Remote sensing image blind deblurring algorithm based on flexible norm constraint

Through the remote sensing image blind defuzzing algorithm based on flexible norm constraints, combined with lP-Q norm constraints and generalized soft threshold algorithm, the remote sensing image blur and noise problems are solved, high-quality image recovery is achieved, and the development of remote sensing technology is improved.

CN118918038APending Publication Date: 2024-11-08CHANGCHUN INST OF OPTICS FINE MECHANICS & PHYSICS CHINESE ACAD OF SCI
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411172293.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The blur distortion and noise caused by atmospheric turbulence, optical system problems, etc. during the imaging process are difficult to effectively remove, affecting the image quality.

Method used

Using the remote sensing image blind defuzzing algorithm based on flexible norm constraints, by establishing an image blind defuzzing model, using projection alternating iteration method and improved generalized soft threshold algorithm, combined with lP-Q norm constraints, sparsely represent image gradients and decompose them. Blur the kernel to restore clear images.

Benefits of technology

It realizes the effect of removing image blur while suppressing noise, obtains high-quality remote sensing image data, and enhances the development significance of remote sensing technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918038A_ABST
    Figure CN118918038A_ABST
Patent Text Reader

Abstract

The invention relates to a remote sensing image blind deblurring algorithm based on flexible norm constraint, and the algorithm comprises the following steps: inputting a fuzzy remote sensing image, and building an image blind deblurring model; initializing model parameters; alternately iterating the clear image L and the blurring kernel k by adopting projection; calculating a clear image L; calculating a fuzzy kernel k; updating the iteration parameters and repeating the steps until a loop termination condition is met; the fuzzy kernel k is output after normalization and non-negative constraint; and performing non-blind deconvolution solution by using the fuzzy kernel k, and outputting a final clear remote sensing image. The characteristics of gradient heavy tail distribution and maximum local gradient are combined, the lP norm is adopted to constrain most gradient elements, the local gradient is implicitly constrained through the lQ norm to be embedded into the potential gradient image, sparse induction is performed in the deblurring iteration process, flexible constraint is achieved, a more effective blurring kernel solution is obtained, and the method has the advantages of being high in robustness and high in robustness. The method can realize acquisition of high-quality remote sensing image data, and is of great significance to development of a remote sensing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a remote sensing image blind deblurring algorithm based on flexible norm constraints. Background Art

[0002] In the process of acquiring and transmitting remote sensing images, various factors, such as atmospheric turbulence effects, diffraction of the optical system in the imaging equipment, aberration, defocus, satellite platform flutter, relative motion between the remote sensor and the ground object, will inevitably cause image distortion. The quality degradation caused by these factors is usually called image degradation. Due to image degradation, the observed image will lose a lot of valuable information, and the contrast will deteriorate. The typical manifestation is that the image appears blurred and distorted and has additional noise. To this end, the degraded image must be processed to estimate the true original image. This process is called image restoration, which includes deblurring and denoising. In practical problems, the degraded point spread function is unknown, and a robust algorithm needs to be designed to obtain the point spread function. Summary of the invention

[0003] The present invention aims to solve the technical problem of degradation and blurring distortion of remote sensing images caused during the imaging process in the prior art, and provides a remote sensing image blind deblurring algorithm based on flexible norm constraints.

[0004] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0005] A remote sensing image blind deblurring algorithm based on flexible norm constraint includes the following steps:

[0006] Step 1: Input the blurred remote sensing image and establish the image blind deblurring model;

[0007] Step 2: Set the initial parameters of the image blind deblurring model and the penalty parameter δ i And the size of blur kernel k and the number of iterations N;

[0008] Step 3: Use projection alternating iteration to estimate the potential clear image L and blur kernel k;

[0009] Step 4: Calculate the sub-problem of the potential clear image L;

[0010] Calculate subproblems t1, t2 and g to get the potential clear image L n+1 ;

[0011] Step 5: Calculate the sub-problem of blur kernel k;

[0012] By clear image L n+1 The subproblem of estimating blur kernel k and normalizing it to obtain blur kernel k n+1 ;

[0013] Step 6: Update the iteration parameters and repeat steps 4 to 5 until the loop termination condition is met;

[0014] Step 7: Blur kernel k n+1 After normalization and non-negative constraints, the final blur kernel k is output;

[0015] Step 8: Use blur kernel k to perform non-blind deconvolution and output the final clear remote sensing image.

