WO2023115801A1

WO2023115801A1 - Point-by-point correction and restoration method and system for large field-of-view degraded image having aero-optical effect

Info

Publication number: WO2023115801A1
Application number: PCT/CN2022/093807
Authority: WO
Inventors: 洪汉玉; 左志潮; 张天序; 时愈; 张耀宗; 吴锦梦; 李琼
Original assignee: 武汉工程大学
Priority date: 2021-12-21
Filing date: 2022-05-19
Publication date: 2023-06-29
Also published as: CN113962908A; CN113962908B

Abstract

Disclosed in the present invention is a point-by-point correction and restoration method for a large field-of-view degraded image having an aero-optical effect. The method comprises: calculating the gradient of an input degraded image, selecting a plurality of large gradient areas, and calculating a blur kernel of each local area; calculating the distances to center points of two nearest local areas point by point, and performing inverse distance weighted interpolation calculation on a blur kernel of each point in the whole image according to the two distances of each pixel point to obtain initial values of the blur kernels of the points in the whole image to form an initial blur kernel matrix; establishing a space-variant degradation model according to the initial blur kernel matrix, and adding a non-negativity constraint regularization item and an adaptive anisotropy variable coefficient-based sparsity constraint regularization item, so that a target image and the blur kernel of each point have non-negativity and spatial adaptability; and solving the space-variant degradation model to obtain the blur kernel of each point and the grayscale value of each point to realize point-by-point correction, and finally outputting a space-variant degradation restored image. According to the present invention, a large field-of-view space-variant degraded image having an aero-optical effect can be corrected and restored.

Description

Method and system for point-by-point correction and restoration of aero-optical effect large field of view degraded image

【Technical field】

The invention relates to the fields of aerospace image processing and aero-optical effect correction, in particular to a method and system for point-by-point correction and restoration of aero-optical effect large-field-of-view degraded images.

【Background technique】

In recent years, my country's high-speed aircraft space detection technology has developed rapidly, and high-speed aircraft equipped with optical imaging systems have been widely used in surveillance, investigation and detection tasks. However, due to the harsh imaging environment of high-speed aircraft during high-speed flight, a complex high-speed flow field is formed between the optical hood and the incoming flow, which produces aero-optical effects, degrades the image quality of the target, and greatly reduces the signal-to-noise ratio and signal-to-clutter ratio. Modern aero-optical effect degradation image restoration technology has not only focused on the restoration, but also has strict requirements on the restoration quality. In the fields of star surface detection and aircraft orbit survey, it is often necessary to obtain high-quality clear images. The clearer the image, the better the information analysis. Therefore, how to counteract or alleviate the adverse effects of aero-optical effects under high-speed flight conditions has become an urgent problem to be solved.

The existing aero-optical effect degraded image restoration methods mainly include: 1. The flow field control method can reduce the influence of the aero-optic effect, but it is far from eliminating this effect; 2. The adaptive optics method can complete the partial restoration of the wavefront difference , Digital image restoration methods can solve the restoration of space-invariant aero-optical effect degraded images, but it is difficult to realize the restoration of aero-optic effect large field of view space-variant degraded images. In fact, the imaging environment of aerospace images has many interferences and is unavoidable. The actual imaging blur kernel is generally space-varying. The existing algorithm will always have deviations in the blur kernel it solves, which will sometimes reduce the quality of image restoration and take time. too long. Therefore, it is necessary to design a dedicated aero-optical effect degradation image restoration algorithm for these application requirements.

【Content of invention】

The technical problem to be solved by the present invention is to provide a method for point-by-point correction and restoration of an aero-optical effect large field of view degraded image for the defects in the prior art.

The technical solution adopted by the present invention to solve its technical problems is:

A method for point-by-point restoration of an aero-optical effect large field of view space-varying degraded image is provided, comprising the following steps:

S1. Calculate the gradient of the input degraded image, select multiple large gradient regions, extract multiple local regions according to the distribution of large gradients on the degraded image, and calculate the blur kernel of each local region;

S2. Calculate the distance to the center points of the two nearest local areas point by point, and perform inverse distance weighted interpolation calculation on the blur kernel of each point in the whole image according to the two distances of each pixel point, and obtain the initial value of the blur kernel of each point in the whole image , forming the initial fuzzy kernel matrix;

S3. Establish a space-varying degradation model based on the initial blur kernel matrix, and add a non-negativity constraint regularization item determined by the gray value of the target image and the initial blur kernel of each point and based on the two initial blur kernels of the target image and each point The sparsity-constrained regularization term of the adaptive anisotropic variable coefficient determined by the gradient value makes the target image and each point blur kernel non-negative and spatially adaptive;

S4. Solve the spatial variation degradation model, obtain the blur kernel of each point and the gray value of each point to realize point-by-point correction, and finally output the restoration image of spatial variation degradation.

