Image restoration method damaged by Cauchy noise and blur based on non-convex variational model

By adopting a non-convex regularization variation model in image restoration, combining the statistical properties of Cauchy noise and sparse gradient information, the problem of insufficient edge features and contrast in the restoration image in the prior art is solved, and better image clarity is achieved and edges to maintain effect.

CN114972068BInactive Publication Date: 2024-11-29HANGZHOU DIANZI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210413631.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-15
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art is difficult to effectively restore images affected by Cauchy noise and blur, especially in maintaining image edge features and contrast.

Method used

Using the image restoration method based on the non-convex regularization variation model, by introducing non-convex regularization terms and image sparse gradient information, combining the statistical properties of Cauchy noise, appropriate data fidelity terms are designed, and the ADMM algorithm is used for iterative solution. .

Benefits of technology

Improved image edge retention, enhanced edge contrast, and improved image clarity, especially in medical image denoising against the influence of Cauchy noise and blur.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114972068B_ABST
    Figure CN114972068B_ABST
Patent Text Reader

Abstract

The present invention discloses an image restoration processing method affected by Cauchy noise and various blurs based on a non-convex regularized variational model, comprising the following steps: S1, collecting images with blur and Cauchy noise through a sensor; S2, building a model for removing Cauchy noise and blur; S3, obtaining an image with blur and Cauchy noise eliminated. A non-convex regularization term with better characterization of image edge sparse features and edge contrast is introduced, and a suitable data fidelity term is designed according to the statistical properties of Cauchy noise. At the same time, the neighboring linearization approximation strategy for solving non-convex subproblems in the past is improved, and a more accurate direct solution is used to ensure the accuracy of iteration. After the improvement, the edge preservation effect of the resulting image is better, and the block area is smoother, which improves the clarity of the image destroyed by Cauchy noise and blur.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of digital image processing, and in particular to an image restoration processing method based on a non-convex regularized variational model that is affected by Cauchy noise and various blurs. Background Art

[0002] At present, most of the literature at home and abroad is about the restoration of images damaged by additive Gaussian noise, so the research on Gaussian noise has been relatively complete. However, in many engineering applications (such as medicine, aviation, etc.), noise is often not Gaussian noise that is easy to handle. Some noises have strong impulse characteristics, such as impulse noise, which can cause serious damage to the image. Impulse noise can be caused by faulty pixels in camera sensors, errors in analog-to-digital converters, and so on. A common type of impulse noise is Cauchy noise, which often appears in atmospheric and underwater acoustic noise, radar and sonar applications, wireless communication systems, biomedical images, and synthetic aperture radar (SAR) images.

[0003] Under the influence of Cauchy noise, the characteristic fluctuation amplitude of the local position of the signal is large and severely damaged, making it difficult to observe the original characteristics of the signal, which brings great challenges to restoring the information destroyed by Cauchy noise. At present, the research on signal and image restoration problems affected by Cauchy noise is relatively limited. Federica Sciacchitano et al. first tried to use a convex variational model to eliminate Cauchy noise, but because its regularization term is a convex function, although there is a certain restoration effect, it still needs to be improved, especially in terms of maintaining image edge information and edge contrast. Using the sparse features of the image under certain transformations, non-convex regularized variational methods play an increasingly important role in maintaining important structural features and edge contrast of the image. Therefore, the present invention proposes a non-convex variational restoration method based on a non-convex regularized potential function and sparse features of the image edge to better solve the problems existing in the prior art. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide an image restoration method based on a non-convex variational model that preserves edge features and edge contrast. The method improves the convex regularization term of the previous variational image restoration model to a non-convex regularization term, combines the sparse gradient information of the image, and provides a more accurate method other than the neighboring linearization approximation to solve the proposed non-convex regularization model.

[0005] In order to solve the above technical problems, the technical solution of the present invention is:

[0006] A method for restoring an image damaged by Cauchy noise and blur based on a non-convex variational model comprises the following steps:

[0007] S1, collect images with blur and Cauchy noise through sensors;

[0008] S2. Build a model to remove Cauchy noise and blur

[0009] S2-1. Construct a variational model of the non-convex regularization term of the image affected only by Cauchy noise. Based on the variational model, introduce and design a suitable non-convex regularization term. The data fidelity term is determined by the statistical properties of Cauchy noise. An additional quadratic term is introduced to make the fidelity term a convex function.

