CN104166961A - Low-rank approximation fuzzy nucleus estimation method for image blind restoration - Google Patents
Low-rank approximation fuzzy nucleus estimation method for image blind restoration Download PDFInfo
- Publication number
- CN104166961A CN104166961A CN201410361709.4A CN201410361709A CN104166961A CN 104166961 A CN104166961 A CN 104166961A CN 201410361709 A CN201410361709 A CN 201410361709A CN 104166961 A CN104166961 A CN 104166961A
- Authority
- CN
- China
- Prior art keywords
- fuzzy core
- image
- fuzzy
- iteration
- cost
- 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 invention discloses a low-rank approximation fuzzy nucleus estimation method for image blind restoration, mainly for solving the problems of how to more accurately realize fuzzy nucleus estimation in a conventional image blind restoration method and how to accordingly restore an ideal image. The realization steps comprise: on one hand, taking a neighbor relation of a gradient image into consideration, and improving an iteration threshold strategy by use of an autoregression (AR) strategy so as to estimate a fuzzy nucleus; on the other hand, enhancing image margin information by use of a heuristic filter to estimate another fuzzy nucleus; afterwards, introducing a low-rank approximation strategy to a fuzzy nucleus estimation process to solve a more reliable fuzzy nucleus; and finally, restoring a clear image by use of an advanced image restoration method. Compared to some conventional methods, the method provided by the invention has the following advantages: PSNR, SSIM and FSIM values are higher, the visual effect is better, fuzziness is effectively removed, more details are maintained, and the designed fuzzy nucleus is also more accurate.
Description
Technical field
The invention belongs to technical field of image processing, relate to the application in blindly restoring image field, a fuzzy core method of estimation of approaching for the low-rank of blindly restoring image specifically, can be used for degraded image fuzzy to various the unknowns and that be subject to slight noise effect and carries out image restoration.
Background technology
Blindly restoring image refers to dispels or alleviates image blurring that the various X factors that are subject in acquired digital picture cause, and meanwhile, the image of this acquisition also can be subject to some inevitable noise effects.Therefore the blind recovery of image is important and challenging research contents during image is processed.It has very important application in many aspects, as aspects such as Medical Image Processing, the processing of material science image, public security, history, the recovery of humane photograph image, monitoring video recovery and scanned document processing.For this problem, researchers have proposed a lot of methods.
More classical method is as the method based on maximum a posteriori probability (MAP), Fergus R, Singh B, Hertzmann A, et al.Removing camera shake from a single photograph[C] //ACM Transactions on Graphics (TOG) .ACM, 2006,25 (3): first 787-794. the method adopts the Bayesian strategy of variation to go to obtain fuzzy core, then with a kind of non-blind image recovery method, remove to solve clearly image.But the mathematical principle of this method is too complicated, even if also can be greatly affected for the estimation that contains the image blurring core of slight noise.Cause image blurring reason can be described as by the method for mathematics a fuzzy matrix, this fuzzy matrix is called fuzzy core.
Other class methods are to utilize image prominent edge information to remove restored image, and these methods think that marginal information role in the process of blindly restoring image of image is greater than other image informations such as homogeneous region.More representational as S.Cho and S.Lee.Fast motion deblurring.ACM Trans.Graph., pages 145:1 – 145:8,2009. the method go to strengthen image border with heuristic wave filter, then use these marginal informations of processing to go to estimate fuzzy core.Owing to too much depending on these didactic wave filters, this method also exists fuzzy core to estimate not accurate enough shortcoming.
Summary of the invention
The object of the invention is to the deficiency for prior art, propose a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image, to improve the accuracy of fuzzy core estimation and then to make the effect of blindly restoring image better.
