CN106296583B - Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps - Google Patents

Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps Download PDF

Info

Publication number
CN106296583B
CN106296583B CN201610620280.5A CN201610620280A CN106296583B CN 106296583 B CN106296583 B CN 106296583B CN 201610620280 A CN201610620280 A CN 201610620280A CN 106296583 B CN106296583 B CN 106296583B
Authority
CN
China
Prior art keywords
image
image block
pairs
resolution
block group
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.)
Expired - Fee Related
Application number
CN201610620280.5A
Other languages
Chinese (zh)
Other versions
CN106296583A (en
Inventor
李映
杨静
钟依彤
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.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical 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 Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN201610620280.5A priority Critical patent/CN106296583B/en
Publication of CN106296583A publication Critical patent/CN106296583A/en
Application granted granted Critical
Publication of CN106296583B publication Critical patent/CN106296583B/en
Expired - Fee Related 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
    • 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 present invention relates to a kind of based on image block group sparse coding and the EO-1 hyperion noisy image ultra-resolution ratio reconstructing method that in pairs maps.This method is divided into two parts of model training and image reconstruction.In model training stage, training set is constructed first with clear image block, and training set image block is grouped and constitutes image block group；Followed by gauss hybrid models training dictionary and pairs of mapping relations.In the image reconstruction stage, image block is grouped first；Then judge the Gauss class belonging to it, select corresponding dictionary；Its sparse coding coefficient finally is adjusted using pairs of mapping method, to acquire without the high spatial resolution high spectrum image made an uproar.

Description

Based on image block group sparse coding and the noisy high spectrum image super-resolution that in pairs maps Rate reconstructing method
Technical field
The invention belongs to Hyperspectral imagery processing fields, and in particular to one kind is based on image block group sparse coding and at mapping The noisy high spectrum image ultra-resolution ratio reconstructing method penetrated.
Background technique
High spectrum image (Hyper-Spectral Images, HSIs) adds spectrum dimension on the basis of two dimensional image, into And form three-dimensional image datacube.While obtaining earth's surface image information, its spectral information is also obtained, it is true for the first time Just accomplishing the combination of spectrum and image.High spectrum image has in multiple fields such as military affairs monitoring, remote sensing and medical diagnosis Important application.However hyperspectral imager has lost spatial information, leads to it while improving image spectrum resolution ratio The reduction of spatial resolution.In addition, in high spectrum image, due to the influence of the factors such as atmospheric interference, hardware device, subwave Section image may contain different degrees of noise, and signal noise ratio (snr) of image is lower, cannot use in practical applications.In order to improve height While the spatial resolution of spectrum picture, it is effectively removed noise contained in image, the noisy high-spectrum of present invention research The super-resolution reconstruction problem of picture.
The process that the denoising of image is two independences and connects each other with super-resolution reconstruction.In conventional methods where, contain Make an uproar image super-resolution reconstruction usually using first with Denoising Algorithm removal noise recycle super-resolution algorithms be reconstructed Mode.However, such method will cause the loss of image detail of the high frequency during denoising, this loss can be serious Influence the effect of super-resolution reconstruction thereafter.To overcome the problems, such as this, in recent years, for the super-resolution of noisy natural image Restructing algorithm is suggested.A kind of effective way is, carries out denoising operation to noisy low-resolution image first and obtains without making an uproar Low-resolution image obtained corresponding secondly respectively to noisy and without low-resolution image for making an uproar carry out super-resolution reconstruction Noisy and without making an uproar high-definition picture, the former contains noise but remains the detailed information of image, and the latter is to have lost image Detail of the high frequency is that cost eliminates noise.Finally, this two images merge both to be remained image The nothing that detailed information has effectively removed noise again is made an uproar high-definition picture.
The either denoising process or super-resolution reconstruction process of image, the self similarity redundancy of image is all can be by The priori knowledge made full use of.Image denoising algorithm based on self similarity redundancy seeks to a series of similar noisy image blocks Carry out smooth noise, such as non-local mean (Non-Local Means, NLM) algorithm is exactly flat using the weighting of similar image block Remove noise.In super-resolution reconstruction algorithm, there are some research achievements to also utilize similar non local noise-free picture Block is to provide more detailed information.
For high spectrum image, the redundancy properties of its own are more obvious.Since hyperspectral imager exists to same atural object It is imaged under different-waveband, to obtain the cube metadata with multiple wave bands.This feature contains only high spectrum image not There is the redundancy of image in wave band, and increase the redundancy of image between wave band, for the joint denoising and oversubscription for realizing high spectrum image Resolution reconstruct provides more utilizable self similarity information.
The present invention carries out super-resolution reconstruction for noisy high spectrum image, makes full use of the distinctive wave of high spectrum image Self similarity redundancy in section between wave band using gauss hybrid models training dictionary, and utilizes the study adjustment of pairs of mapping relations Sparse coding coefficient carries out super-resolution reconstruction to high spectrum image simultaneously in denoising, proposes a kind of sparse based on image block group Coding and the noisy high spectrum image ultra-resolution ratio reconstructing method mapped in pairs.
Summary of the invention
Technical problems to be solved
In order to avoid existing existing high spectrum image ultra-resolution ratio reconstructing method cannot be effectively treated it is noise-containing low The problem of spatial resolution image, the present invention propose a kind of based on sparse coding and the noisy high spectrum image oversubscription that in pairs maps Resolution reconstructing method.
Technical solution
In order to solve problem above, the present invention provides a kind of based on image block group sparse coding and the bloom that in pairs maps Compose noisy image ultra-resolution ratio reconstructing method.This method is divided into two parts of model training and image reconstruction.In model training rank Section constructs training set first with clear image block, and training set image block is grouped and constitutes image block group；Followed by Gauss Mixed model training dictionary and pairs of mapping relations.In the image reconstruction stage, image block is grouped first；Then judge belonging to it Gauss class, select corresponding dictionary；Its sparse coding coefficient finally is adjusted using pairs of mapping method, is made an uproar to acquire nothing High spatial resolution high spectrum image.
It is a kind of based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps, It is characterized in that steps are as follows:
Step 1: construction image block group
Step 1a: input is without the high-resolution HR training image set of blocks X={ x that makes an uproari, the image block in set X is carried out Down-sampling operation obtains corresponding without the low resolution LR training image set of blocks Y={ y that makes an uproari, to obtain pairs of image block Gather { xi,yi}；
Step 1b: these pairs of image blocks are divided into N group according to Euclidean distance, every group contains M to similar image block
Step 1c: by each group of similar image block pairIt is straightened as column vector xn,m∈RP,yn,m∈Rp, Middle P and p is respectively the number of pixel contained by high/low image in different resolution block, calculates its corresponding mean vector μx,μy；
It calculates
Note As pairs of is high/low Image in different resolution block group；
Step 2: gauss hybrid models study
Step 2a: random initializtion K is to Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk), k=1 ..., K, wherein μ, Σ is respectively the mean value and variance of model；
Step 2b: using optimization problem shown in expectation maximization EM algorithm solution formula (2), parameter μ is obtainedxk,Σxk, μyk,Σyk；
Wherein πxk,πykFor weight；
Step 3: constructing high/low resolution ratio dictionary pair
