CN112258397A - Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraints - Google Patents
Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraints Download PDFInfo
- Publication number
- CN112258397A CN112258397A CN202010758248.XA CN202010758248A CN112258397A CN 112258397 A CN112258397 A CN 112258397A CN 202010758248 A CN202010758248 A CN 202010758248A CN 112258397 A CN112258397 A CN 112258397A
- Authority
- CN
- China
- Prior art keywords
- image
- similar
- sensing reconstruction
- residual
- low rank
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012546 transfer Methods 0.000 title abstract description 11
- 238000012937 correction Methods 0.000 claims description 38
- 239000013598 vector Substances 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000012935 Averaging Methods 0.000 claims description 6
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 6
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 14
- 230000000694 effects Effects 0.000 abstract description 7
- 230000006835 compression Effects 0.000 abstract description 2
- 238000007906 compression Methods 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 238000005070 sampling Methods 0.000 description 5
- 238000004088 simulation Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 244000025254 Cannabis sativa Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to the technical field of image processing, in particular to an image compression sensing reconstruction method based on approximate message transfer and double low-rank constraint. The invention solves the reconstruction problem by adopting an approximate message transfer algorithm, and has more ideal compressed sensing reconstruction effect than a common iteration threshold algorithm. According to the non-local self-similarity of the natural image, the problem that the natural image does not have wavelet domain or gradient domain sparse constraint is solved by adopting similar image block low-rank constraint. In addition, residual error low-rank constraint is used on the basis of similar image block low-rank constraint, reconstruction quality is improved, and the problem that partial regions are fuzzy when the similarity of image similar blocks is not high is effectively solved.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraint.
Background
Compressed sensing is a new signal processing method that has appeared in recent years and is now being used more and more widely. The compressed sensing can break through the constraint of the Shannon-Nyquist sampling theorem, sampling is carried out at a bandwidth far less than twice the Nyquist bandwidth, sampling and compression of signals are simultaneously realized, an observed value is obtained by carrying out dimension reduction sampling, an original signal is accurately recovered by utilizing a reconstruction algorithm, and high attention and application are obtained in the fields of medical imaging, wireless communication, radar detection and the like. The compressed sensing mainly comprises three parts of sparse representation, nonlinear measurement and image reconstruction. In the current image reconstruction algorithm, a wavelet domain or gradient domain sparse constraint is usually adopted in an approximate message transfer algorithm based on a sparse signal, a natural image does not have the sparse characteristic but has non-local self-similarity, the adopted general iterative threshold algorithm is not ideal in compressed sensing reconstruction effect, the problem of partial region blurring exists in the image reconstruction process, and meanwhile, the calculated amount is large.
Disclosure of Invention
In order to solve the problems, the invention provides an image compressed sensing reconstruction method based on approximate message transfer and double low rank constraints.
The invention is realized by adopting the following scheme:
an image compressed sensing reconstruction method based on approximate message passing and double low rank constraints comprises the following steps:
estimating an Onsager correction term in residual error updating by adopting a Monte Carlo method and updating the residual error;
step two, calculating a noisy image qt;
Step three, clustering similar image blocks of the noisy images;
step four, low-rank correction is carried out on the similar image blocks;
step five, carrying out low-rank correction on the residual error;
and sixthly, averaging the overlapped image blocks to obtain a reconstructed image.
Further, in the first step, the Onsager correction term in the residual error update is estimated by using a monte carlo method, and the residual error update specifically includes the following steps:
wherein z istIs the residual of the t-th iteration, xtIs an estimate of the original signal x for the tth iteration,in order to be the Onsager correction term,it is the proximity correction operation that is performed,as a function of a threshold valueDerivative of (A)*Representing the conjugate transpose of the observation matrix.
The derivative operation in the Onsager correction term adopts a Monte Carlo random number approximation method:
where E represents the mathematical expectation of solving for a random variable, EbRepresenting the mathematical expectation of finding the random variable b, the random vectors b-N (0, I) are independently and identically distributed, and epsilon is a very small positive number.
Further, the noise-containing image calculated in the second step is specifically as follows:
qt=xt+A*zt。
further, the step three of clustering similar image blocks of the noisy image refers to: noise-containing image qtOverlapping and partitioning, searching similar image blocks in a search window by taking a reference block as a standard, scanning each image block according to rows to form column vectors, and rearranging the column vectors of the similar block groups to obtain a two-dimensional matrix, whereinAnd representing the two-dimensional matrix obtained by rearranging the ith similar block group.
Further, the low rank correction on the similar image blocks in the fourth step is specifically as follows:
to pairPerforming singular value threshold operation onVector of singular values pi=[ρi,1,ρi,2,...,ρi,N]Each element of (1) uses a corresponding weight value wi,jPerforming soft threshold operation by Bayesian threshold method, wiAdaptively setting according to the signal-to-noise ratio of each similar block:
wherein sigmai=[σi,1,σi,2,...,σi,N]The standard deviation vector for a similar block, equal to its singular value squared minus the noise variance:
further, the step five performs low rank correction on the residual error specifically as follows:
wherein, for residual errorPerforming singular value threshold operation onSingular value vector delta ofi=[δi,1,δi,2,...,δi,N]Each element of (1) uses a corresponding weight value gammai,jPerforming soft threshold operation; gamma rayiAnd adaptively setting according to the signal-to-noise ratio of each residual block:
wherein the content of the first and second substances,the standard deviation vector for the residual, equal to its singular value squared minus the noise variance:
further, the averaging of the overlapped blocks in the step 6 is specifically as follows:
compared with the prior art, the invention has the following beneficial effects:
1. the invention solves the reconstruction problem by adopting an approximate message transfer algorithm, and has more ideal compressed sensing reconstruction effect than a common iteration threshold algorithm.
2. According to the invention, the problem that the natural image does not have wavelet domain or gradient domain sparse constraint is effectively solved by adopting similar image block low-rank constraint according to the fact that the natural image has non-local self-similarity.
3. On the basis of the low-rank constraint of the similar image blocks, the residual error low-rank constraint is used, the residual error low-rank constraint has better detail retention capacity, the problem of partial region blurring can be improved, the reconstruction quality is improved, and the problem of partial region blurring when the similarity of the image similar blocks is not high is effectively solved.
Drawings
FIG. 1 is a flowchart of an image compressed sensing reconstruction method based on approximate message passing and double low rank constraints according to the present invention.
Fig. 2 is a comparison graph of simulation effect of the c.man image of the present invention.
Fig. 3 is a comparison of simulated effects of barbarbara images of the invention.
Detailed Description
To facilitate an understanding of the present invention for those skilled in the art, the present invention will be described in further detail below with reference to specific embodiments and accompanying drawings.
Referring to fig. 1, the present invention provides an image compressive sensing reconstruction method based on approximate message passing and dual low rank constraints, and in order to make the present invention more clearly and easily understood, the following describes the reconstruction method further. The compressed sensing reconstruction technology obtains an observed value y which is Ax + v and y belongs to CmMiddle estimation original signal x ∈ CnWherein A ∈ CmIs an observation matrix, v is a variance of σ2Gaussian noise. Compressed perceptual reconstruction is an underdetermined problem since m < n. The problem can be used as a constraint condition through the prior knowledge of the signal to recover the original signal.
From an initial state x0=0,z0Starting with y, the specific implementation iteration steps of the invention are as follows:
estimating an Onsager correction term in residual error updating by adopting a Monte Carlo method and updating the residual error;
step two, calculating the noiseImage qt;
Step three, clustering similar image blocks of the noisy images;
step four, low-rank correction is carried out on the similar image blocks;
step five, carrying out low-rank correction on the residual error;
and sixthly, averaging the overlapped image blocks to obtain a reconstructed image.
In the first step, the approximate message transfer algorithm approximates the residual error by using the central limit theorem and the Taylor expansion, so that the calculation amount is greatly reduced under the high-dimensional condition, and the residual error is one Onsager correction term more than that of the iterative threshold algorithm.
In the first step, the Onsager correction term needs to calculate the derivative of the adjacent correction function, and the adjacent correction function has no definite input-output relation, and the derivative is difficult to calculate, so the derivative operation of the Onsager correction term is estimated by adopting a Monte Carlo random number approximation method.
And thirdly, overlapping and blocking the noisy images, searching similar blocks according to Euclidean distances, and rearranging the groups of the similar image blocks to obtain a two-dimensional matrix.
And step four, using singular value decomposition to the two-dimensional matrix of the similar image block group to enable the similar blocks in the similar image block group to have similar sparse coefficients, and adopting a weighted nuclear norm to express the low-rank constraint of the similar image block in order to reflect the different importance of different sparse coefficients.
And fifthly, singular value threshold operation is carried out on the residual two-dimensional matrix, and similarly, the low-rank constraint of the residual is represented by adopting a weighted kernel norm.
In the first step, a Monte Carlo method is adopted to estimate Onsager correction terms in residual error updating, and the updating residual error is specifically as follows:
wherein z istIs the residual of the t-th iteration, xtIs an estimate of the original signal x for the tth iteration,in order to be the Onsager correction term,it is the proximity correction operation that is performed,as a function of a threshold valueDerivative of (A)*Representing the conjugate transpose of the observation matrix. Residual ztResidual y-Ax of iterative thresholding algorithm(t)Add one more itemI.e., the aforementioned Onsager correction term.
In the specific implementation, the main difficulty lies in the Onsager correction term, because it needs to calculate the proximity correction function l (q)t) Derivative (introduction of an auxiliary variable q)t=xt+A*zt) And l (q)t) The function has no explicit input-output relationship. Therefore, the derivative operation in the Onsager correction term adopts a monte carlo random number approximation method:
where E represents the mathematical expectation of solving for a random variable, EbRepresenting the mathematical expectation of solving the random variable b, and independently and identically distributed random vectors b-N (0, I), wherein epsilon is a positive number, in particular a very small positive number. In the calculation, firstly, L independent and equally distributed random vectors b are generated1,b2,…bL(ii) a Then for each vector bjCalculating an estimateFinally, the average value of the estimated values is calculatedAccording to the weak logarithm theorem, when L → ∞, the estimated value converges to the true derivative value.
The noise-containing image calculated in the second step is specifically as follows:
qt=xt+A*zt。
the step three of clustering similar image blocks of the noisy image refers to: noise-containing image qtOverlapping and partitioning, searching similar image blocks in a search window by taking a reference block as a standard, scanning each image block according to rows to form column vectors, and rearranging the column vectors of the similar block groups to obtain a two-dimensional matrix, whereinAnd representing the two-dimensional matrix obtained by rearranging the ith similar block group. In this embodiment, the size of an image block is 6 × 6, the number of image blocks included in the similar block group is 40, the size of the search window is 30 × 30, and the interval between two reference image blocks is 5 pixels. Of course, the specific implementation can be self-adaptive according to specific requirements.
The low rank correction of the similar image blocks in the fourth step is specifically as follows:
is FiThe weighted kernel norm is equal to the weighted sum of the singular values of the matrix. FiIs QiDenoised values. To pairPerforming singular value threshold operation onVector of singular values pi=[ρi,1,ρi,2,...,ρi,N]Each of (1)Each element uses a corresponding weight wi,jPerforming soft threshold operation by Bayesian threshold method, wiAdaptively setting according to the signal-to-noise ratio of each similar block:
wherein sigmai=[σi,1,σi,2,...,σi,N]The standard deviation vector for a similar block, equal to its singular value squared minus the noise variance:
whereinFor the noise variance, the value is estimated here by means of the properties of an approximate message transfer algorithm, i.e. the residual z considered to be corrected by the Onsager termtIs random noise with a noise variance of ztTo estimate the energy of:
the step five of performing low rank correction on the residual error specifically comprises the following steps:
Riis the residual error after low rank correction, RiIs thatDenoised values. Wherein, for residual errorPerforming singular value threshold operation onSingular value vector delta ofi=[δi,1,δi,2,...,δi,N]Each element of (1) uses a corresponding weight value gammai,jPerforming soft threshold operation; gamma rayiAnd adaptively setting according to the signal-to-noise ratio of each residual block:
wherein the content of the first and second substances,the standard deviation vector for the residual, equal to its singular value squared minus the noise variance:
the averaging of the overlapped blocks in the step 6 is specifically as follows:
in specific implementation, the results of low-rank correction and residual low-rank correction of similar image blocks are correspondingly summed according to pixels. Since the image blocks are divided in an overlapping manner, and the same pixel is divided into a plurality of different similar block groups, the pixels in the overlapping area need to be averaged.
Referring to fig. 2 and 3, in order to verify the effect of the present invention, the present example performed a simulation experiment comparing the reconstruction performance of the present invention with other 4 methods (original AMP method, tree sparsity based method WaTMRI and Turo-AMP, non-local sparsity based method BM 3D-AMP). In simulation, two-dimensional compressed sensing observation is firstly carried out on an image, and then reconstruction is carried out by adopting various methods. The method is characterized in that a commonly used C.man image and a Barbara image are selected as experimental images, the size of the images is 128 x 128, the peak signal-to-noise ratio (PSNR) is used as a performance index, and the higher the PSNR value is, the better the reconstruction performance is.
Setting parameters: the number of iterations of the inventive method, AMP, WaTMRI, and BM3D-AMP methods was set to 50, and the number of iterations of Turo-AMP and BM3D-AMP methods was set to 20 and 100, respectively. The regularization factor α of WaTMRI method is 0.8 and β is 0.35. The size of the similar image blocks of the method is 6 multiplied by 6, one similar block group comprises 40 image blocks, the range of a search window is 30 multiplied by 30, and the reference image blocks are separated by 5 pixels.
And (3) simulation results: fig. 2 and 3 show the reconstruction results of 5 methods at 20% sampling rate for c.man images and Barbara images, respectively. Comparing the results of the reconstructions it can be found that the PSNR values of the proposed method are higher than the other 4 methods, indicating that the method of the invention is superior to the other methods and it can be seen that the reconstruction results of the method of the invention are clearly clearer than the other methods, in particular at details such as grass in c.man images and hood in barbarara images.
The invention solves the reconstruction problem by adopting an approximate message transfer algorithm, and has more ideal compressed sensing reconstruction effect than a common iteration threshold algorithm. According to the invention, the problem that the natural image does not have wavelet domain or gradient domain sparse constraint is effectively solved by adopting similar image block low-rank constraint according to the fact that the natural image has non-local self-similarity. On the basis of the low-rank constraint of the similar image blocks, the residual error low-rank constraint is used, the residual error low-rank constraint has better detail retention capacity, the problem of partial region blurring can be improved, the reconstruction quality is improved, and the problem of partial region blurring when the similarity of the image similar blocks is not high is effectively solved.
In the description of the present invention, it is to be understood that the description is for convenience only and is simplified in order not to indicate or imply that the device or element so referred to must be constructed and operated in a particular orientation or orientation and, therefore, should not be considered as limiting.
Furthermore, the terms "first", "second", etc. are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "connected," "secured," and the like are to be construed broadly, e.g., as meaning permanently attached, removably attached, or integral to one another; can be mechanically or electrically connected; either directly or indirectly through intervening media, either internally or in any other relationship. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
While the invention has been described in conjunction with the specific embodiments set forth above, it is evident that many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the foregoing description. Accordingly, it is intended to embrace all such alternatives, modifications, and variations that fall within the scope of the included claims.
Claims (8)
1. An image compressed sensing reconstruction method based on approximate message passing and double low rank constraints is characterized by comprising the following steps:
estimating an Onsager correction term in residual error updating by adopting a Monte Carlo method and updating the residual error;
step two, calculating a noisy image qt;
Step three, clustering similar image blocks of the noisy images;
step four, low-rank correction is carried out on the similar image blocks;
step five, carrying out low-rank correction on the residual error;
and sixthly, averaging the overlapped image blocks to obtain a reconstructed image.
2. The image compressive sensing reconstruction method based on approximate message passing and double low rank constraints according to claim 1, wherein the first step adopts a monte carlo method to estimate the Onsager correction term in the residual update, and the residual update is as follows:
wherein z istIs the residual of the t-th iteration, xtIs an estimate of the original signal x for the tth iteration,in order to be the Onsager correction term,it is the proximity correction operation that is performed,as a function of a threshold valueDerivative of (A)*Representing the conjugate transpose of the observation matrix.
3. The image compressive sensing reconstruction method based on approximate messaging and double low rank constraints according to claim 2, wherein the derivative operation in the Onsager correction term adopts a Monte Carlo random number approximation method:
where E represents the mathematical expectation of solving for a random variable, EbRepresenting the mathematical expectation of finding the random variable b, the random vectors b-N (0, I) are independently and identically distributed, and epsilon is a very small positive number.
4. The method for compressed sensing reconstruction of images based on approximate messaging and dual low rank constraints according to claim 3, wherein the second step of computing noisy images is as follows:
qt=xt+A*zt。
5. the method for image compressed sensing reconstruction based on approximate message passing and double low rank constraints according to claim 4, wherein the step three for similar image block clustering of noisy images comprises: noise-containing image qtOverlapping and partitioning, searching similar image blocks in a search window by taking a reference block as a standard, scanning each image block according to rows to form column vectors, and rearranging the column vectors of the similar block groups to obtain a two-dimensional matrix, whereinAnd representing the two-dimensional matrix obtained by rearranging the ith similar block group.
6. The method for compressed image sensing reconstruction based on approximate message passing and double low rank constraints according to claim 5, wherein the low rank correction on similar image blocks in the fourth step is specifically as follows:
to pairPerforming singular value threshold operation onVector of singular values pi=[ρi,1,ρi,2,...,ρi,N]Each element of (1) uses a corresponding weight value wi,jPerforming soft threshold operation by Bayesian threshold method, wiAdaptively setting according to the signal-to-noise ratio of each similar block:
wherein sigmai=[σi,1,σi,2,...,σi,N]The standard deviation vector for a similar block, equal to its singular value squared minus the noise variance:
7. the method for image compressive sensing reconstruction based on approximate message passing and double low rank constraints as claimed in claim 6, wherein the step five performs low rank correction on the residual error specifically as follows:
wherein, for residual errorPerforming singular value threshold operation onSingular value vector delta ofi=[δi,1,δi,2,...,δi,N]Each element of (1) uses a corresponding weight value gammai,jPerforming soft threshold operation; gamma rayiAnd adaptively setting according to the signal-to-noise ratio of each residual block:
wherein the content of the first and second substances,the standard deviation vector for the residual, equal to its singular value squared minus the noise variance:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010758248.XA CN112258397B (en) | 2020-07-31 | 2020-07-31 | Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010758248.XA CN112258397B (en) | 2020-07-31 | 2020-07-31 | Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112258397A true CN112258397A (en) | 2021-01-22 |
CN112258397B CN112258397B (en) | 2024-01-05 |
Family
ID=74223889
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010758248.XA Active CN112258397B (en) | 2020-07-31 | 2020-07-31 | Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112258397B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105513026A (en) * | 2015-12-17 | 2016-04-20 | 浙江大学 | Compressed sensing reconstruction method based on image nonlocal similarity |
US20160358316A1 (en) * | 2015-06-05 | 2016-12-08 | North Carolina State University | Approximate message passing with universal denoising |
US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
CN109584319A (en) * | 2018-12-05 | 2019-04-05 | 重庆邮电大学 | A kind of compression of images sensing reconstructing algorithm based on non-local low rank and full variation |
CN110113607A (en) * | 2019-04-25 | 2019-08-09 | 长沙理工大学 | A kind of compressed sensing video method for reconstructing based on part and non-local constraint |
-
2020
- 2020-07-31 CN CN202010758248.XA patent/CN112258397B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160358316A1 (en) * | 2015-06-05 | 2016-12-08 | North Carolina State University | Approximate message passing with universal denoising |
CN105513026A (en) * | 2015-12-17 | 2016-04-20 | 浙江大学 | Compressed sensing reconstruction method based on image nonlocal similarity |
US20190050372A1 (en) * | 2017-08-14 | 2019-02-14 | City University Of Hong Kong | Systems and methods for robust low-rank matrix approximation |
CN109584319A (en) * | 2018-12-05 | 2019-04-05 | 重庆邮电大学 | A kind of compression of images sensing reconstructing algorithm based on non-local low rank and full variation |
CN110113607A (en) * | 2019-04-25 | 2019-08-09 | 长沙理工大学 | A kind of compressed sensing video method for reconstructing based on part and non-local constraint |
Non-Patent Citations (2)
Title |
---|
ZHONG-HUA XIE 等: "a probabilistic model-based method with nonlocal filtering for robust magnetic resonance imaging reconstruction", 《IEEEACESS》, pages 1 - 18 * |
谢中华: "基于图像非局部结构稀疏模型的压缩感知重构", 《中国博士学位论文全文数据库信息科技辑》, pages 138 - 107 * |
Also Published As
Publication number | Publication date |
---|---|
CN112258397B (en) | 2024-01-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109102477B (en) | Hyperspectral remote sensing image recovery method based on non-convex low-rank sparse constraint | |
Dong et al. | Image deblurring and super-resolution by adaptive sparse domain selection and adaptive regularization | |
CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
CN110675347A (en) | Image blind restoration method based on group sparse representation | |
Liu et al. | Multi-filters guided low-rank tensor coding for image inpainting | |
CN107451961B (en) | Method for recovering sharp image under multiple fuzzy noise images | |
CN110830043B (en) | Image compressed sensing reconstruction method based on mixed weighted total variation and non-local low rank | |
Xiao et al. | Bayesian sparse hierarchical model for image denoising | |
CN114202459B (en) | Blind image super-resolution method based on depth priori | |
Wen et al. | The power of complementary regularizers: Image recovery via transform learning and low-rank modeling | |
Gao et al. | A novel fractional‐order reaction diffusion system for the multiplicative noise removal | |
CN108537853B (en) | Underwater sonar image compression transmission method | |
CN112270650B (en) | Image processing method, system, medium, and apparatus based on sparse autoencoder | |
CN111243047B (en) | Image compression sensing method based on self-adaptive nonlinear network and related product | |
Gong et al. | Combining edge difference with nonlocal self-similarity constraints for single image super-resolution | |
CN107590781A (en) | Adaptive weighted TGV image deblurring methods based on primal dual algorithm | |
CN112258397B (en) | Image compressed sensing reconstruction method based on approximate message transfer and double low-rank constraint | |
CN106033595B (en) | Image blind deblurring method based on local constraint | |
CN110266318B (en) | Measurement matrix optimization method based on gradient projection algorithm in compressed sensing signal reconstruction | |
Zhou et al. | Transformed denoising autoencoder prior for image restoration | |
Si-Yao et al. | Understanding kernel size in blind deconvolution | |
CN114998107A (en) | Image blind super-resolution network model, method, equipment and storage medium | |
CN109919857B (en) | Noise image completion method based on weighted Schleiden norm minimization | |
CN113379647A (en) | Multi-feature image restoration method for optimizing PSF estimation | |
CN114820387B (en) | Image recovery method and terminal based on probability induction nuclear norm minimization |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |