CN109767404A - Infrared image deblurring method under a kind of salt-pepper noise - Google Patents
Infrared image deblurring method under a kind of salt-pepper noise Download PDFInfo
- Publication number
- CN109767404A CN109767404A CN201910070557.5A CN201910070557A CN109767404A CN 109767404 A CN109767404 A CN 109767404A CN 201910070557 A CN201910070557 A CN 201910070557A CN 109767404 A CN109767404 A CN 109767404A
- Authority
- CN
- China
- Prior art keywords
- image
- model
- noise
- salt
- formula
- 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.)
- Granted
Links
Landscapes
- Image Processing (AREA)
Abstract
The present invention proposes the novel regularization model of infrared image deblurring under salt-pepper noise, and using OGS-ATV as regular terms, fidelity term uses Lp norm；Based on the basic framework of ADMM method and MM method in optimization algorithm, the step of acceleration is restarted is introduced, the solution efficiency of algorithm is improved；In addition to this, in deblurring treatment process, difference operator is also considered as convolution operator by the present invention, to be handled model in a frequency domain using convolution theory, so as to avoid the calculating of large-scale matrix.
Description
Technical field
The present invention relates to field of image processings, and in particular to infrared image deblurring method under a kind of salt-pepper noise.
Background technique
Infrared image has the characteristics that ambient noise and interference are big, resolution ratio is low.The target of infrared imaging system detection is past
Toward in big visual field, complex background, signal-to-noise ratio it is low under conditions of, detection range is remote, and target pixel shared on imaging surface is very
Few, resolution ratio is low to be caused to lack enough information such as details, shape, color characteristic, so that the surveying tape to target is next difficult.Institute
With the image enhancement of infrared imaging system and the suppression technology of noise are the key technologies in current infrared image processing field
And difficult point.
Main source of the detector as infrared imaging system noise, mechanism of production is very complicated, is to influence infrared system
The principal element of picture quality.The noise of detector itself be it is unavoidable, according to its mechanism of production can be divided into thermal noise,
Shot noise, photon noise etc..These noise on image, which influence bigger part, can be equivalent to white Gaussian noise and green pepper
Salt noise.In addition, during image capture, due to various factors, such as the opposite fortune between defocus, diffraction, detector and object
Dynamic, random atmospheric turbulance and the noise of sensor etc. can cause the degeneration of observation image.
Image restoration is exactly the promotion to degraded image quality, and removal or mitigate occurs during obtaining digital picture
Image quality decrease, to reach the improvement of image visually.Most typical degradation phenomena is exactly fuzzy and noise, master of the present invention
The recovery problem of blurred picture, i.e. deblurring is discussed.
The blurring process of image can be modeled as the convolution of clear image and fuzzy core, add noise, i.e. g=h*f+n,
Wherein * indicates convolution operator, and g expression observes that the blurred picture comprising noise, f indicate original image, and k is fuzzy core, also known as
Point spread function (Point Spread Function, PSF)；N is noise.The inversely processing process of blurred picture is known as image and goes
Convolution, the purpose is to go out clearly image by blur ed image restoration.According to PSF whether it is known that image deconvolution problem is divided into
Image blind deconvolution and non-two class of blind deconvolution of image.
The non-blind deconvolution of image assumes that blurred picture and fuzzy core have all provided, and estimates clear image.In the extensive of image
In multiple processing, the non-blind deconvolution of image is an ill-condition problem, is usually modeled as following energy function using regularization method
Model is minimized to be solved:
Wherein: first item is data fidelity term；Section 2 is regular terms (or bound term, Regularization function)；μ is canonical
Change parameter, for controlling the weight ratio between fidelity term and regular terms.According to the difference of regular terms, different canonical is produced
Change method.Earliest regularization method is the Tikhonov regularization method that Tikhonov in 1977 et al. is proposed, is gone in image
Its regular terms is in fuzzy problemThe regular terms can effectively inhibit noise, but be also easy to produce smooth figure
Picture, so that processing result is still fuzzy.For the disadvantage for overcoming Tikhonov regularization method excessively smooth, Rudin et al. is proposed
Total variation regularization (Total variation, TV) method, its regular terms is in image deblurring problemTV regularization method can inhibit noise, retain the edge of image, but it is only capable of effectively approaching fragment constant letter
Number, so being also easy to produce alias in image smoothing region and reducing the recovering quality of image.In order to weaken the rank for restoring picture
Terraced effect and the marginal information for retaining image, Lysaker etc. propose it is a kind of go to replace with the full variational regularization of second order it is original complete
Variational regularization item.Chan etc. proposes a kind of full variational method of mixing, and single order and the full variation of second order are used in mixed way.Existing skill
Art, which also has, proposes that the image restoration model constructed based on high-order TV regular terms is being restored although can effectively inhibit alias
Smudgy problem is easily led to when the detailed information and important feature of image.Huang etc. is replaced by introducing auxiliary variable
True picture and propose a kind of quickly full variation (Fast-TV) and minimize method.Bredies et al. proposes total GENERALIZED VARIATIONAL
(total generalized variation, TGV) substitutes common TV regular terms, and the image restoration model of building can send out
The advantage of arbitrary order polynomial function can effectively be approached by waving TGV, and alias can be effectively inhibited during image restoration
It generates, protects the material particular information of image, improve the quality of image restoration.
But but when image is by impulse noise effect, foregoing model can not be from degraded image because of the change of assumed condition
Recover important information.According to the statistical nature of impulsive noise, therefore the TV model (2) based on L1 data fidelity term is suggested use
To solve the problems, such as such blur image restoration containing impulsive noise.
Similarly, because TV regular terms has the deficiency for being only capable of effectively approaching fragment constant function, the alias of generation is easy
Reduce the recovering quality of image.For this purpose, the prior art improves the visual quality of image restoration by introducing the modified of TV.It is existing
There is the NLTV model proposed in technology, which can inhibit alias and retain image detail information, but exists and calculate again
It is miscellaneous to spend high problem, it is difficult to be flexibly applied to Practical Project problem；The L1-HTV model that the prior art proposes can effectively overcome
The alias in image smoothing region, but the material particular feature of image cannot be effectively protected.
Liu et al. and Shi etc. restore the image of noise damage using the sparse regularization of overlapping group, in terms of mitigating alias
It is highly effective.Bai etc. proposes a kind of model that full variational regularization is solved based on overlapping direction multiplier method, and the model is to removal green pepper
Salt noise is highly effective, but general for random noise effect.
Summary of the invention
To solve the above-mentioned problems, the present invention proposes infrared figure under a kind of salt-pepper noise with higher algorithm solution efficiency
As deblurring method, include the following steps,
Construct image deblurring model；
Solve the Augmented Lagrangian Functions of image deblurring model.
Further,
Described image deblurring model is,
Symbol * indicates convolution operation symbol, F ∈ R in formulaN×NIndicate the image recovered by denoising model, G ∈ RN×NIt indicates
Image polluted by noise,For fidelity term, ROGSTVIt (F) is the overlapping sparse full variation regular terms of group, μ is fidelity
The coefficient of balance of item and regular terms, | | | |1The L1 norm of representing matrix, is defined asK1=[- 1,1]
Indicate lateral difference convolution kernel,Indicate longitudinal difference convolution kernel, μ is regularization parameter.
Further,
The Augmented Lagrangian Functions are,
Z in formula1,Z2, W, T are the auxiliary variable introduced, F ∈ RN×NIndicate the image recovered by denoising model, Vi(i=
It 1,2,3,4) is Lagrange multiplier, λi> 0, (i=1,2,3) are penalty factor.
Further,
It is described solve image deblurring model Augmented Lagrangian Functions the following steps are included:
Step A1: setting initial pointK=0, λ1,λ2,λ3, γ, μ, the size K of group2,Maximum inner iteration number is NIt；
Step A2: following formula calculating is respectively adoptedW(k+1)、T(k+1)、F(k+1)With
F(k+1)=F-1(rhs·/lhs),
Step A3: executing k plus 1 operation；
Step A4: terminating if meeting termination condition, if being unsatisfactory for termination condition repeats step A2 and step A3.
Termination condition is, | | PSNR (Fk+1,G)-PSNR(Fk,G)||2/||PSNR(Fk,G)||2> tol.
