Self-adaptive stripe noise removal method based on spatial characteristics

Abstract

The invention discloses a self-adaptive stripe noise removal method based on spatial characteristics. A remote sensing imaging detection system is used for acquiring a noise image as an observation image; constructing an adaptive weighting function according to spatial features of the observation image, adaptively controlling the constraint intensity of a penalty term in a variational denoising model at different pixel points, applying a larger constraint at a stripe noise position to remove noise, and applying a smaller constraint at an image texture position to maintain the characteristics of the observation image; meanwhile, the local smoothness in the strip direction is constrained by adopting an L1 norm, and the image gray scale is constrained by adopting an L2 norm; and finally, quickly solving the variational denoising model by using a split Bregman iteration method and an alternating minimization method to obtain a better denoised image. According to the self-adaptive denoising method, the edge detail information of the image is effectively kept while the noise of a plurality of strips with different intensities is removed.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital image processing and relates to an adaptive stripe noise removal method based on spatial characteristics. The stripe noise mainly exists in a remote sensing imaging system in which a focal plane is spliced ​​by multiple CCD chips. Background Art

[0002] Remote sensing technology has the advantages of large detection range, less restrictions on ground conditions, and large amount of information. It has become an important means of obtaining ground information and is widely used in military reconnaissance, cruise guidance, resource exploration, land surveying and mapping, environmental monitoring and other fields. With the development of remote sensing technology and computer technology, large field of view and wide coverage have become important indicators for high-performance remote sensing cameras. However, due to the limitations of technology and process level, the number of imaging pixels of a single sensor charge-coupled device (CCD) chip cannot meet the system requirements. Multiple CCDs are often mechanically spliced ​​to form a large focal plane. However, there are problems such as non-uniform radiation response and inconsistent readout circuits in different rows or columns between multiple CCDs, which leads to stripe noise with alternating light and dark distribution in the acquired image. This noise is different from other noises in remote sensing images and has very significant structured distribution characteristics. It is highly destructive to the original characteristic structure of remote sensing images and directly affects the judgment and subsequent processing of remote sensing images. Therefore, in-depth research on image stripe noise removal methods has important theoretical and practical application value.

[0003] Assume that the image f∈L 2 (Ω), the stripe noise is additive noise, and the image degradation model is:

[0004] f(i,j)=u(i,j)+n(i,j)

[0005] Where f(i,j), i = 1, 2, ..., M, j = 1, 2, ..., N is the actual observation image obtained by the remote sensing camera; u(i,j) is the original image, which is the quantity to be solved; n(i,j) is the strip noise.

[0006] So far, scholars at home and abroad have proposed many methods to remove image stripe noise and solve the original image u, which can be divided into three categories: statistical matching-based methods, filtering-based methods and variational optimization-based methods. Statistical matching-based methods mainly include histogram matching, moment matching and their improved methods. These methods remove stripe noise based on the difference in numerical statistics of each sensor. The denoising effect is seriously dependent on the statistical data set, and there are problems such as incomplete denoising effect and slow convergence speed. Filtering-based methods mainly include Fourier transform filtering and wavelet transform filtering. The Fourier transform filtering method analyzes the frequency of stripe noise and uses a suitable filter to detect and remove stripe noise; the wavelet transform filtering method removes stripe noise by determining the threshold. However, neither Fourier nor wavelet transform can accurately separate the frequency components, resulting in incomplete stripe noise removal or loss of image detail information, and the selection of parameters such as filters or thresholds requires certain skills and experience. The variational optimization-based method constructs an energy functional and minimizes the energy functional to obtain the optimal solution. Marouan et al. (Marouan Bouali and Said Ladjal. Toward optimal destriping of MODIS data using a unidirectional variational model. IEEE Transactions on Geoscience and Remote Sensing. 2011, 49(8): 2924-2935.) established a unidirectional UTV variational denoising model based on the unidirectional characteristics of stripe noise. This model can effectively remove stripe noise and retain edges. However, when there are multiple stripe noises of varying strengths in the image, this model often leaves residual stripes or over-blurs the details. Zhou et al. (Zhou Gang, Fang Houzhang, Yan Luxin, and et al. Removal of stripe noise with spatially adaptive unidirectional total variation. Optik. 2014 (125): 2756-2762.) proposed an adaptive variational denoising model SAUTV based on the spatial characteristics of the image. The model adaptively adjusts the regularization constraint weights, imposes smaller regularization constraints at the edges to maintain the edge features of the image, and imposes larger regularization constraints in flat areas, thereby better removing stripe noise.Zhou et al. (Zhou Gang, Fang Houzhang, Lu Cen, and et al. Robust destriping of MODIS and hyperspectral data using a hybrid unidirectional total variation model. Optik. 2015(126):838-845.) used L based on the UV variational denoising model. 1 The norm constrains the images before and after denoising and proposes the HUTV variational denoising model. The model adaptively selects the fidelity term and its strength of the variational denoising model based on the weight matrix W established according to the image gradient, which can effectively remove the strip noise of different strengths, but will blur the image details. Zhang et al. (Zhang Yaozong, Zhou Gang, Yan Luxin, and Zhang Tianxu. Adestriping algorithm based on TV-stokes and unidirectional total variation model. Optik. 2016 (127): 428-439.) use the TV-Stokes model to effectively reduce L 1 In order to eliminate the staircase effect produced by the norm and better preserve the characteristics of the image edge, a variational model (UTV-Stokes) for removing stripe noise is constructed, which combines one-way variation with TV-Stokes. This model can avoid the loss of image details due to excessive smoothing of stripe noise.

[0007] In general, the variational optimization-based method has strong flexibility and can improve the denoising performance and application scope of the algorithm through model design. Therefore, it has important research significance and prospects in the problem of complex stripe noise removal in images. The variational optimization-based method has a relatively complex calculation process in terms of numerical solution, but the continuous development of optimization algorithms and optimization models has laid a good foundation for the implementation of variational methods. The split Bregman iteration method (Tom Goldstein and Stanley Osher. The split Bregman method for L1-regularized problems. Siam Journal on Imaging Sciences. 2009 (2): 323-343.) is the most effective method for solving variational methods. Its basic idea is to introduce auxiliary variables to replace the more difficult parts of the original variational denoising model, thereby simplifying the solution of the variational denoising model, and then replace the equality constraints by adding penalty terms to ensure that the new linear variational denoising model has the same solution as the original variational denoising model. Finally, the alternating minimum method is used to solve the new linear variational correction model.

[0008] Although the stripe noise removal method based on variational optimization has achieved certain research results, there are still some problems and shortcomings. The UTV, HUTV and UTV-Stokes variational denoising models have the same regularization constraint weights, that is, the variational denoising model assigns the same weight to the image texture area and the stripe noise area, resulting in blurred image details after removing the stripe noise; the SAUTV variational denoising model uses adaptive regularization weights, that is, a smaller regularization constraint is imposed on the image texture area to maintain the image characteristics, and a larger regularization constraint is imposed on the stripe noise area, which can remove multiple stripe noises of the same strength, but the stripe noise will be classified as image texture, resulting in residual stripe noise and loss of image detail information after denoising. Summary of the invention

[0009] (I) Purpose of the invention

[0010] The purpose of the present invention is to propose an adaptive stripe noise removal method based on spatial characteristics to address the deficiencies of the prior art, which can effectively maintain the edge detail information of the image itself while removing multiple stripe noises of different strengths.

[0011] (II) Technical solution

[0012] In order to solve the above technical problems, the present invention provides an adaptive stripe noise removal method based on spatial characteristics, using a remote sensing imaging detection system to obtain a noise image f as an observation image, and the size is M×N; constructing an adaptive weight function w(i, j) according to the spatial characteristics of the observation image; constructing an adaptive variational denoising model E(u) about the original image u, and adaptively controlling the penalty term ||▽ in the variational denoising model E(u) through the weight function w(i, j). y u|| is the constraint strength at different pixel points. A larger constraint is applied at the stripe noise to remove the noise, while a smaller constraint is applied at the image texture to maintain the characteristics of the observed image f. 1 The norm constrains the local smoothness along the strip direction ||▽ x (uf)||, maintain the smoothness of the strip noise n(i,j) in its extension direction; and use L 2 Norm constrains the image grayscale ||uf|| 2 ; Then, according to the split Bregman iteration method, auxiliary variables d1 and d2 are introduced to transform the nonlinear variational denoising model E(u) into a linear variational denoising model E′(u); Finally, the alternating minimization method is used to solve the above linear variational denoising model E′(u) to obtain the optimal solution of the denoised image Wherein M and N are natural numbers, 1≤i≤M, 1≤j≤N, i and j are the coordinate values ​​of a given pixel on the x-axis and y-axis, respectively, both of which are natural numbers.

[0013] Among them, the process of constructing the adaptive weight function w(i,j) according to the spatial characteristics of the observed image is:

[0014] (1.1) Calculate the local absolute difference D(i,j), 1≤i≤M,1≤j≤N

[0015] The local absolute difference D(i,j) is defined as:

[0016]

[0017] The D(i,j) operator is used to distinguish stripe noise from image texture, where G σ is a Gaussian kernel with standard deviation σ; Denotes the convolution operation; D x (i,j) is the absolute difference on the x-axis, calculated as follows:

[0018] D x (i,j)=max{f i+1,j -f i,j |,f i-1,j -f i,j}

[0019] Among them, max{} represents the larger value between the two numbers;

[0020] (1.2) Calculate the balance factor Where max() represents the largest number in D(i,j), and min() represents the smallest number in D(i,j).

[0021] Among them, the local absolute difference scale is 3×7, σ=1.

[0022] Among them, the adaptive weight function constructed Where k is a non-negative parameter that controls the degree of spatial information participation, D(i,j) is the local absolute difference, and S(i,j) is the balance factor.

[0023] Among them, the adaptive weight function w(x,y) is used to construct the adaptive variational denoising model E(u) for the original image u:

[0024]

[0025] Among them, λ1 and λ2 are non-negative regularization parameters, u is the original image to be found, ▽ x represents the x-axis gradient operator, ▽ y represents the gradient operator in the y-axis direction, and w is the adaptive weight function.

[0026] Among them, the split Bregman iteration method is used to calculate the variational denoising model E(u) to obtain the denoised image The process includes:

[0027] (3.1) Due to the regularization constraint ||▽ x (uf)|| and||▽ y u|| is non-smooth and inseparable. Auxiliary variables d1 and d2 are introduced to transform the variational denoising model E(u) into the following constrained variational denoising model:

[0028]

[0029] (3.2) Add two penalty terms to transform the constrained variational denoising model E(u) into the unconstrained variational denoising model E′(u):

[0030]

[0031] Where α1 and α2 are non-negative parameters.

[0032] Among them, the variational denoising model E′(u) is solved by the alternating minimization method. The specific steps are as follows:

[0033] (3.3.1) Initialize λ1, λ2, α1, α2, ε, u 0 =f,

[0034] (3.3.2) Calculation of auxiliary variables

[0035]

[0036] (3.3.3) Calculation of auxiliary variables

[0037]

[0038] (3.3.4) Calculate the denoised image u k+1 / 2 :

[0039] The term u in E′(u) is strictly diagonal and is solved using the Gauss-Seidel method. 1≤i≤M, 1≤j≤N, that is:

[0040]

[0041] (3.3.5) Using the denoised image u k+1 / 2 Update denoised image u k+1 :u k+1 =max{u k+1 / 2 ,0};

[0042] (3.3.6) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters

[0043] (3.3.7) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters

[0044] (3.3.8) until the termination condition ||u k+1 -u k || / ||u k+1 ||≤ε satisfies, otherwise, set the counting parameter k=k+1 and continue to execute (3.3.2);

[0045] In the above formula, b1 and b2 are Bregman auxiliary parameters, and k is the counting parameter; x represents the gradient in the x-axis direction, ▽ y Represents the gradient in the y-axis direction; max{} represents the larger and smaller values ​​of the two numbers.

[0046] Among them, for the denoised image u k+1 Round off to get the denoised image u = uint (u k+1), grayscale image uint() represents 8-bit integer operation, color image uint() represents 8-bit integer operation on R, G and B channels.

[0047] (III) Beneficial effects

[0048] The adaptive stripe noise removal method based on spatial characteristics provided by the above technical solution aims at the shortcomings of the traditional stripe noise removal method, calculates the local absolute difference according to the spatial characteristics of the observed image; then, an adaptive weight function is constructed by using the local absolute difference to adaptively control the constraint strength of the penalty term in the variational denoising model at different pixel points, imposes a larger constraint at the stripe noise to remove the noise, and imposes a smaller constraint at the image texture to maintain the characteristics of the observed image; secondly, L 1 The norm constrains the local smoothness along the strip direction, keeping the strip noise having obvious smooth characteristics in its extension direction; and L 2 norm to constrain the grayscale of the image; finally, the split Bregman iteration method and the alternating minimization method are used to quickly solve the model and obtain better correction results ( Figure 2 f, Figure 3 f, Figure 4 e, Figure 6 e), thereby overcoming the shortcomings of the traditional method of removing stripe noise, such as blurred edges and lost details. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of the adaptive stripe noise removal method based on spatial characteristics of the present invention.

[0050] Figure 2 This is a schematic diagram of the experimental comparison of Lena image simulated degraded stripe noise removal.

[0051] Figure 3 This is a schematic diagram of the comparison of Cameraman image simulation degraded stripe noise removal experiments.

[0052] Figure 4 This is a schematic diagram of the comparison of experiments on removing degraded stripe noise from MODIS image 1.

[0053] Figure 5 yes Figure 4 The original strip image and the corresponding profile diagrams of the denoising results of the four methods.

[0054] Figure 6 This is a schematic diagram of the comparison of MODIS image 2 degraded stripe noise removal experiments.

[0055] Figure 7 yes Figure 6 The original strip image and the corresponding profile diagrams of the denoising results of the four methods. DETAILED DESCRIPTION

[0056] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.

[0057] See also Figure 1 According to the present invention, a noise image f is obtained as an observation image using a remote sensing imaging detection system, and the size is M×N; an adaptive weight function w(i, j) is constructed according to the spatial characteristics of the observation image; an adaptive variational denoising model E(u) about the original image u is constructed, and the penalty term ||▽ in the variational denoising model E(u) is adaptively controlled by the weight function w(i, j). y u|| is the constraint strength at different pixel points. A larger constraint is applied at the stripe noise to remove the noise, while a smaller constraint is applied at the image texture to maintain the characteristics of the observed image f. 1 The norm constrains the local smoothness along the strip direction ||▽ x (uf)||, keep the strip noise n(i,j) with obvious smoothness in its extension direction; and use L 2 Norm constrains the image grayscale ||uf|| 2 ; Then, according to the split Bregman iteration method, auxiliary variables d1 and d2 are introduced to transform the nonlinear variational denoising model E(u) into a linear variational denoising model E′(u). Finally, the alternating minimization method is used to solve the above linear variational denoising model E′(u) to obtain the optimal solution of the denoised image. Wherein M and N are natural numbers.

[0058] The specific process includes:

[0059] (1) Construct an adaptive weight function w(i, j) based on the spatial characteristics of the observed image, 1≤i≤M, 1≤j≤N, where i and j are the coordinate values ​​of the given pixel on the x-axis and y-axis, respectively, and are both natural numbers, including the following process:

[0060] (1.1) Calculate the local absolute difference D(i,j), 1≤i≤M, 1≤j≤N;

[0061] The local absolute difference D(i,j) is defined as:

[0062]

[0063] The D(i,j) operator can effectively distinguish stripe noise and image texture, where G σ is a Gaussian kernel with standard deviation σ; represents the convolution operation, in this patent, the local absolute difference scale is 3×7, σ=1; D x (i,j) is the absolute difference on the x-axis, calculated as follows:

[0064] D x (i,j)=max{f i+1,j -f i,j |,f i-1,j -f i,j}

[0065] Where max{} represents the larger value of the two numbers.

[0066] (1.2) Calculate the balance factor Where max() represents the largest number in D(i,j), and min() represents the smallest number in D(i,j);

[0067] (1.3) Constructing adaptive weight function Where k is a non-negative parameter that controls the degree of spatial information participation, D(i,j) is the local absolute difference, and S(i,j) is the balance factor.

[0068] (2) Using the adaptive weight function w(x,y), an adaptive variational denoising model E(u) is constructed for the original image u:

[0069]

[0070] Where λ1 and λ2 are both non-negative regularization parameters, u is the original image to be obtained, ▽ x represents the x-axis gradient operator, ▽ y represents the gradient operator in the y-axis direction, and w is the adaptive weight function.

[0071] (3) Use the split Bregman iteration method to calculate the variational denoising model E(u) to obtain the denoised image The process includes:

[0072] (3.1) Due to the regularization constraint ||▽ x (uf)|| and||▽ y u|| is non-smooth and inseparable. Auxiliary variables d1 and d2 are introduced to transform the variational denoising model E(u) into the following constrained variational denoising model:

[0073]

[0074] (3.2) Add two penalty terms to transform the constrained variational denoising model E(u) into the unconstrained variational denoising model E′(u):

[0075]

[0076] Where α1 and α2 are non-negative parameters.

[0077] (3.3) The variational denoising model E′(u) is solved by the alternating minimization method. The specific steps are as follows:

[0078] (3.3.1) Initialize λ1, λ2, α1, α2, ε, u 0 =f, Where b1 and b2 are Bregman auxiliary parameters, and k is the counting parameter;

[0079] (3.3.2) Calculate auxiliary variable d1 k+1 :

[0080]

[0081] , where max{} represents the larger value of the two numbers;

[0082] (3.3.3) Calculation of auxiliary variables

[0083]

[0084] (3.3.4) Calculate the denoised image u k+1 / 2 :

[0085] The term u in E′(u) is strictly diagonal and is solved using the Gauss-Seidel method. 1≤i≤M, 1≤j≤N, that is:

[0086]

[0087] (3.3.5) Using the denoised image u k+1 / 2 Update denoised image u k+1 :u k+1 =max{u k+1 / 2 ,0}, where max{} represents the larger value of the two numbers;

[0088] (3.3.6) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters

[0089] (3.3.7) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters

[0090] (3.3.8) until the termination condition ||u k+1 -u k || / ||u k+1||≤ε satisfies, otherwise, set the counting parameter k=k+1 and continue to execute (3.3.2).

[0091] (4) Denoised image u k+1 Round to get the denoised image For grayscale images, uint() represents 8-bit rounding operations, and for color images, uint() represents 8-bit rounding operations on the three channels of R, G, and B.

[0092] Examples:

[0093] Figure 2 This is a schematic diagram of the experimental comparison of Lena image simulation degraded stripe noise removal; Figure 2 a is the original image of Lena; Figure 2 b is the simulated degraded image; Figure 2 c is the denoised image using the UTV method proposed by Marouan et al.; Figure 2 d is the denoised image using the SAUTV method proposed by Zhou et al.; Figure 2 e is the denoised image using the UTV-Stokes method proposed by Zhang et al.; Figure 2 f is the adaptive denoised image of the present invention.

[0094] Figure 3 This is a comparison diagram of the Cameraman image simulation degraded stripe noise removal experiment; Figure 3 a is the original Cameraman image; Figure 3 b is the simulated degraded image; Figure 3 c is the denoised image using the UTV method proposed by Marouan et al.; Figure 3 d is the denoised image using the SAUTV method proposed by Zhou et al.; Figure 3 e is the denoised image using the UTV-Stokes method proposed by Zhang et al.; Figure 3 f is the adaptive denoised image of the present invention.

[0095] Figure 4 This is a comparison diagram of the MODIS image 1 degraded stripe noise removal experiment; Figure 4 a is the MODIS degraded strip image; Figure 4 b is the denoised image using the UTV method proposed by Marouan et al.; Figure 4 c is the denoised image using the SAUTV method proposed by Zhou et al.; Figure 4 d is the denoised image using the UTV-Stokes method proposed by Zhang et al.; Figure 4 e is the adaptive denoised image of the present invention.

[0096] Figure 5 yes Figure 4 The original strip image and the corresponding profile diagrams of the denoising results of the four methods; Figure 5 a is the profile corresponding to the MODIS degraded strip image; Figure 5 b is the profile diagram corresponding to the denoising result of the UTV method proposed by Marouan et al.; Figure 5 c is the profile corresponding to the denoising result of the SAUTV method proposed by Zhou et al.; Figure 5 d is the profile diagram corresponding to the denoising result of the UTV-Stokes method proposed by Zhang et al.; Figure 5 e is the profile diagram corresponding to the denoising result of the adaptive method of the present invention.

[0097] Figure 6 This is a comparison diagram of the MODIS image 2 degraded stripe noise removal experiment; Figure 6 a is the MODIS degraded strip image; Figure 6 b is the denoised image using the UTV method proposed by Marouan et al.; Figure 6 c is the denoised image using the SAUTV method proposed by Zhou et al.; Figure 6 d is the denoised image using the UTV-Stokes method proposed by Zhang et al.; Figure 6 e is the adaptive denoised image of the present invention.

[0098] Figure 7 yes Figure 6 The original strip image and the corresponding profile diagrams of the denoising results of the four methods; Figure 7 a is the profile corresponding to the MODIS degraded strip image; Figure 7 b is the profile diagram corresponding to the denoising result of the UTV method proposed by Marouan et al.; Figure 7 c is the profile corresponding to the denoising result of the SAUTV method proposed by Zhou et al.; Figure 7 d is the profile diagram corresponding to the denoising result of the UTV-Stokes method proposed by Zhang et al.; Figure 7 e is the profile diagram corresponding to the denoising result of the adaptive method of the present invention.

[0099] In this embodiment, Figure 5 and Figure 7 They are shown separately Figure 4 and Figure 6 The original stripe image and the corresponding profiles of the denoising results of the four methods. The profile of the stripe image has a curve that changes randomly over the entire bandwidth, that is, the curve has strong oscillations. It can be seen that after using the four methods to remove the stripes, the curves in the profile corresponding to the restored results become very smooth. Figure 5 (bc) and Figure 7 In (bc), the curve has small peaks and burrs, and the corresponding methods are UTV and SAUTV. Figure 4 and Figure 6The corresponding recovery results show that these small burrs are the manifestation of residual strips. In contrast, the profile curves of UTV-Stokes and the present invention are smoother.

[0100] The present invention introduces two evaluation factors, peak signal-to-noise ratio (PSNR) and structural similarity method (SSIM), to objectively and quantitatively evaluate the simulation experiment. Table 1 is the experimental Figure 2 and Figure 3 objective evaluation results.

[0101] Table 1

[0102]

[0103]

[0104] The objective evaluation index Q index of no-reference images (Zhu Xiang and Peyman Milanfar. Automatic Parameter Selection for Denoising Algorithms Using a No-Reference Measure of Image Content. 2010, 19(12): 3116-3122.) is used to conduct objective quantitative evaluation of the real experiment. Table 2 is the experimental Figure 4 and Figure 6 objective evaluation results.

[0105] Table 2

[0106]

[0107] The present invention can better preserve the original image features, and the intensity of the denoised image is closer to that of the original image, because:

[0108] 1) An adaptive weight function with excellent performance is constructed by using local absolute difference to adaptively control the constraint strength of the penalty term in the variational denoising model at different pixel points. A larger constraint is imposed on the stripe noise to remove the noise, while a smaller constraint is imposed on the image texture to maintain the characteristics of the observed image f;

[0109] 2) Using L 1 The norm constrains the local smoothness along the strip direction, keeping the strip noise having obvious smooth characteristics in its extension direction; and using L 2 Norm constrains the grayscale of the image and can obtain better results ( Figure 2 f, Figure 3 f, Figure 4 e, Figure 6 e).

[0110] In addition, it can be seen from the objective evaluation of the simulation experiment shown in Table 1 that the present invention obtains higher PSNR and SSIM values; and it can be seen from the objective evaluation of the real experiment shown in Table 2 that the present invention obtains a higher Q index. This shows that the results obtained by the present invention can better represent the original clear image in terms of both grayscale information and structural features, which is consistent with the visual comparison results. In short, it has obvious advantages in the comparison of visual effects and quantitative evaluation.

[0111] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An adaptive stripe noise removal method based on spatial characteristics, characterized in that: A noise image f is obtained as the observation image using a remote sensing imaging detection system, and the size is M×N; an adaptive weight function w(i, j) is constructed according to the spatial characteristics of the observation image; an adaptive variational denoising model E(u) is constructed for the original image u, and the penalty term in the variational denoising model E(u) is adaptively controlled by the weight function w(i, j) The constraint strength at different pixel points is to apply a larger constraint to remove the noise at the stripe noise, and to apply a smaller constraint to the image texture to maintain the characteristics of the observed image f; L 1 The norm constrains the local smoothness along the strip direction Keep the smoothness of the strip noise n(i,j) in its extension direction; and use L 2 Norm constrains the image grayscale uf 2 ; Then, according to the split Bregman iteration method, auxiliary variables d1 and d2 are introduced to transform the nonlinear variational denoising model E(u) into a linear variational denoising model E′(u); Finally, the alternating minimization method is used to solve the above linear variational denoising model E′(u) to obtain the optimal solution of the denoised image Wherein M and N are natural numbers, 1≤i≤M, 1≤j≤N, i and j are the coordinate values ​​of a given pixel on the x-axis and y-axis, respectively, both of which are natural numbers.

2. The method for removing adaptive stripe noise based on spatial characteristics according to claim 1, characterized in that: The process of constructing the adaptive weight function w(i,j) according to the spatial features of the observed image is: (1.1) Calculate the local absolute difference D(i,j), 1≤i≤M,1≤j≤N The local absolute difference D(i,j) is defined as: The D(i,j) operator is used to distinguish stripe noise from image texture, where G σ is a Gaussian kernel with standard deviation σ; Denotes the convolution operation; D x (i,j) is the absolute difference on the x-axis, calculated as follows: D x (i,j)=max{f i+1,j -f i,j ,f i-1,j -f i,j } Among them, max{} represents the larger value between the two numbers; (1.2) Calculate the balance factor Where max() represents the largest number in D(i,j), and min() represents the smallest number in D(i,j).

3. The method for removing adaptive stripe noise based on spatial characteristics according to claim 2, characterized in that: The local absolute difference scale is 3×7, σ=1.

4. The method for removing adaptive stripe noise based on spatial characteristics according to claim 3, characterized in that: Constructed adaptive weight function Where k is a non-negative parameter that controls the degree of spatial information participation, D(i,j) is the local absolute difference, and S(i,j) is the balance factor.

5. The method for removing adaptive stripe noise based on spatial characteristics according to claim 4, characterized in that: The adaptive weight function w(x,y) is used to construct the adaptive variational denoising model E(u) for the original image u: Among them, λ1 and λ2 are non-negative regularization parameters, u is the original image to be found, represents the x-axis direction gradient operator, represents the gradient operator in the y-axis direction, and w is the adaptive weight function.

6. The method for removing adaptive stripe noise based on spatial characteristics according to claim 5, characterized in that: The split Bregman iteration method is used to calculate the variational denoising model E(u) to obtain the denoised image u, which includes the following process: (3.1) Due to the regularization constraint and It is non-smooth and inseparable. Auxiliary variables d1 and d2 are introduced to transform the variational denoising model E(u) into the following constrained variational denoising model: (3.2) Add two penalty terms to transform the constrained variational denoising model E(u) into the unconstrained variational denoising model E′(u): Where α1 and α2 are non-negative parameters.

7. The method for removing adaptive stripe noise based on spatial characteristics according to claim 6, characterized in that: The variational denoising model E′(u) is solved by the alternating minimization method. The specific steps are as follows: (3.3.1) initializationλ1,λ2,α1,α2,ε,u 0 =f, k=0; (3.3.2) Calculation of auxiliary variables (3.3.3) Calculation of auxiliary variables (3.3.4) Calculate the denoised image u k+12 : The term u in E′(u) is strictly diagonal and is solved using the Gauss-Seidel method. Right now: (3.3.5) Using the denoised image u k+12 Update denoised image u k+1 :u k+1 =max{u k+12 ,0}; (3.3.6) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters (3.3.7) Using auxiliary variables and denoised image u k+1 Update auxiliary parameters (3.3.8) until the termination condition u k+1 -u k u k+1 ≤ε is satisfied, otherwise, set the counting parameter k = k + 1 and continue to execute (3.3.2); In the above formula, b1 and b2 are Bregman auxiliary parameters, and k is the counting parameter; represents the gradient in the x-axis direction, Represents the gradient in the y-axis direction; max{} represents the larger and smaller values ​​of the two numbers.

8. The method for removing adaptive stripe noise based on spatial characteristics according to claim 7, characterized in that: For the denoised image u k+1 Round to get the denoised image For grayscale images, uint() represents 8-bit rounding operations, and for color images, uint() represents 8-bit rounding operations on the three channels of R, G, and B.

9. An application of the adaptive stripe noise removal method according to any one of claims 1 to 8 in stripe noise removal in a remote sensing imaging system with multiple CCDs stitching focal planes.

10. An application of the adaptive stripe noise removal method according to any one of claims 1 to 8 in the field of digital image processing technology.

Images