Following the above technical solution, the specific method of step S1 is:

S11. Using a multi-scale morphological gradient operator to obtain the gradient of the degraded image;

S12. Use the gradient usefulness index to filter out small structural gradient regions, select large gradient local regions whose length and width directions are greater than a certain value on the filtered gradient image, and then extract multiple local regions according to the distribution of large gradients on the degraded image area;

S13. Estimate the extracted blur kernel of each local region by using a non-negative least squares criterion algorithm based on spatial correlation constraints.

Following the above technical solution, the specific method of step S2 is:

S21. Taking the blur kernel of each local area as the initial value of the blur kernel of each point in the local area;

S22. Calculate the Euclidean distance from each pixel point to the center points of all local regions;

S23. Comparing and obtaining the Euclidean distance from each pixel point to the center points of the two nearest local areas, and regarding the distance as the corresponding weight coefficient, performing inverse distance weighted interpolation calculation on the blur kernel of the corresponding pixel point according to the weight coefficient, and obtaining the corresponding The initial value of the blur kernel for pixels.

Following the above technical solution, in step S4, the Bregman multivariate separation algorithm and the lag fixed-point iterative method are used to solve the fuzzy kernel and the gray value of each point.

The present invention also provides a point-by-point restoration system for large field of view space-varying degraded images with aero-optical effects, including:

The local area screening module is used to calculate the gradient of the input degraded image, select multiple large gradient areas, extract multiple local areas according to the distribution of large gradients on the degraded image, and calculate the blur kernel of each local area;

The initial value calculation module of the blur kernel is used to calculate the distance to the center points of the two nearest local areas point by point, and perform inverse distance weighted interpolation calculation on the blur kernel of each point in the whole image according to the two distances of each pixel point, and obtain the full The initial value of the fuzzy kernel of each point in the graph constitutes the initial fuzzy kernel matrix;

The spatial variation degradation model building block is used to establish the spatial variation degradation model according to the initial fuzzy kernel matrix, and add non-negativity constraint regularization items and sparsity constraint regularization items based on adaptive anisotropy variable coefficients to make the target image and each The point blur kernel is non-negative and spatially adaptive;

The model solving module is used to solve the spatial variation degradation model, obtain the blur kernel of each point and the gray value of each point to realize point-by-point correction, and finally output the restoration image of spatial variation degradation.

Following the above technical solution, the local area screening module specifically includes:

The gradient calculation sub-module is used to obtain the gradient of the degraded image by using a multi-scale morphological gradient operator;

The gradient filtering sub-module is used to filter out the small structural gradient area by using the gradient usefulness index, and select a large gradient local area whose length and width are greater than a certain value on the filtered gradient image, and then according to the large gradient on the degraded image The distribution of extracts multiple local regions;

The regional blur kernel estimation submodule is used for estimating the extracted blur kernel of each local region by using a non-negative least square criterion algorithm based on spatial correlation constraints.

Following the above technical solution, the fuzzy kernel initial value calculation module specifically includes:

The sub-module for determining the fuzzy kernel of each point is used to regard the fuzzy kernel of each local area as the initial value of the fuzzy kernel of each point in the local area;

The distance calculation sub-module is used to calculate the Euclidean distance from each pixel point to the center points of all local areas;

The weighted blur kernel calculation sub-module is used to compare the Euclidean distance between each pixel point and the center points of the two nearest local areas, and regard the distance as the corresponding weight coefficient, and reverse the blur kernel of the corresponding pixel point according to the weight coefficient Calculate the distance weighted interpolation to obtain the initial value of the blur kernel of the corresponding pixel.

Following the above technical solution, the model solution module specifically uses the Bregman multivariate separation solution algorithm and the hysteresis fixed-point iteration method to solve the fuzzy kernel and gray value of each point.

The present invention also provides a computer storage medium, which can be executed by a processor, and has a computer program stored therein, and the computer program executes the point-by-point restoration method for an aero-optical effect large field of view space-varying degraded image described in the above technical solution.

The present invention provides a method for point-by-point correction and restoration of aero-optical effect large-field-of-view degraded images, which includes the following steps:

The beneficial effects produced by the present invention are: the point-by-point restoration method of the aero-optical effect large field of view degraded image of the present invention, under the condition of large field of view, the initial fuzzy kernel matrix is formed by interpolation to meet the law of continuous change of the fuzzy kernel at each point, and further to The space-varying degradation model adds non-negative and sparse regularization items based on adaptive anisotropic variable coefficients, which have obvious effects in suppressing noise and retaining edge features. Iteratively solve the blur kernel and gray value of each point to obtain a clear image. Compared with the empty invariant restoration method, the restoration effect is more accurate.

【Description of drawings】

The present invention will be further described below in conjunction with accompanying drawing and embodiment, in the accompanying drawing:

Fig. 1 is a flowchart of a method for point-by-point correction and restoration of aero-optical effect large field of view degraded image according to an embodiment of the present invention;

Fig. 2 is a schematic diagram of the relationship between image coordinates and blur kernel positions according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a regularization scheme according to an embodiment of the present invention;

Fig. 4 is the embodiment of the present invention

F-curve image;

Fig. 5 is the embodiment of the present invention

F-curve image;

Fig. 6 is an image of the original aero-optical effect degradation of the embodiment of the present invention;

Fig. 7 is the large gradient distribution image of the embodiment of the present invention;

Fig. 8 is a local area blur kernel image according to an embodiment of the present invention;

Fig. 9 is a representative point (1, 1) interpolation estimation space-variant blur kernel image of the embodiment of the present invention;

Fig. 10 is a comparison chart of the space-invariant restoration results and the space-variant restoration results of the aero-optical effect degraded image according to the embodiment of the present invention.

【Detailed ways】

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

According to the embodiment of the present invention, the point-by-point restoration method of the aero-optical effect, large field of view, and space-varying degraded image can be realized by using the C++ program of VC6.0 and the Matlab platform, the operating environment is Windows 10, and the processor is Intel Core i7.

As shown in Figure 1, the point-by-point restoration method of the aero-optical effect large field of view space-varying degraded image according to the embodiment of the present invention includes the following steps:

Include the following steps:

S1. Input a degraded image, calculate the gradient of the degraded image, select multiple regions with large gradients, extract multiple local regions according to the distribution of large gradients on the degraded image, and calculate the blur kernel of each local region;

Further, the specific method of step S1 is:

S13. Estimate the blur kernel of each extracted local region by using a non-negative least squares criterion algorithm based on spatial correlation constraints.

The specific method of step S2 is:

In a preferred embodiment of the present invention, in step S1, the degraded image is recorded as g(x, y), as shown in FIG. 6, and the size of the image is 600×400. First, the gradient of the degraded image is obtained by using the multi-scale morphological gradient operator:

In the formula

is the gradient,

and

represent expansion and erosion operations respectively, (x, y) are the coordinates of each point in the image, x∈(0,600), y∈(0,400), d is the scale, the value is 3, R is the radius of the morphological structure element, S _R is Morphological structuring element side length.

Use the gradient usefulness metric γ to filter out small structured gradient regions:

In the formula, ||·|| ₂ is the l ₂ norm, N _r (x, y) is the rectangular neighborhood with the length and width r=1, 2, 3 centered on the pixel point (x, y). (x', y') is the pixel index in the neighborhood, x'=x±a, y'=y±b, a, b=0,1,2,3;

can be eliminated

The gradient zigzag small peaks in , that is, the small structural gradients in the image gradient.

is the sum of the absolute values of the gradients in N _r (x,y). Calculate γ(x,y) and set a threshold τ. The value of γ(x,y) is less than τ, indicating that the neighborhood of (x,y) is a small structural gradient area, which is filtered out. In the example of the present invention, the constant c takes a value of 0.1, and the threshold τ is set to 0.5.