For every a pair of of Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk) in parameter { Σxk,Σyk, utilization is unusual Value is decomposed svd algorithm and is broken down into:Wherein Λxk,ΛykFor by Σxk, ΣykEigenvalue cluster at diagonal matrix,As K is to required high/low resolution ratio dictionary pair；
Step 4: mapping relations study
Step 4a: each group of high/low image in different resolution block pair is judged by formula (3)Institute is right The Gauss model k answered, wherein c is constant；
Step 4b: being solved the rarefaction representation coefficient of image block, m=1 ..., M by formula (4), and sgn () is sign function；
Step 4c: learn height image block rarefaction representation coefficient mapping relations by formula (5)；
Wherein,The image block pair of different groups if it existsBelong to same Gauss Model k, thenFor its mean value；
Step 5: K is to high/low resolution ratio dictionary for outputAnd pairs of mapping relations
Step 6: image reconstruction
Step 6a: the input noisy high spectrum image Z of low spatial resolution, K is to high/low resolution ratio dictionaryWith And pairs of mapping relationsInitialize Z(0)=Z, t=1, the initialization estimated value of HR high spectrum image GIt is set as Z's S times of bicubic interpolation image；
Step 6b: rule of iterationδ is regularization constant；
Step 6c: estimation noise criteria is poorη is constant；
Step 6d: in high spectrum image Z(t)Each band image in, search for similar image block, construct image block groupAnd calculate each group of mean μz；
Step 6e: it is directed to each group of image blockExecute following operation:
The maximum Gauss model k of posterior probability is selected by formula (6), c is constant；
Noisy image block group is calculated by formula (7)In corresponding dictionaryUnder sparse coding
Noisy low-resolution image block group is solved by formula (8)Accordingly without high-definition picture block group of making an uproar
Step 6f: integration is without high-definition picture block group of making an uproarIt obtains without high-resolution high spectrum image of making an uproar
Step 6g:t=t+1, return step 6b are continued to execute, until reaching maximum number of iterations Tmax, output high spatial point Resolution high spectrum image
δ∈[0.05,0.1]。
η∈(0.5,1.5)。
c∈(0,0.4)。
TmaxIt is taken as 4 to 6 times.
Beneficial effect
The present invention takes full advantage of the redundancy properties in high spectrum image itself wave band between wave band, in the same of removal noise Shi Jinhang super-resolution reconstruction.Invention algorithm searches for similar image block between each band image and constitutes image block group, utilizes height This mixed model obtains pairs of high/low resolution ratio dictionary and carries out sparse coding to image block group, and utilizes mapping adjustment in pairs Its code coefficient effectively increases the visual quality of reconstructed image, can while suppressing noise, airspace more efficiently The structure features such as edge, the texture of reconstructed image.Multiple wave bands of high spectrum image are reconstructed simultaneously, can rapidly be weighed Structure provides the high spectrum image of higher resolution and resolution.
Detailed description of the invention
Flow chart Fig. 1 of the invention
Specific embodiment
Now in conjunction with embodiment, attached drawing, the invention will be further described:
1. gauss hybrid models and the study of pairs of mapping relations
Input: without the high-resolution training image set of blocks X={ x that makes an uproari}
(1) image block group is constructed
Step 1: down-sampling operation being carried out to the image block in high-resolution training image set of blocks X, obtains corresponding nothing Low resolution of making an uproar training image set of blocks Y={ yi, to obtain pairs of image block set { xi,yi}；
Step 2: these pairs of image blocks being divided into N group according to Euclidean distance, every group contains M to (generally taking M=10) Similar image block
Step 3: image block pair similar for each groupIt performs the following operations:
A) by image block to being straightened as column vector xn,m∈RP,yn,m∈Rp(P and p is respectively contained by high-low resolution image block The number of pixel, generally takes p=7), and its corresponding mean vector μ is calculated according to formula (1)x,μy；
B) it calculates
Note As pairs of is high/low Image in different resolution block group.
(2) gauss hybrid models learn
Step 1: random initializtion K is to (K is generally taken as 32) Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk), k =1 ..., K, wherein μ, Σ are respectively the mean value and variance of model；
Step 2: using optimization problem shown in EM algorithm solution formula (2), obtaining parameter μxk,Σxk,μyk,Σyk；
Wherein πxk,πykFor the weight of random initializtion.
(3) high/low resolution ratio dictionary pair is constructed
For every a pair of of Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk) in parameter { Σxk,Σyk, utilization is unusual Value is decomposed (Singular Value Decomposition, SVD) algorithm and is broken down into: Wherein Λxk,ΛykFor by Σxk,ΣykEigenvalue cluster at diagonal matrix,As K pairs Required high/low resolution ratio dictionary pair.
(4) mapping relations learn
To each group of high/low image in different resolution block pairExecute following operation:
Step 1: judging that this group of image block is constant (under normal circumstances, c to corresponding Gauss model k, c by formula (3) ∈(0,0.4))；
Step 2: the rarefaction representation coefficient of image block, m=1 ..., M are solved by formula (4), sgn () is sign function；
Step 3: learning height image block rarefaction representation coefficient mapping relations by formula (5)；
Wherein,The image block pair of different groups if it existsBelong to same Gauss Model k, thenFor its mean value.
Output: K is to high/low resolution ratio dictionaryAnd pairs of mapping relations
2. image reconstruction
Input: the noisy high spectrum image Z of low spatial resolution, K is to high/low resolution ratio dictionaryAnd in pairs Mapping relationsAmplification factor s
Step 1: initialization Z(0)=Z, t=1, the initialization estimated value of HR high spectrum image GIt is set as s times double three of Z Secondary interpolation image；
Step 2: rule of iterationδ is regularization constant (taking δ ∈ [0.05,0.1])；
Step 3: estimation noise criteria is poorη constant often takes η ∈ (0.5,1.5)；
Step 4: in high spectrum image Z(t)Each band image in, search for similar image block, construct image block groupAnd calculate each group of mean μz；
Step 5: being directed to each group of image blockExecute following operation:
A) selecting the maximum Gauss model k of posterior probability, c by formula (6) is constant (generally having c ∈ (0,0.4))；
B) noisy image block group is calculated according to formula (7)In corresponding dictionaryUnder sparse coding
C) noisy low-resolution image block group is solved by formula (8)Accordingly without high-definition picture block group of making an uproar
Step 6: integration is without high-definition picture block group of making an uproarIt obtains without high-resolution high spectrum image of making an uproar
Step 7:t=t+1, return step 2 continue to execute, until reaching maximum number of iterations Tmax(TmaxGenerally be taken as 4 to 6 times).
Output: high spatial resolution high spectrum image

