CN106339996B - A kind of Image Blind deblurring method based on super Laplace prior - Google Patents
A kind of Image Blind deblurring method based on super Laplace prior Download PDFInfo
- Publication number
- CN106339996B CN106339996B CN201610814545.5A CN201610814545A CN106339996B CN 106339996 B CN106339996 B CN 106339996B CN 201610814545 A CN201610814545 A CN 201610814545A CN 106339996 B CN106339996 B CN 106339996B
- Authority
- CN
- China
- Prior art keywords
- image
- intermediate image
- layer
- fuzzy core
- pyramid
- 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 41
- 238000010586 diagram Methods 0.000 claims abstract description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 7
- 238000005315 distribution function Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 2
- 238000011084 recovery Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 abstract description 2
- 238000013459 approach Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 2
- 239000010931 gold Substances 0.000 description 2
- 229910052737 gold Inorganic materials 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of Image Blind deblurring methods based on super Laplace prior, include the following steps：Read in blurred picture；Then image pyramid is constructed to it；Using first layer pyramid diagram picture as initial intermediate image；Conspicuousness edge is solved by intermediate image, then by conspicuousness edge calculations fuzzy core, intermediate image is then solved by fuzzy core；Judge this layer of fuzzy core and whether intermediate image meets the stop condition of iteration update, continue iteration update if being unsatisfactory for, otherwise enter pyramidal next layer, and using upper one layer of intermediate image finally acquired as next layer of initial intermediate image；The recovery of image can be realized based on non-blind deblurring method in conjunction with the optimal fuzzy core that the present invention estimates using pyramid bottom fuzzy core as the corresponding optimal fuzzy core of the blurred picture.The present invention finally improves the quality of image restoration by improving the estimation effect of intermediate image to increase the accuracy of fuzzy kernel estimates.
Description
Technical field：
The present invention relates to field of machine vision, a kind of Image Blind deblurring based on super Laplace prior especially set out
Method.
Background technique：
Image obtains the important sources of information as the mankind, occupys an important position in modern society.However for
For hand-held capture apparatus, jitter phenomenon often occurs during exposure, the image obtained is caused to thicken, this is tight
Ghost image rings the use and its subsequent processing of image.The purpose of image deblurring is exactly only to restore from the blurred picture deteriorated
Image clearly, comprising abundant information out.Fuzzy core refers to that capture apparatus is generated because shake occurs during exposure
Motion profile.Image-deblurring process can be divided into blind deblurring and non-blind deblurring according to whether fuzzy core is known.Wherein,
Image Blind deblurring is the deblurring process in the case where fuzzy core can not know situation, is a serious ill-posed problem, has
Very big challenge.The most important task of Image Blind deblurring is to estimate fuzzy core, later the problem of translated into it is non-blind
Deblurring problem is based ultimately upon the recovery that non-blind deblurring method realizes image.Image deblurring as Digital Image Processing and
The important subject of computer vision field, in photography, optics, astronomy, medical image, monitoring, remote sensing and army
The fields such as thing research are all with a wide range of applications, and a hot subject of academia's common concern in recent years, have
Very important theory significance and realistic meaning.
In recent years, aiming at the problem that blind deblurring serious ill-posedness, domestic and foreign scholars are being based on maximum a posteriori probability frame
New method is constantly proposed under frame.However in the deblurring method proposed under the frame, often there are problems that two.It is first
Effective image prior regular terms cannot be chosen, this will lead in fuzzy core estimation procedure, reduce the estimation matter of intermediate image
Amount, and then cause fuzzy kernel estimates inaccurate.In order to acquire fuzzy core, the method by thick scale to smart scale is generallyd use, every
Image and fuzzy core alternating iteration under a scale update, and finally obtain fuzzy core, the image during this is referred to as middle graph
Picture.Therefore the solution of intermediate image directly influences the quality of fuzzy kernel estimates.The regular terms followed by chosen may will solve
Process becomes the problem of non-convex sparse optimization, causes to be difficult to acquire globally optimal solution, therefore which kind of side selected in this case
Method also becomes a problem to solve.