After filtering out the small structural gradient area, select a large gradient area whose length and width are both greater than 10 on the gradient image, as shown in Figure 7. If the area cannot be selected, the size of the selected area can be appropriately changed, or the image is already a clear image, and there is no need to restore it. Finally, three local areas are extracted according to the distribution of large gradients on the degraded image, and the size of the blur kernel h(m,n) is set to 21×21 (usually the range is 3 to 27 and is an odd number), where m≤21 , n≤21, using the existing basis to use the non-negative least squares criterion algorithm based on spatial correlation constraints to extract the fuzzy kernels h _k of the three local regions:

Using iterative minimization algorithm to solve h _k , let the objective function derivative be zero solution:

Where ||·|| ₁ is the l ₁ norm, A is a blurred point corresponding to an area in the original image, b is a one-dimensional column vector composed of pixels from the degraded image, h is a one-dimensional column vector form the fuzzy kernel. h _i , h _w represent the specific values of the i-th component and the adjacent w-th component in the blur kernel, and the solved blur kernels and 3D displays of three local areas are shown in Figure 8.

In step S2, the blur kernel of each local area can be regarded as the initial value of the blur kernel of each point in the local area. Since only small field of view images can be regarded as space-invariant, regarding the problem of space-variant deblurring of large field of view, the present invention is based on the law that the change of the space-variant blur kernel of the aero-optical effect has continuity in the large field of view, and the interpolation estimation The initial value of each point fuzzy kernel. Take the point (x, y) as an example, compare the Euclidean distance between the point (x, y) and the center point (x _k , y _k ) of the k-th blur kernel estimation area and store it in the set D, k=1, 2, 3 .

Comparing the size of each element in D, the Euclidean distances from the point (x, y) to the centers of the two nearest blur kernel estimation regions are recorded as D ₁ and D ₂ , as shown in Figure 2(a). The present invention regards the distance as a weight coefficient, and performs an inverse distance weighted interpolation operation on the point fuzzy kernel. Calculated as follows:

In the formula, h ₁ (m,n) is the blur kernel of the center point (x ₁ ,y ₁ ) of the local area whose distance to the point (x,y) is D ₁ , and h ₂ (m,n) is the blur kernel to the point (x,y) ,y) is the blur kernel of the central point (x ₂ ,y ₂ ) of the local area of D ₂ , h _(x,y) (m,n) is the blur kernel formed by interpolation at the point (x,y), such as Figure 2(b) shows. By analogy, the initial value of the blur kernel of each point in the whole image is obtained. The representative point (1,1) interpolation blur kernel and 3D display are shown in Figure 9.

In step S3, a space-varying degradation model is established, and a sparsity-constrained regularization term and a non-negativity-constrained regularization term based on adaptive anisotropy regularization variable coefficients are added to constrain the blur kernel of each point of the target image and the degraded image. The process of constructing the model is as follows:

The degraded image formation process can be modeled as the following form:

where g(x,y) is the degraded image, h(m,n) is the blur kernel, f(x,y) is the clear image, and n(x,y) is the noise.

Under the condition of large field of view, the present invention pays attention to the fact that each point of the degraded image has different blurs, and each point of g(x,y) {g(0,0), g(0,1), g(0,2) ,...,g(M-1,N-1)} are piled up with column vector g, then the space-varying degradation process can be expressed in matrix-vector form as:

g＝Hf+n (7)

Among them, g, f, and n are the one-dimensional stacked column vectors of g(x,y), f(x,y), and n(x,y) respectively, H is called the fuzzy kernel matrix, and H is from left to right, from From top to bottom is the blur kernel of each pixel (x, y). When the image is empty and unblurred, H corresponds to the same blur kernel from left to right and from top to bottom, that is, h(m,n) of each point is fixed. When the image is space-variable and blurred, the blur kernel corresponding to different pixel points from left to right and from top to bottom in H, that is, h(m,n) of each point is changed.

Since the fuzzy kernel matrix H of each pixel point has a certain calculation error in the interpolation formation process, the accuracy needs to be further improved. In order to solve the error problem, let

is the known fuzzy kernel matrix containing errors, that is:

in

From left to right, from top to bottom, the fuzzy kernel h _{(x, y) (m, n) formed by the interpolation of each pixel point (x, y)} in turn, δ _H is the error, and the space-varying degradation model is established as follows:

Further adding a reasonable regularization term to formula (9) makes the solution of the model approach the real solution. Most of the traditional space-varying degradation models use fixed regularization parameters to adjust the smoothness of each pixel in the four directions α ₁ , α ₂ , α ₃ , and α ₄ are the same and cannot achieve good results. Due to the directionality of the gradient, when the magnitude of the gradient in different directions of the point is different, the smoothing should be different. Therefore, the size of the regularization parameter should be related to the gradient value of each point. The present invention aims at correcting the degraded image of a large field of view, and proposes different The anisotropic regularization parameter of , respectively adjusts the four directions of each point differently, so as to achieve different smoothing in different directions of each point, as shown in Figure 3.

In fact, there are many different edge features in the observed aero-optical effect large field of view image. The pixel gradient value in the target image and the background area in the target image is small, and the gray value is relatively close, so a large degree of regularization is required. To smooth this part of the area and suppress the influence of noise. At the same time, in order to notice the difference in the gray value of the pixel at the boundary between the target and the background, the regularization of this part of the area must be weakened to maintain the gradient difference. Such regularization parameters should decrease monotonically and sharply in mathematical analysis. Functional form, in order to simplify the calculation, the present invention selects the regularization function of the following form

As the sparsity of the target image, the coefficient of the regularization term is constrained to achieve the purpose of protecting edge features while suppressing noise.

In the formula,

is the gradient value of each point of the target image f(x,y), and the initial value is obtained in step S1

Gradient value at each point. C ₁ is the regularization constant coefficient of the target image, generally set to 1, when

When the value is too large, properly increase the value of C ₁ to increase the smoothness, and vice versa. When C ₁ =1,

The range is between [0, 1], and its function curve is shown in Figure 4. The index n takes a value between 0 and 5, and the value of n is determined by the attenuation. The faster the attenuation, the larger the value of n, and vice versa. Small. When n=0,

No directionality, the same direction, when n≠0, obviously can be determined by the gradient value

To adjust, so that the target image has a certain spatial adaptability.

The fuzzy kernel image has a Gaussian-like shape, not steep, the rise and fall of the gradient of each point is continuous, and the gradient change is mainly reflected in the overall attenuation, and the transition between adjacent points is slow. The present invention integrates this prior knowledge into the model. In order to protect the peak value of the blur kernel in the large gradient area, the regularization parameter should take a smaller value, and in the small gradient area, in order to suppress noise, it should be larger. For the convenience of algorithm implementation, first calculate the gradient of the blur kernel at each point

Choose a regularization function that varies more slowly

Constrains regularization term coefficients as blur kernel sparsity to preserve blur kernel peaks.

In the formula,

The initial value is the blur kernel gradient

Gradient value of each point in , C ₂ is the regularization constant coefficient of the fuzzy kernel, generally set to 1, when

When the value is too large, appropriately increase the value of C ₂ to increase the smoothness, and vice versa. When C ₂ =1,

Its symmetrical monotone descending function curve is shown in Fig. 5 . The parameter ξ is the smoothing control coefficient, and ξ takes 1 under normal circumstances. When ξ is not 1, it is determined by the attenuation. When the attenuation is faster, the value of ξ is between 0 and 1, which protects a large gradient. Otherwise, it can be determined by the gradient value

to adjust to make the blur kernel spatially adaptive.

In addition, considering the non-negative characteristics of the target image and the blur kernel, when a negative value appears in the solution vector, it is necessary to punish the negative value to force the solution to develop in a non-negative direction. In order to reduce the amount of calculation, according to the gray value of each point of the target image and the blur kernel image, the present invention selects the cost function J(f, h) of the following form as the non-negativity constraint regularization item of the target and blur kernel.

is a constant coefficient that penalizes negative values in f,h,

The range of values is between 0 and 10. When the negative value of f and h is too large,

The value is smaller, and vice versa, to ensure that the changes of each point of the target image and the blurred kernel image are gentle. Set to 1 in this example,

is a diagonal matrix related to each element in f, h, namely:

Among them, p=M×N, b=m×n, and the diagonal elements a _i and b _j take values of 1 and 0. The diagonal elements a _i , b _j are respectively determined by the values f _i , h _j corresponding to the i, jth elements in f, h, that is,

Let the variable u = δ _H f, u is related to the blur kernel, based on the above description, if the noise is not considered, the minimization model of space-varying deblurring based on formula (9) is established as follows:

In the formula, _the _first item is the basic data item, ||||| The three items are the sparsity-constrained regularization items based on the adaptive anisotropy variable coefficient of the target and the fuzzy kernel respectively. J(f,h) is the target, fuzzy kernel non-negativity constraint regularization term.

In step S3, the solution process of the present invention to the minimized model constructed is as follows:

Since the formula (16) contains the l ₁ norm, the present invention adopts the Bregman multivariate separation solution method to solve the formula (16), let the variables d ₁ =Wf,d ₂ =f and introduce the Bregman auxiliary variable t ₁ =d ₁ - Wf, t ₂ =d ₂ -f to update the iterative process, that is, formula (16) can be transformed into:

make

It is the result obtained from the iterative solution of the nth step, and the iterative solution process of the n+1 step is as follows:

The update of the Bregman auxiliary variable is:

Until the iteration termination condition is reached during iteration:

ε is to set any small value, or stop iteration when the maximum number of iterations maxIter times is reached. In the present invention, ε is set to 10 ^-6 , and the value of maxIter is 200.

After minimizing the derivative of the variable f ⁽ⁿ⁺¹⁾ in formula (18) to zero, use FFT to quickly solve f ⁽ⁿ⁺¹⁾ in the above formula in the frequency domain, that is:

In the formula, F( ) means Fourier transform, F ^-1 ( ) means inverse Fourier transform,

Represents the Fourier complex conjugate operator, thereby realizing fast point-by-point correction and obtaining a space-variant restoration image, as shown in Figure 10(b). It can be seen from Figure 10 that the effect of space-variant restoration is better than that of space-invariant restoration.

The point-by-point restoration system of the aero-optical effect large-field-of-view space-varying degraded image in the embodiment of the present invention is mainly used to realize the above-mentioned method embodiment, and the system includes:

Among them, the local area screening module specifically includes:

Further, the fuzzy kernel initial value calculation module specifically includes:

Further, the model solving module specifically uses the Bregman multivariate separation and solving algorithm to solve the fuzzy kernel of each point and the gray value of each point.

For the implementation of each module, refer to the method embodiments above, and details are not described here.

The present application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), Magnetic Storage, Magnetic Disk, Optical Disk, Server, App Store, etc., on which computer programs, program When executed by the processor, corresponding functions are realized. The computer-readable storage medium in this embodiment is used to implement the method for point-by-point correction and restoration of an image degraded by aero-optical effect and large field of view in the method embodiment when executed by a processor.

The point-by-point restoration method of the aero-optical effect large field of view degraded image can provide a processing method for blur kernel estimation, blur kernel optimization, and restoration requirements of each point of the aero-optical effect space-varying degraded image; The restoration of space-varying degraded images with large field of view due to optical effects can meet the requirements of the aerospace field for restoring space-varying and degraded images with aero-optical effects in a large field of view.

It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should belong to the protection scope of the appended claims of the present invention.

Claims

A method for point-by-point restoration of an aero-optical effect large field of view space-varying degraded image, characterized in that it comprises the following steps:

S1. Calculate the gradient of the input degraded image, select multiple large gradient regions, extract multiple local regions according to the distribution of large gradients on the degraded image, and calculate the blur kernel of each local region;

S2. Calculate the distance to the center points of the two nearest local areas point by point, and perform inverse distance weighted interpolation calculation on the blur kernel of each point in the whole image according to the two distances of each pixel point, and obtain the initial value of the blur kernel of each point in the whole image , forming the initial fuzzy kernel matrix;

S3. Establish a space-varying degradation model based on the initial blur kernel matrix, and add a non-negativity constraint regularization item determined by the gray value of the target image and the initial blur kernel of each point and based on the two initial blur kernels of the target image and each point The sparsity-constrained regularization term of the adaptive anisotropic variable coefficient determined by the gradient value makes the target image and each point blur kernel non-negative and spatially adaptive;

S4. Solve the spatial variation degradation model, obtain the blur kernel of each point and the gray value of each point to realize point-by-point correction, and finally output the restoration image of spatial variation degradation.
The method for point-by-point restoration of a large-field-of-view space-varying degraded image with dynamic optical effects according to claim 1, wherein the specific method of step S1 is:

S11. Using a multi-scale morphological gradient operator to obtain the gradient of the degraded image;

S12. Use the gradient usefulness index to filter out small structural gradient regions, select large gradient local regions whose length and width directions are greater than a certain value on the filtered gradient image, and then extract multiple local regions according to the distribution of large gradients on the degraded image area;

S13. Estimate the blur kernel of each extracted local region by using a non-negative least squares criterion algorithm based on spatial correlation constraints.
The method for point-by-point restoration of the aero-optical effect large-field-of-view space-varying degraded image according to claim 1, wherein the specific method of step S2 is:

S21. Taking the blur kernel of each local area as the initial value of the blur kernel of each point in the local area;

S22. Calculate the Euclidean distance from each pixel point to the center points of all local regions;

S23. Comparing and obtaining the Euclidean distance from each pixel point to the center points of the two nearest local areas, and regarding the distance as the corresponding weight coefficient, performing inverse distance weighted interpolation calculation on the blur kernel of the corresponding pixel point according to the weight coefficient, and obtaining the corresponding The initial value of the blur kernel for pixels.
The point-by-point restoration method of the aero-optical effect large-field-of-view space-varying degraded image according to claim 1, wherein in step S4, the Bregman multivariate separation solution algorithm and the hysteresis fixed-point iterative method are specifically used to solve the fuzzy kernel sum of each point The gray value of each point.
A point-by-point restoration system for aero-optical effect large-field space-varying degraded images, characterized in that it includes:

The local area screening module is used to calculate the gradient of the input degraded image, select multiple large gradient areas, extract multiple local areas according to the distribution of large gradients on the degraded image, and calculate the blur kernel of each local area;

The initial value calculation module of the blur kernel is used to calculate the distance to the center points of the two nearest local areas point by point, and perform inverse distance weighted interpolation calculation on the blur kernel of each point in the whole image according to the two distances of each pixel point, and obtain the full The initial value of the fuzzy kernel of each point in the graph constitutes the initial fuzzy kernel matrix;

The spatial variation degradation model building block is used to establish the spatial variation degradation model according to the initial fuzzy kernel matrix, and add non-negativity constraint regularization items and sparsity constraint regularization items based on adaptive anisotropy variable coefficients to make the target image and each The point blur kernel is non-negative and spatially adaptive;

The model solving module is used to solve the spatial variation degradation model, obtain the blur kernel of each point and the gray value of each point to realize point-by-point correction, and finally output the restoration image of spatial variation degradation.
The point-by-point restoration system for large-field-of-view space-varying degraded images with dynamic optical effects according to claim 5, wherein the local area screening module specifically includes:

The gradient calculation sub-module is used to obtain the gradient of the degraded image by using a multi-scale morphological gradient operator;

The gradient filtering sub-module is used to filter out the small structural gradient area by using the gradient usefulness index, and select a large gradient local area whose length and width are greater than a certain value on the filtered gradient image, and then according to the large gradient on the degraded image The distribution of extracts multiple local regions;

The regional blur kernel estimation submodule is used for estimating the extracted blur kernel of each local region by using a non-negative least square criterion algorithm based on spatial correlation constraints.
The point-by-point restoration system for aero-optical effect large-field-of-view space-varying degraded images according to claim 5, wherein the blur kernel initial value calculation module specifically includes:

The sub-module for determining the fuzzy kernel of each point is used to regard the fuzzy kernel of each local area as the initial value of the fuzzy kernel of each point in the local area;

The distance calculation sub-module is used to calculate the Euclidean distance from each pixel point to the center points of all local areas;

The weighted blur kernel calculation sub-module is used to compare the Euclidean distance between each pixel point and the center points of the two nearest local areas, and regard the distance as the corresponding weight coefficient, and reverse the blur kernel of the corresponding pixel point according to the weight coefficient Calculate the distance weighted interpolation to obtain the initial value of the blur kernel of the corresponding pixel.
The point-by-point restoration system of the aero-optical effect large field of view space-varying degraded image according to claim 5, wherein the model solution module specifically uses the Bregman multivariate separation solution algorithm and the hysteresis fixed-point iterative method to solve the fuzzy kernel of each point and The gray value of each point.
A computer storage medium, characterized in that it can be executed by a processor, and a computer program is stored therein, and the computer program executes the aero-optical effect and large field of view spatial variation degradation described in any one of claims 1-4 Image point-by-point restoration method.