Improved mixed-order total variation-based image denoising implementation method

Abstract

The invention discloses a mixed-order total variation image denoising implementation method based on improvement. The method comprises the following steps: S1, carrying out gray level conversion and noise addition processing on a collected original image; s2, filtering is carried out by using Gaussian-Laplacian joint transformation; s3, introducing an anisotropic diffusion tensor into the TV1 model, and introducing an image feature extraction function based on a difference curvature into the LLT model; s4, coupling the TV1 model and the LLT model by using an adaptive weighting function to obtain a coupling model; s5, performing post-processing on a de-noised image obtained through the coupling model by using a fast non-local mean filtering algorithm; and S6, obtaining a discrete form of the coupling model by using a discrete difference method, updating a pixel value by using an iteration method for denoising to obtain a denoised image, and verifying the effectiveness and feasibility of the coupling model by using a peak signal-to-noise ratio and a structural similarity index. According to the invention, the integrity of image edge information can be ensured, and the de-noising efficiency is improved.

Description

Technical Field

[0001] The invention relates to the technical field of image denoising, and in particular to an image denoising implementation method based on improved mixed-order total variation. Background Art

[0002] In today's digital age, images have become an important form of information dissemination and storage, and are deeply integrated into many fields such as scientific research, industry, medical care, and entertainment. With the rise of artificial intelligence and computer vision applications, people have higher and higher requirements for image quality. The quality of images directly affects the accuracy of information transmission and the quality of user experience, and has become an important support for the development of various fields. However, in the process of acquisition, transmission, and storage, images will inevitably be interfered by noise and degraded, resulting in the inability of images to accurately present the original essential characteristics of things. In order to restore the image to a state that can truthfully reflect the true appearance of things, it is necessary to denoise the image through image processing technology. The total variation image denoising method is an important branch in the field of image denoising. Among them, the high-order variational LLT denoising model (LLT model) is prone to edge blurring due to excessive smoothing; based on The TV (Total Variation) denoising model (TV1 model) of the norm is an anisotropic diffusion model that can effectively retain edge and smooth noise, but it is easy to produce a "staircase effect" in the smooth area. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide an image denoising implementation method based on an improved mixed-order total variation to solve the problems of insufficient representation of image edge information, unsatisfactory denoising effect and low iteration efficiency.

[0004] Technical solution: A method for image denoising based on improved mixed-order total variation, comprising the following steps: S1, converting the collected original image into grayscale and performing noise processing; S2, filtering using Gaussian-Laplace joint transform; S3, anisotropic diffusion tensor is introduced into TV1 model, and image feature extraction function based on differential curvature is introduced into LLT model; S4, using an adaptive weight function to couple the TV1 model and the LLT model to obtain a coupled model; S5, post-processing the denoised image obtained by the coupling model using a fast non-local mean filtering algorithm; S6, the discrete difference method is used to obtain the discrete form of the coupling model, and the pixel value is updated by the iterative method for denoising to obtain the denoised image. The peak signal-to-noise ratio and structural similarity indicators are used to verify the effectiveness and feasibility of the coupling model.

[0005] Furthermore, after introducing the anisotropic diffusion tensor, the expression of the TV1 model is: , In the formula, is the minimized energy functional of the TV1 model, is the original grayscale image, is the noise image function, is the gradient of the noisy image function, is the weight parameter; represents the anisotropic diffusion tensor, which is expressed as: , , In the formula, the parameters and Used to adjust the strength of the tensor, A is the direction of the image gradient, is the normal vector of the image gradient, T represents the matrix transpose, and e represents the natural constant; After introducing the image feature extraction function based on differential curvature, the expression of the LLT model is: , In the formula, is the minimized energy functional of the LLT model; is an image feature extraction function based on differential curvature, and its expression is: , In the formula, represents the normalized difference curvature, Indicates a minimum value.

[0006] Furthermore, the expression of coupling the TV1 model and the LLT model using the adaptive weight function is: , In the formula, represents the minimized energy functional of the coupled model; the first term is the regularization term, and the second term is the fidelity term; is the noise image function The gradient modulus of is the adaptive weight parameter, which adjusts the weight of the fidelity term; is an adaptive weight function used to adjust the regularization term of the TV1 model and the LLT model; The adaptive weight parameter The expression is: , In the formula, ; The expression is: , In the formula, yes The minimum value of yes The maximum value of is the differential curvature, and its expression is: , In the formula, is the second-order derivative of the image along the gradient direction, is the second-order derivative of the image along the edge direction; its expressions are: , , In the formula, is the noise image function along The first derivative of the direction, is the noise image function along The first derivative of the direction, is the noise image function along The second derivative of the direction, is the noise image function along The second derivative of the direction, is the noise image function The second-order partial derivative of ; The adaptive weight function The expression is: , In the formula, is the arccosine function, which is a decreasing function; is the noise image function The gradient modulus of is the adaptive gradient threshold; , They are thresholds of two different values; is the noise visibility function; The expression is: , In the formula, For images Local variance, for The gray value at the point; is the local mean, and its expression is: , Wherein, h and q represent the coordinate values ​​of the x-axis and y-axis respectively, both of which are variables; P and Q are constants, representing the half width of the image area block in the x-axis and y-axis directions respectively; (2P+1) is the length of the image area block, (2Q+1) is the width of the image area block, and (2P+1)(2Q+1) represents the area size of the image area block.

[0007] Further, in step S5, in the fast non-local means filtering algorithm, for each pixel in the image, a pixel similar to it is searched within a search window centered on the pixel, and the similarity is measured by calculating the difference in grayscale values ​​of the neighborhoods around the two pixels.

[0008] Compared with the prior art, the present invention has the following significant effects: 1. The present invention uses Gaussian-Laplace transform instead of Gaussian filtering, optimizes the image preprocessing process, solves the problem of image edge weakening after Gaussian filtering, highlights edge details, and makes the edge clearer and sharper; 2. The present invention introduces anisotropic diffusion tensor into the TV1 model to accurately adjust the edge diffusion direction, effectively solving key problems in image denoising, such as edge protection, balance between noise removal and detail retention, thereby maximally retaining image details while removing noise; 3. The present invention introduces an image feature extraction function based on differential curvature into the LLT model, effectively adjusting the diffusion strength in different areas of the image, enhancing the smoothing strength at large noise points, and suppressing the "staircase effect" of the total variation model, so that the LLT model is more robust when processing large noise points, can reduce the generation of artifacts, and improve the quality of the denoised image; 4. The present invention couples the TV1 model and the LLT model by improving the adaptive weight function to obtain a corresponding coupling model, which can effectively remove noise while protecting image details, thereby improving the robustness and generalization ability of image processing; 5. The present invention uses a fast non-local mean filtering algorithm to post-process the denoised image obtained by the coupling model, which can restore details and texture information, further suppress residual noise, and thus improve the stability of image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a flow chart of the present invention; Figure 2 are the original images used in the experiment of the present invention: (a) is the first water level map, (b) is the second water level map, (c) is the bird, (d) is the house, (e) is the tiger, and (f) is the arch; Figure 3is the noisy image σ=35 used in the experiment of the present invention, where (a) is the first water level map, (b) is the second water level map, (c) is the bird, (d) is the house, (e) is the tiger, and (f) is the arch; Figure 4 It is a comparison diagram of the effects of using 6 models to denoise the first water scale, among which (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention; Figure 5 It is a comparison diagram of the effects of using 6 models to denoise the second water scale, among which (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention; Figure 6 This is a comparison chart of the effects of using 6 models to denoise a house, where (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention; Figure 7 This is a comparison chart of the effect of using 6 models to denoise the tiger, where (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention; Figure 8 The figure is a comparison of the effects of using 6 models to denoise the arch, where (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention; Fig. 9 This is a comparison chart of the effects of using 6 models for Bird denoising, where (a) is the NLM algorithm, (b) is the LLT model, (c) is the TV2 model, (d) is the TV1 model, (e) is the ZTV model, and (f) is the model of the present invention. DETAILED DESCRIPTION

[0010] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0011] The present invention couples the TV1 model and the LLT model and proposes an image denoising method based on an improved mixed-order total variation. The method has better denoising effect and better edge preservation effect, and suppresses the "step effect" of the TV1 model to a greater extent.

[0012] like Figure 1 As shown in FIG. 1 , a flowchart of an improved mixed-order total variation image denoising method is provided, which includes the following steps: Step 1: Convert the collected original image into grayscale and add noise. The expression is: , In formulas (1) and (2), is the original grayscale image, is the noise image function, is the Gaussian kernel function, is the random noise variance; is the two-dimensional space coordinate of the image domain.

[0013] Preferably, the grayscale conversion uses the rgb2gray function of MATLAB to convert the collected RGB image into a grayscale image; the image noise processing uses the imnoise function of MATLAB to add Gaussian noise to verify the effectiveness of the improved model, and its calling format is ,in, represents the original grayscale image, gaussian represents Gaussian noise in MATLAB, m is the mean of Gaussian noise, and s is the variance of Gaussian noise. Figure 2 The original image shown in the figure is gray-scale converted, and the noise-added image after gray-scale conversion is as follows Figure 3 shown.

[0014] Step 2: Use Gaussian-Laplacian joint transform instead of Gaussian filtering to optimize the filtering process of image preprocessing.

[0015] Preferably, the expression for filtering using Gaussian-Laplace joint transform is: , set up is the noise image function, and its expression is obtained by the commutativity of convolution and differentiation in the linear system: , in, is the Laplace operator, Represents the noisy image function With Gaussian kernel function After the product is Laplace transformed, represents the Gaussian kernel function After Laplace transform and noisy image function The product of .

[0016] Step 3, introduce the anisotropic diffusion tensor into the TV1 model and introduce the image feature extraction function based on differential curvature into the LLT model; Preferably, after introducing the anisotropic diffusion tensor, the expression of the TV1 model is: , In the formula, is the minimized energy functional of the TV1 model, is the noise image function, is the noise image function The gradient of is the weight parameter.

[0017] Represents the anisotropic diffusion tensor, which can accurately adjust the edge diffusion direction and is conducive to maintaining the integrity of edge information. Its expression is: , , In the formula, the parameters and Used to adjust the strength of the tensor, A is the direction of the image gradient, is the normal vector of the image gradient, T represents the matrix transpose, and e represents a natural constant.

[0018] Preferably, after introducing the image feature extraction function based on differential curvature, the expression of the LLT model is: , In the formula, is the minimized energy functional of the LLT model; It is an image feature extraction function based on differential curvature. It can adjust the diffusion strength in different areas of the image and increase the smoothing strength at large noise points. Its expression is: , In the formula, represents the normalized difference curvature, Represents a minimum value to avoid the denominator being 0.

[0019] Step 4, using an adaptive weight function to couple the TV1 model and the LLT model to obtain a coupled model; Preferably, the TV1 model and the LLT model are coupled using an adaptive weight function, and the expression of the coupled model is: , In the formula, is the minimized energy functional of the coupled model; the first term is the regularization term, and the second term is the fidelity term; is the noise image function The gradient modulus of is an adaptive weight parameter, which can adaptively adjust the weight of the fidelity term; is an adaptive weight function used to adjust the regularization term of the TV1 model and the LLT model; Preferably, the adaptive weight parameter The expression is: , In the formula, , to ensure that the adaptive weight parameter is always positive; is the normalized differential curvature, and its function expression is: , In the formula, yes The minimum value of yes The maximum value of represents the differential curvature, and its expression is: , In the formula, is the second-order derivative of the image along the gradient direction, is the second-order derivative of the image along the edge direction; its expressions are: , , In the formula, is the noise image function The first derivative along the x direction, is the noise image function The first derivative along the y direction, is the noise image function The second derivative along the x direction, is the noise image function The second derivative along the y direction, is the noise image function The second-order partial derivative of .

[0020] Preferably, the adaptive weight function The expression is: , In the formula, is the arccosine function, which is a decreasing function; is the noise image function The gradient modulus of is the adaptive gradient threshold; is the noise visibility function, which is used to indicate the intensity of noise. The greater the noise intensity, the greater the visibility function value. , are two thresholds with different values, which are used to solve the optimal threshold. In this embodiment, , ; The regularization term used to couple the TV1 model and the LLT model effectively removes noise while protecting image details, achieving a balance between the two.

[0021] The expression is: , In the formula, For images Local variance, for The gray value at the point; is the local mean, and its expression is: , Wherein, h and q represent the coordinate values ​​of the x-axis and y-axis respectively, both of which are variables; P and Q are constants, representing the half width of the image area block in the x-axis and y-axis directions respectively, (2P+1) is the image area block length, (2Q+1) is the image area block width, and (2P+1)(2Q+1) represents the image area size.

[0022] Step 5, post-processing the denoised image obtained by the coupling model using a fast non-local mean filtering algorithm; Preferably, the fast non-local mean filtering algorithm is an improvement of the non-local mean filtering algorithm. For each pixel in the image, a pixel similar to it is searched in a search window centered on the pixel, and the similarity is measured by calculating the gray value difference of the neighborhood around the two pixels. Post-processing by the fast non-local mean filtering algorithm can effectively restore details and texture information, further suppress residual noise, and improve the stability of image quality.

[0023] Step 6: The discrete difference method is used to obtain the discrete form of the coupling model, and the pixel value is updated by the iterative method to achieve denoising and obtain the denoised image. Specifically, first use equations (6), (9) and (11) to respectively calculate , and The value of is then obtained by formula (22) and The value of and the number of iterations N, , , , Substitute α and N into formula (28), iterate according to formula (28), take the best denoised image of the new model within the iteration number range N, and use the peak signal-to-noise ratio PSNR and structural similarity SSIM to verify the effectiveness and feasibility of the coupling model.

[0024] The expression of peak signal-to-noise ratio PSNR is: , In the formula, Indicates the image size, is the original grayscale image; generally, the higher the PSNR value, the better the image quality.

[0025] The expression of structural similarity SSIM is: , In the formula, for The average value of for The average value of for The variance of for The variance of for and covariance; c1 and c2 are parameters that make the denominator non-zero, and are usually small non-negative constants.

[0026] The SSIM value range is usually between 0 and 1. When two images are visually identical, the SSIM value is 1. The lower the similarity between the two images, the closer the SSIM value is to 0.

[0027] Preferably, the discrete form of the coupling model is expressed as: , In the formula, The expression is: ; Preferably, the difference equations are first introduced to define the variables in the model , The discrete form of : , Where h is the step size of the difference equation and the boundary conditions are: , , Among them, J and I are the upper limits of the index of the discretized grid in two different directions, which are used to define the range of the calculation area in the corresponding dimensions.

[0028] Existing , ; then the boundary conditions are: , , Let p be The descent direction of the gradient descent method, so the finite difference method is used to discretize formula (21), and its discrete form is expressed as: , in, is the time step, is the anisotropic diffusion tensor for n iterations, is the adaptive weight parameter for n iterations, is the discretized original grayscale image exist The pixel value of the position, for The descent direction of the gradient descent method.

[0029] Explain formula (10): First, in the TV1 model, the anisotropic diffusion tensor , whose properties are: A1) It can adaptively adjust the direction and intensity of diffusion according to the local features of the image, such as gradient information. At the edge of the image, the diffusion tensor will make the diffusion process proceed in the direction tangent to the edge, while suppressing the diffusion perpendicular to the edge direction, which can effectively avoid blurring the image edge during the denoising process, thereby better preserving the details and edge information of the image, so that the denoised image can maintain a clear outline, improving the visual quality of the image and the accuracy of subsequent processing.

