CN108681995A - A method of motion blur is gone based on variation Bayesian Estimation - Google Patents
A method of motion blur is gone based on variation Bayesian Estimation Download PDFInfo
- Publication number
- CN108681995A CN108681995A CN201810484731.6A CN201810484731A CN108681995A CN 108681995 A CN108681995 A CN 108681995A CN 201810484731 A CN201810484731 A CN 201810484731A CN 108681995 A CN108681995 A CN 108681995A
- Authority
- CN
- China
- Prior art keywords
- image
- motion blur
- bayesian estimation
- distribution
- variation bayesian
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000009826 distribution Methods 0.000 claims abstract description 46
- 230000009466 transformation Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 4
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 230000015556 catabolic process Effects 0.000 claims description 3
- 238000006731 degradation reaction Methods 0.000 claims description 3
- 235000015170 shellfish Nutrition 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 16
- 238000001914 filtration Methods 0.000 abstract description 2
- 238000011084 recovery Methods 0.000 description 9
- 230000008859 change Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013398 bayesian method Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000003706 image smoothing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000000007 visual effect Effects 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
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
A kind of method for removing motion blur based on variation Bayesian Estimation of the present invention, on the basis of variation Bayesian Estimation and natural image gradient statistical property combine, for the ringing effect occurred in recuperation, the method for proposing to go ring in restoring based on loop boundary average statistical filtering image finally uses Richardson Lucy algorithms to image restoration.Variation Bayesian Estimation core concept is with a more tractable approximate full Posterior distrbutionp come the Posterior probability distribution of approaching to reality.Method provided by the invention solves quick, and robustness is high, can be effectively removed and be generated because of camera shake fuzzy in image, and while keeping image border and details, can preferably reduce the influence of noise and ringing effect to image restoration quality.
Description
Technical field
The invention belongs to technical field of image recovery, are related to a kind of side for removing motion blur based on variation Bayesian Estimation
Method.
Background technology
In recent years, motion blur image deblurring technology is always in a hot research in image restoration technology research
Hold, it not only has important theory significance, also there is active demand in practical applications, and such as intelligent vehicle driving technology criminal is detectd
Look into etc..
There are several factors that can cause the motion blur of image, following two are segmented into according to different fuzzy generation types
Class:The focus adjustment when shake of imaging device is inaccurate, and the object to be imaged is quickly removed.For this kind of motion blur
Image, domestic and international all multi-experts and scholar motion blur image restoration is expanded study and achieve a series of researchs at
Fruit.Motion blur image restoration belongs to blind recovery field, and there are many blind restoration method, wherein the method being concerned at present mainly has
Two classes:Regularization method and method based on Bayes.In the prior art, existing image goes motion blur method there are bright
Aobvious noise phenomenon and ringing effect, affects the quality of restored image.
Invention content
The present invention provides a kind of methods for removing motion blur based on variation Bayesian Estimation, effectively remove image
It is middle to be generated because of camera shake fuzzy, it solves apparent noise phenomenon and ringing effect during restored image, is a kind of
Recovery effect is good, and the blind method for removing motion blur that robustness is high.
To achieve the goals above, this invention takes the following technical solutions:
Step 1:Establish motion blur image degradation function model.The fuzzy process that degrades of image can use following mathematics
Model describes, as shown in the formula (1):
Wherein f indicates that former clear image, h indicate that fuzzy core is the function that degrades, and g indicates that blurred picture, n indicate noise.Root
According to Bayes principle, then there is following formula establishment:
Step 2:Select suitable probabilistic model, including the prior probability model of original image and degrade function and noise
Prior probability model.
What image grey level histogram reflected is the frequency that some gray value occurs in image, it is also assumed that being gradation of image
The estimation of probability density.Gauss hybrid models are exactly accurately to quantify thing with Gaussian probability-density function (normal distribution curve)
Object, it is one and things is decomposed into several models formed based on Gaussian probability-density function (normal distribution curve).Zero
Mean value gauss hybrid models being capable of approximate " heavy-tailed " distribution well.So using zero-mean mixed Gauss model come approximate long-tail
Distribution, it can provide the experience distribution of good approximation, while our algorithm being made to have easy-to-handle estimation procedure.Assuming that
Each element is independent identically distributed in { ▽ f }, then the probability density function of clear image gradient is represented by Gaussian Mixture mould
Type function:
Wherein C indicates the sum of zero-mean gaussian model, vcAnd πcThe variance of the C zero-mean gaussian model is indicated respectively
And weight, N indicate that Gaussian Profile, i indicate the index of pixel in image.Mixed Gauss model uses C (essentially 3 to 5) a height
This model come characterize each pixel in image feature thus herein test in take C=4.vcAnd πcValue use it is classical
EM algorithms (expectation maximization) solve next.