[0016] In the above technical solution, in step 1, a flexible norm l adapted to deblurring is used. P-Q To sparsely represent the image gradient, establish an image blind deblurring model:

[0017]

[0018] Among them, B represents the blurred image, L represents the potential clear image, k represents the blur kernel, and '*' represents the convolution operator symbol. and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, α and β are regularization parameters, l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm combination (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

[0019] In the above technical solution, the initial parameters of the image blind deblurring model in step 2 include: ω1>0, ω2>0, α>0, β=2, L 0 =B, 0<p<1, q=-0.5.

[0020] In the above technical solution, step 4 is specifically as follows:

[0021] The sub-problem of the latent sharp image L is expressed as:

[0022]

[0023] The semi-quadratic splitting method is used to introduce auxiliary variables t1, t2 and g to replace the above formula and Get the formula:

[0024]

[0025] Among them, B represents the blurred image, L represents the potential clear image, and k n represents the nth iteration blur kernel, and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, and α are regularization parameters, δ1, δ2, and δ3 are the corresponding penalty parameters; l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm array (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components;

[0026] Given a latent clear image L, the sub-problem model for solving t1 and t2 is:

[0027]

[0028] The solution is:

[0029]

[0030] Among them, RE represents the residual image BL*k n ; B represents blurred image, L represents potential clear image, k n represents the nth iteration blur kernel, and denote the first-order and second-order gradient operators respectively, ω1 and ω2 are regularization parameters, δ1 and δ2 are corresponding penalty parameters, and l e The norm represents the combination of the l0 and l1 norms (||·|| e =||·||0+||·||1);

[0031] Given a latent clear image L, the sub-problem model for solving g is:

[0032]

[0033] The solution is obtained by combining positive P and negative Q norms with flexible constraints using an improved generalized soft threshold algorithm, where the element value with the largest local gradient is used. Q The norm is solved, and the other element values ​​are minimized using l P Norm solution.

[0034] In the above technical solution, in step 4, for fixed t1, t2 and g, the model of the latent clear image L is transformed into:

[0035]

[0036] By FFT solution:

[0037]

[0038] in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

[0039] In the above technical solution, step five is specifically: the sub-problem of the fuzzy kernel k is expressed as:

[0040]

[0041] Among them, the norm ||·|| e The problem of solving L and the norm when solving L||·|| e Similarly, the sub-problem needs to introduce auxiliary variables t1 and t2, and then solve it through FFT:

[0042]

[0043] in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

[0044] The present invention has the following beneficial effects:

[0045] The remote sensing image blind deblurring algorithm based on flexible norm constraint of the present invention combines the characteristics of gradient heavy tail distribution and local gradient maximum, and adopts l P norm constrains most gradient elements and implicitly through l QThe norm-constrained local gradient is embedded into the latent gradient image, and sparse induction is performed in the deblurring iterative process to achieve flexible constraints. An improved generalized soft threshold method is used to optimize the solution of the gradient subproblem, and finally an effective blur kernel solution is obtained through the Fourier operator.

[0046] The remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention can remove blur in the image while having a certain noise suppression effect, and can realize the acquisition of high-quality remote sensing image data, which is of great significance to the development of remote sensing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0048] Figure 1 The figure is a schematic diagram of the overall process of the remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention.

[0049] Figure 2 A schematic diagram of the process of solving the image gradient with flexible norm constraints is provided to utilize the remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention.

[0050] Figure 3 This is Example 1 of the effect diagram of processing a real remote sensing image using the remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention; wherein (a) is Example 1 of the blurred remote sensing image, and (b) is the deblurred image using the algorithm of the present invention.

[0051] Figure 4 This is Example 2 of the effect diagram of using the remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention to process a real remote sensing image; wherein (a) is Example 2 of the blurred remote sensing image, and (b) is the deblurred image of the algorithm of the present invention. DETAILED DESCRIPTION

[0052] The inventive concept of the present invention is:

[0053] The remote sensing image blind deblurring algorithm based on flexible norm constraint of the present invention adds corresponding constraints to the blurred image, decomposes the degraded point spread function, which can also be called blur kernel, and further uses blur kernel k to restore high-quality clear image. Its restoration mechanism is as follows:

[0054] B(x,y)=k(x,y)*L(x,y)+N(x,y)

[0055] Among them, L(x,y) represents the original image, B(x,y) represents the degraded image, k(x,y) represents the blur kernel, N(x,y) represents the noise, and '*' is the convolution operator symbol.

