A distributed image denoising method based on edge computing
The distributed image denoising method based on edge computing, which utilizes an edge node model and adaptive adjustment of denoising intensity coefficients, solves the problem of incomplete or excessive denoising in existing technologies and achieves efficient image denoising results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANXI QINWAN INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2026-03-25
- Publication Date
- 2026-06-23
AI Technical Summary
Existing image denoising techniques lack targeted distributed image processing, resulting in incomplete or excessive denoising. They cannot adaptively adjust and are prone to incomplete noise suppression or distortion of image details.
Edge computing is used to set up an edge node model, collect target images for preprocessing and morphological classification, extract denoising intensity coefficients, perform distributed denoising processing, and generate target reconstruction images by stitching together image feature sets. The denoising coefficients are adjusted multiple times until they are qualified.
It improves image denoising performance, ensuring thorough denoising without distortion, and achieves adaptive image denoising adjustment.
Smart Images
Figure CN122265081A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edge computing technology, and more specifically to a distributed image denoising method based on edge computing. Background Technology
[0002] Edge computing is a distributed computing framework that refers to an open platform that integrates core capabilities of networking, computing, storage, and applications at the network edge, close to the source of objects or data, to provide edge intelligence services locally. Its applications are initiated at the edge to meet the industry's needs in real-time business, application intelligence, and other aspects. Existing image denoising techniques lack targeted distributed image processing and denoising adjustment. They typically obtain the target denoised image through a single denoising process, which may result in incomplete or excessive denoising. They cannot adaptively adjust the denoising, which can lead to incomplete noise suppression or image detail distortion during the denoising process, resulting in poor image denoising performance. Therefore, in order to improve the denoising effect of images, this invention provides a distributed image denoising method based on edge computing. Summary of the Invention
[0003] The purpose of this invention is to provide a distributed image denoising method based on edge computing to address the shortcomings of the prior art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a distributed image denoising method based on edge computing, the method comprising the following steps: Step S1: Set up an edge node model, acquire the target preprocessed image based on the edge node model, and classify the target preprocessed image into image morphology to obtain distributed morphology images; Step S2: Extract features from the distributed morphological image to obtain the denoising intensity coefficients of the distributed morphological image; then, denoise the distributed image based on the denoising intensity coefficients to obtain the distributed denoised image; Step S3: Obtain the image feature set of the distributed denoised image, and stitch the distributed denoised image together according to the image feature set to generate the target restored image; Step S4: Obtain the denoising coefficients of the target restored image; determine the passability of the denoising coefficients. If they are not passable, perform multiple denoising processes on the target restored image to obtain a passable target restored image and the corresponding number of denoising times.
[0005] Furthermore, the process of setting up an edge node model and acquiring a target preprocessed image based on the edge node model includes: The edge node model includes an edge acquisition node unit and an edge preprocessing node unit. The edge preprocessing node unit is equipped with a noise judgment subunit and a noise preprocessing subunit. The edge acquisition node unit is communicatively connected to various image acquisition terminals, and is used to acquire target images through the image acquisition terminals and send the target images to the noise judgment subunit to obtain the pixel data of the target images. Then, it is judged against a preset pixel data threshold. If the pixel data is less than or equal to the pixel data threshold, the corresponding target image is recorded as an abnormal target image, and an instruction to acquire the target image within the same acquisition range is resent to the image acquisition terminal of the corresponding abnormal target image, which is recorded as a re-acquisition instruction. Otherwise, the target image is sent to the noise preprocessing subunit, which is equipped with a minimum adjustment pixel data, and is used to pre-adjust the pixel data of the target image whose pixel data is less than the minimum adjustment pixel data to the minimum adjustment pixel data, thereby obtaining the target preprocessed image. Otherwise, the corresponding pixel data does not change.
[0006] Furthermore, the process of classifying the target preprocessed image into image morphology to obtain distributed morphological images includes: The image shape of the target preprocessed image is obtained, including the image shape size and the image spatial size; a unit segmentation dataset for the image shape size is set; the unit segmentation dataset includes the unit segmentation shape size and the corresponding image shape size range; the corresponding unit segmentation shape size is obtained by matching the image shape size with the corresponding image shape size range. The target preprocessed image is segmented according to the unit segmentation shape size to obtain unit segmentation images. If the unit segmentation image corresponding to the edge of the target preprocessed image is smaller than the unit segmentation shape size, it is considered as one unit segmentation image. The unit segmentation space size of the unit segmentation image is obtained and compared with a preset unit segmentation space size threshold. If the size of the unit segmentation space is greater than the threshold, the unit segmentation image is divided equally to obtain a distributed morphological image; otherwise, it is recorded as a distributed morphological image.