Since all functions that degrade all are intended to sparsity, that is, most numerical value in the function that degrades is zero, and nonzero value shape
At a kind of path of similar curve, decline faster than Gaussian Profile, is similar to exponential distribution.Therefore referred to herein using mixing
Number distribution models the function that degrades.Assuming that the element in the function that degrades is independent identically distributed, in order to ensure the conservation of energy,
It is required that the value in the function that degrades be all higher than equal to zero and and be 1, then:
Wherein D indicates the sum of exponential distribution model, λdAnd πdThe scale factor and power of d-th of exponential distribution are indicated respectively
Weight, E indicate exponential distribution, and j indicates to degrade the index of element in function.It is found in many experiments, as D=4, effect is more preferable.
In image processing field, usually assume that noise is the Gaussian noise of zero-mean, therefore also use zero-mean high herein
Prior model of this model as noise.Convolution (2) can obtain known to ▽ f and h:
Wherein σ2Indicate that the variance of noise, N indicate that Gaussian Profile, i indicate the index of pixel in image.
In order to solve the estimation problem of formula (2), variational Bayesian method provides a kind of effective solution thinking, basic
Thought is to carry out the Posterior probability distribution p (h, ▽ f | ▽ g) of approaching to reality with a more tractable APPROXIMATE DISTRIBUTION q (h, ▽ f).Mirror
The distance between two distributions can be weighed in KL divergences, therefore by minimizing between APPROXIMATE DISTRIBUTION and true Posterior probability distribution
KL divergences realize.Using the variance of noise as the unknown quantity in variation Bayesian Estimation, therefore APPROXIMATE DISTRIBUTION q (h, ▽ f)
Just it has been rewritten into q (h, ▽ f, σ2).KL divergences between the approximate Posterior distrbutionp of definition and true Posterior probability distribution are:
It notices that p (▽ g) is constant always in entire estimation procedure, therefore can define a cost function CKLTo obtain
The optimal value of APPROXIMATE DISTRIBUTION:
Variation Bayesian Estimation realizes cost function by iterative manner (i.e. variation Bayes expectation maximization theorem)
It minimizes, to estimate the function that degrades.It can be obtained by maximal possibility estimation:
Then EM algorithms are used:
M step:▽g(t+1)←arg▽g max F(qh (t+1)(h),▽g). (9)
It needs that some parameters are previously set:One rectangular block of full-size, blurred picture including fuzzy core, fuzzy side
To the selection with Color Channel.Fuzzy core can be obtained by the estimation of variation Bayes.
Step 3:The method for going ring based on loop boundary image is used to motion blur image.
The algorithm is mainly carried out in two steps, and the first step is the Alpha's mean filter being first modified to image, keeps it full
The smooth requirement of sufficient gradient, obtains extrapolated value image.Second step is using the extrapolated value image acquired in the first step, to fill out
Outer image border is filled, so that the data of the image after extension is had periodically, meets the requirement of Fourier transformation.
3.1 calculate extrapolated value image
Due to ringing effect typically occur in the gradient at image strong edge it is precipitous near, the strong edge of image shows as figure
As the pace of change of gradient, therefore we will allow the image gradient of extension outer boundary to change " slow ".According to this idea, calculate
The extrapolated value image without prominent edge is acquired, the image completion is recycled to extend.Concrete operation step is as follows:
Alpha's mean filter is carried out first, this is to eliminate the noise of image.Then gradient minimisation is carried out.Pass through
Calculate the filling image of extension so that the sum of gradient minimum.
3.2 reflection interpolation fillings
Due to carrying out image restoration using Fourier transformation in frequency domain, observed image g boundary pixels is needed to meet periodically.
Image is h after remembering border extension, and h image sizes are 2m*2n.Image boundary after extension is smoothly continuous, and meets the period and follow
Ring, this helps greatly to inhibit boundary ring.
Step 4:Motion blur image is restored with Richardson-Lucy algorithms.Richardson-Lucy algorithms
It is a kind of Iterative nonlinear restoration algorithm, it is extracted from maximum likelihood formula, and image is modeled with Poisson distribution
, iterative formula is as follows.