[0030] A2) Noise in different directions can be processed differently. It can identify the directional characteristics of noise, and based on the relationship between noise and image structure, perform appropriate diffusion denoising in areas with more severe noise, and reduce diffusion in smooth areas of the image to avoid image information loss caused by over-smoothing. Compared with the traditional isotropic diffusion method, it can remove noise more accurately while retaining the original structure and texture of the image, improving the denoising effect and efficiency.

[0031] Therefore, an anisotropic diffusion tensor can be introduced into the TV1 model , the edge diffusion direction can be adjusted to maintain the integrity of edge information.

[0032] Point 2: In the LLT model, the image feature extraction function based on differential curvature , whose properties are: B1) Differential curvature can quantitatively describe the curvature of curves or surfaces in an image. In an image, different object edges, textures and other details often have unique curvature characteristics. By extracting these features based on differential curvature, local details of the image can be captured more accurately, such as the corners of objects in the image, the curvature changes of lines, etc., so that the LLT model can better distinguish different image structures when processing images.

[0033] B2) The edge of an image usually corresponds to an area with a large change in differential curvature. The feature extraction function based on differential curvature can accurately locate the edge of the image by detecting these curvature changes. In the LLT model, this feature can be used to perform special processing on the edge area during the denoising process, and the denoising intensity and method can be adjusted in a targeted manner to avoid excessive smoothing or blurring of the edge while removing noise, thereby better protecting the edge information of the image and making the denoised image contour clearer and more accurate.

[0034] Therefore, the image feature extraction function based on differential curvature is introduced into the LLT model , you can adjust the diffusion strength in different areas of the image and increase the smoothing strength at large noise points.

[0035] Point 3: Using adaptive weight function The properties of the coupled TV1 model and LLT model are: In formula (16), when the noise point is small, the gradient modulus Larger, and The product is less than When the image is at the edge or at a small noise point, , substituted into formula (10), the image is denoised by TV1 model, which can effectively retain the edge texture details of the image. At the same time, the anisotropic diffusion tensor is introduced into the TV1 model to adaptively adjust the diffusion direction of the image edge and better retain the texture details. When the noise point is large, the gradient modulus Smaller, and The product is greater than When the image is in a smooth area or at a point with large noise, , substituted into formula (10), the image is denoised by the LLT model, which can effectively smooth the noise, and introduce the image feature extraction function of the differential curvature in the LLT model. The differential curvature value is small in the smooth area of ​​the image or at large noise points, so the feature extraction function is large, thereby increasing the smoothing strength at large noise points. Good size, and The product is , In the transition area of ​​the image, , substituting into formula (10), the image is obtained by Function is used to couple the TV1 model and the LLT model to achieve adaptive denoising.

[0036] Therefore, in the edge area or small noise point, , the image is denoised by TV1 model; in smooth areas or large noise points, , the image is denoised by LLT model; in the transition area of ​​the image, , the image is obtained by The TV1 model and LLT model are coupled to achieve adaptive denoising.

[0037] Combine the following Figure 4~Figure 9 Detailed description: Compared with the NLM algorithm, LLT model, TV2 model, TV1 model, and ZTV model, the improved model provided in this specific embodiment has a beneficial effect on image denoising.

[0038] Figure 4~Figure 9 This is a comparison chart of the denoising effects of using the NLM algorithm, LLT model, TV2 model, TV1 model, and ZTV model. Figure 4 Middle (a), Figure 5 Middle (a), Figure 6 Middle (a), Figure 7 Middle (a), Figure 8 Middle (a), Fig. 9 (a) is denoising using the NLM algorithm; the NLM algorithm uses the non-local similarity information in the image to denoise the image. It searches for pixels with similar neighborhood structures to the current pixel in the entire image and estimates the true value of the current pixel by weighted averaging these similar pixels. It can effectively remove various types of noise, especially Gaussian noise. In the denoised image, the details and texture information of the image can be well preserved, making the image look more natural and clear.

[0039] Figure 4 Middle (b), Figure 5 Middle (b), Figure 6 Middle (b), Figure 7 Middle (b), Figure 8 Middle (b), Fig. 9 (b) uses the LLT model for denoising. The LLT model expression is: , In the formula, The expression is .

[0040] Figure 4 Middle (c), Figure 5 Middle (c), Figure 6 Middle (c), Figure 7 Middle (c), Figure 8 Middle (c), Fig. 9 In (c), the TV model based on L2 norm (TV2 model for short) is used for denoising. The TV2 model expression is: , In the formula, is the gradient operator; is the weight coefficient, and the model is an isotropic diffusion model.

[0041] Figure 4 Middle (d), Figure 5 Middle (d), Figure 6 Middle (d), Figure 7 Middle (d), Figure 8 Middle (d), Fig. 9 (d) is the TV model based on L1 norm for denoising (referred to as TV1 model), and the TV1 model expression is: , In the formula, is the gradient operator; is the weight coefficient, the model is an anisotropic diffusion model; u is the noise image function, the expression is ; v is the original grayscale image, expressed as .

[0042] Figure 4 Middle (e), Figure 5 Middle (e), Figure 6 Middle (e), Figure 7 Middle (e), Figure 8 Middle (e), Fig. 9 In (e), the ZTV model is used for denoising. The ZTV model expression is: , The model has the characteristics of TV1 model and TV2 model. is the regular term index, and its expression is: .

[0043] Figure 4 Middle (f), Figure 5 Middle (f), Figure 6 Middle (f), Figure 7 Middle (f), Figure 8 Middle (f), Fig. 9 (f) is denoised using the coupling model constructed by the present invention.

[0044] Depend on Figure 4~Figure 9It can be seen that the NLM algorithm is a denoising algorithm based on the similarity of pixel blocks in the image. Since the algorithm has multiple parameters that need to be set, such as search window size, block size, similarity measurement method, etc. The selection of these parameters has a great influence on the denoising effect. And when the noise intensity in the image is large, it may seriously interfere with the matching process of similar blocks, making the similar blocks found actually not similar, so the NLM model has the worst denoising effect; LLT is a high-order variational denoising model, which can suppress the "step effect" of the TV1 model, but due to the large smoothing force, the texture of the image is blurred, so the denoising effect of the LLT model is poor; the TV2 model is an isotropic diffusion model, which will blur the edge texture information of the image during the denoising process, so the denoising effect of the TV2 model is also poor; although the TV1 model will retain the edge texture information of the image during the denoising process, it will produce false edges in the smooth area of ​​the image, so the denoising effect of the TV1 model is also poor; the ZTV model is an improvement on the regularization term index of the adaptive variational model, and the denoising effect is better; the NEW model is through the function The TV1 model and the LLT model are coupled to achieve adaptive denoising, and the IMOTV model has the best denoising effect.

[0045] In order to further compare the denoising effects of the six models in detail, Tables 1 to 3 show the PSNR, SSIM and Time values ​​of the six images after denoising using the NLM algorithm, LLT model, TV2 model, TV1 model, ZTV model and the coupled model constructed by the present invention when the noise standard deviation σ=15, 20, 25, 30, 50 is added.

[0046] From the PSNR and SSIM values ​​in Tables 1 to 3, it can be seen that the PSNR and SSIM values ​​of the model proposed in this example are significantly higher than those of the NLM model, LLT model, TV2 model, TV1 model, and ZTV model. From the calculation time Time, it can be seen that the calculation time of the NEW model is only about 2.3 times that of the TV1 model with the shortest calculation time, which is shorter than the calculation time of NLM. Therefore, while ensuring the basic denoising efficiency, combined with the two indicators of PSNR and SSIM, the NEW model should require less calculation time.

[0047] Table 1 PSNR (dB) values ​​of images denoised by the coupling model constructed by the present invention and other comparative models

[0048] Table 2 SSIM values ​​of denoised images of the coupling model constructed by the present invention and other comparative models

[0049] Table 3 Time (s) values ​​of images after denoising by the coupling model constructed by the present invention and other comparative models

[0050] From the above results, it can be seen intuitively and effectively that the improved mixed-order total variation image denoising model of the present invention has better visibility effect, effectively suppresses the "step effect" appearing in the smooth area, protects the edge texture information of the image, and at the same time improves the denoising efficiency, which is more conducive to practical application.

[0051] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for image denoising based on improved mixed-order total variation, characterized in that: The steps include: S1, converting the collected original image into grayscale and performing noise processing; S2, filtering using Gaussian-Laplace joint transform; S3, anisotropic diffusion tensor is introduced into TV1 model, and image feature extraction function based on differential curvature is introduced into LLT model; S4, using an adaptive weight function to couple the TV1 model and the LLT model to obtain a coupled model; S5, post-processing the denoised image obtained by the coupling model using a fast non-local mean filtering algorithm; S6, the discrete difference method is used to obtain the discrete form of the coupling model, and the pixel value is updated by the iterative method for denoising to obtain the denoised image. The peak signal-to-noise ratio and structural similarity indicators are used to verify the effectiveness and feasibility of the coupling model.

2. The method for implementing image denoising based on improved mixed-order total variation according to claim 1, characterized in that: After introducing the anisotropic diffusion tensor, the expression of the TV1 model is: In the formula, is the minimized energy functional of the TV1 model, is the original grayscale image, is the noise image function, is the gradient of the noisy image function, is the weight parameter; represents the anisotropic diffusion tensor, which is expressed as: ， ， In the formula, the parameters and Used to adjust the strength of the tensor, A is the direction of the image gradient, is the normal vector of the image gradient, T represents the matrix transpose, and e represents the natural constant; After introducing the image feature extraction function based on differential curvature, the expression of the LLT model is: ， In the formula, is the minimized energy functional of the LLT model; is an image feature extraction function based on differential curvature, and its expression is: ， In the formula, represents the normalized difference curvature, Indicates a minimum value.

3. The method for implementing image denoising based on improved mixed-order total variation according to claim 2, characterized in that: The expression of coupling TV1 model and LLT model using adaptive weight function is: ， In the formula, represents the minimized energy functional of the coupled model; The first term is the regularization term, and the second term is the fidelity term; is the noise image function The gradient modulus of is the adaptive weight parameter, which adjusts the weight of the fidelity term; is an adaptive weight function used to adjust the regularization term of the TV1 model and the LLT model; The adaptive weight parameter The expression is: ， In the formula, ; The expression is: ， In the formula, yes The minimum value of yes The maximum value of is the differential curvature, and its expression is: ， In the formula, is the second-order derivative of the image along the gradient direction, is the second-order derivative of the image along the edge direction; The expressions are: ， ， In the formula, is the noise image function along The first derivative of the direction, is the noise image function along The first derivative of the direction, is the noise image function along The second derivative of the direction, is the noise image function along The second derivative of the direction, is the noise image function The second-order partial derivative of ; The adaptive weight function The expression is: ， In the formula, is the arccosine function, which is a decreasing function; is the noise image function The gradient modulus of is the adaptive gradient threshold; , They are thresholds of two different values; is the noise visibility function; The expression is: ， In the formula, For images Local variance, for The gray value at the point; is the local mean, and its expression is: ， Wherein, h and q represent the coordinate values ​​of the x-axis and y-axis respectively, both of which are variables; P and Q are constants, representing the half width of the image area block in the x-axis and y-axis directions respectively; (2P+1) is the length of the image area block, (2Q+1) is the width of the image area block, and (2P+1)(2Q+1) represents the area size of the image area block.

4. The method for implementing image denoising based on improved mixed-order total variation according to claim 1, characterized in that: In step S5, in the fast non-local means filtering algorithm, for each pixel in the image, a pixel similar to it is searched in a search window centered on the pixel, and the similarity is measured by calculating the difference in grayscale values ​​of the neighborhoods around the two pixels.

Images