CN108648162B - Gradient-related TV factor image denoising and deblurring method based on noise level - Google Patents
Gradient-related TV factor image denoising and deblurring method based on noise level Download PDFInfo
- Publication number
- CN108648162B CN108648162B CN201810470422.3A CN201810470422A CN108648162B CN 108648162 B CN108648162 B CN 108648162B CN 201810470422 A CN201810470422 A CN 201810470422A CN 108648162 B CN108648162 B CN 108648162B
- Authority
- CN
- China
- Prior art keywords
- image
- gradient
- factor
- noise
- deblurring
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000005457 optimization Methods 0.000 claims description 9
- 230000001419 dependent effect Effects 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 5
- 238000009795 derivation Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000001629 suppression Effects 0.000 abstract 2
- 230000003068 static effect Effects 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 8
- 238000003384 imaging method Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000004800 variational method Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 230000002401 inhibitory effect Effects 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000010207 Bayesian analysis Methods 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000021615 conjugation Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
Abstract
The invention discloses a gradient correlation TV factor image denoising and deblurring method based on noise level. The images acquired by daily shooting are degraded to a certain degree, even if a static target is stably imaged, the acquired images also contain pixel-level fuzzy quantity, and noise is inevitable. The restoration operation of the noisy and fuzzy image can improve the image quality, however, contradiction exists between noise suppression and deblurring, and the balance of the noise suppression and the deblurring is required to be considered when the image is restored. The method provided by the invention provides an image de-noising and de-blurring method based on total variation regularization, a primary de-blurred image is obtained by combining the characteristics of a Richardson-Lucy algorithm while the image is subjected to fuzzy restoration, a plurality of regularization terms are used for constraining, the prior distribution characteristics of the gradient of a noise image are utilized, gradient-related distributed adjustment processing is carried out on a regularization weight factor, the noise can be effectively inhibited, the good image edge characteristics are kept, and the image with higher quality is obtained.
Description
Technical Field
The invention belongs to the field of digital image processing, and relates to a gradient-related TV factor image denoising and deblurring method based on noise level.
Background
With the popularization of photographing apparatuses and photographing techniques, photographing has become an important activity in daily life of people, and accordingly, the demand for image quality has been increasing. In the process of forming a natural image, the natural image may be affected by various aspects such as the motion of a target scene, the shake of an imaging device, the noise of an imaging sensor and the like, and finally a blurred and degraded image is obtained. In general, on the premise that the imaging system is linearly shifted unchanged, the process of forming a noisy and blurred image can be simply described as follows:
y(u,v)=k(u,v)*x(u,v)+n(u,v)
wherein: k (u, v) is a point spread function (image blur kernel, PSF); is a convolution sign, x (u, v) represents a sharp image, and n (u, v) represents additive noise. Image blur is a common problem in image capturing, and even with stable imaging of a still object, there is still image blur of a small magnitude, and noise is difficult to avoid.
Fig. 1 shows daily shot images and a detailed enlarged view thereof.
Restoration processing of noisy and blurred images is a common approach. In the process of image restoration, a contradiction exists between deblurring and denoising, and balance is often difficult. In general image processing, a noise removal method for a noise image alone is not few, and a good effect can be achieved, but the high-frequency information of the image is sacrificed. The method is combined with the gradient distribution prior characteristic of the image with noise, is particularly applied to image restoration, and achieves a remarkable effect. The method provided by the invention belongs to the field of image non-blind restoration, and is characterized in that a preliminary deblurred image is obtained by combining the characteristics of a Richardson-Lucy algorithm while the image is subjected to fuzzy restoration, a plurality of regularization terms are used for constraining, and gradient-related distributed adjustment processing is carried out on regular weighting factors by utilizing the prior distribution characteristic of the gradient of a noise image, so that the noise can be effectively inhibited and good image edge characteristics can be kept.
Disclosure of Invention
Fig. 2 shows a comparison between a natural sharp image and a noisy sharp image and their respective gradient distribution characteristic statistical graphs, and the analysis can obtain the gradient distribution rule of a general noisy image: the gradient data is mainly distributed in a low gradient range, and a maximum value exists at a lower gradient position and shows unequal sparse distribution to two sides. The invention aims to obtain a primary restoration result by using different responses of noise gradients and applying an RL algorithm to obtain a blurred image, and apply the blurred image to the subsequent regularization variational method image restoration operation, and perform gradient-related distributed adjustment processing on regularization constraint factors by adding image constraint conditions, thereby finally achieving the restoration effect of inhibiting noise and keeping good image edge characteristics.
In order to achieve the above purpose, the invention adopts the following technical scheme: a gradient correlation TV factor image denoising and deblurring method based on noise level comprises the following steps:
(1) obtaining a preliminary deblurred image I by using Richardson-Lucy (RL) algorithm0;
(2) An energy equation model introduced into a traditional total variation regularization method is as follows:
wherein I is a potential sharp image, k is an image blurring kernel, and lambda is a constraint factor,for the purpose of the first-order derivation operation,for convolution operations, E (I) is the energy of image I;
(3) on the basis of the energy equation model in the step (2), introducing new variables w, v, theta, gamma and beta, and adding a gradient penalty term and a secondary gradient smoothing term to obtain the following optimization model:
wherein Ω is the image integration domain;
converting the continuous optimization model into a discrete model, wherein the solution of the discrete model relative to the image is as follows:
wherein, F and F-1Representing Fourier transform and inverse Fourier transform, F*Representing the complex conjugate operation, x, y represent the two integration directions of the image,is a gradient operator;
meanwhile, the discrete model is decomposed into a w subproblem and a v subproblem:
(4) gradient penalty factor distributed processing, specifically:
(4.1) introduction of gradient-dependent TV penalty factor λsThe expression is as follows:
wherein the content of the first and second substances, variable s determines lambdasIs the overall constraint strength of s ∈ [0,2 ]];A threshold value of the gradient is indicated,the value is generally 0.04; parameter p1、b、d、p2κ together determining λsThe respective ranges are: p 1E [0,5 ]],b∈[50,300],d∈[0,20],p2∈[-10,50],κ∈[0,0.5];
(4.2) by using λsReplacing a constraint factor lambda, and alternately solving a w subproblem, a v subproblem and a clear image I;
(5) and controlling beta and theta parameters in the solving process, and repeatedly and iteratively solving until a clear image I is obtained.
Further, a Richardson-Lucy algorithm is combined with a total variation regularization method, and the Richardson-Lucy algorithm acquires a preliminary deblurred image as a prior result of the total variation regularization method and applies the preliminary deblurred image to a subsequent restoration process.
Further, a gradient dependent TV penalty factor λ is introducedsThe factors considered are: the gradient data of the noisy image is mainly distributed in a low gradient range, and a maximum value exists at the low gradient position and is distributed unevenly and sparsely to two sides.
Further, in the step (3), the continuous optimization model is converted into a discrete model:
further, in the step (4.2), the solution of the w subproblem is:
the invention has the beneficial effects that: and acquiring a blurred image by using different responses of noise gradients and an RL algorithm to obtain a primary restoration result, applying the primary restoration result to subsequent regularization variational method image restoration operation, performing gradient-related distributed adjustment processing on regularization constraint factors by adding image constraint conditions, and finally acquiring a de-noised and de-blurred image. The method is suitable for processing a general noisy blurred image, effectively distinguishes image details and noise in a deblurring process according to the noise gradient characteristic, can effectively inhibit the noise while keeping good image edge characteristic, gives consideration to the deblurring and denoising effects, and can reduce time cost to a certain extent due to the introduction of an RL algorithm.
Drawings
Fig. 1 is a blurred image of daily shooting and a detail enlargement example thereof, in which (a) is a shot image and (b) is a detail enlargement image.
FIG. 2 is a noisy image and its gradient statistical chart, in which (a) is a clear image, (b) is an image result after adding a noise variance of 0.001 of mean value, (c) is a gradient distribution statistical characteristic chart of (a), and (d) is a gradient distribution statistical characteristic chart of (b).
FIG. 3 is a schematic view of the process of the present invention.
FIG. 4 is a schematic diagram showing the detailed process of the method of the present invention.
FIG. 5 shows the parameter set to p1=0.3、b=90、d=12、κ=0.03、p2=50、s=0.5、Time-gradient dependent distributed factor λsThe numerical distribution of (c).
Fig. 6 is a comparison graph of restoration effect, wherein (a) is a blurred image with noise, (b) is an RL algorithm restoration result, (c) is an FTVD algorithm restoration result, and (d) is a restoration result of the method of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings and specific examples.