For achieving the above object, the present invention includes following steps:
(1) length of side being set is 3, and spatial domain standard deviation is 0.6, and codomain standard deviation is 0.7 for pretreated two-sided filter f, then pending degraded image y is carried out to bilateral filtering, obtains edge sharpening and suppresses the image y of noise effect
(1);
(2) initialization correlated condition and parameter, generate gradient image matrix;
(3) to gradient image matrix
at pyramid model, (being about to image is n layer with the method down-sampling of bicubic difference, and the ratio between adjacent two layers is in the present invention
) i (i=1, n) layer use following impact response filter to go to strengthen obvious image border:
(4) according to gradient image matrix
autoregression (AR) coefficient of training current layer (i layer)
(horizontal direction) with
(vertical direction);
(5) threshold value adjusting parameters C ost is upgraded in initialization
before=LS
cost+ Re
cost+ AR
cost:
(6) use the optimization of iteration threshold algorithm (ISTA) optimized algorithm:
(7) calculate the parameters C ost that upgrades iteration threshold algorithm (ISTA) threshold value
aferand judge Cost
aferwhether be greater than 1.12*Cost
beforethen process;
(8) use the least square method (IRLS) of heavily composing weights to go to optimize
and there is k>=0, ∑
jk
j=1.K herein
jrefer to the pixel value that fuzzy core k is ordered at j.
(9) repeating step (5)~(8) are asked for the fuzzy core k estimating for iter time
(i)i=1,2...n, iter is customer parameter in the method, gets 21 in experiment, user can, in 21 its left and right values, then use the way of bilinear interpolation by the fuzzy core k estimating
(i)with x
(i)up-sampling, and using the initial value of its one deck under pyramid;
(10) repeating step (3)~(10) n time, n is the image pyramid number of plies according to the setting of fuzzy core size, its computing method are for to be multiplied by successively from minimum fuzzy core (this method is made as 3)
round, from 1 counting, until fuzzy core size is to the fuzzy core size of setting, can be in the hope of, in addition, in order to suppress noise effect, the last one deck of pyramid is estimated to fuzzy core k
1being less than and in pixel value, being less than 0 value assignment is 0, at the fuzzy core k of the last one deck output estimation of pyramid
1;
(11) pending degraded image y is left intact, identical with step (2), calculate the number of plies n of pyramid model, and use the method convergent-divergent y of bilinear interpolation to the most rough layer (the 1st layer) y
ii=1 uses y simultaneously
ii=1 upgrades
i=1, and the multiple that goes forward one by one of setting between every one deck gradient image size is
upgrade the fuzzy core of the most rough layer of initialization
i=1;
(12) repeated execution of steps (2), (4)~(11), export the 2nd the fuzzy core k estimating
2;
(13) utilize the fuzzy core k estimating
1and k
2ask for last fuzzy core k;
(14) utilize the fuzzy core k estimating, by optimized-type (2), estimate picture rich in detail (iterations that solves (2) formula is set to 200 times)
Wherein x is the picture rich in detail that wish is estimated, y is the degraded image observing, T
fit is toeplitz matrix.
(15) the picture rich in detail x after output processing and the fuzzy core k of estimation.
Technical scheme of the present invention has been considered the neighbor relationships of gradient image, with autoregression (AR) thus policy improvement iteration threshold strategies estimates a fuzzy core; Use heuristic wave filter to strengthen image edge information and go to estimate another fuzzy core; The strategy that low-rank is approached is incorporated into and in the estimation procedure of fuzzy core, goes to solve a fuzzy core more reliably.Finally with a kind of advanced person's image recovery method, restore clearly image.The present invention has the following advantages compared with prior art:
1. the present invention has reduced the inaccuracy that fuzzy core that the factors such as the noise that exists in the observed image that degraded or other singular values cause is estimated;
2. the present invention is not overly dependent upon the marginal information that heuristic wave filter removes to utilize image;
3. the present invention considers the neighbor relationships of gradient image, thereby in the process of iteration optimization, can adaptively go to find more excellent fuzzy core;
4. the present invention has realized the process of going to ask for a fuzzy relatively more accurate and stable fuzzy core by two different fuzzy core that estimate, and experimental result shows that this strategy has produced the result of blindly restoring image better.