[0007] Furthermore, the process of extracting features from the distributed morphological image to obtain the denoising intensity coefficients of the distributed morphological image includes: Set the deployment spacing, set several local extraction nodes of the distributed morphological image according to the deployment spacing, and then extract the node noise feature vector of the local extraction node. The node noise feature vector includes gray-level variance, Gaussian noise parameter and impulse noise density; generate a node noise feature matrix from the node noise feature vectors corresponding to all local extraction nodes of the distributed morphological image. Obtain the matrix values of the node noise feature matrix, denoted as noise intensity values; set the threshold range of noise intensity values and the corresponding denoising intensity coefficients for the noise intensity value range, and match the denoising intensity coefficients of the noise intensity values according to the noise intensity value range.
[0008] Furthermore, the process of obtaining a distributed denoised image by denoising the distributed morphological image based on the denoising intensity coefficient includes: The initial denoising coefficient of the local extraction node is set based on the location data of the local extraction node, and the initial denoising coefficient is compared with the denoising intensity coefficient. If the initial denoising coefficient is less than the denoising intensity coefficient, the initial denoising coefficient of the local extraction node is replaced with the denoising intensity coefficient; otherwise, the initial denoising coefficient of the local extraction node is retained. The vector value of the noise feature vector of the corresponding node is obtained based on the initial denoising coefficient of the local extraction node, and is denoised as the local denoising intensity. Set a unit local spacing, take the locally extracted node as the center of the rectangle, and take the unit local spacing as the distance to the corresponding side of the rectangle to obtain the denoised range image; then, according to the local denoising intensity, perform denoising processing on the node noise feature vector of the denoised range image with the same local denoising intensity to obtain the local denoised image; then, generate a distributed denoised image by stitching together the local denoised images.
[0009] Furthermore, the process of obtaining the image feature set of the distributed denoised image and stitching the distributed denoised image together to generate the target restored image based on the image feature set includes: The image edges of the distributed denoised image are obtained, and then the image feature sets corresponding to the image edges are obtained; the image feature sets include SIFT feature points, HOG features and color features; the image feature sets corresponding to the image edges of each distributed denoised image of the target preprocessed image are aligned and stitched together according to the segmentation order to generate the target restored image.
[0010] Furthermore, the process of obtaining the denoising coefficients of the target restored image includes: The mean value of the local denoising intensity of each part of the distributed denoising image is obtained and denoised as the distributed denoising coefficient of the distributed denoising image; then the mean value of the distributed denoising coefficients of each distributed denoising image of the target restored image is obtained and denoised as the denoising coefficient of the target restored image.
[0011] Furthermore, the passability of the denoising coefficient is determined. If it fails, the target restored image is subjected to multiple denoising processes to obtain a passable target restored image. The process of determining the number of denoising cycles includes: Set the threshold range for the denoising coefficient of the target image and compare it with the denoising coefficient; If the denoising coefficients fall within a threshold range, then the corresponding denoising coefficients are considered acceptable and are marked as acceptable denoising coefficients; subsequently, the corresponding target reconstruction image is marked as an acceptable target reconstruction image. Conversely, if the corresponding denoising coefficient is not qualified, the corresponding target restored image will be marked as an unqualified target restored image; Set the adjustment length, adjust the initial denoising coefficients corresponding to each local extraction node of the unqualified target image according to the location data, obtain the adjusted denoising coefficients, and continue to denoise the unqualified target image according to the adjusted denoising coefficients to obtain the latest distributed denoised image, and then obtain the latest target image. Obtain the denoising coefficient of the latest target reconstruction image again and determine its passability; if it is not passable, continue to adjust the initial denoising coefficient until the denoising coefficient is passable, and record the number of adjustments as the denoising count; otherwise, mark the corresponding latest target reconstruction image as a passable target reconstruction image.
[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention sets up an edge node model, acquires a target preprocessed image based on the edge node model, and classifies the target preprocessed image into a distributed morphological image by image morphology; extracts features from the distributed morphological image to obtain the denoising intensity coefficient of the distributed morphological image; then denoises the distributed image according to the denoising intensity coefficient to obtain a distributed denoised image; then obtains the image feature set of the distributed denoised image, and stitches the distributed denoised image according to the image feature set to generate a target restored image; obtains the denoising coefficient of the target restored image; judges the qualification of the denoising coefficient, and if it is not qualified, the target restored image is denoised multiple times to obtain a qualified target restored image and the corresponding number of denoising times; effectively improving the image denoising effect. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0014] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0015] 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.
[0016] Please see Figure 1 As shown, a distributed image denoising method based on edge computing is described, the method comprising the following steps: Step S1: Set up an edge node model, acquire the target preprocessed image based on the edge node model, and classify the target preprocessed image into image morphology to obtain distributed morphology images; Step S2: Extract features from the distributed morphological image to obtain the denoising intensity coefficients of the distributed morphological image; then, denoise the distributed image based on the denoising intensity coefficients to obtain the distributed denoised image; Step S3: Obtain the image feature set of the distributed denoised image, and stitch the distributed denoised image together according to the image feature set to generate the target restored image; Step S4: Obtain the denoising coefficients of the target restored image; determine the passability of the denoising coefficients. If they are not passable, perform multiple denoising processes on the target restored image to obtain a passable target restored image and the corresponding number of denoising times.
[0017] Step S1 requires further refinement. An edge node model is set up, and the process of acquiring the target image preprocessing image based on the edge node model includes: The edge node model includes an edge acquisition node unit and an edge preprocessing node unit. The edge preprocessing node unit is equipped with a noise judgment subunit and a noise preprocessing subunit. The edge acquisition node unit is communicatively connected to various image acquisition terminals, used to acquire target images through the image acquisition terminals and send the target images to the noise judgment subunit to obtain the pixel data of the target images. The pixel data is then compared with a preset pixel data threshold. If the pixel data is less than or equal to the threshold, the corresponding target image is recorded as an abnormal target image, and a re-sending instruction to acquire the same acquisition range of the target image is sent to the corresponding image acquisition terminal, recorded as a re-acquisition instruction. Conversely, if the pixel data is greater than or equal to the threshold, the target image is sent to the noise preprocessing subunit. The noise preprocessing subunit is equipped with a minimum adjustment pixel data, used to pre-adjust the pixel data of the target image whose pixel data is less than the minimum adjustment pixel data to the minimum adjustment pixel data, thereby obtaining the target preprocessed image. Otherwise, the corresponding pixel data remains unchanged. In the above embodiments, it should be further explained that the minimum adjustment pixel data is used to represent the minimum pixel reference value for edge preprocessing, and the minimum adjustment pixel data is greater than the pixel data threshold, which is used to unify the basic quality of qualified images; for example, if the preset pixel data threshold is a (pixel brightness value, 0-255), and the minimum adjustment pixel data is set to b, then the pixel data of the qualified target image with pixel data < b will be adjusted to b; if the pixel data of the target image with pixel data ≥ b remains unchanged, the final output is a target preprocessed image with pixel data ≥ b.
[0018] It should be further explained that the process of classifying the target preprocessed image into image morphology to obtain distributed morphological images includes: The image shape of the target preprocessed image is obtained, including the image shape size and the image spatial size; a unit segmentation dataset for the image shape size is set; the unit segmentation dataset includes the unit segmentation shape size and the corresponding image shape size range; the corresponding unit segmentation shape size is obtained by matching the image shape size with the corresponding image shape size range. The target preprocessed image is segmented according to the unit segmentation shape size to obtain unit segmentation images. If the unit segmentation image corresponding to the edge of the target preprocessed image is smaller than the unit segmentation shape size, it is considered as one unit segmentation image. The unit segmentation space size of the unit segmentation image is obtained and compared with a preset unit segmentation space size threshold. If the size of the unit segmentation space is greater than the threshold, the unit segmentation image is divided equally to obtain a distributed morphological image; otherwise, it is recorded as a distributed morphological image.
[0019] In the above embodiments, it should be further explained that the image outline size is denoted as n×m, where n and m are both non-zero natural numbers, and the unit is millimeter, centimeter, etc.; the image space size is used to represent the space occupied by the image, denoted as q, where q is a non-zero natural number, and the unit is B (byte), KB (kilobyte), MB (megabyte), etc.; and the unit segment outline size is denoted as a×a, where a is a non-zero natural number.
[0020] Step S2 needs further clarification. The process of extracting features from the distributed morphological image to obtain the denoising intensity coefficients of the distributed morphological image includes: Set the deployment spacing, set several local extraction nodes of the distributed morphological image according to the deployment spacing, and then extract the node noise feature vector of the local extraction node. The node noise feature vector includes gray-level variance, Gaussian noise parameter and impulse noise density; generate a node noise feature matrix from the node noise feature vectors corresponding to all local extraction nodes of the distributed morphological image. In the above embodiments, it should be further explained that the rows of the node noise feature matrix represent the number of locally extracted nodes; the corresponding columns represent the dimensions of the node noise feature vector; the gray-level variance is used to characterize the dispersion of pixel brightness in the image area corresponding to the locally extracted node. The larger the gray-level variance, the stronger the noise interference in the area; the Gaussian noise parameter is the noise variance, used to quantify the distribution characteristics of Gaussian noise in the area; the impulse noise density is used to characterize the proportion of impulse noise (salt and pepper noise) in the area. The higher the impulse noise density, the denser the pixel mutation noise in the area. Obtain the matrix values of the node noise feature matrix and denote them as noise intensity values; set the threshold range of noise intensity values and the corresponding denoising intensity coefficients for the noise intensity value range, and match the denoising intensity coefficients of the noise intensity values according to the noise intensity value range; It should be further explained that the process of obtaining a distributed denoised image by denoising the distributed morphological image based on the denoising intensity coefficient includes: The initial denoising coefficient of the local extraction node is set based on the location data of the local extraction node, and the initial denoising coefficient is compared with the denoising intensity coefficient. If the initial denoising coefficient is less than the denoising intensity coefficient, the initial denoising coefficient of the local extraction node is replaced with the denoising intensity coefficient; otherwise, the initial denoising coefficient of the local extraction node is retained. The vector value of the noise feature vector of the corresponding node is obtained based on the initial denoising coefficient of the local extraction node, and is denoised as the local denoising intensity. Set a unit local spacing, take the locally extracted node as the center of the rectangle, and take the unit local spacing as the distance to the corresponding side of the rectangle to obtain the denoised range image; then, according to the local denoising intensity, perform denoising processing on the node noise feature vector of the denoised range image with the same local denoising intensity to obtain the local denoised image; then, stitch the local denoised images to generate a distributed denoised image. In the above embodiments, it should be further explained that the location data is used to represent the position of the distributed morphological image according to the deployment spacing. The deployment spacing is adaptively set according to the size of the distributed morphological image to ensure that the local extraction nodes can uniformly cover the entire distributed morphological image and avoid denoising blind spots. The setting of the initial denoising coefficient is positively correlated with the location data. For example, the closer the local extraction node is to the center area of the distributed morphological image, the closer its distance to the image center, and the larger the corresponding initial denoising coefficient. Conversely, the closer the local extraction node is to the edge area of the distributed morphological image, the farther its distance to the image center, and the smaller the corresponding initial denoising coefficient.
[0021] Step S3 requires further refinement. The process of obtaining the image feature set of the distributed denoised images and stitching the distributed denoised images together to generate the target restored image based on the image feature set includes: The image edges of the distributed denoised image are obtained, and then the image feature sets corresponding to the image edges are obtained; the image feature sets include SIFT feature points, HOG features and color features; the image feature sets corresponding to the image edges of each distributed denoised image of the target preprocessed image are aligned and stitched together according to the segmentation order to generate the target restored image; In the above embodiments, it should be further explained that the SIFT feature points are used to capture the scale-invariant features of the image corners, ensuring that distributed denoised images with different angles and scaling ratios can be accurately matched; the HOG features are used to characterize the gradient direction distribution of the image corners, helping to confirm the contour shape of the corners and improving the stitching alignment accuracy; the color features are used to characterize the color distribution pattern around the image corners, supplementing the deficiencies of the contour features and ensuring that the color transition of the stitched image is natural; wherein, the target restored image and the target preprocessed image have the same image size.
[0022] Step S4 requires further clarification; the process of obtaining the denoising coefficients of the target restored image includes: The mean value of the local denoising intensity of each part of the distributed denoising image is obtained and denoised as the distributed denoising coefficient of the distributed denoising image; then the mean value of the distributed denoising coefficients of each distributed denoising image of the target restored image is obtained and denoised as the denoising coefficient of the target restored image. It should be further explained that the process of determining the passability of the denoising coefficient, and if it fails, performing multiple denoising processes on the target restored image to obtain a passable target restored image, and the number of denoising operations, includes: Set the threshold range for the denoising coefficient of the target image and compare it with the denoising coefficient; If the denoising coefficients fall within a threshold range, then the corresponding denoising coefficients are considered acceptable and are marked as acceptable denoising coefficients; subsequently, the corresponding target reconstruction image is marked as an acceptable target reconstruction image. Conversely, if the corresponding denoising coefficient is not qualified, the corresponding target restored image will be marked as an unqualified target restored image; Set the adjustment length, adjust the initial denoising coefficients corresponding to each local extraction node of the unqualified target image according to the location data, obtain the adjusted denoising coefficients, and continue to denoise the unqualified target image according to the adjusted denoising coefficients to obtain the latest distributed denoised image, and then obtain the latest target image. Obtain the denoising coefficient of the latest target reconstruction image again and determine its passability; if it is not passable, continue to adjust the initial denoising coefficient until the denoising coefficient is passable. Record the number of adjustments as the denoising count; otherwise, mark the corresponding latest target reconstruction image as a passable target reconstruction image. In the above embodiments, it should be further explained that the initial denoising coefficient is adjusted according to the denoising coefficient threshold range; for example, if the denoising coefficient threshold range is set to [0.4, 0.7], and the denoising coefficient = 0.35, which is less than 0.4, then the denoising is insufficient, and the initial denoising coefficient needs to be increased; if the denoising coefficient = 0.82, which is greater than 0.7, then the denoising is excessive, and the initial denoising coefficient needs to be decreased. The adjustment length can be set according to the actual situation, such as 0.1, 0.11, etc. During the adjustment process, if the initial denoising coefficient has reached the upper or lower limit, no adjustment is made. After the adjustment is completed, the denoising and stitching process is re-executed to obtain the latest target restored image, and the denoising coefficient is recalculated. If it is still not qualified, if the denoising coefficient = 0.62, it is qualified if it is within the threshold range; if it is still <0.4, continue to adjust according to the above rules until it is qualified, and record the number of adjustments.
[0023] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A distributed image denoising method based on edge computing, characterized in that, The method includes the following steps: Step S1: Set up an edge node model, acquire the target preprocessed image based on the edge node model, and classify the target preprocessed image into image morphology to obtain distributed morphology images; Step S2: Extract features from the distributed morphological image to obtain the denoising intensity coefficients of the distributed morphological image; then, denoise the distributed morphological image based on the denoising intensity coefficients to obtain a distributed denoised image; Step S3: Obtain the image feature set of the distributed denoised image, and stitch the distributed denoised image together according to the image feature set to generate the target restored image; Step S4: Obtain the denoising coefficients of the target restored image; determine the passability of the denoising coefficients. If they are not passable, perform multiple denoising processes on the target restored image to obtain a passable target restored image and the corresponding number of denoising times.
2. The distributed image denoising method based on edge computing according to claim 1, characterized in that, The process of setting up an edge node model and acquiring a target image preprocessing image based on the edge node model includes: The edge node model includes an edge acquisition node unit and an edge preprocessing node unit. The edge preprocessing node unit is equipped with a noise judgment subunit and a noise preprocessing subunit. The edge acquisition node unit is communicatively connected to various image acquisition terminals. It is used to acquire target images through the image acquisition terminals and send the target images to the noise judgment subunit to obtain the pixel data of the target images. Then, it is judged against a preset pixel data threshold. If the pixel data is less than or equal to the pixel data threshold, the corresponding target image is recorded as an abnormal target image, and an instruction to acquire the target image within the same acquisition range is resent to the image acquisition terminal of the corresponding abnormal target image, which is recorded as a re-acquisition instruction. Otherwise, the target image is sent to the noise preprocessing subunit. The noise preprocessing subunit is equipped with a minimum adjustment pixel data, which is used to pre-adjust the pixel data of the target image whose pixel data is less than the minimum adjustment pixel data to the minimum adjustment pixel data, thereby obtaining the target preprocessed image. Otherwise, the corresponding pixel data does not change.
3. The distributed image denoising method based on edge computing according to claim 2, characterized in that, The process of classifying the morphology of the preprocessed target image to obtain distributed morphological images includes: The image shape of the target preprocessed image is obtained, and the image shape includes the image outline size and the image spatial size; a unit segmentation dataset for the image outline size is set; the unit segmentation dataset includes the unit segmentation shape size and the corresponding image outline size range; the image outline size is matched with the corresponding image outline size range to obtain the corresponding unit segmentation shape size; The target preprocessed image is segmented according to the unit segmentation shape size to obtain unit segmentation images. If the unit segmentation image corresponding to the edge of the target preprocessed image is smaller than the unit segmentation shape size, it is considered as one unit segmentation image. The unit segmentation space size of the unit segmentation image is obtained and compared with a preset unit segmentation space size threshold. If the size of the unit segmentation space is greater than the threshold, the unit segmentation image is divided equally to obtain a distributed morphological image; otherwise, it is recorded as a distributed morphological image.
4. The distributed image denoising method based on edge computing according to claim 3, characterized in that, The process of extracting features from distributed morphological images to obtain the denoising intensity coefficients of the distributed morphological images includes: Set the deployment spacing, set several local extraction nodes of the distributed morphological image according to the deployment spacing, and then extract the node noise feature vector of the local extraction node. The node noise feature vector includes gray-level variance, Gaussian noise parameter and impulse noise density; generate a node noise feature matrix from the node noise feature vectors corresponding to all local extraction nodes of the distributed morphological image. Obtain the matrix values of the node noise feature matrix, denoted as noise intensity values; set the threshold range of noise intensity values and the corresponding denoising intensity coefficients for the noise intensity value range, and match the denoising intensity coefficients of the noise intensity values according to the noise intensity value range.
5. The distributed image denoising method based on edge computing according to claim 4, characterized in that, The process of obtaining a distributed denoised image by denoising a distributed image based on the denoising intensity coefficient includes: The initial denoising coefficient of the local extraction node is set based on the location data of the local extraction node, and the initial denoising coefficient is compared with the denoising intensity coefficient. If the initial denoising coefficient is less than the denoising intensity coefficient, the initial denoising coefficient of the local extraction node is replaced with the denoising intensity coefficient; otherwise, the initial denoising coefficient of the local extraction node is retained. The vector value of the noise feature vector of the corresponding node is obtained based on the initial denoising coefficient of the local extraction node, and is denoised as the local denoising intensity. Set a unit local spacing, take the locally extracted node as the center of the rectangle, and take the unit local spacing as the distance to the corresponding side of the rectangle to obtain the denoised range image; then, according to the local denoising intensity, perform denoising processing on the node noise feature vector of the denoised range image with the same local denoising intensity to obtain the local denoised image; then, stitch the local denoised images to generate a distributed denoised image.
6. The distributed image denoising method based on edge computing according to claim 5, characterized in that, The process of obtaining the image feature set of the distributed denoised image and stitching the distributed denoised image together to generate the target restored image based on the image feature set includes: The image edges of the distributed denoised image are obtained, and then the image feature sets corresponding to the image edges are obtained; the image feature sets include SIFT feature points, HOG features and color features; the image feature sets corresponding to the image edges of each distributed denoised image of the target preprocessed image are aligned and stitched together according to the segmentation order to generate the target restored image.
7. A distributed image denoising method based on edge computing according to claim 6, characterized in that, The process of obtaining the denoising coefficients of the target restored image includes: The mean value of the local denoising intensity of each part of the distributed denoising image is obtained and denoised as the distributed denoising coefficient of the distributed denoising image; then the mean value of the distributed denoising coefficients of each distributed denoising image of the target restored image is obtained and denoised as the denoising coefficient of the target restored image.
8. A distributed image denoising method based on edge computing according to claim 7, characterized in that, The process of determining the passability of the denoising coefficient, and if it fails, performing multiple denoising processes on the target restored image to obtain a passable target restored image, and the number of denoising steps, includes: Set the threshold range for the denoising coefficient of the target image and compare it with the denoising coefficient; If the denoising coefficients are within the denoising coefficient threshold range, then the corresponding denoising coefficients are qualified and marked as qualified denoising coefficients; then the corresponding target restored image is marked as a qualified target restored image; Conversely, if the corresponding denoising coefficient is not qualified, the corresponding target restored image will be marked as an unqualified target restored image; Set the adjustment length, adjust the initial denoising coefficients corresponding to each local extraction node of the unqualified target image according to the location data, obtain the adjusted denoising coefficients, and continue to denoise the unqualified target image according to the adjusted denoising coefficients to obtain the latest distributed denoised image, and then obtain the latest target image. Obtain the denoising coefficient of the latest target reconstruction image again and determine its passability; if it is not passable, continue to adjust the initial denoising coefficient until the denoising coefficient is passable, and record the number of adjustments as the denoising count; otherwise, mark the corresponding latest target reconstruction image as a passable target reconstruction image.