The present invention problem inaccurate for intermediate image estimation, introduces super Laplace prior and carrys out regularization natural image
The heavytailed distribution characteristic of gradient, and solve the problems, such as that non-convex sparse solution optimizes by introducing broad sense soft-threshold operator, most
It is proposed a kind of Image Blind deblurring method based on super Laplace prior eventually, this method can effectively ambiguous estimation core, mention
The quality of high blur image restoration.
Summary of the invention：
The main object of the present invention is to propose a kind of Image Blind deblurring method based on super Laplace prior, using super
Laplace prior simulates the heavytailed distribution characteristic of natural image gradient, effectively solution intermediate image, and then is conducive to estimate mould
Paste core, the final quality improved to blur image restoration.
To achieve the goals above, the present invention provides the following technical solutions：
Step 1: reading in single width blurred picture B, judge whether B is gray level image, if not gray level image then carries out B
Gray processing processing；
Step 2: constructing image pyramid to BN is total number of plies of image pyramid, and i is image gold word
Tower is currently located layer, and i is initially 1；It willThe intermediate image I initial as first layer1；
Step 3: by intermediate image IiAcquire conspicuousness image border
Step 4: by conspicuousness image borderFuzzy core k is updated, is passed through
It carries out seeking optimal solution k,Indicate the gradient of i-th layer of pyramid diagram picture, γ is weight；
Step 5: updating intermediate image I by fuzzy core ki, pass throughIt is asked
Optimal solution Ii,Indicate the gradient of i-th layer of intermediate image, p indicates the parameter of super Laplce's distribution function, and λ is weight；
Step 6: whether the number j for judging that fuzzy core and intermediate image iteration update in i-th layer reaches default greatest iteration
Number m thens follow the steps seven if reaching, and otherwise the number of iterations j adds 1 and return step three, and the value of j is 1 to m；
Step 7: judging whether number of plies i reaches pyramid n-th layer, if reaching n-th layer, eight are thened follow the steps, otherwise the number of plies
I adds 1 and return step three, and by the intermediate image I of i+1 layeri+1It is initialized as Ii；
Step 8: the k that n-th layer is acquired, as the corresponding optimal fuzzy core K of blurred picture B, selection is suitably non-blind to go
Fuzzy algorithmic approach restores image final out.
Compared with prior art, the invention has the advantages that：
(1) through step 5 when solving intermediate image, the present invention is using super Laplace prior as intermediate image ladder
The regularization constraint item of degree, for compare Laplce or Gauss equal distribution, super laplacian distribution can more preferable simulation
The heavytailed distribution characteristic of intermediate image gradient, therefore intermediate image can be effectively solved, and then effectively solve fuzzy core, finally
Improve the quality of image restoration.
(2) it is directed to the intermediate image method for solving of step 5, present invention uses broad sense soft-threshold operators, well solve
The problem of non-convex sparse optimal value solution caused by due to introducing super Laplace prior, the problem of avoiding local optimum.
Therefore, the present invention is led in photography, optics, astronomy, medical image, monitoring, remote sensing and military field engineering etc.
Domain will be with a wide range of applications.
Detailed description of the invention：
Algorithm flow chart Fig. 1 of the invention；
The pyramid schematic diagram of Fig. 2 building blurred picture；
Fig. 3 natural image gradient distribution and its simulation schematic diagram；
The fuzzy kernel estimates schematic diagram of Fig. 4 legend；
The curve graph of the fuzzy core of Fig. 5 legend square error sum in estimation procedure；
The fuzzy core estimated result figure of image im01 in Fig. 6 Levin database；
Average structure similarity indices analysis chart of Fig. 7 algorithm in Levin database；
Average peak signal to noise ratio index analysis figure of Fig. 8 algorithm in Levin database；
Error rate tracing analysis figure of Fig. 9 algorithm in Levin database；
Recovery effect figure of Figure 10 algorithm for natural blurred picture.
Specific embodiment：
Purpose, specific steps and feature in order to better illustrate the present invention, with reference to the accompanying drawing to the present invention make into
One step detailed description：
With reference to Fig. 1, a kind of Image Blind deblurring method based on super Laplace prior proposed by the present invention mainly includes
Following steps：
Step 1: reading in single width blurred picture B, judge whether B is gray level image, if not gray level image then carries out B
Gray processing processing；
Step 2: constructing image pyramid to BN is total number of plies of image pyramid, and i is image gold word
Tower is currently located layer, and i is initially 1；It willThe intermediate image I initial as first layer1；
Step 3: by intermediate image IiAcquire conspicuousness image border
Step 4: by conspicuousness image borderFuzzy core k is updated, is passed through
It carries out seeking optimal solution k,Indicate the gradient of i-th layer of pyramid diagram picture, γ is weight；
Step 5: updating intermediate image I by fuzzy core ki, pass throughIt is asked
Optimal solution Ii,Indicate the gradient of i-th layer of intermediate image, p indicates the parameter of super Laplce's distribution function, and λ is weight；
Step 6: whether the number j for judging that fuzzy core and intermediate image iteration update in i-th layer reaches default greatest iteration
Number m thens follow the steps seven if reaching, and otherwise the number of iterations j adds 1 and return step three, and the value of j is 1 to m；
Step 7: judging whether number of plies i reaches pyramid n-th layer, if reaching n-th layer, eight are thened follow the steps, otherwise the number of plies
I adds 1 and return step three, and by the intermediate image I of i+1 layeri+1It is initialized as Ii；
Step 8: the k that n-th layer is acquired, as the corresponding optimal fuzzy core K of blurred picture B, selection is suitably non-blind to go
Fuzzy algorithmic approach restores image final out.
In above-mentioned technical proposal, n-layer image pyramid is constructed to blurred picture B in step 2, and n is by giving fuzzy core k
Size determine.With Levin database[1]Illustrate to construct pyramidal process for middle image im01-ker08.Image B's
Having a size of 255 × 255, the size of fuzzy core k is 23 × 23.N can be calculated according to following equation：
Wherein k1And k2The height and width of k are respectively indicated, floor indicates downward floor operation symbol.Here k1=k2=23, generation
Enter equations and obtains n equal to 5.The incremental image pyramid of final building five layers of resolution ratio as shown in Figure 2, resolution ratio are respectively
63 × 63,89 × 89,127 × 127,179 × 179 and 255 × 255, and the corresponding fuzzy core size of each layer is respectively 7 × 7,
9 × 9,13 × 13,17 × 17 and 23 × 23.
In above-mentioned technical proposal, by intermediate image I in step 3iFirst solve conspicuousness structureThen it acquires
Conspicuousness edge, specific solution procedure include：
(1) first from intermediate image IiMiddle extraction primary structure Is：
Wherein x indicates pixel；Ii(x) pixel value at x, I are indicateds(x) I is indicatedi(x) structure division；Ii(x)-Is
(x) I is indicatedi(x) detail section；Indicate gradient differential operator both horizontally and vertically；θ initial guess is set as 1, and
θ is updated according to θ/1.1 after above formula executes 1 time, i.e., with the incremental θ value of the number of iterations between intermediate image and fuzzy core
It is gradually reduced, so that the weight of image detail part gradually increases, therefore the calculating process after guarantee includes more details
Part.θ variable is shared by each layer fuzzy core solution procedure of pyramid, therefore only needs the first time iteration in pyramid first layer
Initialization is primary in the process.ω (x) is adaptive dependent variable, and
ω (x)=exp (- | r (x) |0.8),
Wherein
Nh(x) it is h × h window with central pixel point x, selecting h to be equal to 5, y indicates pixel in window area,Indicate the vector sum of pixel gradient in window area,Indicate window
The sum of the absolute value of pixel gradient in region.The flat region gradient value of image generally relatively, so vector sum compared with
It is small；And generally differ larger in the significant region gradient value of structure, so vector sum is larger.Therefore in contrast, r is flat in image
Smooth region takes smaller value, and takes the larger value in the significant region of picture structure.Can by way of adaptively adjusting weight,
It realizes effective selection to picture structure, i.e., improve effective structure for ambiguous estimation core and reduces the choosing to invalid structure
It takes.
(2) conspicuousness structure is acquired secondly by shock filter
Wherein Δ indicates Laplace operator,Indicate gradient operator, sign indicates sign function.Δ I=Ix 2Ixx+
2IxIyIxy+Iy 2Iyy, wherein IxAnd IyRespectively indicate the horizontal first differential with vertical direction, Ixx, IxyAnd IyyIndicate that second order is micro-
Point.
(3) then pass through conspicuousness structureSolve conspicuousness edge
It indicates unit binaryzation exposure mask function, is defined as follows：
WhereinExpression gradient magnitude, the threshold value of t expression gradient magnitude, initial value design 0.5, and whenever interior circulation
T updates in the way of t/1.1 after executing 1 time, i.e., threshold value is gradually reduced, so that operation later may include more
Structure.T variable is shared by each layer fuzzy core solution procedure of pyramid, therefore only needs the first time iteration in pyramid first layer
Initialization is primary in the process.
In above-mentioned technical proposal, pass through conspicuousness edge in step 4Fuzzy core k is solved, method for solving is：
Wherein γ is weight, γ=10-2, above formula can use iteration weighted least-squares method[3]It solves.
In above-mentioned technical proposal, the fuzzy core k that is acquired in step 5 by step 4 solves intermediate image Ii.Such as Fig. 3
Shown, the gradient distribution of natural image is in heavytailed distribution, and super laplacian distribution is relative to Gaussian Profile and La Pu as seen from the figure
Lars is distributed the gradient distribution situation that can preferably simulate natural image.Therefore intermediate image I can more effectively be solvedi, into
And during being iteratively solved with fuzzy core, it can be in the hope of more preferably fuzzy core.Intermediate image IiIt can be solved by following formula：
Wherein the usual value range of p is between 0.5 to 0.8, and being selected as 0.8, λ here according to experience is weight, λ=4 ×
10-3, gradient can be directed to for formula solutionIntroduce auxiliary variable G=(Gh,Gv)T[4], then above formula is converted into：
Wherein β is the weight of variation, and initial value is set to 2 λ.I is iteratively solved respectively to above formulaiAnd G, meet iteration ends item
After part, I is finally obtainedi, specific iterative step is as follows：
(1) first according to the following formula, I is first fixedi, solve optimal G：
Here solving the process of G is to seek non-convex sparse optimal problem, by introducing broad sense soft-threshold operator (GST)[5]It asks
Solution：
WhereinIndicate the minimum value of G,Indicate threshold value, sgn indicates sign function.
(2) then according to the following formula, fixed G, solves optimal Ii：
It can be solved with the method for least square：
Wherein F () and F-1() respectively indicates Fast Fourier Transform and inverse transformation；Indicate that fast Flourier becomes
The complex conjugate operation changed；WithRespectively indicate IiHorizontal and vertical gradient.If β<βmaxThen continue iteration, and β is become
For 2 β.Until β >=βmaxStop interative computation, obtains final intermediate image Ii, β heremaxIt is set as 105。
In above-mentioned technical proposal, for judging in i-th layer of pyramid, fuzzy core updates step 6 with intermediate image iteration
Whether number j reaches default maximum number of iterations m, thens follow the steps seven if reaching, and otherwise j adds 1 return step three.By a large amount of
Experiment weighed two aspect factor of accuracy rate and time, finally set the number of iterations m as 5.
In above-mentioned technical proposal, the k that step 8 acquires n-th layer is as the corresponding optimal fuzzy core K of blurred picture B, choosing
It selects suitable non-blind deblurring algorithm and restores image final out.The method of present invention selection total variation image reconstruction[6]To obtain
It is final as a result, the image X of final recovery can be solved by following formula：
Indicate the gradient of image X, λfFor weight, λf=10-3。
In above-mentioned technical proposal, Fig. 4 gives algorithm to the fuzzy kernel estimates mistake of Levin database im01-ker08 example
The schematic diagram of journey.Fig. 5 gives the curve graph of fuzzy core square error sum in estimation procedure of the example, estimates through the invention
It the error of the fuzzy core counted out and realistic blur core and is compared, it can be found that the increase error with the number of iterations gradually subtracts
It is few, until approximate convergence.Fig. 6 gives the fuzzy core estimated result comparison diagram that (a) is schemed in Levin database, (b) is original
Fuzzy core, (c) fuzzy core estimated for the present invention, the fuzzy core estimated from the visible present invention of qualitative angle are obscured with original
Core is almost the same.Fig. 7 gives average structure similarity indices figure of the algorithm in Levin database, structural similarity index
Closer to be 1 expression restore image it is higher in configuration aspects and original clear image similarity.Fig. 8 gives algorithm and exists
Average peak signal to noise ratio figure in Levin database, peak value signal-to-noise ratio is bigger, indicates that the quality restored is better.Fig. 9 gives calculation
Error rate of the method in Levin database and success rate curve graph, it is generally recognized that error rate image when being less than or equal to 2 region
It is visually acceptable, it is seen that in all images of database, the present invention has been accounted in this region close to 90%.
Figure 10 gives experimental result of the algorithm on naturally fuzzy image instance, it is seen that fuzzy core track is clear, and restored image knot
Fruit visually sees that clarity significantly improves, and image detail reduction degree is good.It is tested by experiment, experimental result is with qualitatively and quantitatively
It is illustrated, the validity of verification algorithm.
This patent simulates the gradient distribution of natural image with super Laplace prior, and combines and extract saliency
The method of structure obtains conspicuousness edge, effectively solves fuzzy core.This patent is solving intermediate image using fuzzy core
In the process, non-convex sparse optimization is efficiently solved the problems, such as using broad sense soft-threshold operator.By improving intermediate image
Estimation effect, to increase the accuracy of fuzzy kernel estimates, the final quality for improving image restoration.
A specific embodiment of the invention is elaborated above in conjunction with attached drawing, but the present invention is not limited to above-mentioned
Embodiment within the knowledge of a person skilled in the art can also be before not departing from present inventive concept
It puts and makes a variety of changes.
[1]http://www.wisdom.weizmann.ac.il/~levina/papers/
LevinEtalCVPR09Data.rar
[2]Jinshan Pan,Risheng Liu,Zhixun Su,and Xianfeng Gu,Kernel
estimation from salient structure for robust motion deblurring.IEEE Signal
Processing Letters,vol.20,pp.841-844,2013.
[3]Anat Levin,Rob Fergus,Fredo Durand,and Bill Freeman,Image and
depth from a conventional camera with a coded aperture.ACM Transactions on
Graphics,vol.26,no.3,pp.70-78,2007.
[4]Jinshan Pan,Zhe Hu,Zhixun Su,and Ming-Hsuan Yang,Deblurring text
images via L0-regularized intensity and gradient prior.IEEE Conference on
Computer Vision and Pattern Recognition,pp.2901-2908,2014.
[5]Wangmeng Zuo,Deyu Meng,Lei Zhang,Xiangchu Feng,and David Zhang,A
gen-eralized iterated shrinkage algorithm for non-convex sparse coding.IEEE
International Conference on Computer Vision,pp.217-224,2013.
[6]Yilun Wang,Junfeng Yang,Wotao Yin,and Yin Zhang,A new alternating
minimi-zation algorithm for total variation image reconstruction.SIAM
J.Image,vol.1,no.3,pp.248-272,2008.
Claims (2)
1. a kind of Image Blind deblurring method based on super Laplace prior, which is characterized in that include the following steps：
Step 1: reading in single width blurred picture B, judge whether B is gray level image, if not gray level image then carries out gray scale to B
Change processing；
Step 2: constructing image pyramid to BN is total number of plies of image pyramid, and i is image pyramid
It is currently located layer, and i is initially 1；It willThe intermediate image I initial as first layer1；
Step 3: by intermediate image IiAcquire conspicuousness image border
Step 4: by conspicuousness image borderFuzzy core k is updated, is passed through Into
Row seeks optimal solution k,Indicate the gradient of i-th layer of pyramid diagram picture, γ is weight；
Step 5: updating intermediate image I by fuzzy core ki, pass throughIt carries out asking optimal
Solve Ii,Indicate the gradient of i-th layer of intermediate image, p indicates the parameter of super Laplce's distribution function, and λ is weight；
Step 6: whether the number j for judging that fuzzy core and intermediate image iteration update in i-th layer reaches default maximum number of iterations
M thens follow the steps seven if reaching, and otherwise the number of iterations j adds 1 and return step three, and the value of j is 1 to m；
Step 7: judging whether number of plies i reaches pyramid n-th layer, if reaching n-th layer, eight are thened follow the steps, otherwise number of plies i adds 1
And return step three, and by the intermediate image I of i+1 layeri+1It is initialized as Ii；
Step 8: the k that n-th layer is acquired is as the corresponding optimal fuzzy core K of blurred picture B, it is multiple using non-blind deblurring algorithm
Original goes out final image.
2. a kind of Image Blind deblurring method based on super Laplace prior according to claim 1, which is characterized in that
The fuzzy core k that is acquired in step 5 by step 4 solves intermediate image Ii；Intermediate image IiIt can be solved by following formula：
λ is weight, can be directed to gradient for formula solutionIntroduce auxiliary variable G=(Gh,Gv)T, then above formula is converted into：
Wherein β is the weight of variation, and initial value is set to 2 λ；I is iteratively solved respectively to above formulaiAnd G, after meeting stopping criterion for iteration,
Finally obtain Ii, specific iterative step is as follows：
(1) first according to the following formula, I is first fixedi, solve optimal G：
Here solving the process of G is to seek non-convex sparse optimal problem, is solved by introducing broad sense soft-threshold operator (GST)：
WhereinIndicate the minimum value of G,Indicate threshold value, sgn indicates sign function；
(2) then according to the following formula, fixed G, solves optimal Ii：
It can be solved with the method for least square：
Wherein F () and F-1() respectively indicates Fast Fourier Transform and inverse transformation；Indicate Fast Fourier Transform
Complex conjugate operation；WithRespectively indicate IiHorizontal and vertical gradient；If β<βmaxThen continue iteration, and β is become into 2 β；
Until β >=βmaxStop interative computation, obtains final intermediate image Ii。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610814545.5A CN106339996B (en) | 2016-09-09 | 2016-09-09 | A kind of Image Blind deblurring method based on super Laplace prior |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610814545.5A CN106339996B (en) | 2016-09-09 | 2016-09-09 | A kind of Image Blind deblurring method based on super Laplace prior |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106339996A CN106339996A (en) | 2017-01-18 |
| CN106339996B true CN106339996B (en) | 2018-11-30 |
Family
ID=57822963
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610814545.5A Active CN106339996B (en) | 2016-09-09 | 2016-09-09 | A kind of Image Blind deblurring method based on super Laplace prior |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106339996B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194887A (en) * | 2017-05-08 | 2017-09-22 | 广州智慧城市发展研究院 | A kind of image deblurring method and system applied to car plate detection |
| CN107292836B (en) * | 2017-06-02 | 2020-06-26 | 河海大学常州校区 | Image blind deblurring method based on external image block prior information and sparse representation |
| CN108257096A (en) * | 2017-12-28 | 2018-07-06 | 三维通信股份有限公司 | A kind of blind deblurring method based on the collaboration of more priori |
| CN108305230A (en) * | 2018-01-31 | 2018-07-20 | 上海康斐信息技术有限公司 | A kind of blurred picture integrated conduct method and system |
| CN108830802B (en) * | 2018-05-16 | 2022-07-08 | 浙江大学 | Image blur kernel estimation method based on short exposure image gradient guidance |
| CN112581378B (en) * | 2019-09-30 | 2022-09-13 | 河海大学常州校区 | Image blind deblurring method and device based on significance strength and gradient prior |
| CN116091367B (en) * | 2023-04-10 | 2023-07-18 | 中国科学院空天信息创新研究院 | Blind deblurring method, device, equipment and medium for optical remote sensing image |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010122993A (en) * | 2008-11-20 | 2010-06-03 | Panasonic Electric Works Co Ltd | Face authentication system |
| CN102389309A (en) * | 2011-07-08 | 2012-03-28 | 首都医科大学 | Compressed sensing theory-based reconstruction method of magnetic resonance image |
| CN104112261A (en) * | 2014-07-17 | 2014-10-22 | 五邑大学 | Norm ratio regularization based rapid image blind deblurring method |
| CN105279740A (en) * | 2015-11-17 | 2016-01-27 | 华东交通大学 | Image denoising method based on sparse regularization |
| CN105493140A (en) * | 2015-05-15 | 2016-04-13 | 北京大学深圳研究生院 | Image deblurring method and system |
| CN105654069A (en) * | 2016-02-03 | 2016-06-08 | 江南大学 | Increment subspace target tracking method based on Lp norm regularization |
| CN105913392A (en) * | 2016-04-08 | 2016-08-31 | 西安电子科技大学昆山创新研究院 | Degraded image overall quality improving method in complex environment |
-
2016
- 2016-09-09 CN CN201610814545.5A patent/CN106339996B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010122993A (en) * | 2008-11-20 | 2010-06-03 | Panasonic Electric Works Co Ltd | Face authentication system |
| CN102389309A (en) * | 2011-07-08 | 2012-03-28 | 首都医科大学 | Compressed sensing theory-based reconstruction method of magnetic resonance image |
| CN104112261A (en) * | 2014-07-17 | 2014-10-22 | 五邑大学 | Norm ratio regularization based rapid image blind deblurring method |
| CN105493140A (en) * | 2015-05-15 | 2016-04-13 | 北京大学深圳研究生院 | Image deblurring method and system |
| CN105279740A (en) * | 2015-11-17 | 2016-01-27 | 华东交通大学 | Image denoising method based on sparse regularization |
| CN105654069A (en) * | 2016-02-03 | 2016-06-08 | 江南大学 | Increment subspace target tracking method based on Lp norm regularization |
| CN105913392A (en) * | 2016-04-08 | 2016-08-31 | 西安电子科技大学昆山创新研究院 | Degraded image overall quality improving method in complex environment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106339996A (en) | 2017-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106339996B (en) | A kind of Image Blind deblurring method based on super Laplace prior | |
| Chun et al. | Momentum-Net: Fast and convergent iterative neural network for inverse problems | |
| CN109377459B (en) | Super-resolution deblurring method of generative confrontation network | |
| CN107507135B (en) | Image reconstruction method based on coding aperture and target | |
| CN109712209A (en) | The method for reconstructing of PET image, computer storage medium, computer equipment | |
| CN106408524A (en) | Two-dimensional image-assisted depth image enhancement method | |
| CN111861906B (en) | Pavement crack image virtual augmentation model establishment and image virtual augmentation method | |
| CN105976330A (en) | Embedded foggy-weather real-time video image stabilization method | |
| CN110796616B (en) | Turbulence degradation image recovery method based on norm constraint and self-adaptive weighted gradient | |
| CN111275686B (en) | Method and device for generating medical image data for artificial neural network training | |
| CN112365514A (en) | Semantic segmentation method based on improved PSPNet | |
| CN108416803A (en) | A kind of scene depth restoration methods of the Multi-information acquisition based on deep neural network | |
| CN114581560B (en) | Multi-scale neural network infrared image colorization method based on attention mechanism | |
| Ding et al. | U 2 D 2 Net: Unsupervised unified image dehazing and denoising network for single hazy image enhancement | |
| CN104657951A (en) | Multiplicative noise removal method for image | |
| CN104252704A (en) | Total generalized variation-based infrared image multi-sensor super-resolution reconstruction method | |
| Liu et al. | Video frame interpolation via optical flow estimation with image inpainting | |
| CN115578262A (en) | Polarization image super-resolution reconstruction method based on AFAN model | |
| CN113870327B (en) | Medical image registration method based on prediction multi-level deformation field | |
| CN105261043A (en) | Video motion object detection method on the basis of significance detection | |
| CN103020905A (en) | Sparse-constraint-adaptive NLM (non-local mean) super-resolution reconstruction method aiming at character image | |
| CN106934398A (en) | Image de-noising method based on super-pixel cluster and rarefaction representation | |
| CN117097876B (en) | Event camera image reconstruction method based on neural network | |
| CN111401209B (en) | Action recognition method based on deep learning | |
| CN112837220A (en) | Method for improving resolution of infrared image and application thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |