CN106169180A - A kind of non-local sparse based on group represents additive noise minimizing technology - Google Patents
A kind of non-local sparse based on group represents additive noise minimizing technology Download PDFInfo
- Publication number
- CN106169180A CN106169180A CN201610548312.5A CN201610548312A CN106169180A CN 106169180 A CN106169180 A CN 106169180A CN 201610548312 A CN201610548312 A CN 201610548312A CN 106169180 A CN106169180 A CN 106169180A
- Authority
- CN
- China
- Prior art keywords
- group
- image
- local
- block
- sparse
- 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
- 238000005516 engineering process Methods 0.000 title claims abstract description 21
- 239000000654 additive Substances 0.000 title claims abstract description 20
- 230000000996 additive effect Effects 0.000 title claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 36
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims description 11
- 239000000203 mixture Substances 0.000 claims description 4
- 230000000694 effects Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- 235000002566 Capsicum Nutrition 0.000 description 3
- 241000758706 Piperaceae Species 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 235000008534 Capsicum annuum var annuum Nutrition 0.000 description 1
- 240000008384 Capsicum annuum var. annuum Species 0.000 description 1
- 206010008190 Cerebrovascular accident Diseases 0.000 description 1
- 208000006011 Stroke Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000011084 recovery 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
- G06T5/70—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
- G06T2207/20081—Training; Learning
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 non-local sparse based on group and represent additive noise minimizing technology, concrete steps include: S1: obtain natural image in standard picture storehouse, and image carries out image block and trains dictionary with gauss hybrid models;S2: each image carries out plus noise process, the method utilizing rarefaction representation based on group and non-local sparse to represent obtains removing noise model;S3: noisy image is divided into the fritter having overlap, carries out non local Similarity matching for each image block, find it to have the similar image block of identical structure type, some the highest for similarity blocks are put in a group;S4: for each group divided, first carry out singular value decomposition to each group, then eigenvalue is carried out threshold process, obtain the estimated value of each group;S5: utilize the dictionary of training in advance to obtain estimation group sparse coding in dictionary;S6: utilize soft threshold method solving model;S7: obtain denoising image.
Description
Technical field
The present invention relates to digital image processing techniques field, a kind of non-local sparse based on group represents that additivity is made an uproar
Sound minimizing technology.
Background technology
The noise removal technical research of image is to go out to obtain high-quality image from observed image.The master of the research of this technology
Method is wanted to include: method based on wave filter, based on full variational approach, based on non-local method, based on rarefaction representation
Method and method based on image prior etc..
On visual quality and Y-PSNR is the most less desirable for noise removal for traditional method.The present invention is permissible
Make image have the similarity of higher Y-PSNR and image, but also the artefact recovering in image can be suppressed the most fine
Preserve the edge of image and textural characteristics, improve visual effect.
Dong et al. is at " Dong W, Zhang L, Shi G and Li X.Nonlocally centralized
sparse representation for image restoration[J].IEEE Trans.on Image Pro-
Cessing, 2013,22 (4): 1620-1630. " (NCSR model) basic non local concentration sparse representation model.NCSR model makes
Be self-adapting dictionary, remove the less image effect of textural characteristics not good enough, and speed is the slowest.Xu et al. " Xu J,
Zhang L,Zuo W,et al.Patch Group Based Nonlocal Self-Similarity Prior Learning
for Image Denoising[C] IEEE International Conference on Computer Vision.2015:
244-252. " (PGPD model) non local self similarity priori based on block group study image denoising model.PGPD model is with original
The non local priori of image trains dictionary by gauss hybrid models, and image block is carried out packet transaction, achieves very well
Denoising effect.This algorithm takes full advantage of the non local priori of picture rich in detail, but directly carries out noisy image point
Group and sparse coding, easily produce error at Block-matching and coding stage, thus produce more artefact, to this end, it is proposed that
A kind of non-local sparse based on group represents additive noise minimizing technology.
Summary of the invention
It is an object of the invention to provide a kind of non-local sparse based on group and represent additive noise minimizing technology, to solve
The problem proposed in background technology.
For realizing purpose, the present invention provides following technical scheme, a kind of non-local sparse based on group to represent additive noise
Minimizing technology, described non-local sparse based on group represents that additive noise minimizing technology comprises the following steps:
S1: obtain natural image in standard picture storehouse, and image is carried out image block and instructs with gauss hybrid models
Practise handwriting allusion quotation;
S2: each image carries out plus noise process, utilizes the side that rarefaction representation based on group and non-local sparse represent
Method obtains removing noise model;
S3: noisy image is divided into the fritter having overlap, carries out non local Similarity matching for each image block, find it to have
There is the similar image block of identical structure type, some the highest for similarity blocks are put in a group;
S4: for each group divided, first carry out singular value decomposition to each group, then eigenvalue is carried out at threshold value
Reason, obtains the estimated value of each group;
S5: utilize the dictionary of training in advance to obtain estimation group sparse coding in dictionary;
S6: utilize soft threshold method solving model;
S7: obtain denoising image;
Preferably, in S3 step, first the image x that size is N × N is divided into the overlapping block that n size is p × p, fixed
Justice isK=1,2 ..., n. then, for block xk, search is mated most with it in training window c block group, group
Become setIt follows that willIn all of piece composition one size be p2The matrix of × c, is defined as
In all of piece of row as it, thus obtainThe matrix of all analog structures will be comprised
It is referred to as group.
Preferably, in S4 step, carry out initial estimation by low-rank method, to noisy image setsCarry out singular value decomposition
(SVD):WhereinK=min (c, Bs) again to λ iiCarry out threshold process: It is soft-threshold operator,WhereinσwIt is to estimate
The noise criteria of meter is poor, and initial estimation group is:WhereinK
=min (c, Bs).。
Preferably, in S6 step, after the sparse coding of noisy group and estimation group all updates well, utilize soft-threshold side
Method obtains the image after denoising.
Compared with prior art, the invention has the beneficial effects as follows: 1, rarefaction representation is trained dictionary to introduce removal by the present invention
Additive noise, by the information of minimum whole image of element representation, so greatly reduces workload, removes additive noise by learning
Handwriting practicing allusion quotation, so more effective denoising.2, the present invention is with the group of image block for the ultimate unit processed, and makes full use of image similarity
Contact between block, preferably retains the marginal texture of image.3, non local regularization is introduced and removes additive noise by the present invention,
Be conducive to the contact between contact image block, strengthen denoising.4, the present invention uses low-rank decomposition to carry out the non-local estimation of image,
The artefact occurred in image can be reduced, improve the visual effect of image, improve the quality of image.
Accompanying drawing explanation
Fig. 1 is that a kind of non-local sparse based on group of the present invention represents that additive noise minimizing technology is in white Gaussian noise standard
Difference is denoising figure and the partial enlarged drawing of " Peppers " figure of " Peppers " figure of 30;
Fig. 2 is that a kind of non-local sparse based on group of the present invention represents that additive noise minimizing technology is in white Gaussian noise standard
Difference is denoising figure and the partial enlarged drawing of " House " figure of " House " figure of 50.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on
Embodiment in the present invention, those of ordinary skill in the art obtained under not making creative work premise all its
His embodiment, broadly falls into the scope of protection of the invention.
Referring to Fig. 1-2, the present invention provides a kind of technical scheme: below in conjunction with the detailed description of the invention technology to this patent
Scheme is described in more detail.
The present embodiment proposes a kind of non-local sparse based on group and represents additive noise minimizing technology: described based on group
Non-local sparse represents that additive noise minimizing technology comprises the following steps:
S1: obtain natural image in standard picture storehouse, and image is carried out image block and instructs with gauss hybrid models
Practise handwriting allusion quotation;
S2: each image carries out plus noise process, utilizes the side that rarefaction representation based on group and non-local sparse represent
Method obtains removing noise model;
S3: noisy image is divided into the fritter having overlap, carries out non local Similarity matching for each image block, find it to have
There is the similar image block of identical structure type, some the highest for similarity blocks are put in a group;In S3 step, first will
Size is that the image x of N × N is divided into the overlapping block that n size is p × p, is defined asK=1,2 ..., n. then,
For block xk, search is mated most with it in training window c block group, form and gatherIt follows that willIn all of piece
Forming a size is p2The matrix of × c, is defined as In all of piece of row as it, thus obtainThe matrix of all analog structures will be comprisedIt is referred to as group;
S4: for each group divided, first carry out singular value decomposition to each group, then eigenvalue is carried out at threshold value
Reason, obtains the estimated value of each group;In S4 step, carry out initial estimation by low-rank method, to noisy image setsEnter
Row singular value decomposition (SVD):WhereinK=min (c, Bs) again
To λ iiCarry out threshold process: It is soft-threshold operator,WhereinσwBeing that the noise criteria estimated is poor, initial estimation group is:Its
InK=min (c, Bs).;
S5: utilize the dictionary of training in advance to obtain estimation group sparse coding in dictionary;
S6: utilize soft threshold method solving model;In S6 step, the sparse coding at noisy group and estimation group all updates
Image after Hao, after utilizing soft threshold method to obtain denoising;
S7: obtain denoising image.
The enforcement step removed based on the mixed noise that weighting is sparse is as follows:
Step 1, obtains natural image in standard picture storehouse, and gray value is between 0-255, in order to simplify amount of calculation to often
Width figure is all normalized, and again image is carried out piecemeal after normalized, and the size present invention of every piece is divided into P × P,
P=7 during noise variance σ≤30,30 < p=8 during σ≤50,50 < p=9 during σ≤100.By the image after piecemeal, use Gaussian Mixture
Image block is divided into L class by model, obtains the covariance matrix ∑ of each classiAnd it is carried out singular value decomposition ∑i=ΦiΛ
Φi T, wherein ΦiAn orthogonal matrix, Λ be the element on diagonal matrix diagonal be eigenvalue, dictionary be designated as Φ=
[Φ1,Φ2,…,ΦL];
Step 2, carries out plus noise process to each image, if y is ∈ RnFor noisy observation figure, x ∈ RnFor denoising figure, n is
White Gaussian noise, then the mathematical formulae of white Gaussian noise is y=x+n, and we add white Gaussian noise standard respectively to test image
Difference be 30,50,100. then utilize based on group rarefaction representation and non-local sparse represent obtain remove noise model;
Step 3, is first divided into, by the image x that size is N × N, the overlapping block that n size is p × p, is defined asK=1,2 ..., n. then, for block xk, search is mated most with it in training window c block group, form and collect
CloseIt follows that willIn all of piece composition one size be p2The matrix of × c, is defined as Middle institute
Some blocks are as its row, thus obtainThe matrix of all analog structures will be comprisedIt is referred to as
Group;
Step 4, carries out initial estimation by low-rank method.To noisy image setsCarry out singular value decomposition:WhereinK=min (c, Bs) again to λ iiCarry out threshold process: It is soft-threshold operator,WhereinσwIt is to estimate
The noise criteria of meter is poor, and initial estimation group is:WhereinK
=min (c, Bs).;
Step 5, utilizes the dictionary of training in advance to obtain estimation group sparse coding in dictionary;
(5.1) for each groupThe gauss hybrid models that we to train in step 1 is chosen most suitable class,
Probability in which kind of is the biggest the most suitable,The probability belonging to l class is: Corresponding dictionary is exactly the dictionary Φ of the apoplexy due to endogenous wind of maximum probabilityk;
(5.2) it is orthogonal due to dictionary, soSparse coding in dictionary isNon local estimate
The sparse coding of meter is
Step 6, utilizes the method for the proxy function of contraction operator to obtain sparse coding, the α of t+1 step iterationGFor:WhereinIt is soft-threshold operator, τ=λ ii,j/c1,c1It is auxiliary parameter, It is the variance of α-μ,ε is a non-zero constant the least, prevents denominator
It is zero,
Step 7, at sparse codingAfter updating well, after determining, the group recovered isWillRevert to original position, obtain final recovery figure
The effect of the present invention is further illustrated by following emulation.
1. simulated conditions
(1) choose natural image in the standard testing image storehouse that size is 256 × 256 to test;
(2) in experiment, the size of image block is p × p, the p=7 when noise criteria difference σ≤30,30 < p=8 during σ≤50,50
< p=9 during σ≤100, size W=31 of search window, auxiliary parameter c in the number c=10. soft-threshold of each group of image block1
=0.28, γ=0.67, δ=0.65, iterations is all 8 times.The white Gaussian noise standard deviation that the image of test adds is respectively
30、50、100。
2. emulation content and result
Emulation content: utilize Peppers figure and the House of 256 × 256 to scheme, calculate with the NCSR of the present invention with prior art
Method, it is removed additive white Gaussian noise by PGPD algorithm respectively.
Table 1 experimental result Y-PSNR (PSNR)
Table 2 experimental result characteristic similarity (SSIM)
Experimental result:
The present invention obtains good denoising effect than NCSR algorithm and PGPD algorithm on Y-PSNR PSNR and FSIM.
Experimental result is as shown in table 1.As can be seen from the table, the present invention is compared with NCSR algorithm and PGPD algorithm, at different noises
With on test image, major part experiment achieves higher PSNR value and FSIM value.
The present invention obtains more preferable denoising effect than NCSR algorithm and PGPD algorithm in visual effect, experimental result such as figure
1 and as shown in Figure 2.
Fig. 1 be white Gaussian noise standard deviation be the denoising figure of 30, Fig. 1 (a) is original image and partial enlarged drawing, and Fig. 1 (b) is
The denoising figure of NCSR algorithm and partial enlarged drawing, from Fig. 1 (b) it can be seen that the comparison of image procossing obscures, lose at the Fructus Capsici base of a fruit
More details, Fig. 1 (c) is denoising figure and the partial enlarged drawing of PGPD algorithm, although PGPD algorithm is at Y-PSNR and figure
As characteristic similarity is better than NCSR algorithm, the denoising effect of image is also better than NCSR algorithm, but NCSR denoising figure is peppery
The green pepper base of a fruit, still loses detailed information, and Fig. 1 (d) is the denoising figure of this patent algorithm, can be evident that from Fig. 1 (d), this
Proprietary algorithms can be good at preserving the detailed information of image.
Fig. 2 be white Gaussian noise standard deviation be the denoising figure of 50, Fig. 2 (a) is original image and partial enlarged drawing, and Fig. 2 (b) is
The denoising figure of NCSR algorithm and partial enlarged drawing, from Fig. 2 (b) it can be seen that the contour line of the eaves of the denoising figure of NCSR obscures
And inner side has lines to lose, smudgy and still with the presence of noise below eaves, Fig. 2 (c) is the denoising figure of PGPD model, room
Eaves contour edge is good, the most clearly, but occurs in that many artefacts in the overall situation, affects visual effect.Can from Fig. 2 (d)
Going out, compared with the algorithm above, the noise removal image ratio of the present invention is more visible, all processes at edge, house and detail section
Be closer to original image, in the overall situation occur artefact less.
The present invention carries out removing of picture noise by representing based on the rarefaction representation organized and non-local sparse.With other algorithms
Comparing, the present invention has while denoising, can preferably suppress artefact to produce, and has in terms of retaining image key character
There is good performance.Emulation experiment shows, the present invention either still shows excellent at subjective judgement in objective quantification.
Although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, permissible
Understand and these embodiments can be carried out multiple change without departing from the principles and spirit of the present invention, revise, replace
And modification, the scope of the present invention be defined by the appended.
Claims (4)
1. a non-local sparse based on group represents additive noise minimizing technology, it is characterised in that: described non-office based on group
Portion's rarefaction representation additive noise minimizing technology comprises the following steps:
S1: obtain natural image in standard picture storehouse, and image is carried out image block and trains word with gauss hybrid models
Allusion quotation;
S2: each image carries out plus noise process, the method utilizing rarefaction representation based on group and non-local sparse to represent obtains
To removing noise model;
S3: noisy image is divided into the fritter having overlap, carries out non local Similarity matching for each image block, find it to have phase
With the similar image block of structure type, some the highest for similarity blocks are put in a group;
S4: for each group divided, first carry out singular value decomposition to each group, then eigenvalue is carried out threshold process,
To the estimated value of each group;
S5: utilize the dictionary of training in advance to obtain estimation group sparse coding in dictionary;
S6: utilize soft threshold method solving model;
S7: obtain denoising image.
A kind of non-local sparse based on group the most according to claim 1 represents additive noise minimizing technology, and its feature exists
In: in S3 step, first the image x that size is N × N is divided into the overlapping block that n size is p × p, is defined asK=1,2 ..., n. then, for block xk, search is mated most with it in training window c block group, form and gatherIt follows that willIn all of piece composition one size be p2The matrix of × c, is defined as In all
Block as its row, thus obtainThe matrix of all analog structures will be comprisedIt is referred to as group.
A kind of non-local sparse based on group the most according to claim 1 represents additive noise minimizing technology, and its feature exists
In: in S4 step, carry out initial estimation by low-rank method, to noisy image setsCarry out singular value decomposition (SVD):WhereinK=min (c, Bs) again to λiCarry out threshold process: It is soft-threshold operator,WhereinσwIt it is the noise estimated
Standard deviation, initial estimation group is:Wherein
A kind of non-local sparse based on group the most according to claim 1 represents additive noise minimizing technology, and its feature exists
In: in S6 step, after the sparse coding of noisy group and estimation group all updates well, after utilizing soft threshold method to obtain denoising
Image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610548312.5A CN106169180A (en) | 2016-07-13 | 2016-07-13 | A kind of non-local sparse based on group represents additive noise minimizing technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610548312.5A CN106169180A (en) | 2016-07-13 | 2016-07-13 | A kind of non-local sparse based on group represents additive noise minimizing technology |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106169180A true CN106169180A (en) | 2016-11-30 |
Family
ID=58064490
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610548312.5A Pending CN106169180A (en) | 2016-07-13 | 2016-07-13 | A kind of non-local sparse based on group represents additive noise minimizing technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106169180A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107358589A (en) * | 2017-07-17 | 2017-11-17 | 桂林电子科技大学 | A kind of combination histogram of gradients and the denoising method of low-rank constraint |
CN108537752A (en) * | 2018-03-30 | 2018-09-14 | 广东工业大学 | Image processing method based on non local self-similarity and rarefaction representation and device |
CN112801884A (en) * | 2020-11-26 | 2021-05-14 | 四川长虹电器股份有限公司 | Image denoising method based on external non-local self-similarity and improved sparse representation |
CN112908420A (en) * | 2020-12-02 | 2021-06-04 | 中山大学 | Multi-mathematical data integration method and system based on denoising network regularization |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105447836A (en) * | 2015-12-30 | 2016-03-30 | 南京信息工程大学 | Non-local sparse representation image de-noising method of coupling clustering center bound term |
-
2016
- 2016-07-13 CN CN201610548312.5A patent/CN106169180A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105447836A (en) * | 2015-12-30 | 2016-03-30 | 南京信息工程大学 | Non-local sparse representation image de-noising method of coupling clustering center bound term |
Non-Patent Citations (6)
Title |
---|
J ZHANG 等: "Group-based sparse representation for image restoration", 《IEEE TRANSACTIONS ON IMAGE PROCESSING A PUBLICATION OF THE IEEE SIGNAL PROCESSING SOCIETY》 * |
JUN XU等: "Patch Group Based Nonlocal Self-Similarity Prior Learning for Image Denoising", 《IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
SG CHANG 等: "Adaptive wavelet thresholding for image denoising and compression.", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
WEISHENG DONG 等: "Nonlocal Image Restoration With Bilateral Variance Estimation A Low-Rank Approach", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
赛朋飞: "基于稀疏表示与非局部相似的图像去噪算法研究", 《万方数据》 * |
陈利霞 等: "组约束与非局部稀疏的图像去噪算法", 《微电子学与计算机》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107358589A (en) * | 2017-07-17 | 2017-11-17 | 桂林电子科技大学 | A kind of combination histogram of gradients and the denoising method of low-rank constraint |
CN107358589B (en) * | 2017-07-17 | 2019-11-26 | 桂林电子科技大学 | A kind of denoising method of combination histogram of gradients and low-rank constraint |
CN108537752A (en) * | 2018-03-30 | 2018-09-14 | 广东工业大学 | Image processing method based on non local self-similarity and rarefaction representation and device |
CN108537752B (en) * | 2018-03-30 | 2022-06-24 | 广东工业大学 | Image processing method and device based on non-local self-similarity and sparse representation |
CN112801884A (en) * | 2020-11-26 | 2021-05-14 | 四川长虹电器股份有限公司 | Image denoising method based on external non-local self-similarity and improved sparse representation |
CN112801884B (en) * | 2020-11-26 | 2022-04-05 | 四川长虹电器股份有限公司 | Image denoising method based on external non-local self-similarity and improved sparse representation |
CN112908420A (en) * | 2020-12-02 | 2021-06-04 | 中山大学 | Multi-mathematical data integration method and system based on denoising network regularization |
CN112908420B (en) * | 2020-12-02 | 2023-07-04 | 中山大学 | Multi-group chemical data integration method and system based on denoising network regularization |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114140353B (en) | Swin-Transformer image denoising method and system based on channel attention | |
CN107492070B (en) | A kind of single image super-resolution calculation method of binary channels convolutional neural networks | |
Yin et al. | Highly accurate image reconstruction for multimodal noise suppression using semisupervised learning on big data | |
CN106204482B (en) | Based on the mixed noise minimizing technology that weighting is sparse | |
CN110796625B (en) | Image compressed sensing reconstruction method based on group sparse representation and weighted total variation | |
CN102184533B (en) | Non-local-restriction-based total variation image deblurring method | |
CN107358589B (en) | A kind of denoising method of combination histogram of gradients and low-rank constraint | |
CN102567973B (en) | Image denoising method based on improved shape self-adaptive window | |
CN109948573A (en) | A kind of noise robustness face identification method based on cascade deep convolutional neural networks | |
CN106169180A (en) | A kind of non-local sparse based on group represents additive noise minimizing technology | |
CN105279740A (en) | Image denoising method based on sparse regularization | |
WO2018227882A1 (en) | A priori constraint and outlier suppression based image deblurring method | |
CN103208097B (en) | Filtering method is worked in coordination with in the principal component analysis of the multi-direction morphosis grouping of image | |
CN102393955A (en) | Perfect information non-local constraint total variation method for image recovery | |
CN105046664A (en) | Image denoising method based on self-adaptive EPLL algorithm | |
CN112270654A (en) | Image denoising method based on multi-channel GAN | |
CN110390646B (en) | Detail-preserving image denoising method | |
CN110519595A (en) | A kind of jpeg compressed image restored method based on frequency domain quantization estimated amount of damage | |
CN104616249A (en) | Curvature variation based wavelet image denoising algorithm | |
CN109493295A (en) | A kind of non local Haar transform image de-noising method | |
CN107169932A (en) | A kind of image recovery method based on Gauss Poisson mixed noise model suitable for neutron imaging system diagram picture | |
CN105913383A (en) | Image noise reduction method based on image block prior estimation mixed framework | |
CN113592728A (en) | Image restoration method, system, processing terminal and computer medium | |
CN107292855A (en) | A kind of image de-noising method of the non local sample of combining adaptive and low-rank | |
CN107085834A (en) | A kind of image de-noising method based on image rotation and piecemeal singular value decomposition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161130 |