Accompanying drawing explanation
Fig. 1 is general flow chart of the present invention;
Fig. 2 is the synthetic fuzzy degraded image containing slight noise of a width that emulation of the present invention is used, and the image lower right corner is synthetic fuzzy core used;
Fig. 3 is by the existing method (MAP) based on maximum a posteriori probability, Fergus ' 06, be Fergus R, Singh B, Hertzmann A, et al.Removing camera shake from a single photograph[C] //ACM Transactions on Graphics (TOG) .ACM, 2006,25 (3): the result that the method in 787-794. is restored;
Fig. 4 is that Cho ' 09 by the existing method based on edge, i.e. S.Cho and S.Lee.Fast motion deblurring.ACM Trans.Graph., pages 145:1 – 145:8, the result that the method in 2009. is restored;
Fig. 5 implements the result of blind deblurring to Fig. 2 with the present invention.
Fig. 6 is that three kinds of methods are implemented the partial result comparison diagram of blind deblurring to Fig. 2.
Embodiment
With reference to Fig. 1, specific implementation step of the present invention is as follows:
Step 1, it is 3 that the length of side is set, and spatial domain standard deviation is 0.6, and codomain standard deviation is 0.7 for pretreated two-sided filter f, then pending degraded image y is carried out to bilateral filtering, obtains edge sharpening and suppresses the image y of noise effect
(1);
Step 2, initialization correlated condition and parameter, generate gradient image matrix;
2a) utilize default fuzzy core size (being the size of fuzzy core matrix) k
sizecalculate the number of plies n of pyramid model, according to size and the k of the most rough layer (the 1st layer) fuzzy core
sizeproportionate relationship, use the method convergent-divergent y of bilinear interpolation
(1)to the most rough layer (the 1st layer)
i=1, and the multiple that goes forward one by one of setting between every one deck gradient image size is
the fuzzy core that initialization is simultaneously the most rough layer
i=1;
2b) by horizontal gradient operator dx=[-1,1; 0,0] with vertical dy=[-1,0; 1,0] with pyramid respectively with i tomographic image
carry out convolution and obtain the image of gradient field
with
i=1,2...n, and gradient image matrix is merged in connection
i=1,2...n;
Step 3, to gradient image matrix
obvious image border in the following impact response filter of the i (i=1, n) of pyramid model layer use goes to strengthen image:
3a) I in (1) is replaced into gradient image matrix
with
with Δ I
trefer to respectively utilize Laplace operator and gradient operator to gradient image matrix
with
carry out computing;
3b) the middle weakening factor that dt is each iteration is set, gets 0.08 in this method, iteration (1) formula 5 times, merges the gradient image matrix after processing to upgrade
Step 4, according to gradient image matrix
horizontal direction autoregression (AR) coefficient of training current layer (i layer)
with vertical direction autoregression (AR) coefficient
4a) getting the required window size of training autoregression as above (AR) coefficient is 3 * 3, and by gradient image matrix
with
around border, a pixel is shone upon processing.Simultaneously will according to 3 * 3 window sizes
with
pull into m
i* 9 matrix
with
i=1,2;
4b) extract matrix
with
the 5th row, be designated as Y
1with Y
2, according to
calculate respectively
(horizontal direction) with
(vertical direction) i=1,2;
Step 5, initialization is upgraded threshold value and is regulated parameters C ost
before=LS
cost+ Re
cost+ AR
cost:
Step 6, use the optimization of iteration threshold algorithm (ISTA) optimized algorithm:
Step 7, calculates the parameters C ost that upgrades iteration threshold algorithm (ISTA) threshold value
aferand judge Cost
aferwhether be greater than 1.12*Cost
beforethen process;
7a) iteration is upgraded to the x that goes out
(i)be decomposed into
with
window according to 3 * 3 decomposes, and is pulled into m
i* 9 matrix
with
i=1,2...n, extracts
with
the 5th row, be designated as
with
i=1,2...n;
7b) basis
with
i=1,2, wherein γ is 10, calculates and upgrades
i=1,2;
7c) calculate and upgrade after iteration
With calculating
I=1,2...n wherein λ is taken as 90 in likelihood item coefficient this method;
7d) calculate the threshold value of upgrading after iteration and regulate parameters C ost
afer=LS
cost+ Re
cost+ AR
cost;
If 7e) Cost
afer>1.12*Cost
before, the threshold value that regulates ISTA next iteration is ought be last time 0.62 times;
If 7f) Cost
afer<1.12*Cost
before, the threshold value of ISTA next iteration is constant, stops ISTA optimization and jumps into step (8);
Step 8, is used the least square method (IRLS) of heavily composing weights to go to optimize
and there is k>=0, ∑
jk
j=1.K herein
jrefer to the pixel value that fuzzy core k is ordered at j.
Step 9, repeating step (5)~(8) are asked for the fuzzy core k estimating for iter time
(i)i=1,2...n, iter is customer parameter in the method, generally gets 21; Then use the way of bilinear interpolation by the fuzzy core k estimating
(i)with x
(i)up-sampling, and using the initial value of its one deck under pyramid;
Step 10, repeating step (3)~(10) n time, the value of n is identical with step 2, is the pyramidal number of plies, in addition, in order to suppress noise effect, the last one deck of pyramid is estimated to fuzzy core k
1in be less than pixel value 0 pixel value assignment be 0, at the fuzzy core k of the last one deck output estimation of pyramid
1;
Step 11, is left intact to pending degraded image y, identical with step (2), calculates the number of plies n of pyramid model, and uses the method convergent-divergent y of bilinear interpolation to the most rough layer (the 1st layer) y
ii=1 uses y simultaneously
ii=1 upgrades
i=1, and the multiple that goes forward one by one of setting between every one deck gradient image size is
upgrade the fuzzy core of the most rough layer of initialization
i=1;
Step 12, repeated execution of steps (2), (4)~(11), export the 2nd the fuzzy core k estimating
2;
Step 13, utilizes the fuzzy core k estimating
1and k
2ask for last fuzzy core k;
13a) by the fuzzy core k estimating
1and k
2after pulling into respectively row and get up to be designated as D
k, then utilize Go-dec algorithm to carry out low-rank decomposition;
13b) low-rank after decomposing is partly designated as to ker, ker is restored to k according to the size of default fuzzy core
size* k
sizefuzzy core k;
Step 14, utilizes the fuzzy core k estimating, and by optimized-type (2), estimates picture rich in detail (iterative loop solves number of times and is set to 200 times)
Wherein x is the picture rich in detail that wish is estimated, y is the degraded image observing, T
fit is Teoplitz (toeplitz) matrix.
Step 15, the picture rich in detail x after output is processed and the fuzzy core k of estimation.
Effect of the present invention can further illustrate by experiment simulation below:
1, experiment condition and method
Hardware platform is: Intel Core2 Duo CPU E6550@2.33GHZ, 2GB RAM;
Software platform is: MATLAB R2013a;
Experimental technique: the method that is respectively the present invention and existing Fergus ' 06 is that the method for (1) and Cho ' 09 is (2), and this is all the representative method in this area.
2, emulation content and result
By the method for the present invention and existing Fergus ' 06 and the method for Cho ' 09, respectively the Sailing blurred picture shown in Fig. 2 is carried out to blind recovery emulation, this figure lower right is synthetic true fuzzy core used; The blind recovery result of Sailing wherein obtaining with the present invention is as Fig. 5, and the fuzzy core estimating is shown in this figure lower right; Use the blind recovery result of Sailing that the method for existing Fergus ' 06 obtains as Fig. 3, the fuzzy core estimating is shown in this figure lower right; The blind recovery result of Sailing obtaining by the existing method based on edge is as Fig. 4, and the fuzzy core estimating is shown in this figure lower right.
In emulation experiment, application peak value noise (PSNR), structural similarity sex index (SSIM) and structural similarity (FSIM) are as evaluation index.
Evaluation result is as shown in table 1, and wherein, Alg1 is method of the present invention, Alg2 be Fergus ' 06 method, Alg3 is the method for Cho ' 09.
Table 1. the present invention and two kinds of PSNR values (unit is dB) that control methods obtains in emulation experiment, and SSIM, FSIM
3. interpretation
Blind recovery result Fig. 3 of method of blind recovery result Fig. 5 Fergus ' 06 that contrast the present invention obtains and recovery result Fig. 4 that the method for Cho ' 09 obtains can find out, the blind recovery result of Sailing that the present invention shown in Fig. 5 obtains has obtained with respect to other two kinds of better experimental results of method, Fig. 5 has not only removed fuzzy effectively, but also having retained more image detail, estimated fuzzy core is out also more close to real fuzzy core; The blind recovery result that the method for Fergus ' 06 shown in Fig. 3 obtains could not effectively be removed blooming when being subject to noise effect, and produces the product of the serious artifact effect of knowing clearly; The method of Cho ' 09 shown in Fig. 4 can effectively be removed fuzzy, but could not play a role the impact of noise, and has lost the details of some images.
As can be seen from Table 1, the present invention has higher PSNR than other two kinds of control methodss, and SSIM and FSIM value visually also have better effect not only effectively to remove fuzzy, kept more details, and the fuzzy core estimating are also more accurate.
Claims (6)
1. a fuzzy core method of estimation of approaching for the low-rank of blindly restoring image, comprises the steps:
Step 1: it is 3 that the length of side is set, and spatial domain standard deviation is 0.6, codomain standard deviation is 0.7 for pretreated two-sided filter f, then pending degraded image y is carried out to bilateral filtering, obtains edge sharpening and suppresses the image y of noise effect
(1);
Step 2: initialization correlated condition and parameter, generate gradient image matrix;
Step 3: to gradient image matrix
at the i (i=1, n) of pyramid model layer, use following impact response filter to go to strengthen obvious image border:
Step 4: according to gradient image matrix
horizontal direction autoregression (AR) coefficient of training current layer (i layer)
with vertical direction autoregression (AR) coefficient
Step 5: initialization is upgraded threshold value and regulated parameters C ost
before=LS
cost+ Re
cost+ AR
cost:
Step 6: use the optimization of iteration threshold algorithm (ISTA) optimized algorithm:
Step 7: calculate the parameters C ost that upgrades iteration threshold algorithm (ISTA) threshold value
aferand judge Cost
aferwhether be greater than 1.12*Cost
beforethen process;
Step 8: use the least square method (IRLS) of heavily composing weights to go to optimize
and there is k>=0, ∑
jk
j=1, k herein
jrefer to the pixel value that fuzzy core k is ordered at j;
Step 9: repeating step (5)~(8) are asked for the fuzzy core k estimating for iter time
(i)i=1,2...n, iter is customer parameter in the method, generally gets 21; Then use the way of bilinear interpolation by the fuzzy core k estimating
(i)with x
(i)up-sampling, and using the initial value of its one deck under pyramid;
Step 10, repeating step (3)~(10) n time, the value of n is identical with step 2, is the pyramidal number of plies, in addition, in order to suppress noise effect, the last one deck of pyramid is estimated to fuzzy core k
1in be less than pixel value 0 pixel value assignment be 0, at the fuzzy core k of the last one deck output estimation of pyramid
1;
Step 11: pending degraded image y is left intact, identical with step (2), calculate the number of plies n of pyramid model, and use the method convergent-divergent y of bilinear interpolation to the most rough layer (the 1st layer) y
ii=1 uses y simultaneously
ii=1 upgrades
i=1, and the multiple that goes forward one by one of setting between every one deck gradient image size is
, upgrade the initialization fuzzy core of layer the most roughly
i=1;
Step 12: repeated execution of steps (2), (4)~(11), export the 2nd the fuzzy core k estimating
2;
Step 13: utilize the fuzzy core k estimating
1and k
2ask for last fuzzy core k;
Step 14: utilize the fuzzy core k estimating, estimate picture rich in detail by optimized-type (2)
Step 15: the picture rich in detail x after output is processed and the fuzzy core k of estimation.
2. a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image according to claim 1, wherein initialization correlated condition and parameter, generate comprising of gradient image matrix:
2a) utilize default fuzzy core size k
sizecalculate the number of plies n of pyramid model, according to size and the k of the most rough layer (the 1st layer) fuzzy core
sizeproportionate relationship, use the method convergent-divergent y of bilinear interpolation
(1)to the most rough layer (the 1st layer)
i=1, and the multiple that goes forward one by one of setting between every one deck gradient image size is 2, initialization is simultaneously the fuzzy core of rough layer
i=1;
2b) by horizontal gradient operator dx=[-1,1; 0,0] with vertical dy=[-1,0; 1,0] with pyramid respectively with i tomographic image
carry out convolution and obtain the image of gradient field
with
i=1,2...n, and gradient image matrix is merged in connection
i=1,2...n.
3. a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image according to claim 1, wherein (1) of step 3 comprises:
3a) I in (1) is replaced into gradient image matrix
with
with Δ I
trefer to respectively utilize Laplace operator and gradient operator to gradient image matrix
with
carry out computing;
3b) the middle weakening factor that dt is each iteration is set, gets 0.08 in this method, iteration (1) formula 5 times, merges the gradient image matrix after processing to upgrade
4. a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image according to claim 1, wherein step 4 comprises:
4a) getting the required window size of training autoregression (AR) coefficient is 3 * 3, and by gradient image matrix
with
around border, a pixel is shone upon processing, simultaneously will according to 3 * 3 window sizes
with
for pulling into m
i* 9 matrix
with
i=1,2;
4b) extract matrix
with
the 5th row, be designated as Y
1with Y
2, according to
calculate respectively
(horizontal direction) with
(vertical direction) i=1,2.
5. a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image according to claim 1, wherein step 7 specifically comprises:
7a) iteration is upgraded to the x that goes out
(i)be decomposed into
with
window according to 3 * 3 decomposes, and is pulled into m
i* 9 matrix
with
i=1,2...n, extracts
with
the 5th row, be designated as
with
i=1,2...n;
7b) basis
with
i=1,2, wherein γ is 10, calculates and upgrades
i=1,2;
7c) calculate and upgrade after iteration
With calculating
I=1,2...n wherein λ is taken as 90 in likelihood item coefficient this method;
7d) calculate the threshold value of upgrading after iteration and regulate parameters C ost
afer=LS
cost+ Re
cost+ AR
cost;
If 7e) Cost
afer>1.12*Cost
before, the threshold value that regulates ISTA next iteration is ought be last time 0.62 times;
If 7f) Cost
afer<1.12*Cost
before, the threshold value of ISTA next iteration is constant, stops ISTA optimization and jumps into step (8).
6. a kind of fuzzy core method of estimation of approaching for the low-rank of blindly restoring image according to claim 1, wherein step 13 specifically comprises:
13a) by the fuzzy core k estimating
1and k
2after pulling into respectively row and get up to be designated as D
k, then utilize Go-dec algorithm to carry out low-rank decomposition;
13b) low-rank after decomposing is partly designated as to ker, ker is restored to k according to the size of default fuzzy core
size* k
sizefuzzy core k.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410361709.4A CN104166961B (en) | 2014-07-28 | 2014-07-28 | The fuzzy core method of estimation that low-rank for blindly restoring image approaches |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410361709.4A CN104166961B (en) | 2014-07-28 | 2014-07-28 | The fuzzy core method of estimation that low-rank for blindly restoring image approaches |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104166961A true CN104166961A (en) | 2014-11-26 |
CN104166961B CN104166961B (en) | 2018-03-16 |
Family
ID=51910754
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410361709.4A Active CN104166961B (en) | 2014-07-28 | 2014-07-28 | The fuzzy core method of estimation that low-rank for blindly restoring image approaches |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104166961B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106373107A (en) * | 2016-12-06 | 2017-02-01 | 四川长虹电器股份有限公司 | Automatic image deblurring system and automatic image deblurring method of smart phone |
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 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101930601A (en) * | 2010-09-01 | 2010-12-29 | 浙江大学 | Edge information-based multi-scale blurred image blind restoration method |
CN102254305A (en) * | 2011-06-30 | 2011-11-23 | 西安电子科技大学 | Image restoring method based on three dictionary block matching |
CN102393955A (en) * | 2011-07-18 | 2012-03-28 | 西安电子科技大学 | Perfect information non-local constraint total variation method for image recovery |
CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
CN103839233A (en) * | 2014-01-20 | 2014-06-04 | 左旺孟 | Restoration method for blurred image caused by camera shaking |
-
2014
- 2014-07-28 CN CN201410361709.4A patent/CN104166961B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101930601A (en) * | 2010-09-01 | 2010-12-29 | 浙江大学 | Edge information-based multi-scale blurred image blind restoration method |
CN102254305A (en) * | 2011-06-30 | 2011-11-23 | 西安电子科技大学 | Image restoring method based on three dictionary block matching |
CN102393955A (en) * | 2011-07-18 | 2012-03-28 | 西安电子科技大学 | Perfect information non-local constraint total variation method for image recovery |
CN103337058A (en) * | 2013-07-05 | 2013-10-02 | 西北工业大学 | Image blind restoration method based on blurred noise image pair joint optimization |
CN103839233A (en) * | 2014-01-20 | 2014-06-04 | 左旺孟 | Restoration method for blurred image caused by camera shaking |
Non-Patent Citations (2)
Title |
---|
KRISHNAN D等: "Blind deconvolution using a normalized sparsity measure", 《COMPUTER VISION AND PATTERN RECOGNITION(CVPR),2011 IEEE CONFERENCE ON》 * |
SUNGHYUN CHO等: "Fast motion deblurring", 《ACM TRANSACTIONS ON GRAPHICS》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106373107A (en) * | 2016-12-06 | 2017-02-01 | 四川长虹电器股份有限公司 | Automatic image deblurring system and automatic image deblurring method of smart phone |
CN106373107B (en) * | 2016-12-06 | 2019-03-19 | 四川长虹电器股份有限公司 | Smart phone automated graphics deblurring system and method |
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 |
Also Published As
Publication number | Publication date |
---|---|
CN104166961B (en) | 2018-03-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Efficient single image dehazing and denoising: An efficient multi-scale correlated wavelet approach | |
Pan et al. | Blind image deblurring using dark channel prior | |
Pan et al. | Deblurring text images via L0-regularized intensity and gradient prior | |
Xu et al. | Image smoothing via L 0 gradient minimization | |
US9245328B2 (en) | Algorithm for minimizing latent sharp image cost function and point spread function with a spatial mask in a fidelity term | |
CN105493140B (en) | Image deblurring method and system | |
US9262815B2 (en) | Algorithm for minimizing latent sharp image cost function and point spread function cost function with a spatial mask in a regularization term | |
Dong et al. | Image reconstruction with locally adaptive sparsity and nonlocal robust regularization | |
CN104599242B (en) | Use the fuzzy core method of estimation of multiple dimensioned non local canonical | |
Singh et al. | Super-resolving noisy images | |
US9928579B2 (en) | Multiple phase method for image deconvolution | |
Yang | Semantic filtering | |
CN105184744A (en) | Prior fuzzy kernel estimation method based on standardized sparse measurement image block | |
CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
US8611690B2 (en) | Real-time video deblurring | |
US9569684B2 (en) | Image enhancement using self-examples and external examples | |
Nejati et al. | Denoising by low-rank and sparse representations | |
Dong et al. | Blur kernel estimation via salient edges and low rank prior for blind image deblurring | |
CN103761710A (en) | Image blind deblurring method based on edge self-adaption | |
CN103337058A (en) | Image blind restoration method based on blurred noise image pair joint optimization | |
CN107451973A (en) | Motion blur image restoration method based on the extraction of abundant fringe region | |
CN102800055B (en) | Low-order decomposition method for blind deblurring of images | |
Hu et al. | Non-local means algorithm with adaptive patch size and bandwidth | |
CN112163994A (en) | Multi-scale medical image fusion method based on convolutional neural network | |
Suryanarayana et al. | Infrared super-resolution imaging using multi-scale saliency and deep wavelet residuals |
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 |