An adaptive weighted Gaussian curvature filtering method based on image edge indication function

By using edge indicator function and adaptive weighted Gaussian curvature filtering method in image denoising, the problems of noise and detail balance are solved, and efficient denoising and detail retention are achieved, avoiding the problems of excessive edge smoothing and incomplete denoising.

CN114359076BActive Publication Date: 2024-11-01WUHAN UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111576705.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-22
Publication Date
2024-11-01
Estimated Expiration
2041-12-22

AI Technical Summary

Technical Problem

Existing image denoising methods are difficult to find a balance between removing noise and retaining details, especially in the case of high noise intensity, where edges are overly smoothed or incompletely denoised.

Method used

Adaptive weighted Gaussian curvature filtering method based on image edge indicator function is used to decompose the image space into different regions, calculate the edge indicator value and weight, and construct the adaptive fractional-integer order energy functional and local weighted projection operator for precise control. Iterative updates to avoid edge diffusion and noise points affect each other.

Benefits of technology

While improving the denoising performance, it maintains the details of the image texture and quickly realizes image denoising, avoids incomplete denoising or excessive edge smoothing, and improves the operating efficiency and sparsity of the algorithm.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114359076B_ABST
    Figure CN114359076B_ABST
Patent Text Reader

Abstract

The present invention provides an adaptive weighted Gaussian curvature filtering method based on an image edge indicator function, improves the Gaussian curvature filtering algorithm in the curvature filtering algorithm, adopts the edge indicator function to construct an adaptive fractional-integer order energy functional and a local weighted projection operator, and adaptively and accurately controls the iterative update of the weighted projection operator by minimizing the energy functional, thereby improving the denoising performance and better maintaining the texture details of the image. The present invention uses an adaptive method to adjust the fractional order and the weights of each item in the energy functional, can accurately control the iterative update of the projection operator, and effectively prevents incomplete image denoising or over-smoothing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of digital image processing, and in particular to an adaptive weighted Gaussian curvature filtering method based on an image edge indication function. Background Art

[0002] Noise often appears as isolated pixels or blocks of pixels that cause strong visual effects on images, directly affecting subsequent image analysis and understanding. The purpose of image denoising is to reduce the noise in the image while keeping the edge texture details of the image as intact as possible. However, denoising and edge preservation are often contradictory processes. How to find a better balance between removing noise and retaining details is the focus of image denoising research.

[0003] At present, the main image denoising methods include those based on deep learning, transform domain, sparse dictionary, non-local self-similar prior and partial differential equation. The image denoising method based on deep learning has good effect, but it requires a large number of noisy image and clean image sample pairs as training data. When it is difficult to obtain clean images, the denoising performance will be greatly affected. The denoising method based on transform domain depends on the transform model and threshold, and it is easy to cause excessive smoothing of edges and textures. The denoising method based on sparse dictionary needs to construct an overcomplete dictionary, and the quality of the dictionary directly affects the denoising performance. The denoising method based on non-local self-similar prior usually uses the redundant information in the image to remove noise. This type of algorithm has a good denoising effect on Gaussian noise when the noise intensity is known, but a poor denoising effect on salt and pepper noise. In addition, the algorithm complexity is high and the denoising efficiency is low.

[0004] The denoising method based on partial differential equations constrains image information by minimizing the energy function. It has the characteristics of local adaptability and easy numerical implementation, and has become a popular image denoising method in recent years. Among them, the model based on anisotropic diffusion can better solve the contradiction between edge preservation and denoising, but the denoising effect is poor for isolated noise or when the noise intensity is large, and the noise removal is incomplete. The denoising method based on the total variation model is relatively simple and can effectively protect the edge texture while removing noise. Among them, the classic total variation (TV) method regards the image as a piecewise continuous function in a bounded variation space, and achieves the purpose of smooth denoising by minimizing the image energy function. The energy functional of the subsequent total variation denoising model adds fractional total variation regularization terms, integer total variation regularization terms, and tight frame sparse regularization terms to improve the sparsity of the denoised image. Although the total variation denoising model can better characterize the image edge, this type of method is time-consuming and the denoised image is prone to staircase artifacts. In order to improve the denoising performance, some algorithms use the total variation model of Gaussian curvature or differential curvature, such as using the local weighted Gaussian curvature as the variation regularization term, or using the fast Fourier transform to solve the partial differential equations involved in curvature filtering. However, this type of algorithm needs to explicitly calculate the gradient of the energy functional and converges slowly when minimizing the energy functional. Subsequently, some algorithms improve the denoising efficiency by obtaining approximate solutions to the variational problem, such as constructing a filter projection operator to implicitly use the image curvature information to perform curvature filtering on the image. Subsequently, some algorithms modify the projection operator and use the local variance to correct the energy function of the regularization term to enhance the denoising ability. Some algorithms use the median grayscale similarity function to weight the curvature projection operator to quickly reduce the curvature filtering energy. The above algorithms can achieve fast image blind denoising without knowing the noise intensity, but they do not distinguish between noise points and edge pixels, so that noise points and edge pixels affect each other. At the same time, when the noise intensity is large, this type of algorithm has the problem of incomplete denoising or excessive edge smoothing. Summary of the invention

[0005] In view of the above-mentioned technical problems, the purpose of the present invention is to provide an adaptive weighted Gaussian curvature filtering method based on an image edge indicator function, to improve the Gaussian curvature filtering algorithm in the curvature filtering algorithm, to adopt the edge indicator function to construct an adaptive fractional-integer order energy functional and a local weighted projection operator, to adaptively and accurately control the iterative update of the weighted projection operator by minimizing the energy functional, to improve the denoising performance while better maintaining the texture details of the image.

[0006] The technical solution of the present invention is an adaptive weighted Gaussian curvature filtering method based on an image edge indication function, comprising the following steps:

[0007] Step 1, decompose the current noise image into four types of non-intersecting and staggered areas, calculate the edge indicator value for each pixel, and calculate the weight of the neighborhood pixels of the central pixel according to the edge indicator value;

[0008] Step 2, select the triangular tangent plane of the central pixel point to construct a Gaussian weighted projection operator, and calculate the weighted projection distance from the central pixel point to M tangent planes according to the pixel weight in step 1;

[0009] Step 3: According to the weighted projection distance calculated in step 2, the minimum projection distance is selected to update the central pixel. When all the pixels in the image are updated, the image is updated once.

[0010] Step 4: According to the denoised image obtained in step 3, the total energy is calculated using the adaptive Gaussian energy functional formula. If the total energy is less than the total energy of the image before updating, go to step 1; otherwise, stop updating and output the denoised image.

[0011] Furthermore, in step 1, the curvature filtering algorithm is used to decompose the input noise image U into image space, dividing the image into white circles Ω WC 、Black circle Ω BC 、White triangle Ω WT and black triangle Ω BT Four types of disjoint and staggered regions are used to eliminate the dependencies between adjacent pixels.

[0012] Furthermore, the two-dimensional image edge indicator value D of the pixel point p in each region defined in step 1 is p and the normalized edge indicator value for:

[0013]

[0014] Among them, image U is the noise image, Represents the second-order derivative of image U in the gradient direction, U x is the first-order derivative of the image in the x direction, U y is the first-order derivative of the image in the y direction, U xx is the second-order derivative of the image in the x direction, U yy is the second-order derivative of the image in the y direction, U xy is the second-order partial derivative of the image in the xy direction; represents the second-order derivative of the image U in the vertical direction; |·| represents the absolute value; Ω represents the set of pixels in the entire image, Represents the maximum value of the edge indication values ​​of all pixels in the Ω area.

[0015] Furthermore, in step 1, the weights of the neighborhood pixels of the central pixel are obtained as follows;

[0016] When denoising an image, a weighted projection operator is used to update the pixel point p, and the weight of its neighborhood pixel point q is defined as:

[0017]

[0018] Among them, p = (i, j) is the center pixel; is the normalized edge indicator value of p; is the normalized edge indicator value of the neighborhood pixel q;

[0019] If p is an edge pixel, when q is an edge pixel, then w q The value is large. When q is a noise or flat area pixel, w q The value is small; if p is a noise pixel or a flat area pixel, when q is an edge pixel, w q The value is small. When q is a noise or flat area pixel, w q The value is larger; through w q It can suppress the diffusion of surrounding pixels to the edge, accelerate the diffusion of noise points, and better maintain the texture details of the image while improving the denoising performance. By defining the weight of the neighborhood pixel point q, the weight matrix calculated using a 3×3 pixel neighborhood is as follows:

[0020]

[0021] Among them, w i,j is the weight of the central pixel p=(i,j), and the other values ​​are the weights of the neighboring pixels of the central pixel.