Claims (5)

1. it is a kind of based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps, it is special Sign is that steps are as follows:
Step 1: construction image block group
Step 1a: input is without the high-resolution HR training image set of blocks X={ x that makes an uproari, down-sampling is carried out to the image block in set X Operation obtains corresponding without the low resolution LR training image set of blocks Y={ y that makes an uproari, to obtain pairs of image block set {xi,yi}；
Step 1b: these pairs of image blocks are divided into N group according to Euclidean distance, every group contains M to similar image block
Step 1c: by each group of similar image block pairIt is straightened as column vector xn,m∈RP,yn,m∈Rp, wherein P It is respectively the number of pixel contained by high/low image in different resolution block with p, calculates its corresponding mean vector μx,μy；
It calculates
Note As pairs of high/low resolution Rate image block group；
Step 2: gauss hybrid models study
Step 2a: random initializtion K is to Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk), k=1 ..., K, wherein μ, Σ divide Not Wei model mean value and variance；
Step 2b: using optimization problem shown in expectation maximization EM algorithm solution formula (2), parameter μ is obtainedxk,Σxk,μyk, Σyk；
Wherein πxk,πykFor weight；
Step 3: constructing high/low resolution ratio dictionary pair
For every a pair of of Gauss model { Nx(μxk,Σxk),Ny(μyk,Σyk) in parameter { Σxk,Σyk, utilize singular value point Solution svd algorithm is broken down into:Wherein Λxk,ΛykFor by Σxk,Σyk Eigenvalue cluster at diagonal matrix,As K is to required high/low resolution ratio dictionary pair；
Step 4: mapping relations study
Step 4a: each group of high/low image in different resolution block pair is judged by formula (3)Corresponding Gauss model k, wherein c is constant；
Step 4b: being solved the rarefaction representation coefficient of image block, m=1 ..., M by formula (4), and sgn () is sign function；
Step 4c: learn height image block rarefaction representation coefficient mapping relations by formula (5)；
Wherein,The image block pair of different groups if it existsBelong to same Gauss model K, thenFor its mean value；
Step 5: K is to high/low resolution ratio dictionary for outputAnd pairs of mapping relations
Step 6: image reconstruction
Step 6a: the input noisy high spectrum image Z of low spatial resolution, K is to high/low resolution ratio dictionaryAnd at To mapping relationsInitialize Z(0)=Z, t=1, the initialization estimated value of HR high spectrum image GIt is set as s times of Z Bicubic interpolation image；
Step 6b: rule of iterationδ is regularization constant；
Step 6c: estimation noise criteria is poorη is constant；
Step 6d: in high spectrum image Z(t)Each band image in, search for similar image block, construct image block groupAnd calculate each group of mean μz；
Step 6e: it is directed to each group of image blockExecute following operation:
The maximum Gauss model k of posterior probability is selected by formula (6), c is constant；
Noisy image block group is calculated by formula (7)In corresponding dictionaryUnder sparse coding
Noisy low-resolution image block group is solved by formula (8)Accordingly without high-definition picture block group of making an uproar
Step 6f: integration is without high-definition picture block group of making an uproarIt obtains without high-resolution high spectrum image of making an uproar
Step 6g:t=t+1, return step 6b are continued to execute, until reaching maximum number of iterations Tmax, export high spatial resolution High spectrum image
2. according to claim 1 a kind of super based on image block group sparse coding and the EO-1 hyperion noisy image that in pairs maps Resolution reconstruction method, it is characterised in that δ ∈ [0.05,0.1].
3. according to claim 1 a kind of super based on image block group sparse coding and the EO-1 hyperion noisy image that in pairs maps Resolution reconstruction method, it is characterised in that η ∈ (0.5,1.5).
4. according to claim 1 a kind of super based on image block group sparse coding and the EO-1 hyperion noisy image that in pairs maps Resolution reconstruction method, it is characterised in that c ∈ (0,0.4).
5. according to claim 1 a kind of super based on image block group sparse coding and the EO-1 hyperion noisy image that in pairs maps Resolution reconstruction method, it is characterised in that TmaxIt is taken as 4 to 6 times.
CN201610620280.5A 2016-08-01 2016-08-01 Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps Expired - Fee Related CN106296583B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610620280.5A CN106296583B (en) 2016-08-01 2016-08-01 Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610620280.5A CN106296583B (en) 2016-08-01 2016-08-01 Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps

Publications (2)

Publication Number Publication Date
CN106296583A CN106296583A (en) 2017-01-04
CN106296583B true CN106296583B (en) 2019-06-04

Family

ID=57663875

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610620280.5A Expired - Fee Related CN106296583B (en) 2016-08-01 2016-08-01 Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps

Country Status (1)

Country Link
CN (1) CN106296583B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108335264B (en) * 2018-01-30 2021-08-03 武汉大学 Image super-resolution method based on mixed resolution sparse dictionary learning
CN109214997A (en) * 2018-08-30 2019-01-15 中国科学院遥感与数字地球研究所 A kind of remote sensing images denoising method based on increment dictionary learning
CN109741254B (en) * 2018-12-12 2022-09-27 深圳先进技术研究院 Dictionary training and image super-resolution reconstruction method, system, equipment and storage medium
CN110288525B (en) * 2019-05-21 2022-12-02 西北大学 Multi-dictionary super-resolution image reconstruction method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102298768A (en) * 2010-06-24 2011-12-28 江南大学 High-resolution image reconstruction method based on sparse samples
CN103077505A (en) * 2013-01-25 2013-05-01 西安电子科技大学 Image super-resolution reconstruction method based on dictionary learning and structure clustering
CN103093445A (en) * 2013-01-17 2013-05-08 西安电子科技大学 Unified feature space image super-resolution reconstruction method based on joint sparse constraint

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102298768A (en) * 2010-06-24 2011-12-28 江南大学 High-resolution image reconstruction method based on sparse samples
CN103093445A (en) * 2013-01-17 2013-05-08 西安电子科技大学 Unified feature space image super-resolution reconstruction method based on joint sparse constraint
CN103077505A (en) * 2013-01-25 2013-05-01 西安电子科技大学 Image super-resolution reconstruction method based on dictionary learning and structure clustering

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于空间-光谱字典的不完备高光谱图像重构;练秋生 等;《仪器仪表学报》;20130131;第34卷(第1期);第112-118页

Also Published As

Publication number Publication date
CN106296583A (en) 2017-01-04

Similar Documents

Publication Publication Date Title
CN110119780B (en) Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network
Guo et al. An efficient SVD-based method for image denoising
CN109035142B (en) Satellite image super-resolution method combining countermeasure network with aerial image prior
Zhussip et al. Training deep learning based image denoisers from undersampled measurements without ground truth and without image prior
CN103473740B (en) Based on the non local denoising method of rarefaction representation and low-rank double constraints
CN106296583B (en) Based on image block group sparse coding and the noisy high spectrum image ultra-resolution ratio reconstructing method that in pairs maps
CN104574336B (en) Super-resolution image reconstruction system based on adaptive sub- mould dictionary selection
CN113177882B (en) Single-frame image super-resolution processing method based on diffusion model
CN107993194B (en) Super-resolution reconstruction method based on stationary wavelet transform
CN112488924A (en) Image super-resolution model training method, reconstruction method and device
CN103854262A (en) Medical image noise reduction method based on structure clustering and sparse dictionary learning
CN111986108A (en) Complex sea-air scene image defogging method based on generation countermeasure network
CN106600533B (en) Single image super resolution ratio reconstruction method
He et al. Remote sensing image super-resolution using deep–shallow cascaded convolutional neural networks
CN110830043A (en) Image compressed sensing reconstruction method based on mixed weighted total variation and non-local low rank
CN107301631B (en) SAR image speckle reduction method based on non-convex weighted sparse constraint
CN105184742B (en) A kind of image de-noising method of the sparse coding based on Laplce's figure characteristic vector
Cao et al. CS-MRI reconstruction based on analysis dictionary learning and manifold structure regularization
CN104200439B (en) Image super-resolution method based on adaptive filtering and regularization constraint
Bao et al. New image denoising algorithm using monogenic wavelet transform and improved deep convolutional neural network
CN104732504A (en) Image fusion method based on compressed sensing and WBCT
Wen et al. The power of complementary regularizers: Image recovery via transform learning and low-rank modeling
CN107146206B (en) High-spectrum remote sensing denoising method based on the filtering of four-dimensional Block- matching
CN117593199A (en) Double-flow remote sensing image fusion method based on Gaussian prior distribution self-attention
CN116612009A (en) Multi-scale connection generation countermeasure network medical image super-resolution reconstruction method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20190604