Further,
It is described solve image deblurring model Augmented Lagrangian Functions the following steps are included:
Step B1: setting initial pointK=0, λ1,λ2,λ3, γ, μ, η, the size K of group2,Maximum inner iteration number is NIt；
Step B2: following formula calculating is respectively adoptedT(k+1)、F(k+1)With
Step B3: it calculates
If
Then,
Otherwise,
D1 is to accelerate threshold value in formula, and η is that threshold value updates coefficient, and α 1 is accelerator coefficient.
Step B4: it is calculated separately using formula identical with step B3
Step B5: executing K plus 1 operation；
Step B6: terminating if meeting termination condition, otherwise repeats step B2, step B3, step B4, step B5.
Termination condition is, | | PSNR (Fk+1,G)-PSNR(Fk,G)||2/||PSNR(Fk,G)||2> tol.
The beneficial effects of the invention are as follows inventive algorithms to increase the constraint of Lp norm on the basis of overlapping group is sparse, will
The characteristics of overlapping sparse canonical constraint of group makes full use of neighborhood combination gradient to improve the otherness of smooth region and fringe region with
The ability of portraying that Lp enhances image gradient combines, and has obtained better image and has rebuild effect, in numerical result and vision
It all has a clear superiority in effect.
Detailed description of the invention
Fig. 1 is one embodiment of the invention flow chart.
Fig. 2 is the sparsity schematic diagram of one embodiment of the invention Lp normal form.
Specific embodiment
The present invention solve invention thinking of problems in background technique first is that, gone using infrared image under salt-pepper noise
Fuzzy novel regularization model, using OGS-ATV as regular terms, fidelity term uses Lp norm；Based on the ADMM in optimization algorithm
The basic framework of method and MM method, introduces the step of acceleration is restarted, and further improves the solution efficiency of algorithm；Except this it
Outside, in deblurring treatment process, difference operator is also considered as convolution operator by the present invention, thus using convolution theory, by model
It is handled in a frequency domain, so as to avoid the calculating of large-scale matrix.
The overlapping sparse regular terms (Overlapping Group Sparse Total Variation, OGSTV) of group.It is one
The non-separation regular terms of kind, can preferably keep the sparsity of objective function.The overlapping sparse regular terms of group not only considers image
The sparsity in difference domain has also excavated each neighborhood of a point difference information, so that the structural sparse for having excavated image gradient is special
Property.The difference of smooth region and borderline region can be improved by overlapping combination gradient, to inhibit the alias of TV model.
One-dimensional overlapping group of sparse regular terms can be extended to the overlapping sparse regular terms of group of two dimension, and be introduced into the full variation of anisotropy
Model, in the denoising reconciliation convolution problem of salt-pepper noise.Overlapping group of sparse regular terms can be used for Speckle noise
Removal.
In conventional model, Quan Bianfen is that but have many non-convex reconstruction models low in practice based on L1 norm
It is better than L1 norm sparse constraint reconstruction model under sample rate.It can be the non-convex of objective function with Lp norm minimum (0 < p < 1)
Optimization problem.Compared to L1 problem, since Lp problem is non-convex, Non-smooth surface under 0 < p <, 1 situation.Its solution is more multiple
It is miscellaneous.The algorithm for solving Lp problem mainly has three classes: iteration weighting L1 algorithm, iteration assign power least square method again and iteration threshold is calculated
Method.Lp (0 < p < 1) norm based on model parameter is same as L1 norm constraint to have the ability for promoting solution sparse.Under Lp norm
Limited equidistant characteristics constraint condition, greatly relaxes the limitation to observing matrix.
Therefore, the present invention is on the basis of overlapping group is sparse, using non-convex model to infrared image under salt-pepper noise into
Row denoising and deconvolution.
For convenience those skilled in the art understand that the present invention, is below illustrated OGS-TV:
If additive noise is salt-pepper noise, since it is with sparse statistical property, then data fidelity term is needed with L1 norm
It portrays, then as follows based on overlapping sparse full variation (OGS) the deblurring model modeling of group:
Wherein, symbol * indicates convolution operation symbol, F ∈ RN×NIndicate the image recovered by denoising model, G ∈ RN×NIt indicates
Image polluted by noise,For fidelity term, ROGSTVIt (F) is the overlapping sparse full variation regular terms of group, μ is fidelity
The coefficient of balance of item and regular terms.||·||1The L1 norm of representing matrix, is defined as
Wherein ROGSTV(F) it is defined as follows:
Wherein, Kh=[- 1,1],Horizontal and vertical difference convolution kernel is respectively indicated,
Gradient is combined for solving,It is defined as follows,
Wherein, K is combined value size.It represents less than or whole equal to the maximum of x
Numerical value.
It can be seen that, gradient is combined from formulaThe gradient information of neighborhood of pixel points is fully considered,
The gradient information of these neighborhood territory pixel points is recombinated in a manner of two norms, to improve smooth region and image border area
Otherness between domain.
When calculating pixel gradient by overlapping combination mode, the height that can highlight smooth region is made an uproar points of contamination and frontier district
The otherness of domain pixel, to be denoised more robustly.
Combination gradient is handled using L21 norm contraction operator or optimization minimizes (Majorization-
Minimization, MM) algorithm, the overlapping sparse denoising model of group can be solved:
WhereinIndicate the overlapping sparse regular terms of group,Expression scale is overlapping group of K × K
Sparse matrix.
According to MM algorithm, minimizes P (V) and need first to find a function Q (V, U), which has Q to all V, U
(V, U) >=P (V), and equal sign is set up when U=V.Accordingly, Q (V, the U) minimum value calculated every time is the excellent of P (V)
The calculating of change value, formula can be converted to following iteration,
In view of organizing the particularity of sparse regular terms, and notice the presence of following inequality,
Wherein equal sign is set up when U=V.
Formula is observed, and combines formula, it is availableOptimization item be shown below,
Formula can be write,
Wherein v is the vector form of matrix V, and C (U) is unrelated with V, can be considered the constant term about V.
It is a diagonal matrix, diagonal element is defined as follows,
In conjunction with formula and formula, can convert formula to following iteration optimization is topic,
Its iteration optimal solution is as follows,
V(k+1)=mat { (I+ γ D2(V(k)))-1v0},\*MERGEFORMAT (13)
WhereinIndicate unit matrix, v0It is V0Vector form, mat indicate moment of a vector array operator.
Therefore, the present invention can be integrated into the algorithm of following solution formula:
The MM algorithm of solution formula is illustrated below
1 initialization: initial point v=v0, γ, group size K2,ε, maximum number of iterations
NIt, k=0.
2 iteration:
V(k+1)=mat { (I+ γ D2(V(k)))-1v0},
K=k+1
Until | | V(k+1)-V(k)||2/||V(k)||2< ε or k > NIt.
3 obtain V(k)
For convenience those skilled in the art understand that the present invention, is below illustrated Lp pseudonorm.
Lp pseudonorm increases one degree of freedom compared to L1 norm, can preferably portray sparse gradient information.Fig. 2 is provided
The full variation contour of anisotropy based on L2, L1, Lp pseudonormSchematic diagram,
Middle L2 and L1 norm is the special case of Lp norm, and Lp norm is defined asLp pseudonorm is defined asIn figure by taking denoising as an example, it is assumed that image is polluted by standard deviation by the Gaussian noise of σ.It can be found that L2 model
Number contour and fidelity termIntersection point it is not sparse, and L1 norm contour and the intersection point of fidelity term are sparse, still
Vulnerable to noise pollution.The contour of Lp pseudonorm is then more robust to noise.
In summary consider, the full variation of anisotropy based on L1 norm is extended to as shown in formula, and by the model of parameter p
It encloses and is limited to 0 < p < 1
Wherein TVp(F) the sparse regular terms of anisotropy total variation based on Lp pseudonorm is indicated.
Referring to Fig.1, technical solution of the present invention is illustrated below.
The present invention, using Lp norm constraint, proposes a kind of new image deblurring on the overlapping sparse full Basis of Variational of group
Model:
For convenience of calculation, by KhAnd KvK is used respectively1And K2It indicates.Then formula is writeable are as follows:
Introduce auxiliary variable Z1,Z2, W, T, then formula can be changed into restricted problem:
Then obtain following Augmented Lagrangian Functions:
Wherein, Vi(i=1,2,3,4) is Lagrange multiplier, λi> 0, (i=1,2,3) are penalty factor.According to ADMM
Algorithm givesIt can calculate and change in next step in the following manner
Generation
1, fixed F=F(k),W=W(k), T=T(k), and by formula about Z1And Z2It minimizes.Then
Minimum value can be obtained by following formula:
Above formula can be calculated by MM algorithm 1.
