denoising path result based on the adaptive weight to obtain the enhanced image representation as follows:

[0104] in, Indicates an enhanced image. As a soft protection weight, To ensure the accuracy of the path results, The result is the denoising path for the non-protected area.

[0105] like Figure 4 The diagram shows a bright spot protection fusion mechanism using a soft protective mask. The soft protective mask is used to divide the image into a protected area (real bright spot) and a non-protected area (background noise area). Differentiated weight allocation strategies are used for the two areas.

[0106] Protected area: The background-corrected image is given a large weight and a small amount of the denoised result image is introduced to prioritize the preservation of the original bright target signal; Unprotected area: The denoised result image is used with a large weight and only a small amount of background-corrected image information is retained with a small weight to fully suppress background noise.

[0107] The background correction image and the denoised image are weighted and blended pixel-by-pixel according to the aforementioned region weighting rules to obtain the enhanced image. A slight sharpening effect can be applied to the blended image to compensate for edge details, ultimately outputting the complete enhanced result.

[0108] The weak photon image denoising method based on bright spot protection and variance stabilization provided in this embodiment can effectively preserve real hot spots, small-scale fluorescence structures and local edge information through geometric screening, protective masking and protective fusion, reducing the risk of signal being accidentally erased. Through soft protective fusion and a small amount of original information re-injection, it can effectively avoid the texture being too smooth and edge distortion caused by single denoising output.

[0109] In one specific embodiment, such as Figure 5The diagram illustrates the complete process. First, candidate bright spot masks are obtained. Noisy false bright spots are eliminated through morphological constraint screening. Then, Gaussian smoothing is performed to generate a gradient soft protective mask. Image regions are divided using the soft protective mask into protected and unprotected pixel regions. Differential weighting is applied to the protected regions, using a high weight to retain the background-corrected image; the unprotected regions are introduced into the denoised image with a high weight, retaining only a small weight for the background-corrected image information. Weighted image fusion is performed, mixing the background-corrected image and the denoised image pixel-by-pixel according to the aforementioned partition weights, outputting the fused enhanced image. Optional post-processing and output include slight sharpening of the fused image to compensate for edge sharpness, ultimately outputting the final enhanced result. It should be noted that the pre-smoothing scale, bright spot threshold factor, background smoothing scale, fusion weight, and sharpening intensity can all be adjusted according to specific application scenarios; the stabilization domain denoising module can also be replaced with other filters with similar functions.

[0110] like Figure 6 The diagram illustrates the effect of the weak photon image denoising method provided in this embodiment. Noisy false bright spots are eliminated by morphological constraint screening of candidate bright spot masks, followed by Gaussian smoothing to generate a gradient soft protective mask. This accurately distinguishes between real targets and the background, eliminating the jagged edges and grayscale discontinuities caused by binary hard masks. The soft protective mask is used to divide the protected and unprotected areas for differentiated processing: the protected area largely retains the original image signal after background correction, maximizing the preservation of key details such as fluorescent spots and small targets, and significantly suppressing the weakening effect of noise reduction on bright spots; the unprotected area primarily uses the denoised image, retaining only a small amount of original image information, effectively filtering out Poisson noise and gradually changing background undulations, balancing noise suppression capability and image naturalness. After weighted fusion, a slight sharpening step can be added to compensate for edge blurring caused by multiple rounds of smoothing and improve the recognition of target contours. The entire process works together to completely eliminate low-frequency background drift and suppress image noise, while fully preserving weak and fine structures without obvious processing artifacts. The final enhanced image output has higher contrast and more complete details, and can be directly used for subsequent image quantitative detection and feature recognition, adapting to the high-precision application requirements of low-light and low signal-to-noise ratio imaging scenarios.

[0111] This embodiment also provides a weak photon image denoising device based on bright spot protection and variance stabilization. This device is used to implement the above embodiments and preferred embodiments, and 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 device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0112] This embodiment provides a weak photon image denoising device based on bright spot protection and variance stabilization, such as... Figure 7As shown, it includes: The soft protective mask generation module 701 is used to perform bright spot detection on the image to be processed and generate a soft protective mask based on morphological constraints.

[0113] Background correction module 702 is used to perform background correction on the image to be processed using a soft protective mask to obtain a background corrected image, and to perform Anscombe transform on the background corrected image to obtain a standard image.

[0114] The stabilization domain denoising module 703 is used to denoise the standard image in the variance stabilization domain to obtain a stabilization domain image, and to perform an inverse Anscombe transform on the stabilization domain image to obtain a denoised image of the image to be processed.

[0115] The image fusion module 704 is used to perform bright spot protection fusion on the background correction image and the denoised image based on a soft protective mask to obtain an enhanced image.

[0116] In some alternative implementations, the soft protective mask generation module 701 includes: The image preprocessing unit is used to convert the image to be processed into a grayscale image if the image to be processed is a color image, and then use the unified data format as the input image.

[0117] The smoothing unit performs Gaussian smoothing on the input image to obtain a pre-smoothed image, constructs high-frequency components based on the pre-smoothed image, and estimates the noise intensity using median absolute deviation to obtain a coarse noise estimate.

[0118] The image segmentation unit is used to construct a segmentation threshold based on the global mean and coarse noise estimate of the pre-smoothed image, and generate candidate bright spot masks based on the segmentation threshold.

[0119] The bright spot screening and morphology expansion unit is used to screen and expand candidate bright spot masks based on morphological constraints, and generate a soft protective mask through Gaussian smoothing.

[0120] In some alternative implementations, the background correction module 702 includes: Replacement filling units are used to replace and fill the areas covered by the soft protective mask using the median filtering results, resulting in a filled image.

[0121] The large-scale Gaussian smoothing unit is used to perform large-scale Gaussian smoothing on the filled image to obtain a gradually changing background image.

[0122] The background subtraction unit is used to subtract the gradually changing background image from the image to be processed, and truncate the negative value after subtraction to zero to obtain the background corrected image.

[0123] In some alternative implementations, the stabilization domain denoising module 703 includes: The adaptive denoising unit is used to perform denoising in the variance stabilization domain using non-local means to obtain a stabilized domain image, and to estimate the smoothing intensity parameter based on the local high-frequency differences in the stabilized domain image for adaptive denoising.

[0124] The inverse transform unit is used to perform an inverse Anscombe transform on the stabilized domain image to obtain a denoised image of the image to be processed.

[0125] In some alternative implementations, the image fusion module 704 includes: The path result calculation unit is used to calculate the bright spot fidelity path result and the non-bright spot denoised path result based on the original background correction information retention coefficients of the bright spot protected area and the non-bright spot protected area, using the background correction image and the denoised image.

[0126] The bright spot protection fusion unit is used to perform bright spot protection fusion based on the soft protective mask, the bright spot fidelity path result and the non-bright spot region denoising path result to obtain an enhanced image.

[0127] The weak photon image denoising device based on bright spot protection and variance stabilization provided in this embodiment of the invention can execute the weak photon image denoising method based on bright spot protection and variance stabilization provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of 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.

[0128] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0129] The following is a detailed reference. Figure 8 This 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.) 801, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 802 or a program loaded from memory 808 into random access memory (RAM) 803. The RAM 803 also stores various programs and data required for the operation of the electronic device. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0130] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 807 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 808 including, for example, magnetic tapes, hard disks, etc.; and communication devices 809. Communication device 809 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 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.

[0131] 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 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by the processor 801, it performs the functions defined in the weak photon image denoising method based on bright spot protection and variance stabilization according to embodiments of the present invention.

[0132] Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0133] 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 recordable 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 weak photon image denoising method based on bright spot protection and variance stabilization shown in the above embodiments is implemented.

[0134] 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 such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A weak photon image denoising method based on bright spot protection and variance stabilization, characterized in that, The method includes: Bright spot detection is performed on the image to be processed, and a soft protective mask is generated based on morphological constraints; The background correction of the image to be processed is performed using the soft protective mask to obtain a background-corrected image. Anscombe transform is then applied to the background-corrected image to obtain a standard image. This process includes: replacing and filling the area covered by the soft protective mask using median filtering to obtain a filled image; performing large-scale Gaussian smoothing on the filled image to obtain a gradually changing background image; and subtracting the gradually changing background image from the image to be processed, truncating the negative values ​​after subtraction to zero to obtain the background-corrected image. The standard image is denoised in the variance stabilization domain to obtain a stabilized domain image, and the stabilized domain image is then subjected to an inverse Anscombe transform to obtain a denoised image of the image to be processed. This process includes: denoising the standard image in the variance stabilization domain using nonlocal means to obtain a stabilized domain image, estimating smoothing intensity parameters based on local high-frequency differences in the stabilized domain image, and performing adaptive denoising; and then performing an inverse Anscombe transform on the stabilized domain image to obtain the denoised image of the image to be processed. The background-corrected image and the denoised image are fused together using the soft protective mask to obtain an enhanced image.

2. The method according to claim 1, characterized in that, The method further includes: The enhanced image is then lightly sharpened to obtain the final image; Calculate the index parameter values ​​of the final image, and evaluate the image processing effect based on the index parameter values.

3. The method according to claim 1, characterized in that, Bright spot detection is performed on the image to be processed, and a soft protective mask is generated based on morphological constraints, including: If the image to be processed is a color image, then the image to be processed is converted into a grayscale image and the data format is unified before being used as the input image; The input image is Gaussian smoothed to obtain a pre-smoothed image, and high-frequency components are constructed based on the pre-smoothed image. The noise intensity is estimated using median absolute deviation to obtain a coarse noise estimate. A segmentation threshold is constructed based on the global mean of the pre-smoothed image and the coarse estimate of the noise, and a candidate bright spot mask is generated based on the segmentation threshold; Based on morphological constraints, the candidate bright spot masks are screened for bright spots and their morphology is expanded. A soft protective mask is then generated by Gaussian smoothing.

4. The method according to claim 3, characterized in that, Based on morphological constraints, the candidate bright spot masks are screened for bright spots and their shapes are expanded. A soft protective mask is then generated using Gaussian smoothing. This includes: Extract the morphological parameters of each position of the candidate bright spot mask, and filter the candidate bright spot regions based on morphological constraints; Based on the candidate bright spot region, the candidate bright spot mask is subjected to morphological dilation processing to obtain an inflated protective mask, and the inflated protective mask is then smoothed to obtain a soft protective mask.

5. The method according to claim 1, characterized in that, Based on the soft protective mask, the background-corrected image and the denoised image are fused with highlight protection to obtain an enhanced image, including: Based on the original background correction information retention coefficients of the bright spot protection region and the non-bright spot protection region, the bright spot fidelity path result and the non-bright spot region denoising path result are calculated using the background correction image and the denoised image; Based on the soft protective mask, the bright spot fidelity path result, and the non-bright spot region denoising path result, bright spot protection fusion is performed to obtain an enhanced image.

6. A weak photon image denoising device based on bright spot protection and variance stabilization, characterized in that, The device includes: The soft protective mask generation module is used to detect bright spots in the image to be processed and generate a soft protective mask based on morphological constraints. The background correction module is used to perform background correction on the image to be processed using the soft protective mask to obtain a background corrected image, and to perform Anscombe transform on the background corrected image to obtain a standard image. This includes: replacing and filling the area covered by the soft protective mask using median filtering results to obtain a filled image; performing large-scale Gaussian smoothing on the filled image to obtain a gradually changing background image; and subtracting the gradually changing background image from the image to be processed, truncating the negative values ​​after subtraction to zero to obtain the background corrected image. The stabilization domain denoising module is used to denoise the standard image in the variance stabilization domain to obtain a stabilized domain image, and to perform an inverse Anscombe transform on the stabilized domain image to obtain a denoised image of the image to be processed. This includes: performing denoising in the variance stabilization domain using non-local means to obtain a stabilized domain image, estimating smoothing intensity parameters based on local high-frequency differences in the stabilized domain image, and performing adaptive denoising; and performing an inverse Anscombe transform on the stabilized domain image to obtain a denoised image of the image to be processed. The image fusion module is used to perform highlight protection fusion on the background correction image and the denoised image based on the soft protective mask to obtain an enhanced image.

7. An electronic device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 5.

Images