New partial differential image noise reduction method and system based on anisotropic diffusion

Abstract

The invention discloses a new partial differential image noise reduction method and system based on anisotropic diffusion, and the method comprises the steps: constructing a new texture detection operator, recording the new texture detection operator as a T operator, replacing a gradient model in a PM model with the T operator, and obtaining a first model; based on the Gaussian curvature and the noise intensity, obtaining an anti-noise coefficient, and combining the anti-noise coefficient with the first model to obtain a second model; constructing a trend fidelity term based on Gaussian curvature, combining with the second model to obtain an image noise reduction model, and processing the model by adopting a finite difference algorithm to obtain a final image noise reduction model; and inputting the to-be-denoised image into the final image denoising model to obtain a denoised image. According to the method, the change condition of the diagonal is increased, more texture features can be expressed, the problem that a PM model has large noise points is solved, and texture detail information and structural integrity are effectively protected.

Description

Technical Field

[0001] The present invention relates to the technical field of image restoration, and in particular to a new partial differential image denoising method and system based on anisotropic diffusion. Background Art

[0002] Image restoration is a crucial processing step because images are subject to varying degrees of noise contamination during acquisition, transmission, and display. Image restoration techniques can be categorized into two main categories. The first category is the recently emerging deep learning-based denoising technology, which utilizes neural network architectures to reduce image noise. While these methods offer good noise reduction performance, they are demanding on equipment and take a long time to complete. The second category is traditional image denoising methods. These methods have many subcategories, including image transform domain denoising and spatial domain denoising. Among transform domain denoising methods, popular ones include wavelet transforms. These methods take into account the high-frequency nature of image noise and transform the image from the spatial domain to the transform domain. A threshold is set in the transform domain to distinguish noise information and achieve noise reduction. However, if the threshold is poorly set, these methods will fail to distinguish noise, resulting in a loss of texture detail. Summary of the Invention

[0003] The purpose of the present invention is to provide a new partial differential image denoising method and system based on anisotropic diffusion, which effectively protects small texture detail information and structural integrity, and significantly improves the peak signal-to-noise ratio and structural similarity indicators.

[0004] The present invention adopts the following technical solution: a new partial differential image denoising method based on anisotropic diffusion, comprising the following steps:

[0005] S1. Construct a new texture detection operator, denoted as T operator, and use T operator to replace the gradient modulus in the PM model to obtain the first model.

[0006] S2. Based on the Gaussian curvature and the noise intensity, an anti-noise coefficient is obtained, and the anti-noise coefficient is combined with the first model to obtain a second model.

[0007] S3. Construct a trend fidelity term based on Gaussian curvature, combine it with the second model to obtain an image denoising model, and use a finite difference algorithm to process the model to obtain the final image denoising model.

[0008] S4. Input the image to be denoised into the final image denoising model to obtain a denoised image.

[0009] Furthermore, in step S1, obtaining the first model includes the following contents:

[0010] The expression of the T operator is:

[0011]

[0012] wherein T i,j denotes the new texture detection operator of the denoised image with pixel coordinate (i,j), u i,j-1 denotes the denoised image with pixel coordinate (i,j-1), u i,j denotes the denoised image with pixel coordinate (i,j), u i,j+1 denotes the denoised image with pixel coordinate (i,j+1), u i-1,j denotes the denoised image with pixel coordinate (i-1,j), u i+1,j denotes the denoised image with pixel coordinate (i+1,j), u i-1,j-1 denotes the denoised image with pixel coordinate (i-1,j-1), u i+1,j+1 denotes the denoised image with pixel coordinate (i+1,j+1), u i-1,j+1 denotes the denoised image with pixel coordinate (i-1,j+1), u i+1,j-1 denotes the denoised image with pixel coordinate (i+1,j-1).

[0013] An adaptive T-operator threshold k T is constructed based on the average value of the T-operator, and the specific expression is as follows:

[0014] k T = 0.5*(mean(T));

[0015] wherein mean() denotes the average value operation, and T denotes the new texture detection operator.

[0016] The expression of the first model is as follows:

[0017]

[0018] wherein u denotes the denoised image, t denotes the t-th moment, div denotes the divergence operator, g() denotes the diffusion function, denotes the gradient operator.

[0019] Further, the second model obtained in step S2 includes the following contents:

[0020] The expression of the Gaussian curvature is as follows:

[0021]

[0022] wherein G denotes the Gaussian curvature, u ηη denotes the second-order derivative of the denoised image along the gradient direction, u xx denotes the second-order derivative of the denoised image along the horizontal axis direction, u xydenotes the mixed partial derivative of the denoised image along the horizontal and vertical axes, u x denotes the first-order derivative of the denoised image along the horizontal axis, u y denotes the first-order derivative of the denoised image along the vertical axis, u yy denotes the second-order derivative of the denoised image along the vertical axis, denotes the second-order derivative of the denoised image along the edge direction,

[0023] A judgment condition based on the Gaussian curvature and the noise intensity is constructed, and the normalized Gaussian curvature is segmented using the judgment condition to obtain an anti-noise coefficient, and the specific expression is:

[0024] GY = (σ n > 25) & ((1-norm(G))≤0.1);

[0025]

[0026] wherein, GY denotes the judgment condition, σ n denotes the noise intensity, norm() denotes the normalization processing, and gs denotes the anti-noise coefficient.

[0027] The expression of the second model is:

[0028]

[0029] wherein,

[0030] Further, in step S3, the final image denoising model includes the following contents:

[0031] The expression of the trend fidelity term coefficient based on the Gaussian curvature is:

[0032]

[0033] The expression of the image denoising model is:

[0034]

[0035] wherein, gb denotes the trend fidelity term coefficient based on the Gaussian curvature,. * denotes the dot product operation between matrices, u0 denotes the noise original image, and Δ denotes the Laplace operator.

[0036] The finite difference algorithm is used to discretize the image denoising model to obtain a discrete result, and the specific expression is:

[0037]

[0038] wherein, x denotes the horizontal axis direction, y denotes the vertical axis direction, Δx and Δy both denote the space step.

[0039] Based on the discrete results, the final image denoising model is obtained, and the specific expression is:

[0040]

[0041] wherein, represents the (n+1)th denoising image of the pixel coordinate (i,j) in the iteration process, represents the nth denoising image of the pixel coordinate (i,j) in the iteration process, dt represents a time step, (a,b) represents the pixel coordinates of the cross-shaped four-neighbor domain located at the pixel coordinate (i,j), and Γ (i,j) represents the upper and lower and left and right pixel range of the pixel coordinate (i,j), represents the diffusion function of the nth denoising image of the pixel coordinate (a,b) in the iteration process, represents the diffusion function of the nth denoising image of the pixel coordinate (i,j) in the iteration process, represents the nth denoising image of the pixel coordinate (a,b) in the iteration process.

[0042] Further, in step S4, obtaining the denoised image comprises the following contents:

[0043] The image to be denoised is input into the final image denoising model for iteration operation, and the iteration operation is stopped when the iteration time stopping condition is met, and the denoised image is output.

[0044] The expression of the iteration time stopping condition is:

[0045]

[0046] wherein, (u-u0) represents a residual image, and σ represents the standard deviation of the image, represents the noise intensity of the image.

[0047] Further, the present application also proposes a new partial differential image denoising method based on anisotropic diffusion system, comprising:

[0048] The first model acquisition module is used to construct a new texture detection operator, denoted as T operator, and the T operator is used to replace the gradient module in the PM model to obtain the first model.

[0049] The second model acquisition module is used to obtain an anti-noise coefficient based on the Gaussian curvature and the noise intensity, and the second model is obtained by combining the coefficient and the first model.

[0050] The denoising model acquisition module is configured to construct a trend fidelity term based on Gaussian curvature, combine the second model, and obtain an image denoising model.

[0051] The denoised image acquisition module is configured to input the image to be denoised into the final image denoising model to obtain a denoised image.

[0052] Further, the present application also provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the steps of the new partial differential image denoising method based on anisotropic diffusion when executing the computer program.

[0053] Further, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the new partial differential image denoising method based on anisotropic diffusion.

[0054] Compared with the prior art, the present application has the following technical effects:

[0055] The present application identifies large noise points through Gaussian curvature and noise intensity, and increases the smoothing strength at the large noise points. In addition, the gradient value in the diagonal direction of the image better represents small texture information, solves the problem of large noise points in the PM (Perona-Malik) model, effectively protects small texture detail information and structural integrity, and significantly improves the peak signal-to-noise ratio and structural similarity index. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is the overall implementation flowchart of the present application.

[0057] Figure 2 is the result image after the image containing a photographer is denoised by using different models in the embodiment of the present application.

[0058] Figure 3 is the result image after the image containing a pepper is denoised by using different models in the embodiment of the present application.

[0059] Figure 4 is the result image after the image containing a starfish is denoised by using different models in the embodiment of the present application. DETAILED DESCRIPTION

[0060] The present application will be further described below in combination with the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0061] To achieve the above objectives, the present invention proposes a new partial differential image denoising method based on anisotropic diffusion, such as Figure 1 The specific steps are as follows:

[0062] S1. Construct a new texture detection operator, denoted as T operator, and use T operator to replace the gradient norm in PM (Perona-Malik) model to obtain the first model. The specific content is:

[0063] The expression of PM model is:

[0064]

[0065] Among them, u represents the denoised image, t represents the tth moment, div represents the divergence operator, g() represents the diffusion function, represents the gradient operator.

[0066] when hour, Increases the smoothing noise strength in flat areas; when hour, Reduce the diffusion strength at the edge area to protect the edge texture.

[0067] Based on the PM model, considering that the gradient modulus operator is insufficient in expressing image texture details, the T operator is constructed. The T operator makes up for the deficiency of the gradient modulus operator that only represents the changes in the x and y directions, adds the changes in the diagonal directions, and can express more texture features.

[0068] The expression of the T operator is:

[0069]

[0070] Among them, T i,j Represents the new texture detection operator of the denoised image with pixel coordinates (i, j), u i,j-1 represents the denoised image with pixel coordinates (i, j-1), u i,j represents the denoised image with pixel coordinates (i, j), u i,j+1 represents the denoised image with pixel coordinates (i, j+1), u i-1,j Represents the denoised image with pixel coordinates (i-1, j), u i+1,j represents the denoised image with pixel coordinates (i+1, j), u i-1,j-1 represents the denoised image with pixel coordinates (i-1, j-1), u i+1,j+1 represents the denoised image with pixel coordinates (i+1,j+1), u i-1,j+1 represents the denoised image with pixel coordinates (i-1, j+1), u i+1,j-1The denoised image of which the pixel coordinate is (i+1, j-1).

[0071] The adaptive T-operator threshold k is constructed based on the average value of the T-operator T The specific expression is:

[0072] k T = 0.5*(mean(T));

[0073] Wherein, mean() represents the average value operation, and T represents the new texture detection operator.

[0074] The expression of the first model is:

[0075]

[0076] S2, based on the Gaussian curvature and the noise intensity, an anti-noise coefficient is obtained, and the second model is obtained by combining the coefficient and the first model. The specific content is:

[0077] The expression of the Gaussian curvature is:

[0078]

[0079] Wherein, G represents the Gaussian curvature, u ηη represents the second derivative of the denoised image along the gradient direction, u xx represents the second derivative of the denoised image along the horizontal axis direction, u xy represents the mixed partial derivative of the denoised image along the horizontal axis and the vertical axis direction, u x represents the first derivative of the denoised image along the horizontal axis direction, u y represents the first derivative of the denoised image along the vertical axis direction, u yy represents the second derivative of the denoised image along the vertical axis direction, represents the second derivative of the denoised image along the edge direction,

[0080] At the noise point, is larger, u ηη is larger, the value of the Gaussian curvature is larger; in the edge region of the image, is smaller, u ηη is larger, the value of the Gaussian curvature is second; in the flat region of the image, is smaller, u ηη is smaller, the value of the Gaussian curvature is smaller. Therefore, the Gaussian curvature can distinguish the noise point.

[0081] To solve the problem that the T operator in the first model is sensitive to large noise points, resulting in a small diffusion value of the diffusion function at the large noise points, the Gaussian curvature can be used to distinguish the properties of the noise points, a judgment condition based on the Gaussian curvature and noise intensity is constructed, and the normalized Gaussian curvature is segmented using the judgment condition to obtain an anti-noise coefficient, and the specific expression is:

[0082] GY=(σ n >25)&((1-norm(G))≤0.1);

[0083]

[0084] wherein GY represents the judgment condition, σ n represents the noise intensity, norm() represents the normalization processing, and gs represents the anti-noise coefficient.

[0085] 1-(norm(G)) tends to 0 at the noise points, and 1-(norm(G)) has a large value at the edges and flat areas. Considering that the first model cannot smooth the large noise points, a judgment condition based on the noise intensity is proposed, and in the image area meeting the condition, the anti-noise coefficient is an increasing function, and in the remaining area, the anti-noise coefficient is equal to 1. Multiplying the anti-noise coefficient with the T operator can effectively reduce the influence of large noise points on the model.

[0086] The expression of the second model is:

[0087]

[0088] wherein,

[0089] The smoothing degree of the second model at the large noise points is increased, and the diffusion degree in the texture detail area is unchanged, so that the large noise points can be smoothed.

[0090] S3, a trend fidelity term based on the Gaussian curvature is constructed, combined with the second model to obtain an image denoising model, and a finite difference algorithm is used to process the model to obtain a final image denoising model. The specific content is:

[0091] Considering that the value of the Gaussian curvature at the flat area of the image is small and tends to 0, a trend fidelity term based on the Gaussian curvature is set. The expression of the trend fidelity term coefficient based on the Gaussian curvature is:

[0092]

[0093] At the flat area, gb→1, the second model has a trend fidelity term, which can suppress the staircase effect; at the texture detail area, gb→0, the second model has no trend fidelity term, and the diffusion term is used to smooth the noise and protect the image texture and edges.

[0094] The expression of the image denoising model is:

[0095]

[0096] Wherein, gb represents the trend fidelity term coefficient based on Gaussian curvature,. represents the dot product operation between matrices, u0 represents the noise original image, and Δ represents the Laplace operator.

[0097] The finite difference algorithm is used to discretize the image denoising model to obtain a discrete result, and the specific expression is:

[0098]

[0099] Wherein, x represents the horizontal axis direction, y represents the vertical axis direction, Δx and Δy both represent the space step, and Δx = Δy = 1.

[0100] Based on the discrete result, the final image denoising model is obtained, and the specific expression is:

[0101]

[0102] Wherein, represents the (n+1)th denoising image of the pixel coordinate (i,j) in the iteration process, represents the nth denoising image of the pixel coordinate (i,j) in the iteration process, dt represents the time step, and (a,b) represents the pixel coordinate of the cross-shaped four-neighbor domain located at the pixel coordinate (i,j), Γ (i,j) represents the upper and lower and left and right pixel range of the pixel coordinate (i,j), represents the diffusion function of the nth denoising image of the pixel coordinate (a,b) in the iteration process, represents the diffusion function of the nth denoising image of the pixel coordinate (i,j) in the iteration process, represents the nth denoising image of the pixel coordinate (a,b) in the iteration process.

[0103] S4, input the image to be denoised into the final image denoising model to obtain the denoised image. The specific content is:

[0104] Since the final image denoising model is a partial differential model, an iteration stopping condition is needed to control the iteration stopping time. Considering that the standard deviation of the residual image and the noise intensity contained in it will become larger and larger as the iteration proceeds, and through experiments it is found that the standard deviation of the residual image is first less than the noise intensity and then greater than the noise intensity as the iteration proceeds. This shows that the image texture contained in the residual image will also become more and more as the iteration proceeds. In order to balance the strength of smoothing image noise and protecting image texture, the difference between the standard deviation of the residual image and the noise intensity is taken as the standard, and when this difference is greater than 0, the iteration is stopped.

[0105] The image to be denoised is input into the final image denoising model for iteration operation, and the iteration is stopped when the iteration time stopping condition is met, and the denoised image is output.

[0106] The expression of the iteration time stopping condition is:

[0107]

[0108] where (u-u0) represents the residual image, (u0) represents the image to be denoised, and (u) represents the denoised image. σ represents the standard deviation of the image, represents the noise intensity of the image.

[0109] In order to further illustrate the denoising ability of the final image denoising model, the model (referred to as the new model) is compared with the first model (referred to as the first model), the second model (referred to as the second model), the PM model, the regularized C model (referred to as the C model), the TV model (Total Variation Model), the YK model (YK fourth-order Model), the R model proposed by Riya scholars (referred to as the R model), and the Te model proposed by Tebin scholars (referred to as the Te model) for denoising. The PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) indicators are used to objectively verify the denoising ability of the model.

[0110] The parameters of the PM model, the C model, and the YK model are set to be the same, let dt=0.02, and the adaptive gradient threshold expression is used for the gradient threshold; the parameters of the TV model are set, let dt=0.2; the parameters of the final image denoising model, the first model, the second model, the R model, and the Te model are set to be the same, let dt=0.02.

[0111] Three images containing photographer, pepper and starfish were selected for verification, Gaussian white noise with standard deviation σ = 10, 20, 30 was added respectively, and the above nine models were used for noise reduction processing, and table 1 is the PSNR and SSIM values of the images after noise reduction obtained by different models.

[0112]

[0113] As can be seen from table 1, the PSNR and SSIM values of the new model are the highest. Therefore, in terms of objective noise reduction index value, the new model has the best edge preservation effect in noise reduction.

[0114] Gaussian white noise with standard deviation σ = 20 was added, and the above nine models were verified again by selecting images containing photographer, pepper and starfish respectively. The image containing the photographer is relatively smooth, the image containing the pepper has prominent edges, and the image containing the starfish has rich texture.

[0115] Figure 2 (a) of figure 1 is an original image containing a photographer, Figure 2 (b) of figure 1 is a noise image containing a photographer, Figure 2 (c) of figure 1 is a noise reduction image containing a photographer processed by the PM model, Figure 2 (d) of figure 1 is a noise reduction image containing a photographer processed by the C model, Figure 2 (e) of figure 1 is a noise reduction image containing a photographer processed by the YK model, Figure 2 (f) of figure 1 is a noise reduction image containing a photographer processed by the TV model, Figure 2 (g) of figure 1 is a noise reduction image containing a photographer processed by the R model, Figure 2 (h) of figure 1 is a noise reduction image containing a photographer processed by the Te model, Figure 2 (i) of figure 1 is a noise reduction image containing a photographer processed by the one model, Figure 2 (j) of figure 1 is a noise reduction image containing a photographer processed by the two model, Figure 2 (k) of figure 1 is a noise reduction image containing a photographer processed by the new model, and Figure 2 As can be seen from figure 1, the noise reduction image processed by the new model has fewer noise points and clearer texture, and has better noise reduction effect.

[0116] Figure 3 (a) of figure 2 is an original image containing a pepper, Figure 3 (b) of figure 2 is a noise image containing a pepper, Figure 3 (c) of figure 2 is a noise reduction image containing a pepper processed by the PM model, Figure 3 (d) of figure 2 is a noise reduction image containing a pepper processed by the C model, Figure 3 (e) of figure 2 is a noise reduction image containing a pepper processed by the YK model, Figure 3(f) is a noise-reduced image containing a pepper image processed by a TV model, Figure 3 (g) is a noise-reduced image containing a pepper image processed by an R model, Figure 3 (h) is a noise-reduced image containing a pepper image processed by a Te model, Figure 3 (i) is a noise-reduced image containing a pepper image processed by a one model, Figure 3 (j) is a noise-reduced image containing a pepper image processed by a two model, Figure 3 (k) is a noise-reduced image containing a pepper image processed by a new model, from Figure 4 It can be seen from the above that the noise-reduced image processed by the new model has no obvious noise points and rich texture, and has good noise reduction effect.

[0117] Figure 4 (a) is an original image containing a starfish image, Figure 4 (b) is a noise image containing a starfish image, Figure 4 (c) is a noise-reduced image containing a starfish image processed by a PM model, Figure 4 (d) is a noise-reduced image containing a starfish image processed by a C model, Figure 4 (e) is a noise-reduced image containing a starfish image processed by a YK model, Figure 4 (f) is a noise-reduced image containing a starfish image processed by a TV model, Figure 4 (g) is a noise-reduced image containing a starfish image processed by an R model, Figure 4 (h) is a noise-reduced image containing a starfish image processed by a Te model, Figure 4 (i) is a noise-reduced image containing a starfish image processed by a one model, Figure 4 (j) is a noise-reduced image containing a starfish image processed by a two model, Figure 4 (k) is a noise-reduced image containing a starfish image processed by a new model, from ​ It can be seen from the above that the noise-reduced image processed by the new model has no obvious noise points and rich texture, and has good noise reduction effect.

[0118] The embodiment of the present application also proposes a new partial differential image denoising system based on anisotropic diffusion, which comprises a first model acquisition module, a second model acquisition module, a denoising model acquisition module, a denoised image acquisition module and a computer program executable on a processor. It should be noted that each module in the above system corresponds to the specific steps of the method provided by the embodiment of the present application, has the corresponding function modules and beneficial effects of the execution method. Technical details not described in detail in the present embodiment can be referred to the method provided by the embodiment of the present application.

[0119] The embodiment of the present application further provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor. It should be noted that the processor executes the computer program to correspond to the specific steps of the method provided by the embodiment of the present application, has the function modules and beneficial effects corresponding to the execution method. The technical details not described in detail in the embodiment can be referred to the method provided by the embodiment of the present application.

[0120] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. It should be noted that the computer program is run by the processor to correspond to the specific steps of the method provided by the embodiment of the present application, has the function modules and beneficial effects corresponding to the execution method. The technical details not described in detail in the embodiment can be referred to the method provided by the embodiment of the present application.

[0121] The above is only the preferred embodiment of the present application, it should be pointed out that, for the ordinary skilled in the art, without departing from the technical principles of the present application, can make a number of improvements and deformation, these improvements and deformation also should be considered as the protection scope of the present application.

Claims

1. A new partial differential image denoising method based on anisotropic diffusion, characterized in that: include: S1. Construct a new texture detection operator and use it to replace the gradient modulus in the PM model to obtain the first model; S2. Obtain an anti-noise coefficient based on the Gaussian curvature and the noise intensity, and combine the coefficient with the first model to obtain a second model; S3. Construct a trend fidelity term based on Gaussian curvature, combine it with the second model, obtain an image denoising model, and process the model using a finite difference algorithm to obtain a final image denoising model; S4. Input the image to be denoised into the final image denoising model to obtain a denoised image.

2. The new partial differential image denoising method based on anisotropic diffusion according to claim 1, characterized in that: In step S1, obtaining the first model includes the following contents: The expression of the new texture detection operator is: Among them, T i,j Represents the new texture detection operator of the denoised image with pixel coordinates (i, j), u i,j-1 represents the denoised image with pixel coordinates (i, j-1), u i,j represents the denoised image with pixel coordinates (i, j), u i,j+1 represents the denoised image with pixel coordinates (i, j+1), u i-1,j Represents the denoised image with pixel coordinates (i-1, j), u i+1,j represents the denoised image with pixel coordinates (i+1, j), u i-1,j-1 represents the denoised image with pixel coordinates (i-1, j-1), u i+1,j+1 represents the denoised image with pixel coordinates (i+1,j+1), u i-1,j+1 represents the denoised image with pixel coordinates (i-1, j+1), u i+1,j-1 represents the denoised image with pixel coordinates (i+1, j-1); Based on the average value of the new texture detection operator, the adaptive new texture detection operator threshold k is constructed. T , the specific expression is: k T =0.5*(mean(T)); Among them, mean() represents the average operation, and T represents the new texture detection operator; The expression of the first model is: Among them, u represents the denoised image, t represents the tth moment, div represents the divergence operator, g() represents the diffusion function, represents the gradient operator.

3. The new partial differential image denoising method based on anisotropic diffusion according to claim 1, characterized in that: In step S2, the second model is obtained, including the following contents: The expression for Gaussian curvature is: Among them, G represents Gaussian curvature, u ηη represents the second-order derivative of the denoised image along the gradient direction, u xx Represents the second-order derivative of the denoised image along the horizontal axis, u xy Represents the mixed partial derivative of the denoised image along the horizontal and vertical axes, u x Represents the first-order derivative of the denoised image along the horizontal axis, u y Represents the first-order derivative of the denoised image along the vertical axis, u yy represents the second-order derivative of the denoised image along the vertical axis, represents the second-order derivative of the denoised image along the edge direction, A judgment condition based on Gaussian curvature and noise intensity is constructed, and the normalized Gaussian curvature is segmented using the judgment condition to obtain the noise reduction coefficient. The specific expression is: GY=(σ n >25)&((1-norm(G))≤0.1); Among them, GY represents the judgment condition, σ n Indicates noise intensity, norm() indicates normalization processing, and gs indicates anti-noise coefficient; The expression of the second model is: Among them, u represents the denoised image, t represents the tth moment, div represents the divergence operator, g() represents the diffusion function, T represents the new texture detection operator, represents the gradient operator, k T Represents the adaptive new texture detection operator threshold.

4. The new partial differential image denoising method based on anisotropic diffusion according to claim 1, characterized in that: In step S3, the final image denoising model is obtained, which includes the following contents: The expression of the trend fidelity term coefficient based on Gaussian curvature is: Among them, gb represents the trend fidelity coefficient based on Gaussian curvature, and G represents Gaussian curvature; The expression of the image denoising model is: Among them, u represents the denoised image, t represents the tth moment, div represents the divergence operator, g() represents the diffusion function, T represents the new texture detection operator, gs represents the anti-noise coefficient, represents the gradient operator, .* represents the dot product operation between matrices, u0 represents the noise preimage, and Δ represents the Laplace operator; The finite difference algorithm is used to discretize the image denoising model and obtain the discrete result. The specific expression is: Among them, x represents the horizontal axis direction, y represents the vertical axis direction, Δx and Δy both represent the spatial step length; Based on the discrete results, the final image denoising model is obtained, and the specific expression is: in, represents the n+1th denoised image with pixel coordinates (i, j) during the iteration process, represents the nth denoised image with pixel coordinates (i, j) during the iteration process, dt represents the time step, (a, b) represents the pixel coordinates of the four cross-shaped neighborhood located at the pixel coordinate (i, j), Γ (i,j) Indicates the upper, lower, left, and right pixel ranges of pixel coordinates (i, j). represents the diffusion function of the nth denoised image with pixel coordinates (a, b) during the iteration process, represents the diffusion function of the nth denoised image with pixel coordinates (i, j) during the iteration process, Represents the nth denoised image with pixel coordinates (a, b) during the iteration process.

5. The new partial differential image denoising method based on anisotropic diffusion according to claim 1, characterized in that: In step S4, the image after noise reduction is obtained, including the following contents: The image to be denoised is input into the final image denoising model for iterative operation until the iteration time stop condition is met, and the denoised image is output; The expression of the iteration time stopping condition is: Among them, (u-u0) represents the residual graph, () σ Represents the standard deviation of the image, ()σ n Indicates the noise intensity of the image.

6. A system for the new partial differential image denoising method based on anisotropic diffusion according to claim 1, characterized in that: include: A first model acquisition module is used to construct a new texture detection operator, and use the operator to replace the gradient modulus in the PM model to obtain a first model; A second model acquisition module is used to obtain an anti-noise coefficient based on Gaussian curvature and noise intensity, and combine the coefficient with the first model to obtain a second model; A denoising model acquisition module is used to construct a trend fidelity term based on Gaussian curvature, combine it with the second model, obtain an image denoising model, and process the model using a finite difference algorithm to obtain the final image denoising model; The denoised image acquisition module is used to input the image to be denoised into the final image denoising model to obtain the denoised image.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the new partial differential image denoising method based on anisotropic diffusion according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the new partial differential image denoising method based on anisotropic diffusion according to any one of claims 1 to 5 is executed.

Images