[0056] This problem is a classic inverse problem, the key lies in the acquisition of k(x,y) in the deconvolution process and the suppression of noise.

[0057] In order to obtain an effective and clear solution, it is necessary to impose constraints on the above problems and design a decoupling algorithm. In the framework of maximum a posteriori probability, the present invention adds the first-order and second-order gradients of the constrained data fitting terms, considers the heavy-tailed distribution of the gradient and combines the characteristics of the maximum local gradient, and proposes a flexible norm l suitable for deblurring. P-Q To sparsely represent the image gradient, an optimization model for blind image deblurring is established:

[0058]

[0059] The first and second terms in the model are based on the norm l2 and the sparse norm l e The weighted sum of is used to constrain the data fitting term (BL*k), which is used to model complex noise to constrain and ensure that the convolution output of the latent image L and the blur kernel k decomposed from the restored image is similar to the observed blurred image B. The third term is the proposed l P-Q The joint norm of this term ensures that the algorithm prefers clear images to blurred images, retains large gradients and sharp edges and removes small details. The fourth term uses the l2 norm to constrain the solution of the blur kernel, which can be efficiently calculated using the Fast Fourier Transform (FFT).

[0060] Among them, B represents the blurred image, L represents the potential clear image, k represents the blur kernel, and '*' represents the convolution operator symbol. and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, α and β are regularization parameters, l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm combination (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

[0061] The above optimization problem can be solved by alternately iterating the potential clear image L and the blur kernel k using the projection alternating minimization strategy. The steps are as follows:

[0062] 1. The sub-problem of the potential clear image L can be expressed as:

[0063]

[0064] The semi-quadratic splitting method is used to introduce auxiliary variables t1, t2 and g to replace the above formula and Then we can get the formula:

[0065]

[0066] Among them, B represents the blurred image, L represents the potential clear image, and k n represents the nth iteration blur kernel, and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, and α are regularization parameters, δ1, δ2, and δ3 are the corresponding penalty parameters. l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm array (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

[0067] Given a latent clear image L, the sub-problem model for solving t1 and t2 is:

[0068]

[0069] The solution is:

[0070]

[0071] For convenience, RE is used here to represent the residual image BL*k n Among them, B represents the blurred image, L represents the potential clear image, and k nrepresents the nth iteration blur kernel, and denote the first-order and second-order gradient operators respectively, ω1 and ω2 are regularization parameters, δ1 and δ2 are corresponding penalty parameters, and l e The norm represents the combination of the l0 and l1 norms (||·|| e =||·||0+||·||1).

[0072] Given a latent clear image L, the sub-problem model for solving g is:

[0073]

[0074] This problem is solved by combining positive P and negative Q norms with flexible constraints using an improved generalized soft thresholding algorithm, where the element value with the largest local gradient is used. Q The norm is solved, and the other element values ​​are minimized using l P The specific solution process is as follows: Figure 2 shown.

[0075] For fixed t1, t2 and g, the model of L becomes:

[0076]

[0077] By FFT solution:

[0078]

[0079] in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

[0080] 2. The sub-problem of blur kernel k is to solve it in the gradient domain, which can be expressed as:

[0081]

[0082] Similarly, similar to the subproblem of solving the latent clear image L, the norm ||·|| e The solution of the problem requires the introduction of auxiliary variables t1 and t2, and then the sub-problem of the fuzzy kernel k can be solved by FFT:

[0083]

[0084] in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

[0085] The constraint model is used to iterate the potential clear image L and the blur kernel k, and the optimal value is approached continuously until the loop cutoff condition is met. The blur kernel k calculated by the output image is normalized and non-negatively constrained, and then the non-blind deconvolution algorithm is used to obtain the final clear image.

[0086] The present invention is described in detail below with reference to the accompanying drawings.

[0087] like Figure 1 As shown, the remote sensing image blind deblurring algorithm based on flexible norm constraint of the present invention comprises the following steps:

[0088] Step 1: Input the blurred remote sensing image and establish the image blind deblurring model;

[0089] Step 2: Set the initial model parameters ω1>0, ω2>0, α>0, β=2, L 0 =B, 0<p<1, q=-0.5, penalty parameter δ i And the size of blur kernel k and the number of iterations N;

[0090] Step 3: Use projection alternating iteration to estimate the potential clear image L and blur kernel k;

[0091] Step 4: Calculate the sub-problem of the potential clear image L;

[0092] Calculate subproblems t1, t2 and g to get the potential clear image L n+1 ;

[0093]

[0094] Step 5: Calculate the sub-problem of blur kernel k;

[0095] By L n+1 The subproblem of estimating the blur kernel k and normalizing it to get k n+1 ;

[0096]

[0097] Step 6: Update the iteration parameters and repeat steps 4 to 5 until the loop termination condition is met;

[0098] Step 7: Blur kernel k n+1 After normalization and non-negative constraints, the final blur kernel k is output;

[0099] Step 8: Use blur kernel k to perform non-blind deconvolution and output the final clear remote sensing image.

[0100] In step 1, a flexible norm l is adapted to deblur P-Q To sparsely represent the image gradient, establish an image blind deblurring model:

[0101]

[0102] Among them, B represents the blurred image, L represents the potential clear image, k represents the blur kernel, and '*' represents the convolution operator symbol. and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, α and β are regularization parameters, l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm combination (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

[0103] Step 4 is as follows:

[0104] The sub-problem of the latent sharp image L is expressed as:

[0105]

[0106] The semi-quadratic splitting method is used to introduce auxiliary variables t1, t2 and g to replace the above formula and Get the formula:

[0107]

[0108] Among them, B represents the blurred image, L represents the potential clear image, and k n represents the nth iteration blur kernel, and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, and α are regularization parameters, δ1, δ2, and δ3 are the corresponding penalty parameters. l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l PRepresents l0 and l p Norm array (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

[0109] Given a latent clear image L, the sub-problem model for solving t1 and t2 is:

[0110]

[0111] The solution is:

[0112]

[0113] Among them, RE represents the residual image BL*k n ; B represents a blurred image, L represents a potential clear image, k n represents the nth iteration blur kernel, and denote the first-order and second-order gradient operators respectively, ω1 and ω2 are regularization parameters, δ1 and δ2 are corresponding penalty parameters, and l e The norm represents the combination of the l0 and l1 norms (||·|| e =||·||0+||·||1).

[0114] Given a latent clear image L, the sub-problem model for solving g is:

[0115]

[0116] The solution is obtained by combining positive P and negative Q norms with flexible constraints using an improved generalized soft threshold algorithm, where the element value with the largest local gradient is used. Q The norm is solved, and the other element values ​​are minimized using l P Norm solution.

[0117] The blind deblurring algorithm for remote sensing images based on flexible norm constraints of the present invention is used to process real remote sensing images. Figure 3 and Figure 4 It can be seen that the two images have become clearer overall. Figure 3 Looking at the area in the upper left corner, the diffuse pixels at the edge of the building are significantly reduced, and the texture details are also enhanced; Figure 4The plane in the picture becomes clearer, the surrounding textures have a stronger visual texture, and the buildings below the picture have more layers.

[0118] The remote sensing image blind deblurring algorithm based on flexible norm constraint of the present invention combines the characteristics of gradient heavy tail distribution and local gradient maximum, and adopts l P norm constrains most gradient elements and implicitly through l Q The norm-constrained local gradient is embedded into the latent gradient image, and sparse induction is performed in the deblurring iterative process to achieve flexible constraints. An improved generalized soft threshold method is used to optimize the solution of the gradient subproblem, and finally an effective blur kernel solution is obtained through the Fourier operator.

[0119] The remote sensing image blind deblurring algorithm based on flexible norm constraints of the present invention can remove blur in the image while having a certain noise suppression effect, and can realize the acquisition of high-quality remote sensing image data, which is of great significance to the development of remote sensing technology.

[0120] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A remote sensing image blind deblurring algorithm based on flexible norm constraints, characterized in that: The following steps are involved: Step 1: Input the blurred remote sensing image and establish the image blind deblurring model; Step 2: Set the initial parameters of the image blind deblurring model and the penalty parameter δ i And the size of blur kernel k and the number of iterations N; Step 3: Use projection alternating iteration to estimate the potential clear image L and blur kernel k; Step 4: Calculate the sub-problem of the potential clear image L; Calculate subproblems t1, t2 and g to get the potential clear image L n+1 ; Step 5: Calculate the sub-problem of blur kernel k; By clear image L n+1 The subproblem of estimating blur kernel k and normalizing it to obtain blur kernel k n+1 ; Step 6: Update the iteration parameters and repeat steps 4 to 5 until the loop termination condition is met; Step 7: Blur kernel k n+1 After normalization and non-negative constraints, the final blur kernel k is output; Step 8: Use blur kernel k to perform non-blind deconvolution and output the final clear remote sensing image.

2. The remote sensing image blind deblurring algorithm based on flexible norm constraint according to claim 1, characterized in that: In step 1, a flexible norm l is adapted to deblur P-Q To sparsely represent the image gradient, establish an image blind deblurring model: Among them, B represents the blurred image, L represents the potential clear image, k represents the blur kernel, and '*' represents the convolution operator symbol. and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, α and β are regularization parameters, l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm combination (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components.

3. The remote sensing image blind deblurring algorithm based on flexible norm constraint according to claim 1, characterized in that: The initial parameters of the image blind deblurring model in step 2 include: ω1>0, ω2>0, α>0, β=2, L 0 =B, 0<p<1, q=-0.

5.

4. The remote sensing image blind deblurring algorithm based on flexible norm constraint according to claim 2, characterized in that: Step 4 is as follows: The sub-problem of the latent sharp image L is expressed as: The semi-quadratic splitting method is used to introduce auxiliary variables t1, t2 and g to replace the above formula and Get the formula: Among them, B represents the blurred image, L represents the potential clear image, and k n represents the nth iteration blur kernel, and represents the first-order gradient operator, represents the second-order gradient operator, ω1, ω2, and α are regularization parameters, δ1, δ2, and δ3 are the corresponding penalty parameters; l represents the norm ||·||, and its different subscripts represent different norms, l e represents the combination of l0 and l1 norms (||·|| e =||·||0+||·||1), l P Represents l0 and l p Norm array (||·|| P =||·||0+||·|| p ), l Q Represents l0 and l q The combination of norms (||·|| Q =||·||0+||·|| q ), l P-Q Indicates the use of norm l P and l Q Jointly constrain different image gradient components; Given a latent clear image L, the sub-problem model for solving t1 and t2 is: The solution is: Among them, RE represents the residual image BL*k n ; B represents a blurred image, L represents a potential clear image, k n represents the nth iteration blur kernel, and denote the first-order and second-order gradient operators, ω1 and ω2 are regularization parameters, δ1 and δ2 are the corresponding penalty parameters, and l e The norm represents the combination of the l0 and l1 norms (||·|| e =||·||0+||·||1); Given a latent clear image L, the sub-problem model for solving g is: The solution is obtained by combining positive P and negative Q norms with flexible constraints using an improved generalized soft threshold algorithm, where the element value with the largest local gradient is used. Q The norm is solved, and the other element values ​​are minimized using l P Norm solution.

5. The remote sensing image blind deblurring algorithm based on flexible norm constraint according to claim 4, characterized in that: In step 4, for fixed t1, t2 and g, the model of the latent clear image L is transformed into: By FFT solution: in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

6. The remote sensing image blind deblurring algorithm based on flexible norm constraint according to claim 2, characterized in that: Step 5 is specifically: the sub-problem of fuzzy kernel k, expressed as: Similarly, similar to the subproblem of finding the latent clear image L, the norm ||·|| e The solution of the problem requires the introduction of auxiliary variables t1 and t2, and then the sub-problem of the fuzzy kernel k can be solved by FFT: in, and They are the direct Fourier operator and the inverse Fourier operator respectively.

Citation Information

Patent Citations

  • DR blurred image blind deconvolution restoration method and system

    CN114092416A

  • Image processing method, device and equipment and readable storage medium

    CN116797466A

  • Light field refocusing image deblurring algorithm

    CN117237226A