[0022] Furthermore, in step 2, 4 half-window tangent planes containing 5 pixels and 4 triangular tangent planes containing 3 pixels are selected, and a total of 8 tangent planes are used to construct a weighted projection operator.

[0023] Furthermore, in step 2, the weighted projection distance from the central pixel to the eight tangent planes is calculated as follows;

[0024] According to the weight of the central pixel neighborhood obtained in step 1, calculate the weighted projection distance d from the central pixel to the eight tangent planes i (i=1,2,…,8), where d1 to d4 are the weighted projection distances from the center pixel to the four half-window tangent planes, d5 to d8 are the weighted projection distances from the center pixel to the four triangular tangent planes, and the weighted projection operator of the pixel point p=(i,j) is as shown in the formula:

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033] Among them, w · is the weight of the neighborhood pixel calculated by formula (3) in step 2, U i,j is the pixel value of the pixel at coordinate (i, j); sum i , (i=1,…,8) is the required neighborhood pixel weight sum; Normalize the weights; d1,…,d8 are weighted projection distances.

[0034] Furthermore, step 3 is implemented by selecting the projection distance d that causes the smallest change in the intensity of the central pixel from the weighted projection distances of the developable approximate tangent planes of the M local pixel neighborhoods calculated in step 2. min :

[0035] |d min |=min{|d1|,|d2|,…,|d M |}

[0036] Update the pixel value of point p to Among them U p is the original pixel value of point p, The updated pixel value is repeated for all pixels in the image to complete an image update.

[0037] Furthermore, the energy functional of the adaptive weighted Gaussian curvature filtering algorithm in step 4 is defined as follows:

[0038]

[0039] in, and are the normalized denoised image and input image respectively; E GC is the total Gaussian energy; is the adaptive fractional-order data fitting term; is the Gaussian regularization term; is the tight frame regularization term; is the order of the adaptive fractional order of the data fitting term, represents the normalized edge indicator value; is the weight of the Gaussian regularization term; represents Gaussian curvature; is the weight of the tight frame regularization term; W is the tight frame operator, which satisfies W T W = I; |·| represents the absolute value, p represents the pixel point, and Ω represents the set of pixels in the entire image;

[0040] calculate The formulas are:

[0041]

[0042]

[0043]

[0044]

[0045] in, for The first derivative in the x direction, for The first derivative in the y direction, for The second derivative in the x direction, for The second derivative in the y direction, for Second-order partial derivatives in the xy direction; adaptive fractional order of the data fitting term Gaussian regularization term coefficient Tight frame regularization term coefficient The adjustment coefficients μ1 and μ2 are constants.

[0046] The present invention has the following advantages and positive effects:

[0047] (1) The present invention uses an edge indicator function to determine whether a central pixel is an edge pixel, a noise point, or a flat point, and calculates the weights of the neighboring pixels according to the types of the central pixel and the neighboring pixels, thereby suppressing the diffusion of the edge and accelerating the diffusion of the noise points, thereby achieving denoising while maintaining edge texture details;

[0048] (2) The present invention uses a developable surface to locally approximate the image to reduce the energy of the regularization term, and adopts a weighted projection operator to update the image, which can quickly perform image denoising and improve the operation efficiency of the algorithm;

[0049] (3) The energy functional of the present invention includes three parts: fractional-order data fitting term, variational regularization term and tight frame regularization term, which effectively improves the sparsity of the image;

[0050] (4) The present invention uses an adaptive method to adjust the fractional order and the weights of each term in the energy functional, which can accurately control the iterative update of the projection operator and effectively prevent incomplete image denoising or over-smoothing. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of image space decomposition according to an embodiment of the present invention.

[0052] Figure 2 FIG. 4 is a schematic diagram of a Gaussian filtering cutting plane according to an embodiment of the present invention.

[0053] Figure 3 Schematic diagram of the distance from the central pixel point to the cutting plane according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0055] The technical solution of the present invention can use computer software technology to realize the automatic operation process. The process of the adaptive weighted curvature filtering algorithm based on the image edge indication function provided in the embodiment includes the following steps in sequence:

[0056] 1. Perform image space decomposition on the input noisy image, calculate the edge indicator value of the pixel and assign weights to the neighborhood pixels of the central pixel.

[0057] The present invention uses the curvature filtering algorithm to first decompose the input noise image U into image space, and divides the image into white circles Ω WC 、Black circle Ω BC 、White triangle Ω WT and black triangle Ω BT There are four types of disjoint and staggered regions to eliminate the dependencies between adjacent pixels, such as Figure 1 shown.

