Image blending deblurring method

Abstract

The application provides an image mixed degradation recovery method. The method comprises the following steps: obtaining an observation image, constructing a deconvolution model according to the observation image, correcting geometric unsharpness of the observation image, constructing a low-rank denoising model, and denoising the observation image, and obtaining an original image according to the deconvolution model and the low-rank denoising model. The recovery method provided by the application can well remove the mixed noise and spatial geometric unsharpness of a photographic image, and achieves a better recovery effect.

Description

Technical Field

[0001] This invention relates to the field of neutron photography technology, and in particular to a method for image hybridization and degradation restoration. Background Technology

[0002] Neutron radiography, a technique that utilizes neutron imaging, is widely used in the nondestructive testing industry. However, due to limitations in neutron sources and imaging conditions, the quality of images obtained through neutron radiography is often poor, especially with small neutron radiography devices, resulting in blurred images and high geometrical sharpness. Existing image restoration methods are only effective at removing single, specific types of noise. In practical applications of neutron radiography, the presence of mixed noise in the neutron image further hinders image restoration. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide an image hybridization degradation restoration method that overcomes or at least partially solves the above problems, aiming to solve the problem that existing image restoration methods cannot effectively remove hybrid noise, thereby improving the accuracy of non-destructive testing.

[0004] Specifically, the present invention provides the following technical solution:

[0005] A method for restoring image blending degradation includes:

[0006] Obtain the observed image;

[0007] Based on the observed image, a deconvolution model is constructed to correct the geometric unsharpness of the observed image; and a low-rank denoising model is constructed to denoise the observed image.

[0008] The original image is obtained based on the deconvolution model and the low-rank denoising model.

[0009] Optionally, obtaining the original image based on the deconvolution model and the low-rank denoising model includes:

[0010] The output of the deconvolution model is used as the input of the low-rank denoising model, and the output of the low-rank denoising model is used as the input of the deconvolution model to perform double-loop degradation correction.

[0011] Optionally, constructing a deconvolution model based on the observed image includes:

[0012] The average pixel value of the dark field of the observed image is obtained; and the average pixel value of the flat field of the observed image is obtained.

[0013] The observed image is normalized based on the average pixel value of the dark field and the average pixel value of the flat field.

[0014] A masking operation is performed on the normalized observed image to obtain an initial denoised image.

[0015] Optionally, constructing a deconvolution model based on the observed image further includes:

[0016] Obtain the geometric unsharpness of the observed image;

[0017] Based on the geometric non-sharpness, the point spread function is obtained;

[0018] Using the point spread function as the convolution kernel, an anticonvolution function is obtained.

[0019] Optionally, obtaining the point spread function based on the geometric unsharpness includes:

[0020] The two-dimensional Cauchy distribution function is selected as the point spread function;

[0021] The geometric unsharpness is approximated as the full width at half maximum (FWHM) of the two-dimensional Cauchy distribution function to obtain the dispersion parameter;

[0022] The two-dimensional Cauchy distribution function is obtained based on the dispersion parameter.

[0023] Optionally, the construction of the low-rank denoising model includes:

[0024] The observed image is simulated as a Poisson noise distribution of the original image to obtain the KL divergence function.

[0025] Optionally, the construction of the low-rank denoising model further includes:

[0026] Matrix partitioning, wherein the matrix partitioning includes:

[0027] Multiple rectangular blocks of equal size are selected on the observed image, wherein each pixel of the observed image is located in at least two of the rectangular blocks; a search window is set within each rectangular block, and multiple sub-blocks are set within the search window; and

[0028] Similar block matching, wherein similar block matching includes:

[0029] Search for blocks of other search windows within the area surrounding the block near the center of one search window and perform similarity calculations; based on the similarity calculation results, select multiple blocks as similar blocks and generate a similarity matrix from the multiple similar blocks in sequence;

[0030] Low-rank approximation of similar blocks, wherein the low-rank approximation of similar blocks includes:

[0031] A low-rank approximation solution function is obtained using the nonlocal mean algorithm;

[0032] The optimal solution is obtained based on the KL divergence function and the low-rank approximation solution function.

[0033] Optionally, the construction of the low-rank denoising model further includes image reconstruction, wherein the image reconstruction includes:

[0034] Pixel overlay, wherein the pixel overlay includes:

[0035] The similar block matrix is ​​superimposed according to the index of the block at the center corresponding to the search window that generates the similar block matrix;

[0036] Mean processing, the mean processing includes:

[0037] For each superimposed pixel, its grayscale is divided by the number of times it was superimposed when it was in the similar block matrix.

[0038] Optionally, obtaining the original image based on the deconvolution model and the low-rank denoising model includes:

[0039] The deconvolution function is processed using Laplacian sharpness as the evaluation criterion.

[0040] Optionally, obtaining the optimal solution based on the KL divergence function and the low-rank approximation solution function includes:

[0041] The KL divergence function and the low-rank approximation solution function are regularized.

[0042] The optimal solution is obtained based on the regularized KL divergence function and the regularized low-rank approximation solution function.

[0043] The image restoration method for mixed degradation provided in this application, tailored to the characteristics of photographic systems, corrects the geometric unsharpness of images caused by the photographic system by constructing a deconvolution model. It also denoises the image by constructing a low-rank denoising model, taking advantage of the self-similar information inherent in the observed image itself. The restoration method provided in this application can effectively remove mixed noise and spatial geometric unsharpness from photographic images, achieving good restoration results. Applying the restoration method provided in this application to mixed degradation images with low signal-to-noise ratio and low geometric unsharpness, especially to images from small neutron radiography systems, can achieve good image restoration and improve the accuracy of non-destructive testing in small neutron radiography systems.

[0044] Furthermore, the restoration method provided in this application forms a dual loop in the image correction process, where the deconvolution model and the low-rank denoising model mutually enhance the restoration effect to obtain the optimal original image. Moreover, during the correction process, no manual adjustment is required; the denoising and deblurring parameters can be adaptively adjusted, further improving the efficiency and effectiveness of image degradation restoration.

[0045] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0046] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0047] Figure 1 This is a flowchart of a restoration method according to an embodiment of the present invention;

[0048] Figure 2 This is a flowchart of a restoration method according to an embodiment of the present invention, which obtains the original image based on a deconvolution model and a low-rank denoising model.

[0049] Figure 3 This is a flowchart of the reconstruction method for constructing a deconvolution model according to an embodiment of the present invention;

[0050] Figure 4 This is a flowchart of the reconstruction method for constructing a deconvolution model according to an embodiment of the present invention;

[0051] Figure 5 This is a flowchart of the reconstruction method for constructing a low-rank denoising model according to an embodiment of the present invention;

[0052] Figure 6 This is a flowchart of a restoration method according to an embodiment of the present invention, which obtains the original image based on a deconvolution model and a low-rank denoising model.

[0053] Figure 7 This is a schematic diagram of matrix partitioning in a restoration method according to an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of similar block matching in a restoration method according to an embodiment of the present invention;

[0055] Figure 9 This is a flowchart of similar block matching in a restoration method according to an embodiment of the present invention. Detailed Implementation

[0056] The following reference Figures 1 to 9 The following is a flowchart describing the restoration method according to an embodiment of the present invention. In this description, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature, that is, include one or more of that feature. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. When a feature "includes or contains" one or more of the features it encompasses, unless otherwise specifically described, this indicates that other features are not excluded and may be further included.

[0057] Unless otherwise expressly specified and limited, the terms "set up," "install," "connect," "link," "fix," and "couple" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art should be able to understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0058] Furthermore, in the description of this embodiment, "above" or "below" the second feature can include direct contact between the first and second features, or it can include contact between the first and second features through another feature between them. That is, in the description of this embodiment, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "below" of the second feature can mean the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0059] In the description of this embodiment, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0060] Figure 1This is a flowchart of an image blending degradation restoration method according to an embodiment of the present invention, which is described below in conjunction with... Figure 1-9 This paper provides a detailed description of the restoration method described in this application.

[0061] like Figure 1 As shown, a restoration method according to an embodiment of the present invention includes:

[0062] S100 acquires observation image 1;

[0063] S200 constructs a deconvolution model based on observed image 1 to correct the geometric unsharpness of observed image 1;

[0064] S300 constructs a low-rank denoising model to denoise the observed image 1;

[0065] S400 obtains the original image based on the deconvolution model and the low-rank denoising model.

[0066] In this embodiment, for observation image 1 obtained by a photographic system, which simultaneously exhibits mixed noise and geometric unsharpness, such as observation image 1 obtained by a small neutron radiography system, this invention proposes a restoration method. Specifically, on one hand, a deconvolution model is constructed to correct the geometric unsharpness of observation image 1. The deconvolution model can effectively correct geometric unsharpness based on the structure or parameters of the photographic system, improving image sharpness. On the other hand, a low-rank denoising model is constructed to denoise observation image 1. The low-rank denoising model can fully utilize the redundant information of the image itself, i.e., the non-local self-similarity information of the image, for noise removal. Furthermore, in this embodiment, both the deconvolution model and the low-rank denoising model are combined to restore observation image 1, which can significantly improve the signal-to-noise ratio and geometric unsharpness of the image. In particular, when the restoration method of this embodiment is applied to images from a small neutron radiography system, it can achieve better image restoration and improve the accuracy of non-destructive testing of the small neutron radiography system.

[0067] In some embodiments of the restoration method of the present invention, such as Figure 2 As shown, the methods for obtaining the original image based on the deconvolution model and the low-rank denoising model include:

[0068] S441 uses the output of the deconvolution model as the input of the low-rank denoising model;

[0069] S442 simultaneously uses the output of the low-rank denoising model as the input of the deconvolution model for double-loop degradation correction.

[0070] In this embodiment, a loop is established between the deconvolution model and the low-rank denoising model. The output of the deconvolution model is used as the input of the low-rank denoising model, and vice versa. This fully utilizes the advantages of both models, and through multiple iterations, they mutually improve the restoration effect to obtain the optimal original image. Furthermore, by establishing a loop, no manual adjustment is needed during the image restoration process; the denoising and deblurring parameters can be adaptively adjusted, further improving the efficiency and effectiveness of image blending degradation restoration.

[0071] In some embodiments of the restoration method of the present invention, such as Figure 3 As shown in Figure 1, the methods for constructing a deconvolution model include:

[0072] S201 obtains the average pixel value of the dark field of observed image 1;

[0073] S202 obtains the average pixel value of the flat field of observed image 1;

[0074] S203 normalizes the observed image 1 based on the average pixel value of the dark field and the average pixel value of the flat field;

[0075] S204 performs a masking operation based on the normalized observation image 1 to obtain the initial denoised image.

[0076] The observation image 1, obtained directly by the photographic system, suffers from inconsistent grayscale values ​​due to source intensity instability during the photographic process, such as fluctuations in light source intensity, neutron source intensity, and ultrasonic source intensity. In this embodiment, preprocessing is required to correct the inconsistency in grayscale values ​​caused by source intensity fluctuations. Specifically, the average pixel value of all dark-field images is calculated. Calculate the pixel average of all flat field images The observed image 1 is normalized based on the average pixel value of the dark field image and the average pixel value of the flat field image. After normalization, observed image 1 still contains pixels with grayscale anomalies, which need to be masked. Specifically, a mask matrix is ​​defined for each normalized observed image 1:

[0077]

[0078] In the above formula, g j Let j represent the j-th observed image 1.

[0079] The normalized observation image 1 is initialized and denoised using the mask matrix as follows:

[0080]

[0081] In the above formula, fj Let represent the initial denoised image corresponding to the j-th observation image 1.

[0082] The initial denoised image can be considered as the observed image obtained by a photographic system with a stable source strength 1.

[0083] In some embodiments of the restoration method of the present invention, such as Figure 4 As shown in Figure 1, methods for constructing deconvolution models also include:

[0084] S210 obtains the geometric unsharpness of the observed image 1;

[0085] S220 obtains the point spread function based on geometric non-sharpness;

[0086] S230 uses a point spread function as the convolution kernel to obtain an anti-convolution function.

[0087] In this embodiment, the observed image 1 can be the observed image 1 directly obtained by the imaging system, or it can be the observed image 1 after normalization and initial denoising processing. The deconvolution model is a non-blind deconvolution model. For some imaging systems, the geometric unsharpness of the imaging system can be directly obtained through the structure or parameters of the imaging system itself. For example, in a neutron imaging system, the geometric unsharpness can be calculated by applying the principle of similar triangles based on the geometric parameters of the imaging system structure. Then, based on the geometric unsharpness, it is approximated as the full width at half maximum (FWHM) of the point spread function. The dispersion parameter of the point spread function is obtained through the FWHM, thus obtaining the point spread function. Through the point spread function, a deconvolution function can be generated, thereby constructing a deconvolution model to correct the geometric unsharpness of the observed image 1.

[0088] In some embodiments of the restoration method of the present invention, the method for obtaining the point spread function based on geometric unsharpness includes:

[0089] S221 selects the two-dimensional Cauchy distribution function as the point spread function;

[0090] S222 approximates the geometric unsharpness as the full width at half maximum of a two-dimensional Cauchy distribution function to obtain the dispersion parameter;

[0091] S223 obtains the two-dimensional Cauchy distribution function based on the dispersion parameter.

[0092] There are many distribution models for the point spread function. In this embodiment, the two-dimensional Cauchy distribution function is chosen as the point spread function. The two-dimensional Cauchy distribution function can better reflect the imaging characteristics of some imaging systems, such as those of small neutron imaging systems. The dispersion parameter of the two-dimensional Cauchy distribution function is an unknown quantity and needs to be solved. The solution method is to approximate the geometric unsharpness as the full width at half maximum (FWHM) of the two-dimensional Cauchy distribution function, and then the dispersion parameter of the two-dimensional Cauchy distribution function can be calculated. Substituting the dispersion parameter into the two-dimensional Cauchy distribution function, the two-dimensional Cauchy distribution function can be obtained. Specifically, the two-dimensional Cauchy distribution function can be expressed as:

[0093]

[0094] In the above formula, ξ represents the dispersion parameter.

[0095] In some embodiments of the restoration method of the present invention, the method for constructing a low-rank denoising model includes:

[0096] S310 simulates the Poisson noise distribution of the observed image 1 as the original image and obtains the KL divergence function.

[0097] When constructing the low-rank denoising model, it is necessary to first analyze the noise in the observed image 1. The noise in some imaging systems is mainly Poisson noise, such as in small neutron radiography systems. Therefore, this embodiment focuses primarily on the removal of Poisson noise. In constructing the low-rank denoising model, the observed image 1 is first simulated as the Poisson noise distribution of the original image to obtain the KL divergence function. The KL divergence function is used to quantify the difference between the original image and the observed image 1 after passing through a preset Poisson noise distribution with preset parameters. When a certain preset parameter is set, if the image obtained after simulating the Poisson noise distribution of the preset original image is less different from the observed image 1, it can be considered that the preset original image under this preset parameter is closer to the real original image, meaning the restoration effect is good. Specifically, simulating the Poisson noise distribution of the original image in the observed image 1 yields the following function:

[0098] g i =Poisson(u i +b i )

[0099] In the above formula, g i Let u represent the i-th observed image 1. i Let b represent the i-th original image. i This represents the background pixel value.

[0100] Using the above equation as parameters, the KL divergence function can be obtained:

[0101]

[0102] In the above formula, D KL This represents the KL divergence function.

[0103] In some embodiments of the restoration method of the present invention, such as Figure 5 As shown, methods for constructing low-rank denoising models also include:

[0104] S321 matrix is ​​divided into 5 blocks, such as Figure 7-8 As shown, matrix block 5 includes:

[0105] Select multiple rectangular blocks 2 of equal size on the observed image 1, wherein each pixel of the observed image 1 is located in at least two rectangular blocks 2; set a search window 3 within each rectangular block 2, and set multiple sub-blocks 5 within the search window 3. For example... Figure 7 As shown, Figure 7 The same observed image 1 was divided into nine matrix blocks 5, resulting in nine rectangular blocks 2. These nine rectangular blocks 2 have overlapping portions.

[0106] S322 similar block matching, such as Figure 8-9 As shown, similar block matching includes:

[0107] Search for blocks 5 of other search windows 3 within the range of the block 4 near the center of a search window 3, and perform similarity calculations; based on the similarity calculation results, select multiple blocks 5 as similar blocks, and generate a similarity matrix from the multiple similar blocks in sequence.

[0108] S323 Low-rank approximation of similar blocks, including:

[0109] A low-rank approximation solution function is obtained using the nonlocal mean algorithm;

[0110] The optimal solution is obtained by using the KL divergence function and the low-rank approximation solution function.

[0111] In this embodiment, based on the nonlocal mean algorithm and the KL divergence function, the following expression can be obtained:

[0112]

[0113] In the above formula, u i Let g represent the i-th original image. i Let f represent the i-th observed image 1, f represent the denoised image, and R represent the denoised image. j f represents the matrix composed of corresponding similar blocks in the observed image, F j ω represents a low-rank matrix consisting of corresponding similar blocks in the observed image. j This is the weight matrix.

[0114] In some embodiments of the restoration method of the present invention, such as Figure 5As shown, methods for constructing low-rank denoising models also include:

[0115] S324 image reconstruction, image reconstruction includes:

[0116] Pixel overlay, which includes:

[0117] The similar block matrices are superimposed according to the index of block 4, which is the center of the search window that generates the similar block matrix.

[0118] Mean processing, which includes:

[0119] For each superimposed pixel, divide its grayscale by the number of times it was superimposed when it was in a similar block matrix.

[0120] During the pixel overlay process, since the pixels in block 5 overlap and each block is selected a different number of times, each pixel in the overlay image needs to be divided by the number of times the pixel itself has been selected, which is equivalent to the mean approximation of each pixel.

[0121] In some embodiments of the restoration method of the present invention, the method for obtaining the original image based on the deconvolution model and the low-rank denoising model includes:

[0122] S410 uses Laplacian sharpness as the evaluation criterion to process the deconvolution function.

[0123] Specifically, the deconvolution function based on the two-dimensional Cauchy distribution function is first expressed as:

[0124] f*Cauchy(x, y) T

[0125] In the above formula, f represents the denoised image, and * represents the deconvolution operation.

[0126] In deconvolution models, some noise may be amplified, affecting image restoration. Therefore, it is necessary to process the deconvolution function using Laplacian sharpness as an evaluation criterion. The processing expression can be represented as:

[0127] Laplace(f*Cauchy(x, y) T )

[0128] In the above formula, Laplace represents the Laplace operator. In this embodiment, for a small neutron radiography system, the Laplace operator is provided by the following formula:

[0129]

[0130] After processing with the deconvolution function, the correction of the geometric unsharpness of observed image 1 is transformed into solving the following equation:

[0131] min(Laplace(f*Cauchy T ))

[0132] In the above formula, f represents the denoised image, and * represents the deconvolution operation. The sharpness evaluation function is provided by the following formula:

[0133]

[0134] In the above formula, L(i,j) represents the result of convolving the image with the Laplace operator.

[0135] In some embodiments of the restoration method of the present invention, the method for obtaining the optimal solution based on the KL divergence function and the low-rank approximation solution function includes:

[0136] S420 performs regularization on the KL divergence function and the low-rank approximation solution function;

[0137] S430 obtains the optimal solution based on the regularized KL divergence function and the regularized low-rank approximation solution function.

[0138] In this embodiment, in order to find the optimal solution under the same standard, the KL divergence function and the low-rank approximation solution function need to be regularized. The regularized result is as follows:

[0139]

[0140] In the above formula, α represents the regularization parameter, and η represents the regularization parameter.

[0141] The denoising correction for observed image 1 is transformed into solving the following equation:

[0142]

[0143] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A method for restoring image degradation caused by image blending, characterized in that, include: Obtain the observed image; Based on the observed image, a deconvolution model is constructed to correct the geometric unsharpness of the observed image; Furthermore, a low-rank denoising model is constructed to denoise the observed image; Based on the deconvolution model and the low-rank denoising model, at least the output of the deconvolution model is used as the input of the low-rank denoising model, and the output of the low-rank denoising model is used as the input of the deconvolution model, so as to perform double-loop degradation correction based on multiple iterations to obtain the original image. The construction of a deconvolution model based on the observed image includes: The average pixel value of the dark field of the observed image is obtained; and the average pixel value of the flat field of the observed image is obtained. The observed image is normalized based on the average pixel value of the dark field and the average pixel value of the flat field. Based on the normalized observed image, a masking operation is performed to obtain an initial denoised image, which can be regarded as the observed image obtained by a photographic system with a stable source strength. The construction of the low-rank denoising model includes: The observed image is simulated as a Poisson noise distribution of the original image to obtain the KL divergence function.

2. The restoration method according to claim 1, characterized in that, The method of constructing a deconvolution model based on the observed image further includes: Obtain the geometric unsharpness of the observed image; Based on the geometric non-sharpness, the point spread function is obtained; Using the point spread function as the convolution kernel, an anticonvolution function is obtained.

3. The restoration method according to claim 2, characterized in that, The method of obtaining the point spread function based on the geometric non-sharpness includes: The two-dimensional Cauchy distribution function is selected as the point spread function; The geometric unsharpness is approximated as the full width at half maximum (FWHM) of the two-dimensional Cauchy distribution function to obtain the dispersion parameter; The two-dimensional Cauchy distribution function is obtained based on the dispersion parameter.

4. The restoration method according to claim 3, characterized in that, The construction of the low-rank denoising model also includes: Matrix partitioning, wherein the matrix partitioning includes: Multiple rectangular blocks of equal size are selected on the observed image, wherein each pixel of the observed image is located in at least two of the rectangular blocks; a search window is set within each rectangular block, and multiple sub-blocks are set within the search window; and Similar block matching, wherein similar block matching includes: Search for blocks of other search windows within the area surrounding the block near the center of one search window and perform similarity calculations; based on the similarity calculation results, select multiple blocks as similar blocks and generate a similar block matrix in sequence from the multiple similar blocks; Low-rank approximation of similar blocks, wherein the low-rank approximation of similar blocks includes: A low-rank approximation solution function is obtained using the nonlocal mean algorithm; The optimal solution is obtained based on the KL divergence function and the low-rank approximation solution function.

5. The restoration method according to claim 4, characterized in that, The construction of the low-rank denoising model also includes image reconstruction, wherein the image reconstruction includes: Pixel overlay, wherein the pixel overlay includes: The similar block matrix is ​​superimposed according to the index of the block centered in the search window that generates the similar block matrix; Mean processing, the mean processing includes: For each superimposed pixel, its grayscale is divided by the number of times it was superimposed when it was in the similar block matrix.

6. The restoration method according to claim 2, characterized in that, The process of obtaining the original image based on the deconvolution model and the low-rank denoising model includes: The deconvolution function is processed using Laplacian sharpness as the evaluation criterion.

7. The restoration method according to claim 5, characterized in that, The process of obtaining the optimal solution based on the KL divergence function and the low-rank approximation solution function includes: The KL divergence function and the low-rank approximation solution function are regularized. The optimal solution is obtained based on the regularized KL divergence function and the regularized low-rank approximation solution function.

Images