WhereinConvolution is represented, g is degraded image, and h represents point spread function, and f represents the image of estimation, and k is iteration time
Number, reasonably value 5 to 30 will produce a large amount of noise if iterations are excessive, be unfavorable for the recovery of image.
Compared with prior art, the advantageous effect of technical solution of the present invention is:Robustness is high, can be effectively removed image
It is middle generated because of camera shake it is fuzzy, and while keeping image border and details, can preferably reduce noise and
Influence of the ringing effect to image restoration quality.
Description of the drawings
The overview flow chart of Fig. 1 present invention;
The detailed step flow chart of Fig. 2 present invention;
The algorithm schematic diagram of Fig. 3 variation Bayesian Estimations;
Fig. 4 border extension schematic diagrames;
The blurred picture and the image contrast figure after recovery of the test used in Fig. 5 (a) and Fig. 5 (b) present invention.
Specific implementation mode
The present invention provides a kind of methods for removing motion blur based on variation Bayesian Estimation, effectively remove image
It is middle to be generated because of camera shake fuzzy, it solves apparent noise phenomenon and ringing effect during restored image, is a kind of
Recovery effect is good, and the blind method for removing motion blur that robustness is high.
To better understand the objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and specific real
Mode is applied the present invention is further described in detail.It should be noted that the certain components of attached drawing have omission, amplification or contracting
It is small, do not represent the size of actual product;The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
As shown in Fig. 2, a kind of method for removing motion blur based on variation Bayesian Estimation, specifically includes following steps:
Step 1:Establish motion blur image degradation function model.The fuzzy process that degrades of image can use following mathematics
Model describes, as shown in the formula (1):
Wherein f indicates that former clear image, h indicate that fuzzy core is the function that degrades, and g indicates that blurred picture, n indicate noise.Root
According to Bayes principle, then there is following formula establishment:
Observation type (2) is it is found that one of the critical issue of the blindly restoring image algorithm based on Bayes principle is how to select
Select the prior probability model of the prior probability model of suitable original image and the prior probability model for the function that degrades and noise.
Step 2:Select suitable probabilistic model, including the prior probability model of original image and degrade function and noise
Prior probability model.
What image grey level histogram reflected is the frequency that some gray value occurs in image, it is also assumed that being gradation of image
The estimation of probability density.Gauss hybrid models are exactly accurately to quantify thing with Gaussian probability-density function (normal distribution curve)
Object, it is one and things is decomposed into several models formed based on Gaussian probability-density function (normal distribution curve).Zero
Mean value gauss hybrid models being capable of approximate " heavy-tailed " distribution well.So using zero-mean mixed Gauss model come approximate long-tail
Distribution, it can provide the experience distribution of good approximation, while our algorithm being made to have easy-to-handle estimation procedure.Assuming that
Each element is independent identically distributed in { ▽ f }, then the probability density function of clear image gradient is represented by Gaussian Mixture mould
Type function:
Wherein C indicates the sum of zero-mean gaussian model, vcAnd πcThe variance of the C zero-mean gaussian model is indicated respectively
And weight, N indicate that Gaussian Profile, i indicate the index of pixel in image.Mixed Gauss model uses C (essentially 3 to 5) a height
This model come characterize each pixel in image feature thus herein test in take C=4.vcAnd πcValue use it is classical
EM algorithms (expectation maximization) solve next.
Since all functions that degrade all are intended to sparsity, that is, most numerical value in the function that degrades is zero, and nonzero value shape
At a kind of path of similar curve, decline faster than Gaussian Profile, is similar to exponential distribution.Therefore referred to herein using mixing
Number distribution models the function that degrades.Assuming that the element in the function that degrades is independent identically distributed, in order to ensure the conservation of energy,
It is required that the value in the function that degrades be all higher than equal to zero and and be 1, then:
Wherein D indicates the sum of exponential distribution model, λdAnd πdThe scale factor and power of d-th of exponential distribution are indicated respectively
Weight, E indicate exponential distribution, and j indicates to degrade the index of element in function.It is found in many experiments, as D=4, effect is more preferable.
In image processing field, usually assume that noise is the Gaussian noise of zero-mean, therefore also use zero-mean high herein
Prior model of this model as noise.Convolution (2) can obtain known to ▽ f and h:
Wherein σ2Indicate that the variance of noise, N indicate that Gaussian Profile, i indicate the index of pixel in image.
In order to solve the estimation problem of formula (2), variational Bayesian method provides a kind of effective solution thinking, basic
Thought is to carry out the Posterior probability distribution p (h, ▽ f | ▽ g) of approaching to reality with a more tractable APPROXIMATE DISTRIBUTION q (h, ▽ f).Mirror
The distance between two distributions can be weighed in KL divergences, therefore by minimizing between APPROXIMATE DISTRIBUTION and true Posterior probability distribution
KL divergences realize.Using the variance of noise as the unknown quantity in variation Bayesian Estimation, therefore APPROXIMATE DISTRIBUTION q (h, ▽ f)
Just it has been rewritten into q (h, ▽ f, σ2).KL divergences between the approximate Posterior distrbutionp of definition and true Posterior probability distribution are:
It notices that p (▽ g) is constant always in entire estimation procedure, therefore can define a cost function CKLTo obtain
The optimal value of APPROXIMATE DISTRIBUTION:
Variation Bayesian Estimation realizes cost function by iterative manner (i.e. variation Bayes expectation maximization theorem)
It minimizes, to estimate the function that degrades.It can be obtained by maximal possibility estimation:
Then EM algorithms are used:
M step:▽g(t+1)←arg▽g max F(qh (t+1)(h),▽g). (9)
The algorithm schematic diagram of variation Bayesian Estimation is as shown in Figure 3.
Step 3:The method for going ring based on loop boundary image is used to motion blur image.
The algorithm is mainly carried out in two steps, and the first step is first to carry out adaptive median filter to image, it is made to meet gradient
Smooth requirement obtains extrapolated value image.Second step is using the extrapolated value image acquired in the first step, to fill image
Outer boundary makes the data of the image after extension have periodically, meets the requirement of Fourier transformation.
3.1 calculate extrapolated value image
Due to ringing effect typically occur in the gradient at image strong edge it is precipitous near, the strong edge of image shows as figure
As the pace of change of gradient, therefore we will allow the image gradient of extension outer boundary to change " slow ".According to this idea, calculate
The extrapolated value image without prominent edge is acquired, the image completion is recycled to extend.Concrete operation step is as follows:
The Alpha's mean filter being modified first, this is to eliminate the noise of image.Modified Alpha's mean value
Filtering is one kind of sort method filter, and specific algorithm is as follows:
Assuming that in neighborhood SxyInside remove the d/2 of the d/2 and highest gray value of g (s, t) lowest gray value.Allow gr(s, t) generation
The remaining mn-d pixel of table.The filter formed by the average value of these residual pixels is known as modified Alpha's mean filter
Device:
Wherein the value range of d can be 0 to mn-1.Then gradient minimisation is carried out.By calculating the filling image of extension,
So that the sum of gradient minimum.
3.2 reflection interpolation fillings
Due to carrying out image restoration using Fourier transformation in frequency domain, observed image g boundary pixels is needed to meet periodically.
Therefore fuzzy observation image boundary is extended to the image block of 3m*3n sizes according to mode shown in Fig. 4 border extensions by we.Fig. 1
Middle A, B, C are the filling image of m*n sizes, then take expanded images top left co-ordinate point (m/2+1, n/2+1) and the lower right corner
The image block of the composition of point (5m/2,5n/2), the image block meet loop cycle continuity and extension outer boundary image smoothing.Note
Image is h after border extension, and h image sizes are 2m*2n.Image boundary after extension is smoothly continuous, and meets loop cycle
Property, this helps greatly to inhibit boundary ring.
Step 4:Motion blur image is restored with Richardson-Lucy algorithms.Richardson-Lucy algorithms
It is a kind of Iterative nonlinear restoration algorithm, it is extracted from maximum likelihood formula, and image is modeled with Poisson distribution
, iterative formula is as follows.
WhereinConvolution is represented, g is degraded image, and h represents point spread function, and f represents the image of estimation, and k is iteration time
Number, reasonably value 5 to 30 will produce a large amount of noise if iterations are excessive, be unfavorable for the recovery of image.Here I
K values be 20.
As shown in Fig. 5 (a) and Fig. 5 (b), it is somebody's turn to do the experiment effect of the method for removing motion blur based on variation Bayesian Estimation
Fruit.Fig. 5 (a) is pending blurred picture, and Fig. 5 (b) blind remove motion blur treated as a result, can from experimental result by this
To find out, in terms of the whole visual effect of image and the recovery extent two of local detail, the recovery effect of this paper algorithms compares
It is ideal.From the point of view of specific, in the restoration result of this paper algorithms, the face and figure of personage are more clear, and sharp outline is alleviated and made an uproar
The influence of sound and ringing effect to restoration result.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
All any modification, equivalent and improvement etc., should be included in the claims in the present invention made by within bright spirit and principle
Within protection domain.
Claims (6)
1. a kind of method for removing motion blur based on variation Bayesian Estimation, which is characterized in that include the following steps:
Step 1:Establish motion blur image degradation function model:
Wherein, f indicates that former clear image, h indicate fuzzy core, that is, degrade function,gIndicate blurred picture, n indicates noise, according to shellfish
This principle of leaf, then have following formula establishment:
Step 2:Suitable probabilistic model is selected, the elder generation of the prior probability model of original image and degrade function and noise is included
Test probabilistic model;
Step 3:Motion blur image is used, ring is gone based on loop boundary image;
Step 4:Motion blur image is restored using Richardson-Lucy algorithms.
2. the method according to claim 1 that remove motion blur based on variation Bayesian Estimation, which is characterized in that step
Prior model of the zero-mean gaussian model as noise is used in 2.
3. the method according to claim 1 that remove motion blur based on variation Bayesian Estimation, which is characterized in that step
To carry out the Posterior probability distribution p (h, ▽ f | ▽ g) of approaching to reality using APPROXIMATE DISTRIBUTION q (h, ▽ f) in 2, pass through minimize it is approximate
The KL divergences between true Posterior probability distribution are distributed to realize;Using the variance of noise as in variation Bayesian Estimation not
The amount of knowing, the APPROXIMATE DISTRIBUTION q (h, ▽ f) have just been rewritten into q (h, ▽ f, σ2), define approximate Posterior distrbutionp and true posterior probability
KL divergences between distribution are:
The p (▽ g) is constant always in entire estimation procedure, can define a cost function CKLTo obtain APPROXIMATE DISTRIBUTION
Optimal value:
4. the method according to claim 1 that remove motion blur based on variation Bayesian Estimation, which is characterized in that step
The minimum for realizing cost function in 2 by variation Bayes's expectation maximization theorem, to estimate the function that degrades, by maximum
Possibility predication can obtain:
Then EM algorithms are used:
5. the method according to claim 1 that remove motion blur based on variation Bayesian Estimation, which is characterized in that step
Ring is gone to specifically include based on loop boundary image in 3:
The first step, the Alpha's mean filter being modified to image, make it meet the smooth requirement of gradient, obtain extrapolated value figure
Picture;
Second, the extrapolated value image acquired in the first step make the data of the image after extension to fill outer image border
With periodicity, meet the requirement of Fourier transformation.
6. the method according to claim 1 that remove motion blur based on variation Bayesian Estimation, which is characterized in that step
Motion blur image is restored specially using Richardson-Lucy algorithms in 4:Richardson-Lucy algorithms are
Iterative nonlinear restoration algorithm is extracted from maximum likelihood formula, and image is modeled with Poisson distribution, iteration
Formula is as follows.
Wherein,Convolution is represented, g is degraded image, and h represents point spread function, and f represents the image of estimation, and k is iterations.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810484731.6A CN108681995A (en) | 2018-05-20 | 2018-05-20 | A method of motion blur is gone based on variation Bayesian Estimation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810484731.6A CN108681995A (en) | 2018-05-20 | 2018-05-20 | A method of motion blur is gone based on variation Bayesian Estimation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108681995A true CN108681995A (en) | 2018-10-19 |
Family
ID=63805388
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810484731.6A Pending CN108681995A (en) | 2018-05-20 | 2018-05-20 | A method of motion blur is gone based on variation Bayesian Estimation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108681995A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110186464A (en) * | 2019-05-30 | 2019-08-30 | 西安电子科技大学 | A kind of X-ray pulsar navigation TOA estimation method based on Bayesian Estimation |
CN111340702A (en) * | 2020-02-24 | 2020-06-26 | 江南大学 | Sparse reconstruction method for high-frequency ultrasonic microscopic imaging of tiny defects based on blind estimation |
CN111882579A (en) * | 2020-07-03 | 2020-11-03 | 湖南爱米家智能科技有限公司 | Large infusion foreign matter detection method, system, medium and equipment based on deep learning and target tracking |
CN111986121A (en) * | 2020-09-24 | 2020-11-24 | 南京航空航天大学 | Based on Framellet l0Norm-constrained fuzzy image non-blind restoration method |
CN112132758A (en) * | 2020-08-05 | 2020-12-25 | 浙江工业大学 | Image restoration method based on asymmetric optical system point spread function model |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106952240A (en) * | 2017-03-29 | 2017-07-14 | 成都信息工程大学 | A kind of image goes motion blur method |
WO2017190432A1 (en) * | 2016-05-03 | 2017-11-09 | 北京大学深圳研究生院 | Image deblurring method based on bright fringe information in image |
-
2018
- 2018-05-20 CN CN201810484731.6A patent/CN108681995A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017190432A1 (en) * | 2016-05-03 | 2017-11-09 | 北京大学深圳研究生院 | Image deblurring method based on bright fringe information in image |
CN106952240A (en) * | 2017-03-29 | 2017-07-14 | 成都信息工程大学 | A kind of image goes motion blur method |
Non-Patent Citations (1)
Title |
---|
SHAN CAO ETC.: ""Single image motion deblurring with reduced ringing effects using variational Bayesian estimation"", 《SIGNAL PROCESSING》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110186464A (en) * | 2019-05-30 | 2019-08-30 | 西安电子科技大学 | A kind of X-ray pulsar navigation TOA estimation method based on Bayesian Estimation |
CN111340702A (en) * | 2020-02-24 | 2020-06-26 | 江南大学 | Sparse reconstruction method for high-frequency ultrasonic microscopic imaging of tiny defects based on blind estimation |
CN111882579A (en) * | 2020-07-03 | 2020-11-03 | 湖南爱米家智能科技有限公司 | Large infusion foreign matter detection method, system, medium and equipment based on deep learning and target tracking |
CN112132758A (en) * | 2020-08-05 | 2020-12-25 | 浙江工业大学 | Image restoration method based on asymmetric optical system point spread function model |
CN112132758B (en) * | 2020-08-05 | 2024-06-21 | 浙江工业大学 | Image restoration method based on asymmetric optical system point spread function model |
CN111986121A (en) * | 2020-09-24 | 2020-11-24 | 南京航空航天大学 | Based on Framellet l0Norm-constrained fuzzy image non-blind restoration method |
CN111986121B (en) * | 2020-09-24 | 2024-01-09 | 南京航空航天大学 | Based on Framelet l 0 Non-blind restoration method for fuzzy image with norm constraint |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108681995A (en) | A method of motion blur is gone based on variation Bayesian Estimation | |
Remez et al. | Deep convolutional denoising of low-light images | |
Mao et al. | Non rigid geometric distortions correction-application to atmospheric turbulence stabilization | |
Fadili et al. | Total variation projection with first order schemes | |
Wen et al. | Primal-dual algorithms for total variation based image restoration under Poisson noise | |
CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
CN102231204A (en) | Sequence image self-adaptive regular super resolution reconstruction method | |
Lee et al. | High-quality non-blind image deconvolution with adaptive regularization | |
Dong et al. | Blur kernel estimation via salient edges and low rank prior for blind image deblurring | |
Anantrasirichai | Atmospheric turbulence removal with complex-valued convolutional neural network | |
Shukla et al. | Adaptive fractional masks and super resolution based approach for image enhancement | |
Zhang et al. | High-quality image restoration from partial random samples in spatial domain | |
Chierchia et al. | Epigraphical splitting for solving constrained convex formulations of inverse problems with proximal tools | |
Toledo et al. | Image denoising based on genetic algorithm | |
Goto et al. | Learning-based super-resolution image reconstruction on multi-core processor | |
Cao et al. | Single image motion deblurring with reduced ringing effects using variational Bayesian estimation | |
CN110121016B (en) | Video deblurring method and device based on double exposure prior | |
Ma et al. | Nonlocal total variation regularization with Shape Adaptive Patches for image denoising via Split Bregman method | |
Fang et al. | Multiframe blind image deconvolution with split Bregman method | |
Janjušević et al. | Fast and interpretable nonlocal neural networks for image Denoising via group-sparse convolutional dictionary learning | |
Biradar et al. | Object Removal using Recursive Median Filter | |
Yen et al. | Deblurring a Camera-Shake image using a Thinning Kernel | |
Nawaz et al. | Dense Optical Flow Estimation Using Sparse Regularizers from Reduced Measurements | |
Zhang et al. | Image inpainting based on directional Gaussian graph model using multi-head reference | |
Shuli et al. | Nonlocal total variation regularization with Shape Adaptive Patches for image denoising via Split Bregman method |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181019 |
|
WD01 | Invention patent application deemed withdrawn after publication |