The method utilizes different responses of noise gradients, applies an RL algorithm to obtain a fuzzy image to obtain a preliminary restoration result, applies the preliminary restoration result to subsequent regularization variational method image restoration operation, and performs gradient-related distributed adjustment processing on regularization constraint factors by adding image constraint conditions, thereby finally achieving the restoration effect of inhibiting noise and keeping good image edge characteristics.
The process of the invention is shown in the attached figures 3 and 4, and mainly comprises the steps of obtaining a preliminary deblurred image by an RL algorithm, introducing an image regularization energy equation, introducing a new variable and solving an optimization equation, performing gradient penalty factor distributed processing, performing iterative computation and the like.
wherein, K*Is the adjoint of K, and t is the number of iterations. Although the RL algorithm is derived from the poisson noise model, it is equally applicable to other types of noise models. The preliminary sharper image obtained by the RL algorithm will be applied to the subsequent regularized variational model.
in the above formula, I represents a latent image, I0Representing the image to be processed, h representing the image blur kernel, λxIn order to be a constraint factor, the method comprises the following steps,for the purpose of the first-order derivation operation,is a convolution operation.
Step 3, introducing new variables and solving an optimization equation, wherein the method comprises the following steps:
3-1, introducing new variables w, v, theta and gamma on the basis of a traditional energy equation model, adding a gradient penalty term and a secondary gradient smoothing term, and having the following model:
wherein Ω is the image integration domain; when β approaches 0 and θ approaches 0, the solution of the above model converges to the solution of the model (3).
The 3-2 model (4) is a continuous total variation model which can be converted into a discrete model as follows:
the above formula is an optimization function of I, w, v, which has a closed form of the optimal solution with respect to I, and can be obtained equivalently after Fourier transform:
wherein, F and F-1Representing Fourier transform and inverse Fourier transform, F*Represents a complex conjugation operation;
3-3 regarding the solution of w and v, since w and v are independent in the respective functions, their values can be solved sequentially by an alternating minimization method, thereby solving the problem into two w sub-problems and v sub-problems. When the w subproblem is processed, v takes a fixed value and is converted into a quadratic problem with solving independent variable w, and the solving formula is as follows:
when the v subproblem is processed, w takes a fixed value and is converted into a minimum value solving problem with an independent variable of v:
4-1 was investigated to conclude the following: gradient data of the noisy and clear image are mainly distributed in a low gradient range, and a maximum value exists at a lower gradient position and is distributed unevenly and sparsely to two sides. Applying this conclusion to a specific restoration process, taking into account that the noise constraints for different gradient values should be changed, and at the same time, not making too much limitation on useful edge information of the image, a gradient correlation factor constraint is introduced, in this case:
wherein the content of the first and second substances,variable s determines lambdasIs the overall constraint strength of s ∈ [0,2 ]];A threshold value of the gradient is indicated,the value is generally 0.04; parameter p1、b、d、p2κ together determining λsThe respective ranges are: p 1E [0,5 ]],b∈[50,300],d∈[0,20],p2∈[-10,50],κ∈[0,0.5](ii) a FIG. 5 shows p1=0.3、b=90、d=12、κ=0.03、p2=50、s=0.5、Time lambdasDistribution of values as a function of gradient.
4-2 solving for (7) is:
solving (8) can obtain:
and 5, controlling the beta and theta parameters in the iteration process, and calculating and obtaining numerical values according to the formulas (6), (10) and (11) by changing the values of the beta and theta parameters to obtain the image of the iteration. Through the selection and the change of the parameters, the gradual convergence is realized in the iterative process, and finally, a satisfactory de-noising and de-blurring image is obtained.
To illustrate the effect of the method of the present invention on restoring a noisy and blurred image, a degraded image (fig. 6(a), gaussian blur, with a blur size of 9 × 9, a standard deviation of 1.5, and a noise of 0 mean and 0 variance of 0.001) is taken as an example, and the result of the method of the present invention is compared with the results of the degraded image, the RL method, and the FTVD method, as shown in fig. 6. Wherein, FIG. 6(b) is the RL restoration result, FIG. 6(c) is the FTVD method result, and FIG. 6(d) is the method result of the present invention. It can be seen from the figure that the method greatly inhibits noise while deblurring, gives consideration to noise removal and detail retention, achieves an ideal visual effect, and is superior to RL and FTVD methods. To further illustrate the superiority of the method, the above results were evaluated by objective evaluation criteria such as Structural Similarity (SSIM), peak signal-to-noise ratio (PSNR) and algorithm time, as shown in table 1 (MATLAB R2014b platform).
TABLE 1 Objective evaluation index Performance for different recovery methods
Claims (5)
1. A gradient correlation TV factor image denoising and deblurring method based on noise level is characterized by comprising the following steps:
(1) obtaining a preliminary deblurred image I by using Richardson-Lucy algorithm0;
(2) An energy equation model introduced into a traditional total variation regularization method is as follows:
wherein I is a potentially sharp image, I0Representing the image to be processed, k is an image blurring kernel, lambda is a constraint factor,for the purpose of the first-order derivation operation,for convolution operations, E (I) is the energy of image I;
(3) on the basis of the energy equation model in the step (2), introducing new variables w, v, theta, gamma and beta, and adding a gradient penalty term and a secondary gradient smoothing term to obtain the following optimization model:
wherein Ω is the image integration domain;
converting the continuous optimization model into a discrete model, wherein the solution of the discrete model relative to the image is as follows:
wherein, F and F-1Representing Fourier transform and inverse Fourier transform, F*RepresentsComplex conjugate operation, x, y represent two integration directions of the image,is a gradient operator;
meanwhile, the discrete model is decomposed into a w subproblem and a v subproblem:
(4) gradient penalty factor distributed processing, specifically:
(4.1) introduction of gradient-dependent TV penalty factor λsThe expression is as follows:
wherein the content of the first and second substances,variable s determines lambdasIs the overall constraint strength of s ∈ [0,2 ]];A threshold value of the gradient is indicated,parameter p1、b、d、p2κ together determining λsThe respective ranges are: p 1E [0,5 ]],b∈[50,300],d∈[0,20],p2∈[-10,50],κ∈[0,0.5];
(4.2) by using λsReplacing a constraint factor lambda, and alternately solving a w subproblem, a v subproblem and a clear image I;
(5) and controlling beta and theta parameters in the solving process, and repeatedly and iteratively solving until a clear image I is obtained.
2. The noise level-based gradient-correlation TV factor image denoising and deblurring method according to claim 1, wherein a Richardson-Lucy algorithm is combined with a total variation regularization method, and the Richardson-Lucy algorithm acquires a preliminary deblurred image as a prior result of the total variation regularization method and applies the preliminary deblurred image to a subsequent restoration process.
3. The method of claim 1, wherein a gradient-dependent TV factor image de-noising and deblurring method based on noise level is characterized by introducing a gradient-dependent TV penalty factor λsThe factors considered are: the gradient data of the noisy image is mainly distributed in a low gradient range, and a maximum value exists at the low gradient position and is distributed unevenly and sparsely to two sides.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810470422.3A CN108648162B (en) | 2018-05-16 | 2018-05-16 | Gradient-related TV factor image denoising and deblurring method based on noise level |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810470422.3A CN108648162B (en) | 2018-05-16 | 2018-05-16 | Gradient-related TV factor image denoising and deblurring method based on noise level |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108648162A CN108648162A (en) | 2018-10-12 |
CN108648162B true CN108648162B (en) | 2021-07-09 |
Family
ID=63756172
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810470422.3A Active CN108648162B (en) | 2018-05-16 | 2018-05-16 | Gradient-related TV factor image denoising and deblurring method based on noise level |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108648162B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109559286B (en) * | 2018-11-19 | 2022-12-06 | 电子科技大学 | Variance gradient constraint method infrared image edge preserving denoising method |
CN110121016B (en) * | 2019-05-08 | 2020-05-15 | 北京航空航天大学 | Video deblurring method and device based on double exposure prior |
CN111611929A (en) * | 2020-05-22 | 2020-09-01 | 天津市大清河管理中心 | River flood risk point identification method, device, server and storage medium based on LiDAR and InSAR technologies |
CN113628200A (en) * | 2021-08-19 | 2021-11-09 | 国网上海市电力公司 | Power transmission line picture deblurring method based on blind deconvolution |
CN114237000B (en) * | 2021-12-15 | 2023-05-23 | 中国工程物理研究院激光聚变研究中心 | Off-axis digital holographic optimization reconstruction method and system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102147915A (en) * | 2011-05-06 | 2011-08-10 | 重庆大学 | Method for restoring weighting sparse edge regularization image |
CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
CN106251315A (en) * | 2016-08-23 | 2016-12-21 | 南京邮电大学 | A kind of image de-noising method based on full variation |
CN106485671A (en) * | 2016-09-10 | 2017-03-08 | 天津大学 | Multi-direction Weighted T V based on edge and self-similarity constraint image deblurring method |
CN107133923A (en) * | 2017-03-02 | 2017-09-05 | 杭州电子科技大学 | A kind of blurred picture non-blind deblurring method based on self-adaption gradient sparse model |
CN107369139A (en) * | 2017-06-27 | 2017-11-21 | 五邑大学 | A kind of noise image deblurring method being fitted based on blended data with weighting total variation |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20120011694A (en) * | 2010-07-29 | 2012-02-08 | 삼성전자주식회사 | Method and apparatus of processing image and medical image system employing the same |
US9659351B2 (en) * | 2014-03-12 | 2017-05-23 | Purdue Research Foundation | Displaying personalized imagery for improving visual acuity |
-
2018
- 2018-05-16 CN CN201810470422.3A patent/CN108648162B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102147915A (en) * | 2011-05-06 | 2011-08-10 | 重庆大学 | Method for restoring weighting sparse edge regularization image |
CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
CN106251315A (en) * | 2016-08-23 | 2016-12-21 | 南京邮电大学 | A kind of image de-noising method based on full variation |
CN106485671A (en) * | 2016-09-10 | 2017-03-08 | 天津大学 | Multi-direction Weighted T V based on edge and self-similarity constraint image deblurring method |
CN107133923A (en) * | 2017-03-02 | 2017-09-05 | 杭州电子科技大学 | A kind of blurred picture non-blind deblurring method based on self-adaption gradient sparse model |
CN107369139A (en) * | 2017-06-27 | 2017-11-21 | 五邑大学 | A kind of noise image deblurring method being fitted based on blended data with weighting total variation |
Also Published As
Publication number | Publication date |
---|---|
CN108648162A (en) | 2018-10-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108648162B (en) | Gradient-related TV factor image denoising and deblurring method based on noise level | |
Teodoro et al. | Image restoration and reconstruction using variable splitting and class-adapted image priors | |
Cai et al. | Linearized Bregman iterations for frame-based image deblurring | |
US9262815B2 (en) | Algorithm for minimizing latent sharp image cost function and point spread function cost function with a spatial mask in a regularization term | |
Sun et al. | Postprocessing of low bit-rate block DCT coded images based on a fields of experts prior | |
CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
CN110796616B (en) | Turbulence degradation image recovery method based on norm constraint and self-adaptive weighted gradient | |
CN109360157B (en) | TV and wavelet regularization-based spatial variation blurred image restoration method | |
Javaran et al. | Non-blind image deconvolution using a regularization based on re-blurring process | |
Wang et al. | An improved image blind deblurring based on dark channel prior | |
Teranishi et al. | Improvement of robustness blind image restoration method using failing detection process | |
Motohashi et al. | A study on blind image restoration of blurred images using R-map | |
CN102682437A (en) | Image deconvolution method based on total variation regularization | |
Chen et al. | Inexact alternating direction method based on proximity projection operator for image inpainting in wavelet domain | |
CN113658061A (en) | Hybrid constraint condition-oriented blurred image restoration method and terminal | |
Sola et al. | Blind image deblurring based on multi-resolution ringing removal | |
Zuo et al. | Image denoising using nonlocal means with shape-adaptive patches and new weights | |
Kaur et al. | Study of Image Denoising and Its Techniques | |
Chen et al. | A structure-preserving image restoration method with high-level ensemble constraints | |
Zhang et al. | Parametric model for image blur kernel estimation | |
Choudhury et al. | A denoising inspired deblurring framework for regularized image restoration | |
Chalkov et al. | Post-processing algorithm for reducing ringing artefacts in deblurred images | |
Han et al. | Local Motion Blurred Image Restoration Based on the Reciprocal of DCT High-Frequency Mean Incremental Prior | |
Li et al. | Hybrid order L 0-regularized blur kernel estimation model for image blind deblurring | |
Simi et al. | Dependency of Optimum Value of Regularization Parameter in Total Variation Filter on Noise Statistics-A MR Phantom Study |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |