CN106651770A - Method for reconstructing multispectral super-resolution imaging based on Lapras norm regularization - Google Patents

Method for reconstructing multispectral super-resolution imaging based on Lapras norm regularization Download PDF

Info

Publication number
CN106651770A
CN106651770A CN201610832828.2A CN201610832828A CN106651770A CN 106651770 A CN106651770 A CN 106651770A CN 201610832828 A CN201610832828 A CN 201610832828A CN 106651770 A CN106651770 A CN 106651770A
Authority
CN
China
Prior art keywords
image
matrix
similar block
resolution
value
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
Application number
CN201610832828.2A
Other languages
Chinese (zh)
Other versions
CN106651770B (en
Inventor
刘丹华
郭宇飞
高大化
牛毅
石光明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201610832828.2A priority Critical patent/CN106651770B/en
Publication of CN106651770A publication Critical patent/CN106651770A/en
Application granted granted Critical
Publication of CN106651770B publication Critical patent/CN106651770B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
    • G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a method for reconstructing multispectral super-resolution imaging based on Lapras norm regularization, which mainly solves a problem that a reconstructed image is low in reliability and has wrong color blocks in the prior art. The technical scheme comprises the steps of 1, performing interpolation processing on a low-resolution image so as to acquire an initial image; 2, performing iteration on the initial image according to a set iterative formula; 3, blocking the iterated image and matching similar blocks; 4, combining the matched similar blocks into image matrix blocks; 5, performing singular value decomposition on the image matrix blocks, and updating the similar blocks by using an iterative shrinkage method; 6, combining the similar blocks to acquire updated image matrix blocks; and 6, combining the updated image matrix blocks into an image, and implementing super-resolution reconstruction for the image by iterations. The method disclosed by the invention reduces the wrong color blocks, effectively improves the reliability and the space resolution of the reconstructed image, and can be applied to reconstructing a high-resolution image from a low-resolution image.

Description

Multispectral super-resolution imaging reconstructing method based on Laplce's norm regularization
Technical field
The invention belongs to technical field of image processing, more particularly to a kind of multispectral super-resolution imaging reconstructing method, are used for High-definition picture is reconstructed from low-resolution image.
Background technology
The reconstruct of super-resolution imaging, is to reconstruct high-definition picture by specific method to low-resolution image.It is logical Often, the spatial high resolution and the high confidence level of multispectral image of image are obtained by the reconstruct of super-resolution imaging.Although existing Super-resolution imaging reconstructing method can obtain higher spatial resolution image.But, existing super-resolution imaging reconstructing method Occurs the color lump that has of mistake in the high resolution gray figure for reconstructing, such as accompanying drawing 4 (a), 4 (b), 4 (c) is shown.And for one It is a little to require the aspects such as more accurate work, such as material analysis, identification discriminating, precise classification, need to improve spatial resolution With multispectral image confidence level.
At present, without good solution on multispectral confidence level is improved.The method for improving spatial resolution is main There are two kinds of effective methods.
First method is directly to carry out super resolution image reconstruct in spatial domain, and with this spatial high resolution figure is reconstructed Picture.But because the reconstructing method have ignored the Spectral correlation of high-order, understand that the solution after reconstruct is suboptimal solution by mathematical analysis, Therefore the method can not effectively improve the confidence level of image.
Second method is, first to image the conversion of RGB to YUV is carried out in Image Reconstruction, then carries out space oversubscription Reconstruct is distinguished, spatial high resolution image is reconstructed with this.But this method Structure adaptation and between spectral coverage in terms of interband decorrelation Property aspect Shortcomings, therefore the confidence level of image can not be effectively improved, also, in the conversion process by RGB to YUV, meeting Making the corresponding locus of each component of R, G, B of image cannot align, and finally occur mistake in gray-scale map has color lump, causes Image after reconstruct includes error message, and then can not accurately realize the reconstruct of image.
The content of the invention
The purpose of the present invention is, for above-mentioned the deficiencies in the prior art, to propose a kind of based on LaplceNorm regularization Super-resolution reconstructing method, with reduce existing method reconstructed image appearance error message, improve image spatial resolution And confidence level.
Embodiment of the present invention is so completed：
One kind is based on LaplceThe Image Super-resolution reconstructing method of norm regularization, comprises the steps：
(1) bicubic interpolation is carried out to low-resolution image, obtains initial image X(0)；
(2) iterative formula is setWhereinL changes for maximum Generation number,ForImage after secondary iteration, δ is iteration regular parameter；
(3) to initial pictures X(0)The figure X after first time iteration is obtained using the iterative formula of above-mentioned setting(1)；
(4) by the image X after first time iteration(1)It is divided into M blocks, and S is obtained using block matching method to i-th pieceiIt is individual similar Block matrix, remembers x(i,j)For i-th piece of j-th similar block matrix, then by this SiIndividual similar block is merged into i-th image array Xi, its Middle i=1,2 ...., M, j=1,2 ..., Si；
(5) to image array XiUsing formula [U, Σ ', V]=SVD (Xi) singular value decomposition is carried out, obtain U, Σ ', V tri- Individual split-matrix, wherein U are and image array XiRelated left orthogonal matrix, Σ ' is comprising image array XiSingular value it is unusual Matrix, V is and image array XiRelated right orthogonal matrix；
(6) U obtained according to step (5), Σ ', V these three matrixes, using public Xi=USμ(∑′)VTUpdate image array Xi, whereinIt is the soft-threshold computing to singular matrix Σ ', VTRepresent to right orthogonal matrix V's Transposition, μlTake the third-largest characteristic value, k in singular value matrix Σ '1To set the parameter of first regular terms, max () is represented to it Maximizing；
(7) similar block matrix x is updated using based on the method for reconstructing in full variational regularization(i,j)Obtain and update matrix
(8) calculate and update matrixMaximum γ of orderi,j；
(9) formula is utilizedTo updating matrixSingular value decomposition is carried out, U is obtained1, Σ1, V1Three Split-matrix, wherein U1It is and updates matrixRelated left orthogonal matrix, Σ1It is comprising renewal matrixSingular value it is unusual Matrix, V1It is and updates matrixRelated right orthogonal matrix；
(10) shrinkage operation formula is setTo similar block matrix x(i,j)Be updated, in formula, x be with Similar block matrix x(i,j)Closest contraction matrix, H (∑s1) it is to make constraint Rank (x(i,j))≤ri,jThe hard threshold set up Value computing, V1 TRepresentative pair and similar block matrix x(i,j)Related right orthogonal matrix V1Transposition, Rank (x(i,j)) represent similar block Matrix x(i,j)Order；
(11) by i-th piece of SiIndividual similar block matrix merges, and obtains i-th image array Xi；
(12) by M image array XiMerge, obtain image X(1), return to step (2), repeat the above steps, until passing through Super-resolution reconstructed image X is exported after L iteration(L)。
The present invention has compared with prior art advantages below
First：Improve the spatial resolution and confidence level of reconstructed image.
The present invention is on the basis of existing reconstructing method such as A+, BCSR, NCSR method, it is contemplated that the correlation on space structure Property, update similar block matrix using iterative shrinkage method so that this method is in Y-PSNR PSNR and structural dependence SSIM is performed better than, and effectively improves the spatial resolution and confidence level of image.
Second：Effectively reducing mistake has color lump.
The image that existing method is reconstructed occurs the color lump that has of mistake, and the present invention is special using second order Laplacian space Property, and to similar block matrix addition of constraints, can effectively reduce mistake has color lump.
3rd：Image after reconstruct is more accurate.
There is any discrepancy for the spectral reflectivity curve of full resolution pricture that existing method is reconstructed and former full resolution pricture, and this Bright method can well be fitted the spectral reflectivity curve of artwork, thus the image after reconstruct is more accurate.
Description of the drawings
Fig. 1 is the flowchart of the present invention；
Fig. 2 is the original spectrum image that emulation is used；
Fig. 3 is to visualize the low-resolution image used when contrast experiment emulates；
Fig. 4 is respectively that Fig. 3 is reconstructed with existing A+ methods, BSSC methods, NSCR methods and the inventive method Simulation result comparison diagram；
The original image that Fig. 5 is used when being trust verification experiment simulation
Fig. 6 is in trust verification experiment, with existing A+ methods, BSSC methods, NSCR methods and the inventive method The spectral reflectivity curve figure drawn after being reconstructed to Fig. 5.
Specific embodiment
Below in conjunction with the accompanying drawings the present invention is described in detail with example
With reference to Fig. 1, the present invention's realizes that step is as follows：
Step 1, initialization.
Down-sampling is carried out to the original spectrum image shown in Fig. 2 and processes acquisition low-resolution image Y as shown in Figure 3；
Bicubic interpolation is carried out to low-resolution image Y, i.e., the interlacing of low-resolution image matrix is compressed every row extraction To sampling matrix, the sampling matrix of acquisition is utilized respectively in line direction and column directionCarry out three times Interpolation obtains initial image X(0), in formula, n be interpolation point number, CkIt is k-th antiderivative value, h (x-xk) it is interpolation base letter Number, the highest power of the Interpolation-Radix-Function is three times, and in domain of definition basic function single order. second dervative is continuous.
Initial pictures are carried out after first time iteration by step 2, and piecemeal, merging treatment are made successively, obtain image array.
(2a) iterative formula is setWherein,L is maximum Iterations,ForImage after secondary iteration, δ is Iteration Regularized coefficient, and its value is 0.22；D (x) is represented under making to x The function of sampling processing；DTX () represents the transposition of D (x)；
(2b) the initial pictures X that step 1 is obtained(0)Utilize setting formula in (2a) to be iterated, obtain first time iteration Image X afterwards(1)；
(2c) by the image X after first time iteration(1)It is divided into M blocks, current block is calculated with adjacent block most using block matching method Near block, obtains SiIndividual similar block matrix, remembers x(i,j)For i-th piece of j-th similar block matrix, then by this SiIndividual similar block is merged into I-th image array Xi, wherein, i=1,2 ...., M, j=1,2 ..., Si。
Step 3, to image array XiSingular value decomposition is carried out, and updates the image array.
(3a) to image array XiUsing formula [U, Σ ', V]=SVD (Xi) singular value decomposition is carried out, obtain U, Σ ', V tri- Individual split-matrix, wherein U are and image array XiRelated left orthogonal matrix, Σ ' is comprising image array XiSingular value it is unusual Matrix, V is and image array XiRelated right orthogonal matrix；
(3b) U obtained according to step 3a, Σ ', V these three matrixes, using formula Xi=USμ(∑′)VTMore new images Matrix Xi, wherein：
It is the soft-threshold computing to singular matrix Σ ',
k1To set the parameter of first regular terms, value is 0.5,
VTThe transposition to right orthogonal matrix V is represented,
μlTake the third-largest characteristic value in singular value matrix Σ ' so that the image array X for being obtainediLow-rank,
Max () is represented to its maximizing.
Step 4, using the method for reconstructing based on full variational regularization similar block matrix x is updated(i,j)。
Implementing for this step is formula in method for reconstructing according to full variational regularization：
Minimum problems are solved, is obtained and similar block matrix x(i,j)Away from From nearest renewal matrixIn formula：
X is independent variable；Argmin () represents the function for making certain functional obtain minimum of a value；ρlIt is weight coefficient, value is 1；k2To set the parameter of second regular terms, value is 0.59；▽2X () is that the square that second order Laplace's operation is obtained is done to x Battle array；||·||1,2RepresentNorm,RepresentNorm square.
Step 5, estimation updates matrixMaximum order.
Using inequality constraintsEstimation obtains updating matrixMaximum order γi,j；Wherein, γkRepresent similar block matrix x(i,j)K-th singular value；Γ is given threshold value, and its value is second largest singular value and the third-largest The mean value of singular value.
Step 6, to updating matrixSingular value decomposition is carried out, and updates similar block matrix x(i,j)。
(6a) formula is utilizedTo updating matrixSingular value decomposition is carried out, U is obtained1, Σ1, V1Three Individual split-matrix, wherein U1It is and updates matrixRelated left orthogonal matrix, Σ1It is comprising renewal matrixSingular value it is strange Different matrix, V1It is and updates matrixRight orthogonal matrix；
(6b) shrinkage operation formula is setTo similar block matrix x(i,j)It is updated, wherein：
It is and similar block matrix x(i,j)Closest contraction matrix,
H(Σ1) it is to make constraint Rank (x(i,j))≤ri,jThe hard -threshold computing set up,
Rank(x(i,j)) represent similar block matrix x(i,j)Order,
V1 TRepresentative pair and similar block matrix x(i,j)Related right orthogonal matrix V1Transposition.
Step 7, successively merges similar block x(i,j), image array Xi, L output super-resolution reconstructed image of iteration.
(7a) by the image X in step 2(1)It is divided into M blocks, for i-th piece, circulation obtains S j timeiIndividual similar block matrix x(i,j), i-th image array X is obtained after being mergedi, by M image array XiImage X is obtained after merging(1), wherein i=1, 2 ...., M, j=1,2 ..., Si；
(7b) by image X(1)Substitute into the iterative formula set in step 2 and obtain image X(2), to image X(2)Execution step 2 ～(7a), obtains new image X(2), step 2 is returned again to, perform successivelySecondary iteration, obtains theImage after secondary iteration Wherein,L values are 200；
(7c) image is judgedIteration L time, if so, stops iteration, the full resolution pricture X after output reconstruct(L), if it is not, then continue to repeat step 2～(7a), until
The effect of the present invention can be further illustrated by following emulation
1. simulated conditions
The hardware test platform of this experiment is：Intel Core i7 CPU, dominant frequency 3.40GHz, internal memory 8GB；Software emulation Platform is：Windows 7,64 bit manipulation systems and Matlab 2013b.
2. emulation experiment
Emulation 1：Contrast experiment,
The validity of method in order to verify the present invention, using the first eight spectrum picture disclosed in Colombia as experiment institute The original spectrum image for needing, down-sampling is carried out to it and obtains eight low-resolution image A～H.
With the inventive method and existing A+ methods, NCSR methods, BSSC methods low-resolution image A～H is entered respectively Line reconstruction obtains the high-definition picture a～h after eight reconstruct, respectively in σ2=0 and σ2Under the noise of=25 varying levels, than The size and the size of structural similarity SSIM of signal to noise ratio PSNR of the high-definition picture a～h after relatively reconstructing, as a result in table Shown in 1.
Table 1
As can be seen from Table 1：Under the noise of varying level, the size of signal to noise ratio PSNR of the inventive method is than A+ side Method average energy increases 0.67dB～1.08dB, averagely increases 0.7dB～2.22dB than BCSR method, averagely increases than NCSR method 0.24dB～0.96dB, structural dependence SSIM are from table it can also be seen that the structural dependence size of the inventive method compares Additive method will height.
Emulation 2：Visualization contrast experiment
There is color lump in order to the more visual prominent present invention can be reduced effectively, down-sampling process is carried out to Fig. 2 The low-resolution image shown in Fig. 3 is obtained, to Fig. 3 by the inventive method and existing A+ methods, NCSR methods, BSSC methods It is reconstructed, the coloured image after the reconstruct represented such as Fig. 4 is obtained, as a result as illustrated, wherein：
4 (a) is to be reconstructed rear coloured image by A+ methods,
4 (b) is to be reconstructed rear coloured image by BSSC methods,
4 (c) is to be reconstructed rear coloured image by NCSR methods,
4 (d) is to be reconstructed rear coloured image by the inventive method.
Fig. 2 artworks are contrasted from Fig. 4 (a)～4 (d), it can be seen that the full resolution pricture meeting of A+, BCSR, NCSR method reconstruct Different degrees of wrong color lump is presented.And the high-definition picture after present invention reconstruct effectively reduces mistake and has color lump.
Emulation 3：Trust verification is tested,
In order to illustrate superiority of the inventive method in confidence level, from Fig. 5 chosen materials are identical, spectral signature identical 3 points：I.e. 1: 1, second point 2, thirdly 3.
Down-sampling is carried out to Fig. 5 and processes acquisition low-resolution image Y1, to low-resolution image Y1By the inventive method and Existing A+ methods, NCSR methods, BSSC methods are reconstructed, the high-definition picture after being reconstructed.
The high-definition picture midpoint 1 after reconstruct, point 2, the spectral reflectivity curve figure of point 3 are drawn, while drawing Fig. 5's Spectral reflectivity curve figure, as a result as shown in fig. 6, wherein,
Fig. 6 (a) represents 1: 1 spectral reflectivity curve figure,
Fig. 6 (b) represents the spectral reflectivity curve figure of second point 2,
Fig. 6 (c) represents thirdly 3 spectral reflectivity curve figure.
3 spectral reflectivity curves shown in contrast Fig. 6, it can be clearly seen that the height of A+, BCSR, NCSR method reconstruct Image in different resolution has larger discrepancy with the spectral reflectivity curve of original image, and the inventive method can very well be fitted artwork The curve of spectral reflectivity, realizes accurate reconstruction, it is ensured that the confidence level of the high-definition picture of reconstruct.

Claims (3)

1. a kind of Image Super-resolution reconstructing method based on Laplce's norm regularization, comprises the steps：
(1) bicubic interpolation is carried out to low-resolution image, obtains initial image X(0)；
(2) iterative formula is setWhereinL is greatest iteration time Number,ForImage after secondary iteration, δ is iteration regular coefficient；
(3) to initial pictures X(0)The image X after first time iteration is obtained using the iterative formula of above-mentioned setting(1)；
(4) by the image X after first time iteration(1)It is divided into M blocks, and S is obtained using block matching method to i-th pieceiIndividual similar block square Battle array, remembers x(i,j)For i-th piece of j-th similar block matrix, then by this SiIndividual similar block is merged into i-th image array Xi, wherein i= 1,2 ...., M, j=1,2 ..., Si；
(5) to image array XiUsing formula [U, Σ ', V]=SVD (Xi) singular value decomposition is carried out, obtain U, tri- points of Σ ', V Dematrix, wherein U are and image array XiRelated left orthogonal matrix, Σ ' is comprising image array XiThe unusual square of singular value Gust, V is and image array XiRelated right orthogonal matrix；
(6) U obtained according to step (5), Σ ', V these three matrixes, using formula Xi=USμ(Σ′)VTUpdate image array Xi, WhereinIt is the soft-threshold computing to singular matrix Σ ', VTRepresentative turns to right orthogonal matrix V Put, μlTake the third-largest characteristic value, k in singular value matrix Σ '1To set the parameter of first regular terms, max () is represented and it is asked Maximum；
(7) similar block matrix x is updated using based on the method for reconstructing in full variational regularization(i,j)Obtain and update matrix
(8) calculate and update matrixMaximum γ of orderi,j；
(9) formula is utilizedTo updating matrixSingular value decomposition is carried out, U is obtained1, Σ1, V1Three decomposition Matrix, wherein U1It is and updates matrixRelated left orthogonal matrix, Σ1It is comprising renewal matrixSingular value unusual square Battle array, V1It is and updates matrixRelated right orthogonal matrix；
(10) shrinkage operation formula is setTo similar block matrix x(i,j)It is updated, in formula,Be to it is similar Block matrix x(i,j)Closest contraction matrix, H (Σ1) it is to make constraint Rank (x(i,j))≤ri,jThe hard -threshold fortune set up Calculate, V1 TRepresentative pair and similar block matrix x(i,j)Related right orthogonal matrix V1Transposition, Rank (x(i,j)) represent similar block matrix x(i,j)Order；
(11) by i-th piece of SiIndividual similar block matrix merges, and obtains i-th image array Xi；
(12) by M image array XiMerge, obtain image X(1), return to step (2), repeat the above steps, until changing through L time For rear output super-resolution reconstructed image X(L)。
2. method according to claim 1, is wherein updated in step (7) using the method for reconstructing based on full variational regularization Similar block matrix x(i,j), carried out by equation below：
x ^ = argmin x | | ▿ 2 ( x ) | | 1 , 2 + ρ l , k 2 | | x - x ( i , j ) | | 2 2 ,
In formula, x is independent variable；For the matrix after renewal；Arg min () represents the function for making certain functional obtain minimum of a value；ρl It is weight coefficient, value is 1；k2To set the parameter of second regular terms, value is 0.59；It is to do second order to x to draw general The matrix that Lars computing is obtained；||·||1,2RepresentNorm,RepresentNorm square.
3. method according to claim 1, wherein calculates similar block matrix x in step (8)(i,j)Maximum order γi,j, It is by inequality constraintsEstimation is obtained, wherein, Σ represents summation symbol；γkRepresent similar block Matrix x(i,j)K-th singular value；Γ is given threshold value, and its value is the average of second largest singular value and the third-largest singular value Value.
CN201610832828.2A 2016-09-19 2016-09-19 Multispectral super-resolution imaging reconstructing method based on Laplce's norm regularization Active CN106651770B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610832828.2A CN106651770B (en) 2016-09-19 2016-09-19 Multispectral super-resolution imaging reconstructing method based on Laplce's norm regularization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610832828.2A CN106651770B (en) 2016-09-19 2016-09-19 Multispectral super-resolution imaging reconstructing method based on Laplce's norm regularization

Publications (2)

Publication Number Publication Date
CN106651770A true CN106651770A (en) 2017-05-10
CN106651770B CN106651770B (en) 2019-07-23

Family

ID=58853152

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610832828.2A Active CN106651770B (en) 2016-09-19 2016-09-19 Multispectral super-resolution imaging reconstructing method based on Laplce's norm regularization

Country Status (1)

Country Link
CN (1) CN106651770B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109685125A (en) * 2018-12-14 2019-04-26 大连海事大学 Daily behavior feature mining and calculation method based on frequent Sensor Events sequence
CN111257833A (en) * 2019-12-24 2020-06-09 重庆大学 Sound source identification method based on Laplace norm for fast iterative shrinkage threshold
CN111311729A (en) * 2020-01-18 2020-06-19 西安电子科技大学 Natural scene three-dimensional human body posture reconstruction method based on bidirectional projection network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102156878A (en) * 2011-04-08 2011-08-17 南京邮电大学 Sparse embedding with manifold information-based human face identification method
CN103473740A (en) * 2013-08-31 2013-12-25 西安电子科技大学 Sparse representation and low-rank double restraints-based nonlocal denoising method
CN104200439A (en) * 2014-09-05 2014-12-10 西安电子科技大学 Image super-resolution method based on adaptive filtering and regularization constraint
CN105021280A (en) * 2015-08-13 2015-11-04 西安电子科技大学 Panchromatic-and-compressed-spectral-imaging-based image super-resolution system and method
CN105590296A (en) * 2015-12-07 2016-05-18 天津大学 Dual-dictionary learning-based single-frame image super-resolution reconstruction method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102156878A (en) * 2011-04-08 2011-08-17 南京邮电大学 Sparse embedding with manifold information-based human face identification method
CN103473740A (en) * 2013-08-31 2013-12-25 西安电子科技大学 Sparse representation and low-rank double restraints-based nonlocal denoising method
CN104200439A (en) * 2014-09-05 2014-12-10 西安电子科技大学 Image super-resolution method based on adaptive filtering and regularization constraint
CN105021280A (en) * 2015-08-13 2015-11-04 西安电子科技大学 Panchromatic-and-compressed-spectral-imaging-based image super-resolution system and method
CN105590296A (en) * 2015-12-07 2016-05-18 天津大学 Dual-dictionary learning-based single-frame image super-resolution reconstruction method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YAOHAI LIN 等: "High-resolution spectral imaging based on coded dispersion", 《EUROPE PMC》 *
李娟 等: "基于自学习的稀疏正则化图像超分辨率方法", 《仪器仪表学报 》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109685125A (en) * 2018-12-14 2019-04-26 大连海事大学 Daily behavior feature mining and calculation method based on frequent Sensor Events sequence
CN111257833A (en) * 2019-12-24 2020-06-09 重庆大学 Sound source identification method based on Laplace norm for fast iterative shrinkage threshold
CN111257833B (en) * 2019-12-24 2023-08-01 重庆大学 Sound source identification method based on Laplace norm rapid iteration shrinkage threshold
CN111311729A (en) * 2020-01-18 2020-06-19 西安电子科技大学 Natural scene three-dimensional human body posture reconstruction method based on bidirectional projection network
CN111311729B (en) * 2020-01-18 2022-03-11 西安电子科技大学 Natural scene three-dimensional human body posture reconstruction method based on bidirectional projection network

Also Published As

Publication number Publication date
CN106651770B (en) 2019-07-23

Similar Documents

Publication Publication Date Title
CN110119780B (en) Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network
CN109035142B (en) Satellite image super-resolution method combining countermeasure network with aerial image prior
Liang et al. Details or artifacts: A locally discriminative learning approach to realistic image super-resolution
CN106952228B (en) Super-resolution reconstruction method of single image based on image non-local self-similarity
CN103093444B (en) Image super-resolution reconstruction method based on self-similarity and structural information constraint
CN110443768B (en) Single-frame image super-resolution reconstruction method based on multiple consistency constraints
CN106934766A (en) A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation
Mahdizadehaghdam et al. Sparse generative adversarial network
CN105513033B (en) A kind of super resolution ratio reconstruction method that non local joint sparse indicates
CN105046672A (en) Method for image super-resolution reconstruction
CN102156875A (en) Image super-resolution reconstruction method based on multitask KSVD (K singular value decomposition) dictionary learning
CN104657962B (en) The Image Super-resolution Reconstruction method returned based on cascading linear
Yang et al. Image super-resolution based on deep neural network of multiple attention mechanism
CN111640059B (en) Multi-dictionary image super-resolution method based on Gaussian mixture model
CN105550989A (en) Image super-resolution method based on nonlocal Gaussian process regression
CN106886978A (en) A kind of super resolution ratio reconstruction method of image
CN106651770A (en) Method for reconstructing multispectral super-resolution imaging based on Lapras norm regularization
Xue et al. Dynamic coarse‐to‐fine ISAR image blind denoising using active joint prior learning
CN104252704A (en) Total generalized variation-based infrared image multi-sensor super-resolution reconstruction method
Cao et al. CS-MRI reconstruction based on analysis dictionary learning and manifold structure regularization
CN103093431A (en) Compressed sensing reconstruction method based on principal component analysis (PCA) dictionary and structural priori information
CN104200439B (en) Image super-resolution method based on adaptive filtering and regularization constraint
Wang et al. SAR images super-resolution via cartoon-texture image decomposition and jointly optimized regressors
CN106296583B (en) Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps
Zhou et al. Multi-scale dilated convolution neural network for image artifact correction of limited-angle tomography

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