[0010] S2-2, further construct a non-convex regularization term variational model that simultaneously removes the effects of Cauchy noise and blur;

[0011] S2-3. Design augmented Lagrangian functional and solution algorithm based on variational model;

[0012] S2-4, iteratively solve according to the design algorithm, and use ADMM algorithm for iterative solution;

[0013] S3. Obtain an image with blur and Cauchy noise removed.

[0014] Preferably, the variational model of the non-convex regularization term constructed in step S2-1 introduces and designs a suitable non-convex regularization term on the basis of the variational model, the data fidelity term is determined by the statistical properties of Cauchy noise, and an additional quadratic term is introduced to make the fidelity term a convex function.

[0015] Preferably, the image non-convex regularization term variational model only affected by Cauchy noise is as follows:

[0016]

[0017] In the above model, is the data fidelity item, is a non-convex regularization term, u is the final restored result close to the real image, f is the observed image collected by the sensor, u0 is the result of the preliminary restoration of the observed image using the median filter, is the gradient operator, is a non-convex potential function, λ is a parameter that balances the regularization term and the data fidelity term, and μ is a parameter that ensures that the second term in the model is a convex function.

[0018] Preferably, in step S2-2, a variational model for removing the Cauchy noise and blurring-affected non-convex regularization term is specifically as follows:

[0019]

[0020] Where K is the blur kernel.

[0021] Preferably, the specific method of step S2-3 is:

[0022] For the non-convex regularization term variational model of images affected by Cauchy noise and blur, a new variable is introduced v=Ku, and then construct the augmented Lagrangian functional:

[0023]

[0024] Preferably, in step S2-4, the specific expression of the ADMM algorithm is as follows:

[0025]

[0026] The final iterative result u is the restored image after removing the Cauchy noise and blur.

[0027] Preferably, when solving the non-convex and non-smooth p-subproblem in the ADMM algorithm, a new function is constructed:

[0028]

[0029] Then find the root s of x(s) * As a result of one iteration of p, that is

[0030] The present invention has the following characteristics and beneficial effects:

[0031] A non-convex regularization term that can better characterize the sparse features and edge contrast of the image edge is introduced, and a suitable data fidelity term is designed according to the statistical properties of Cauchy noise. At the same time, the neighboring linearization approximation strategy for solving non-convex subproblems in the past is improved, and a more accurate direct solution scheme is used to ensure the accuracy of iteration. After the improvement, the edge preservation effect of the resulting image is better, and the block area is smoother, which improves the clarity of the image destroyed by Cauchy noise and blur. The suitable non-convex regularization term and data fidelity term designed according to the Cauchy noise type and image prior information have achieved superior results in restoring various images, especially in the piecewise constant image, which is superior to the traditional noise processing method. In particular, it also has great advantages in denoising medical images affected by noise and blur. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0033] Figure 1, is a flow chart of the method of the present invention.

[0034] Figure 2 , is a comparison chart of the experimental results of Example 1 of the present invention.

[0035] Figure 3 , is a comparison chart of the experimental results of Example 2 of the present invention.

[0036] Figure 4 , is a comparison chart of the experimental results of Example 3 of the present invention. DETAILED DESCRIPTION

[0037] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0038] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0039] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0040] The present invention provides a method for restoring an image damaged by Cauchy noise and blur based on a non-convex variational model. Figure 1 As shown, the following steps are included:

[0041] S1, collect images with blur and Cauchy noise through sensors;

[0042] S2. Build a model to remove Cauchy noise and blur

[0043] S2-1. Construct a variational model of the non-convex regularization term of the image affected only by Cauchy noise. Based on the variational model, introduce and design a suitable non-convex regularization term. The data fidelity term is determined by the statistical properties of Cauchy noise. An additional quadratic term is introduced to make the fidelity term a convex function.

[0044] Specifically, a variational model of non-convex regularization term is constructed. On the basis of the variational model, a suitable non-convex regularization term is introduced and designed. The data fidelity term is determined by the statistical properties of Cauchy noise, and an additional quadratic term is introduced to make the fidelity term a convex function.

[0045] The non-convex regularization term variational model of the image only affected by Cauchy noise is as follows:

[0046]

[0047] In the above model, is the data fidelity item, is a non-convex regularization term, u is the final restored result close to the real image, f is the observed image collected by the sensor, u0 is the result of the preliminary restoration of the observed image using the median filter, is the gradient operator, is a non-convex potential function, λ is a parameter that balances the regularization term and the data fidelity term, and μ is a parameter that ensures that the second term in the model is a convex function.

[0048] S2-2, further construct a non-convex regularization term variational model that simultaneously removes the effects of Cauchy noise and blur;

[0049] Specifically, in step S2-2, the variational model for removing the Cauchy noise and blurring-affected non-convex regularization term is as follows:

[0050]

[0051] Where K is the blur kernel.

[0052] S2-3. Design augmented Lagrangian functional and solution algorithm based on variational model;

[0053] In the above technical solution, in order to solve the variational model that removes both the Cauchy noise and the blur-affected non-convex regularization term, a new variable is introduced v = Ku to solve the problem that the coupling between variables is difficult to solve, and then the augmented Lagrangian functional can be constructed:

[0054]

[0055] S2-4, iteratively solve according to the design algorithm, and use ADMM algorithm for iterative solution;

[0056] Specifically, the specific expression of the ADMM algorithm is as follows:

[0057]

[0058] The final iterative result u is the restored image after removing the Cauchy noise and blur.

[0059] Furthermore, when solving the non-convex and non-smooth p-subproblem in the ADMM algorithm, Instead of using linear function approximation to solve the problem, a new function is designed:

[0060]

[0061] Then find the root s of x(s) * As a result of one iteration of p, that is At this time, the iterative result is more accurate than the linear function approximation method.

[0062] S3. Obtain an image with blur and Cauchy noise removed.

[0063] In order to further explain and illustrate the above technical solution, a specific embodiment is given below:

[0064] Example 1

[0065] For the first set of experimental images, we simulated the situation where the medical images "NCAT", "SheppLogan", "CT" and the astronomical image "Salellite" were only contaminated by Cauchy noise. Among them, Cauchy noise is simulated by the ratio of two different Gaussian noises. The specific code for applying Cauchy noise is:

[0066] gc=0.02;

[0067] randn('state',-34)

[0068] v1 = randn(size(I));

[0069] randn('state',94)

[0070] v2 = randn(size(I));

[0071] f = I + gc * v1. / v2;

[0072] Where I is the real image not corrupted by noise.

[0073] like Figure 2As shown, the visual results and peak signal-to-noise ratio values ​​of four observed images contaminated by Cauchy noise are compared using median filtering, TV convex variational model and the non-convex variational method of the present invention for restoration. The higher the visual result and the peak signal-to-noise ratio value, the better the effect. Therefore, the parameters of each method are adjusted to the optimal state.

[0074] Example 2

[0075] For the second set of experimental images, we simulated the situation where the medical images "NCAT", "SheppLogan", "CT" and the astronomical image "Salellite" were contaminated by Cauchy noise and Gaussian blur at the same time. The specific code for applying Cauchy noise and Gaussian blur is as follows:

[0076] gc=0.02;

[0077] randn('state',-34)

[0078] v1 = randn(size(I));

[0079] randn('state',94)

[0080] v2 = randn(size(I));

[0081] H=fspecial('gaussian',[9,9],1);

[0082] f=imfilter(I,H,'circular','conv')+gc*v1. / v2;

[0083] Where I is the unaffected true image.

[0084] like Figure 3 As shown, the visual results and peak signal-to-noise ratio values ​​of four observed images contaminated by noise and Gaussian blur are compared using median filtering, TV convex variational model and the non-convex variational method of the present invention for restoration. The higher the visual result and the peak signal-to-noise ratio value, the better the effect. Therefore, the parameters of each method are adjusted to the optimal state.

[0085] Example 3

[0086] For the third set of experimental images, we simulate the situation where the medical images "NCAT", "SheppLogan", "CT" and the astronomical image "Salellite" are contaminated by Cauchy noise and motion blur at the same time. The specific code for applying Cauchy noise and motion blur is as follows:

[0087] gc=0.02;

[0088] randn('state',-34)

[0089] v1 = randn(size(I));

[0090] randn('state',94)

[0091] v2 = randn(size(I));

[0092] H=fspecial('motion',15,30);

[0093] f=imfilter(I,H,'circular','conv')+gc*v1. / v2;

[0094] Where I is the unaffected true image.

[0095] like Figure 4 As shown, the visual results and peak signal-to-noise ratio values ​​of four observed images contaminated by noise and motion blur are compared using median filtering, TV convex variational model and the non-convex variational method of the present invention for restoration. The higher the visual result and the peak signal-to-noise ratio value, the better the effect. Therefore, the parameters of each method are adjusted to the optimal state.

[0096] "NCAT" is a medical image with the characteristics of a sliced ​​constant image. In each group of experiments, after the restoration method of the present invention, each area is more clean and uniform after enlargement; while other images have color blocks with slightly different color numbers mixed in various areas. On the edge, the restoration method of the present invention has clear boundaries and high contrast; while other restored images can be seen that some boundaries have color infiltrating to the other side. In addition, compared with other methods, the method of the present invention can obtain a higher peak signal-to-noise ratio (the higher the peak signal-to-noise ratio, the better the restoration result). Therefore, whether it is from a visual intuitive observation or a numerical peak signal-to-noise ratio restoration result, the non-convex regularized model of the present invention can perform better in the restoration work. "SheppLogan" and "CT" are two other medical images. The block areas of these two images are small and the restoration difficulty is high. In the restoration experiment affected by noise and blur at the same time, the restoration result of the method of the present invention is obviously superior in peak signal-to-noise ratio, and the image texture details are more completely preserved.

[0097] "Salellite" is an astronomical image with more complex texture details. Figure 2 and Figure 3 In the experiment, it is slightly better or similar to the TV convex variational model; Figure 4 In the experiment, the visual restoration results of the method of the present invention are obviously superior, and the texture details and boundary contrast of the image are clearer.

[0098] According to the above experiments, in medical and astronomical scenes where Cauchy noise is prone to appear, the method of the present invention has advantages in edge and contrast preservation, has the best denoising effect in flat areas, and has a better restoration effect in terms of peak signal-to-noise ratio.

[0099] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A method for restoring an image damaged by Cauchy noise and blur based on a non-convex variational model, characterized in that: The steps include: S1, collect images with blur and Cauchy noise through sensors; S2. Build a model to remove Cauchy noise and blur S2-1. Construct a variational model of the non-convex regularization term of the image affected only by Cauchy noise. Based on the variational model, introduce and design a suitable non-convex regularization term. The data fidelity term is determined by the statistical properties of Cauchy noise. An additional quadratic term is introduced to make the fidelity term a convex function. The variational model of the non-convex regularization term constructed in step S2-1, based on the variational model, introduces and designs a suitable non-convex regularization term, the data fidelity term is determined by the statistical properties of Cauchy noise, and an additional quadratic term is introduced to make the fidelity term a convex function; The non-convex regularization term variational model of the image only affected by Cauchy noise is as follows: In the above model, is the data fidelity term, is the non-convex regularization term, u is the finally restored result approximating the real image, f is the observed image collected by the sensor, and u0 is the result of initially restoring the observed image using median filtering. is the gradient operator, is the non-convex potential function, where 0 < r < 1, λ is the parameter for balancing the regularization term and the data fidelity term, and μ is the parameter for ensuring that the second term in the model is a convex function. S2-2, further construct a non-convex regularization term variational model that simultaneously removes the effects of Cauchy noise and blur; S2-3. Design augmented Lagrangian functional and solution algorithm based on variational model; S2-4, iteratively solve according to the design algorithm, and use ADMM algorithm for iterative solution; S3. Obtain an image with blur and Cauchy noise removed.

2. The image restoration method damaged by Cauchy noise and blur based on non-convex variational model according to claim 1, characterized in that: In step S2-2, a variational model for removing the Cauchy noise and blurring-affected non-convex regularization term is specifically as follows: Where K is the blur kernel.

3. The image restoration method damaged by Cauchy noise and blur based on non-convex variational model according to claim 2, characterized in that: The specific method of step S2-3 is: For the non-convex regularization term variational model of images affected by Cauchy noise and blur, a new variable is introduced v=Ku, and then construct the augmented Lagrangian functional:

4. The image restoration method damaged by Cauchy noise and blur based on non-convex variational model according to claim 3, characterized in that: In step S2-4, the specific expression of the ADMM algorithm is as follows: The final iterative result u is the restored image after removing the Cauchy noise and blur.

5. The image restoration method damaged by Cauchy noise and blur based on non-convex variational model according to claim 4, characterized in that: When solving the non-convex and non-smooth p-subproblem in the ADMM algorithm, a new function is constructed: Then find the root s of x(s) * As a result of one iteration of p, that is

Citation Information

Patent Citations

  • Image restoration method under Cauchy noise and application thereof

    CN112233046A

  • Image restoration method and system, processing terminal and computer medium

    CN113592728A