A blind image deblurring method robust to impulse noise

Abstract

A blind image deblurring method robust to impulse noise belongs to the field of computer vision and image processing technology. It solves the technical problem that when the blur kernel is unknown and the image is contaminated by impulse noise, traditional methods have difficulty balancing noise suppression and detail preservation, resulting in distorted restored images. A coarse-to-fine blur kernel estimation method is used to establish a coarse-to-fine image pyramid #imgabs1# for the blurred image #imgabs0#. The blur kernel #imgabs3# of the #imgabs2#th layer of the degraded image is estimated, #imgabs4# is upsampled to obtain the initial value of the blur kernel of the #imgabs5#th layer, and the blur kernel #imgabs7# of #imgabs6# is updated. Alternating iterative optimization is repeated. By imposing sparse logarithmic prior constraints in the blind image deblurring model, the noise detection process is stabilized to avoid the influence of extreme weight values. Combined with FFT acceleration, the time complexity is reduced. The overall method improves the accuracy of blur kernel estimation in complex scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision and image processing, and in particular relates to a blind image deblurring method that is robust to impulse noise. Background Art

[0002] Traditional blind image deblurring methods rely on edge detection or Gaussian noise assumptions, are sensitive to impulse noise, and easily lead to blur kernel estimation bias. Existing methods suppress noise through weight matrices, but the regularization term design is unstable (for example, the constraint fails when the weight is close to 0 or 1), and some algorithms require multiple iterative optimizations, which is too time-consuming and difficult to meet real-time requirements.

[0003] When the blur kernel is unknown and the image is contaminated by impulse noise, traditional methods find it difficult to balance noise suppression and detail preservation, resulting in distortion in the restored image. Summary of the Invention

[0004] In order to solve the technical problem that when the blur kernel is unknown and the image is contaminated by impulse noise, traditional methods have difficulty in balancing noise suppression and detail preservation, resulting in distortion of the restored image, the present invention provides a blind image deblurring method that is robust to impulse noise. The method comprises the following steps:

[0005] S1. Create a blurred image containing impulse noise Coarse-to-fine image pyramid ;

[0006] S2. Initialize blur kernel ,right The alternating optimization algorithm is used to solve The corresponding blur kernel ;

[0007] S3, yes Up-sample, The upsampling result is Initialized blur kernel, The alternating optimization algorithm is used to solve The corresponding blur kernel ;

[0008] S4, using the same method as step S3, gradually obtain The corresponding blur kernel ;

[0009] S5, through blur kernel Solve for the potential clear image.

[0010] Furthermore, the alternating optimization algorithm is specifically as follows:

[0011] S21, setting a blind image deblurring model, wherein the model includes a data fitting term, an image regularization term, a blur kernel regularization term, a weight matrix sparse constraint term, and a weight matrix distribution constraint term;

[0012] S22. Initialize the weight matrix of the blind image deblurring model ;

[0013] S23, fixed blur kernel , update the latent sharp image by fast Fourier transform ;

[0014] S24. Fixing the potential clear image , update the blur kernel , get the updated blur kernel ;

[0015] S25, through the potential clear image and the updated blur kernel Update the weight matrix .

[0016] Furthermore, the blind image deblurring model is specifically:

[0017]

[0018] ;

[0019] in, express norm, express norm, represents the Hadamard product, represents the gradient operator, 、 、 and represents the weight term, Represents a coarse-to-fine image pyramid Any one of .

[0020] Furthermore, the latent sharp image is updated by fast Fourier transform ,pass:

[0021] conduct;

[0022] in, 、 and They are FFT operator, inverse FFT operator and complex conjugate operator respectively. ,in, represents iterative reweighted least squares method, Represents a potential clear image The negative 1.2 power of the gradient modulus,

[0023] represents the number of iterations of the iterative reweighted least squares method, is a positive number, The purpose is to avoid the situation where the image gradient value is 0.

[0024] Further, update the blur kernel , get the updated blur kernel ,pass:

[0025] conduct.

[0026] Further, through the potential clear image and the updated blur kernel Update the weight matrix ,pass:

[0027] conduct.

[0028] Furthermore, through the blur kernel Solve for the potential sharp image by:

[0029] conduct;

[0030] in, Represents The corresponding weight matrix, Represents The corresponding potential clear image.

[0031] The beneficial effects of this method are:

[0032] The method of the present invention uses a coarse-to-fine blur kernel estimation method to establish a blurred image Coarse-to-fine image pyramid ; Estimate blur kernel of layer-degraded image ,Will Upsample to get The initial value of the layer blur kernel is Update its blur kernel , repeating the alternating iterative optimization until the blur kernel with the highest fineness and its corresponding clear image are obtained, which can balance noise suppression and detail preservation; and by constraining the sparse logarithmic prior in the blind image deblurring model, stabilizing the noise detection process and avoiding the influence of extreme weight values, and combining FFT accelerated calculation to reduce time complexity, the overall method improves the blur kernel estimation accuracy in complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 1. A comparison of a clear image (a) corresponding to a picture on the Levin dataset in an embodiment of the present invention, a blurred image (b) containing 10% noise, and the results of applying various deblurring methods to the blurred image (b). DETAILED DESCRIPTION

[0034] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0035] Example 1

[0036] This embodiment provides a blind image deblurring method that is robust to impulse noise, the method comprising the following steps:

[0037] S1. Create a blurred image containing impulse noise Coarse-to-fine image pyramid ;

[0038] S2. Initialize blur kernel ,right The alternating optimization algorithm is used to solve The corresponding blur kernel ;

[0039] S3, yes Up-sample, The upsampling result is Initialized blur kernel, The alternating optimization algorithm is used to solve The corresponding blur kernel ;

[0040] S4, using the same method as step S3, gradually obtain The corresponding blur kernel ;

[0041] S5, through blur kernel Solve for the potential clear image.

[0042] Example 2

[0043] This embodiment further limits the embodiment 1, and the alternating optimization algorithm is specifically as follows:

[0044] S21, setting a blind image deblurring model, wherein the model sequentially includes a data fitting term, an image regularization term, a blur kernel regularization term, a weight matrix sparse constraint term, and a weight matrix distribution constraint term;

[0045] S22. Initialize the weight matrix of the blind image deblurring model ;

[0046] S23, fixed blur kernel (The blur kernel here It can be understood as the initialization blur kernel in Example 1 , in the right In the process of continuously solving the potential clear image, the blur kernel and weight matrix are iterating), and the potential clear image is updated through fast Fourier transform. ;

[0047] S24. Fixing the potential clear image , update the blur kernel , get the updated blur kernel ;

[0048] S25, through the potential clear image and the updated blur kernel Update the weight matrix .

[0049] The blind image deblurring model is specifically:

[0050]

[0051] ;

[0052] in, express norm, express norm, represents the Hadamard product, represents the gradient operator, 、 、 and represents the weight term, Represents a coarse-to-fine image pyramid Any one of .

[0053] Update the latent sharp image via Fast Fourier Transform ,pass:

[0054] conduct;

[0055] in, 、 and They are FFT operator, inverse FFT operator and complex conjugate operator respectively. ,in, represents iterative reweighted least squares method, Represents a potential clear image The negative 1.2 power of the gradient modulus,

[0056] represents the number of iterations of the iterative reweighted least squares method, is a positive number, The purpose is to avoid the situation where the image gradient value is 0.

[0057] Update blur kernel , get the updated blur kernel ,pass:

[0058] conduct.

[0059] By potentially clear images and the updated blur kernel Update the weight matrix ,pass:

[0060] conduct.

[0061] Through blur kernel Solve for the potential sharp image by:

[0062] conduct;

[0063] in, Represents The corresponding weight matrix, Represents The corresponding potential clear image.

[0064] Example 3

[0065] This example further illustrates the beneficial effects of the present invention. The method of the present invention improves PSNR and SSIM by 2.5%-5.2% and 0.7%-2.7% respectively on Levin and natural image datasets. Furthermore, the method is highly robust and can restore clear details even under 10% impulse noise. The method of the present invention converges quickly, and both blind and non-blind deblurring processes converge within a limited number of iterations.

[0066] In this embodiment, the parameters are set as follows: .

[0067] Results: On the Levin dataset, the PSNR reached 27.69 d (a 2.7% improvement over the second-best method).

[0068] like Figure 1 The figure shows a clear image (a) corresponding to a picture on the Levin dataset and its blurred image (b) with 10% noise (PSNR=15.4855, SSIM=0.1333), as well as a comparison of the results of various deblurring methods for the blurred image (b). The deblurring method used for image (c) (PSNR=9.1338, SSIM=0.0277) is the method described in the paper "Blind image deblurring with Gaussian curvature of the image surface" (Source: SignalProcessing: Image Communication); the deblurring method used for image (d) (PSNR=8.7570, SSIM=0.0136) is the method described in the paper "Graph-based blind image deblurring from a single photograph" (Source: IEEE transactions on image processing); the deblurring method used for image (e) (PSNR=23.4329, SSIM=0.7845) is the method described in the paper "Blind image deblurring without The deblurring method used for image (f) (PSNR=26.3822, SSIM=0.8577) is the method described in the paper "Oid: Outlier identifying and discarding in blind image deblurring" (Source: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXV 16. Springer International Publishing); the deblurring method used for image (g) (PSNR=27.6918, SSIM=0.8729) is the method described in the present invention.

[0069] The above comparison shows that the method described in this paper excels in preserving image details, outperforming other methods in visual quality, and producing clearer and more natural restored images. The method described in this paper achieved the highest average PSNR and SSIM values of 26.5575 dB and 0.8289, respectively. Compared to the next-best method's average PSNR and SSIM values of 25.8576 dB and 0.8232, these improvements represent 2.7% and 0.7% improvements in PSNR and SSIM, respectively.

Claims

1. A blind image deblurring method robust to impulse noise, characterized in that: The method comprises the following steps: S1. Create a blurred image containing impulse noise Coarse-to-fine image pyramid ; S2. Initialize blur kernel ,right The alternating optimization algorithm is used to solve The corresponding blur kernel ; S3, yes Up-sample, The upsampling result is Initialized blur kernel, The alternating optimization algorithm is used to solve The corresponding blur kernel ; S4, using the same method as step S3, gradually obtain The corresponding blur kernel ; S5, through blur kernel Solve for the potential clear image; The alternating optimization algorithm is specifically: S21, setting a blind image deblurring model, wherein the model includes a data fitting term, an image regularization term, a blur kernel regularization term, a weight matrix sparse constraint term, and a weight matrix distribution constraint term; S22. Initialize the weight matrix of the blind image deblurring model ; S23, fixed blur kernel , update the latent sharp image by fast Fourier transform ; S24. Fixing the potential clear image , update the blur kernel , get the updated blur kernel ; S25, through the potential clear image and the updated blur kernel Update the weight matrix ; The blind image deblurring model is specifically: ； in, express norm, express norm, represents the Hadamard product, represents the gradient operator, 、 、 and represents the weight term, Represents a coarse-to-fine image pyramid Any one of .

2. The blind image deblurring method robust to impulse noise according to claim 1, characterized in that Update the latent sharp image via Fast Fourier Transform ,pass: conduct; in, 、 and They are FFT operator, inverse FFT operator and complex conjugate operator respectively. ,in, represents iterative reweighted least squares method, Represents a potential clear image The negative 1.2 power of the gradient modulus, represents the number of iterations of the iterative reweighted least squares method, is a positive number, The purpose is to avoid the situation where the image gradient value is 0.

3. The blind image deblurring method robust to impulse noise according to claim 2, characterized in that: Update blur kernel , get the updated blur kernel ,pass: conduct.

4. The blind image deblurring method robust to impulse noise according to claim 3, characterized in that: By potentially clear images and the updated blur kernel Update the weight matrix ,pass: conduct.

5. The blind image deblurring method robust to impulse noise according to claim 4, characterized in that: Through blur kernel Solve for the potential sharp image by: conduct; in, Represents The corresponding weight matrix, Represents The corresponding potential clear image.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images