It can be obtained with soft-threshold shrinking calculation:
Wherein
For
It is solved are as follows:
Wherein PΩTo be defined in set omega={ F ∈ R for normalized imageN×N| 0≤F≤1 } on projection operator,
For F problem,
To above formula derivation, and enabling it is 0, then has
F(k+1)=F-1(rhs·/lhs),\*MERGEFORMAT (31)
Finally multiplier is updated, is obtained:
Therefore, the corresponding all problems of model are all addressed.
The specific embodiment of the present invention realization above process are as follows:
Step A1: initial pointK=0, λ1,λ2,λ3, γ, μ, the size K of group2,Maximum inner iteration times N It
Step A2: it is calculated using formulaWith
W is calculated using formula(k+1)；
T is calculated using formula(k+1)；
Using formula and calculate F(k+1)；
Multiplier is updated using formula
Step A3: step A3: k being executed plus 1 operates；
Step A4: terminating if meeting termination condition, if being unsatisfactory for termination condition repeats step A2 and step A3.
The acceleration ADMM proposed in another embodiment according to Goldstein is theoretical, and the present invention can introduce acceleration variableVariableAnd dual variable
So the present invention uses the embodiment for accelerating to restart ADMM mode are as follows:
Step B1: setting initial pointK=0, λ1,λ2,λ3, γ, μ, η, the size K of group2,Maximum inner iteration number is NIt；
Step B2: following formula calculating is respectively adoptedT(k+1)、F(k+1)With
Step B3: it calculates
If
Then,
Otherwise,
Step B4: it is calculated separately using formula identical with step B3
Step B5: executing K plus 1 operation；
Step B6: terminating if meeting termination condition, otherwise repeats step B2, step B3, step B4, step B5.
The advantages of below by specific embodiment come to the present invention compared with the existing technology, is illustrated:
Mentioned method and the OGSATVL1 method of Liu are 30%~60% in salt-pepper noise grade by the present invention, and mould
Pasting core is respectively Gaussian Blur core fspecial (' gaussian ', 7,5), fspecial (' gaussian ', 15,5) and mean value
Under the conditions of three kinds of fuzzy core, the processing result of 6 width test picture is compared.For OGSATVL1 method, the present invention is to protect
Maximum PSNR value can be obtained by demonstrate,proving every width figure, its parameter is separately provided.The parameter of the mentioned method of the present invention is still according to preceding
The method of stating is configured.
Firstly, setting salt-pepper noise grade is distinguished under the conditions of Gaussian Blur core fspecial (' gaussian ', 7,5)
It is 30%~60%, PSNR, SSIM and ReE numerical value of 6 width test images is analyzed, numerical result is as shown in table 1.
Data comparison, which can be carried out, from table 1 finds out that the method for the present invention is to all test images in all noise grades
Processing result be all substantially better than OGSATVL1 method as a result, and as noise grade steps up, gap also incrementally increases.
1 Gaussian Blur core of table is fspecial (' gaussian ', 7,5)
Secondly, setting salt-pepper noise grade is distinguished under the conditions of Gaussian Blur core special (' gaussian ', 15,5)
It is 30%~60%, PSNR, SSIM and ReE numerical value of 6 width test images is analyzed, numerical result is as shown in table 2.
In terms of data result, all 6 test pictures, PSNR the and ReE value of the method for the present invention are all substantially better than
OGSATVL1 method as a result, but 2 open test picture in noise grade 30%, 4 open test picture when 40%, and 2 open when 50%
Picture is tested, the 2 SSIM value for opening test picture when 60%, the result of OGSATVL1 method is better than the mentioned method of the present invention, consideration
To the present invention when carrying out parameter selection mainly with PSNR value to compare, so SSIM value only makes reference herein.
2 Gaussian Blur core of table is fspecial (' gaussian ', 15,5)
Finally, by the method for the present invention and OGSATVL1 in 7 × 7 mean value fuzzy cores and item of the noise grade from 30%~60%
The comparison of ambiguity solution treatment effect is carried out under part.In terms of numerical analysis, the method for the present invention is in all 6 width pictures in different noise grades
Under the conditions of processing structure be all substantially better than OGSATVL1's as a result, and increase with noise grade, the processing result of the two is poor
Away from obviously becoming larger.
3 7*7 mean value fuzzy core data comparison of table
The beneficial effects of the present invention are:
Inventive algorithm increases the constraint of Lp norm on the basis of overlapping group is sparse, and overlapping group of sparse canonical is constrained
The characteristics of making full use of the otherness of neighborhood combination gradient raising smooth region and fringe region enhances image gradient with Lp
The ability of portraying combines, and has obtained better image and has rebuild effect, all has on numerical result and visual effect obvious excellent
Gesture.
For L1 algorithm using OGS-TV as regular terms, fidelity term is L1 norm, no matter from Gaussian Blur core or mean value fuzzy core,
And different grades of salt-pepper noise comparing result is seen, all there is certain gap in treatment effect and the method for the present invention, especially exist
Can be seen that L1 algorithm in visual effect, there are the alias of part in image edge processing.
Two kinds of algorithms of ITV and ATV are superimposed the processing of Gaussian Blur core and salt-pepper noise under fidelity term L1 norm condition,
Entirety is seen all undesirable, and processing result is poor in several comparison algorithms, generally existing edge clear but image it is integrally fuzzy or
Excessively smooth phenomenon causes image to be integrally distorted seriously.
L0 algorithm fidelity term is L0 normal form, for superposition Gaussian Blur core and salt-pepper noise image procossing from numerical result
Look down in L1 method and the method for the present invention；In terms of visual effect, in processing result, the alias at edge and smooth region
Blocking artifact is all more obvious.
In addition, the ADMM step for accelerating to restart is also introduced in inventive algorithm, so that there is the number of iterations in treatment process
It is decreased obviously, improves the operational efficiency of algorithm.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it is still
Technical solution documented by foregoing embodiments is modified, or is equally replaced to some or all of the technical features
It changes；And these are modified or replaceed, the model for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution
It encloses, should all cover within the scope of the claims and the description of the invention.
Claims (5)
1. infrared image deblurring method under a kind of salt-pepper noise, which is characterized in that include the following steps,
Construct image deblurring model；
Solve the Augmented Lagrangian Functions of image deblurring model.
2. infrared image deblurring method under a kind of salt-pepper noise as described in claim 1, which is characterized in that described image is gone
Fuzzy model is,
Symbol * indicates convolution operation symbol, F ∈ R in formulaN×NIndicate the image recovered by denoising model, G ∈ RN×NIt indicates by noise
The image of pollution,For fidelity term, ROGSTVIt (F) is the overlapping sparse full variation regular terms of group, μ is for fidelity term and just
The then coefficient of balance of item, | | | |1The L1 norm of representing matrix, is defined asK1=[- 1,1] indicate horizontal
To difference convolution kernel,Indicate longitudinal difference convolution kernel, μ is regularization parameter.
3. infrared image deblurring method under a kind of salt-pepper noise as claimed in claim 2, which is characterized in that the augmentation is drawn
Ge Lang function is,
Z in formula1,Z2, W, T are the auxiliary variable introduced, F ∈ RN×NIndicate the image recovered by denoising model, Vi(i=1,2,
It 3,4) is Lagrange multiplier, λi> 0, (i=1,2,3) are penalty factor.
4. infrared image deblurring method under a kind of salt-pepper noise as claimed in claim 3, which is characterized in that the solution figure
As deblurring model Augmented Lagrangian Functions the following steps are included:
Step A1: setting initial pointK=0, λ1,λ2,λ3, γ, μ, the size K of group2, Vi (0)=0, i=1,2,
3,4, maximum inner iteration number is NIt；
Step A2: following formula calculating is respectively adoptedW(k+1)、T(k+1)、F(k+1)And Vi (k+1), i=1,2,3,
4,
F(k+1)=F-1(rhs·/lhs),
Step A3: executing k plus 1 operation；
Step A4: terminating if meeting termination condition, if being unsatisfactory for termination condition repeats step A2 and step A3.
5. infrared image deblurring method under a kind of salt-pepper noise as claimed in claim 3, which is characterized in that the solution figure
As deblurring model Augmented Lagrangian Functions the following steps are included:
Step B1: setting initial pointK=0, λ1,λ2,λ3, γ, μ, η, the size K of group2, Vi (0)=0, i=1,
2,3,4, maximum inner iteration number is NIt；
Step B2: following formula calculating is respectively adoptedT(k+1)、F(k+1)And Vi (k+1), i=1,2,3,
4,
Step B3: it calculatesIf d1 (k +1)< η d1 (k),
Then,
Otherwise,
D in formula1It is to accelerate threshold value, η is that threshold value updates coefficient, α1For accelerator coefficient.
Step B4: it is calculated separately using formula identical with step B3
Step B5: executing K plus 1 operation；
Step B6: terminating if meeting termination condition, otherwise repeats step B2, step B3, step B4, step B5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910070557.5A CN109767404B (en) | 2019-01-25 | 2019-01-25 | Infrared image deblurring method under salt and pepper noise |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910070557.5A CN109767404B (en) | 2019-01-25 | 2019-01-25 | Infrared image deblurring method under salt and pepper noise |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109767404A true CN109767404A (en) | 2019-05-17 |
| CN109767404B CN109767404B (en) | 2023-03-31 |
Family
ID=66455191
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910070557.5A Active CN109767404B (en) | 2019-01-25 | 2019-01-25 | Infrared image deblurring method under salt and pepper noise |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109767404B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233046A (en) * | 2020-11-16 | 2021-01-15 | 山东科技大学 | Image restoration method under Cauchy noise and application thereof |
| CN112991194A (en) * | 2021-01-29 | 2021-06-18 | 电子科技大学 | Infrared thermal wave image deblurring method based on depth residual error network |
| CN114299004A (en) * | 2021-12-25 | 2022-04-08 | 宁波凯德科技服务有限公司 | Image semantic segmentation method based on pepper salt noise total variation data augmentation |
| CN114663304A (en) * | 2022-03-16 | 2022-06-24 | 中国科学院光电技术研究所 | Sparse prior-based optical synthetic aperture system imaging enhancement method |
| CN117132505A (en) * | 2023-10-20 | 2023-11-28 | 武夷山市玉琼生物科技有限公司 | Motion deblurring method and device based on auto-correlation of texture image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709877A (en) * | 2016-11-11 | 2017-05-24 | 天津大学 | Image deblurring method based on multi-parameter regular optimization model |
| CN108198149A (en) * | 2018-01-24 | 2018-06-22 | 闽南师范大学 | A kind of image deblurring method |
| CN108416723A (en) * | 2018-02-07 | 2018-08-17 | 南京理工大学 | Based on full variational regularization and variable divide without lens imaging method for fast reconstruction |
| CN108416750A (en) * | 2018-03-08 | 2018-08-17 | 闽南师范大学 | A kind of image recovery method |
| CN109146797A (en) * | 2018-06-15 | 2019-01-04 | 闽南师范大学 | A kind of impulsive noise ancient book image inpainting method sparse based on Lp pseudonorm and overlapping group |
-
2019
- 2019-01-25 CN CN201910070557.5A patent/CN109767404B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709877A (en) * | 2016-11-11 | 2017-05-24 | 天津大学 | Image deblurring method based on multi-parameter regular optimization model |
| CN108198149A (en) * | 2018-01-24 | 2018-06-22 | 闽南师范大学 | A kind of image deblurring method |
| CN108416723A (en) * | 2018-02-07 | 2018-08-17 | 南京理工大学 | Based on full variational regularization and variable divide without lens imaging method for fast reconstruction |
| CN108416750A (en) * | 2018-03-08 | 2018-08-17 | 闽南师范大学 | A kind of image recovery method |
| CN109146797A (en) * | 2018-06-15 | 2019-01-04 | 闽南师范大学 | A kind of impulsive noise ancient book image inpainting method sparse based on Lp pseudonorm and overlapping group |
Non-Patent Citations (3)
| Title |
|---|
| 彭真明，等.: "基于稀疏表示及正则约束的图像去噪方法综述", 《数据采集与处理》 * |
| 胡悦，等.: "基于增广拉格朗日乘子的快速高阶全变分图像去噪方法", 《系统工程与电子技术》 * |
| 黄志强.: "基于低秩表达的图像去噪算法研究", 《信息科技》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233046A (en) * | 2020-11-16 | 2021-01-15 | 山东科技大学 | Image restoration method under Cauchy noise and application thereof |
| CN112991194A (en) * | 2021-01-29 | 2021-06-18 | 电子科技大学 | Infrared thermal wave image deblurring method based on depth residual error network |
| CN112991194B (en) * | 2021-01-29 | 2022-06-24 | 电子科技大学 | Infrared thermal wave image deblurring method based on depth residual error network |
| CN114299004A (en) * | 2021-12-25 | 2022-04-08 | 宁波凯德科技服务有限公司 | Image semantic segmentation method based on pepper salt noise total variation data augmentation |
| CN114663304A (en) * | 2022-03-16 | 2022-06-24 | 中国科学院光电技术研究所 | Sparse prior-based optical synthetic aperture system imaging enhancement method |
| CN114663304B (en) * | 2022-03-16 | 2023-09-19 | 中国科学院光电技术研究所 | Sparse prior-based optical synthetic aperture system imaging enhancement method |
| CN117132505A (en) * | 2023-10-20 | 2023-11-28 | 武夷山市玉琼生物科技有限公司 | Motion deblurring method and device based on auto-correlation of texture image |
| CN117132505B (en) * | 2023-10-20 | 2023-12-22 | 武夷山市玉琼生物科技有限公司 | Motion deblurring method and device based on auto-correlation of texture image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109767404B (en) | 2023-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109767404A (en) | Infrared image deblurring method under a kind of salt-pepper noise | |
| Chen et al. | Robust image and video dehazing with visual artifact suppression via gradient residual minimization | |
| CN108122197B (en) | Image super-resolution reconstruction method based on deep learning | |
| CN107025632B (en) | Image super-resolution reconstruction method and system | |
| CN106920220B (en) | The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method | |
| CN109146797B (en) | A kind of impulsive noise ancient book image inpainting method sparse based on Lp pseudonorm and overlapping group | |
| CN110796616B (en) | Turbulence degradation image recovery method based on norm constraint and self-adaptive weighted gradient | |
| CN107993208A (en) | It is a kind of based on sparse overlapping group prior-constrained non local full Variational Image Restoration method | |
| CN110830043B (en) | Image compressed sensing reconstruction method based on mixed weighted total variation and non-local low rank | |
| CN106709879A (en) | Spatial variation point diffusion function smoothing method based on simple lens calculating imaging | |
| CN111260571B (en) | Depth image restoration method based on non-convex low-rank low gradient | |
| CN113284061B (en) | Underwater image enhancement method based on gradient network | |
| CN102236887A (en) | Motion-blurred image restoration method based on rotary difference and weighted total variation | |
| CN112233046B (en) | Image restoration method under Cauchy noise and application thereof | |
| CN108629741B (en) | Fuzzy kernel estimation method based on L0 and L1 regular terms | |
| CN115358953B (en) | Turbulence removing method based on image registration and dynamic target fusion | |
| CN108171654A (en) | Chinese character image super resolution ratio reconstruction method with AF panel | |
| Tsutsui et al. | Halo artifacts reduction method for variational based realtime retinex image enhancement | |
| Liu et al. | Total variation with overlapping group sparsity and Lp quasinorm for infrared image deblurring under salt-and-pepper noise | |
| JP2020061080A (en) | Image processing device, imaging device, and image processing method | |
| Wang et al. | An improved image blind deblurring based on dark channel prior | |
| CN111462022A (en) | Underwater image sharpness enhancement method | |
| CN105654431B (en) | It is a kind of to there are the image deblurring methods of circumstance of occlusion | |
| CN109636735B (en) | Rapid video defogging method based on space-time consistency constraint | |
| CN116342443A (en) | Near infrared and visible light image fusion method and system |
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 |