[0058] The embodiment of the invention effectively distinguishes the pixel points in the image as edge points, isolated noise points and flat points based on the two-dimensional image edge indication function of the differential curvature, and defines the two-dimensional image edge indication value D of the pixel point p. p and the normalized edge indicator value for:

[0059]

[0060] in Represents the second-order derivative of image U in the gradient direction, U x is the first-order derivative of the image in the x direction, U y is the first-order derivative of the image in the y direction, U xxis the second-order derivative of the image in the x direction, U yy is the second-order derivative of the image in the y direction, U xy is the second-order partial derivative of the image in the xy direction; represents the second-order derivative of the image U in the vertical direction; |·| represents the absolute value; Ω represents the set of pixels in the entire image, Represents the maximum value of the edge indication values ​​of all pixels in the Ω area.

[0061] Experimental results show that if p is an edge pixel, |U hh |The value is very large, |U ξξ | value is small, so The value is very large; if p is a pixel point in a flat area, |U hh |and|U ξξ The values ​​of | are all small, so The value is very small; if p is a noise point, |U hh |and|U ξξ The value of | is large and the difference is small, so The value is very small. The size of the value can better distinguish the image edges, flat areas and noise points.

[0062] The embodiment defines the weight of the neighborhood pixel q of the central pixel p as follows:

[0063]

[0064] Among them, p = (i, j) is the center pixel; is the normalized edge indicator value of p; is the normalized edge indicator value of the neighborhood pixel q.

[0065] The experimental results show that if p is an edge pixel and q is an edge pixel, then w q The value is large. When q is a noise or flat area pixel, w q The value is small; if p is a noise pixel or a flat area pixel, when q is an edge pixel, w q The value is small. When q is a noise or flat area pixel, w q The value is larger. q It can suppress the diffusion of surrounding pixels to the edge and accelerate the diffusion of noise points, so that the algorithm can better maintain the texture details of the image while improving the denoising performance. The weight matrix calculated by the 3×3 pixel neighborhood in this embodiment is as follows:

[0066]

[0067] Among them, w i,jis the weight of the central pixel p=(i,j), and the other values ​​are the weights of the neighboring pixels of the central pixel.

[0068] 2. Use the weighted Gaussian curvature filtering algorithm to calculate the weighted projection distance from the pixel to the tangent plane.

[0069] In the curvature filtering process, in order to avoid explicitly calculating Gaussian curvature and differential curvature, the pixel value of each point is directly adjusted to be located on the tangent plane of the neighboring pixels to meet the segmented expandability. In view of the problem that the Gaussian curvature filtering algorithm does not maintain the edge texture details of the denoised image ideally, the present invention first selects 8 tangent planes to calculate the corresponding weighted projection operator. Figure 2 Select 4 half-window tangent planes containing 5 pixels and 4 triangular tangent planes containing 3 pixels, a total of 8 tangent planes to enhance the projection operator. Then use the edge indicator function to calculate the weighted projection distance from the center pixel to the 8 tangent planes, where d1 to d4 are the weighted projection distances from the center pixel to the 4 half-window tangent planes, and d5 to d8 are the weighted projection distances from the center pixel to the 4 triangular tangent planes. The weighted projection operator of the pixel point p = (i, j) is shown as follows:

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078] Among them, w · is the weight of the neighborhood pixel calculated by formula (3) in step 2, U i,j is the pixel value of the pixel at coordinate (i, j); sum i , (i=1,…,8) is the required neighborhood pixel weight sum; Normalize the weights; d1,…,d8 are weighted projection distances.

[0079] 3. Find the minimum projection distance from the pixel to the tangent plane and update the image.

[0080] Any developable surface can be locally approximated by its tangent plane, so the present invention uses a local developable surface to approximate the image to reduce the energy of the regularization term. Among the projection distances of the developable approximate tangent planes of the 8 local pixel neighborhoods selected in step 2, the projection distance with the smallest intensity change to the center pixel point is selected, that is, the minimum absolute distance is used as the minimum projection distance from the center pixel point to the tangent plane. The minimum projection distance is defined as:

[0081] |d min |=min{|d1|,|d2|,…,|d8|} (12)

[0082] The present invention updates the pixel value of point p to be Among them U p is the original pixel value of point p, is the updated pixel value. The image can be updated once by finding the minimum projection distance for all pixels in the image and updating them.

[0083] The experimental results show that when the Gaussian noise intensity reaches 50, the image denoising task can be completed through at most 6 iterative updates, and when the salt and pepper noise intensity reaches 0.1, the denoising task can be completed through at most 5 iterative updates, which proves that the algorithm has a faster convergence speed and maintains good edge texture details of the image while denoising.

[0084] 4. Calculate the total energy according to the adaptive Gaussian energy functional formula, compare the total energy of the updated image with the total energy before the update, and control the filtering of the image by minimizing the energy functional.

[0085] In the curvature filtering process, it is necessary to find a minimized energy functional to control the iterative update of the curvature filtering to avoid incomplete denoising or excessive smoothing of the image. The present invention is based on the edge indicator function and adopts an adaptive weighted Gaussian curvature energy functional, whose energy functional is defined as follows:

[0086]

[0087] in, and are the normalized denoised image and input image respectively; E GC is the total Gaussian energy; is the adaptive fractional-order data fitting term; is the Gaussian regularization term; is the tight frame regularization term; is the order of the adaptive fractional order of the data fitting term, represents the normalized edge indicator value; is the weight of the Gaussian regularization term; represents Gaussian curvature; is the weight of the tight frame regularization term; W is the tight frame operator, which satisfies W T W=I; |·| represents the absolute value, p represents a pixel, and Ω represents the set of pixels in the entire image.

[0088] calculate The formulas are:

[0089]

[0090]

[0091]

[0092]

[0093] in, for The first derivative in the x direction, for The first derivative in the y direction, for The second derivative in the x direction, for The second derivative in the y direction, for Second-order partial derivatives in the xy direction; adaptive fractional order of the data fitting term Gaussian regularization term coefficient Tight frame regularization term coefficient Experimental results show that when the adaptive Gaussian energy functional removes noise, in the image filtering iterative process, The value of remains unchanged or increases, and The value of remains unchanged or decreases. According to the adaptive Gaussian energy functional proposed in this paper, when the noise is removed, As the value increases, the energy of the data fitting term will be suppressed. The rise of As the value increases, the regularization term energy will increase. and The decrease in the total energy E GC is on a downward trend; when the image is over-smoothed, and As the value decreases, the energy of the data fitting term will be accelerated. The rise of and The decrease in the total energy E GC Therefore, according to the change of the above total energy, the update of the image U can be effectively controlled to ensure better noise removal and prevent the image from being over-smoothed.

[0094] In this embodiment, when calculating the total energy using the energy functional, the adjustment coefficients μ1 and μ2 are constants, and their values ​​are 35 and 10.5 respectively. Total Energy and the total energy E before update GC (U) size, if the total energy Less than the total energy E before update GC (U), then let Repeat the above steps to update the image again, otherwise stop the iteration and output the denoised image Experiments have shown that the adaptive weighted Gaussian curvature energy functional proposed in the present invention can accurately control the update iteration of the image, ensure better noise removal, and prevent the image from being over-smoothed.

[0095] Experimental results show that this technical solution can effectively remove noise from images and better preserve the edge texture details of the image. Compared with the existing curvature filtering method, this algorithm improves the peak signal-to-noise ratio and structural similarity index of the denoised image, and has adaptability and high operating efficiency.

[0096] The specific embodiments described herein are merely examples of the present invention. A person skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner without departing from the spirit of the present invention or exceeding the scope defined by the appended claims.

Claims

1. An adaptive weighted Gaussian curvature filtering method based on an image edge indication function, characterized in that: The following steps are involved: Step 1, decompose the current noise image into four types of non-intersecting and staggered areas, calculate the edge indicator value for each pixel, and calculate the weight of the neighborhood pixels of the central pixel according to the edge indicator value; The two-dimensional image edge indicator value D of the pixel point p in each area defined in step 1 p and the normalized edge indicator value for: Among them, image U is the noise image, Represents the second-order derivative of image U in the gradient direction, U x is the first-order derivative of the image in the x direction, U y is the first-order derivative of the image in the y direction, U xx is the second-order derivative of the image in the x direction, U yy is the second-order derivative of the image in the y direction, U xy is the second-order partial derivative of the image in the xy direction; represents the second-order derivative of the image U in the vertical direction; |·| represents the absolute value; W represents the set of pixels in the entire image, Represents the maximum value of the edge indication values ​​of all pixels in the two-dimensional image in the W area; In step 1, the weights of the neighborhood pixels of the central pixel are obtained as follows: When denoising an image, a weighted projection operator is used to update the pixel point p, and the weight of its neighborhood pixel point q is defined as: Among them, p = (i, j) is the center pixel; is the normalized edge indicator value of p; is the normalized edge indicator value of the neighborhood pixel q; If p is an edge pixel, when q is an edge pixel, then w q The value is large. When q is a noise or flat area pixel, w q The value is small; if p is a noise pixel or a flat area pixel, when q is an edge pixel, w q The value is small. When q is a noise or flat area pixel, w q The value is larger; through w q It can suppress the diffusion of surrounding pixels to the edge, accelerate the diffusion of noise points, and better maintain the texture details of the image while improving the denoising performance. By defining the weight of the neighborhood pixel point q, the weight matrix calculated using the 3′3 pixel neighborhood is as follows: Among them, w i,j is the weight of the central pixel p=(i,j), and the other values ​​are the weights of the neighboring pixels of the central pixel; Step 2, select the triangular tangent plane of the central pixel point to construct a Gaussian weighted projection operator, and calculate the weighted projection distance from the central pixel point to M tangent planes according to the pixel weight in step 1; In step 2, the weighted projection distance from the central pixel to the eight tangent planes is calculated as follows; According to the weight of the central pixel neighborhood obtained in step 1, calculate the weighted projection distance d from the central pixel to the eight tangent planes i (i=1,2,…,8), where d1 to d4 are the weighted projection distances from the center pixel to the four half-window tangent planes, d5 to d8 are the weighted projection distances from the center pixel to the four triangular tangent planes, and the weighted projection operator of the pixel point p=(i,j) is as shown in the formula: Among them, w · is the weight of the neighborhood pixel calculated by formula (3) in step 2, U i,j is the pixel value of the pixel at coordinate (i, j); sum i , (i=1,…,8) is the required neighborhood pixel weight sum; Normalize the weights; d1,…,d8 are weighted projection distances; Step 3: According to the weighted projection distance calculated in step 2, the minimum projection distance is selected to update the central pixel. When all the pixels in the image are updated, the image is updated once. Step 4: According to the denoised image obtained in step 3, the total energy is calculated using the adaptive Gaussian energy functional formula. If the total energy is less than the total energy of the image before updating, go to step 1; otherwise, stop updating and output the denoised image. The energy functional of the adaptive weighted Gaussian curvature filter algorithm in step 4 is defined as follows: in, and are the normalized denoised image and input image respectively; E GC is the total Gaussian energy; is the adaptive fractional-order data fitting term; is the Gaussian regularization term; is the tight frame regularization term; is the order of the adaptive fractional order of the data fitting term, represents the normalized edge indicator value; is the weight of the Gaussian regularization term; represents Gaussian curvature; is the weight of the tight frame regularization term; W is the tight frame operator, which satisfies W T W = I; |·| represents the absolute value, p represents the pixel point, and W represents the set of pixel points in the entire image; calculate The formulas are: in, for The first derivative in the x direction, for The first derivative in the y direction, for The second derivative in the x direction, for The second derivative in the y direction, for Second-order partial derivatives in the xy direction; adaptive fractional order of the data fitting term Gaussian regularization term coefficient Tight frame regularization term coefficient The adjustment coefficients μ1 and μ2 are constants.

2. The adaptive weighted Gaussian curvature filtering method based on image edge indication function according to claim 1, characterized in that: In step 1, the curvature filtering algorithm is used to decompose the input noise image U into image space, dividing the image into white circles W WC 、Black circle W BC 、White triangle W WT and black triangle W BT Four types of disjoint and staggered regions are used to eliminate the dependencies between adjacent pixels.

3. The adaptive weighted Gaussian curvature filtering method based on image edge indication function according to claim 1, characterized in that: In step 2, four half-window tangent planes containing five pixels and four triangular tangent planes containing three pixels are selected, for a total of eight tangent planes to construct a weighted projection operator.

4. The adaptive weighted Gaussian curvature filtering method based on image edge indication function according to claim 1, characterized in that: Step 3 is implemented by selecting the projection distance d that results in the smallest change in the intensity of the central pixel from the weighted projection distances of the developable approximate tangent planes of the M local pixel neighborhoods calculated in step 2. min : |d min |=min{d1|,|d2|,…,|d M } Update the pixel value of point p to Among them U p is the original pixel value of point p, The updated pixel value is repeated for all pixels in the image to complete an image update.