CN109840888A - A kind of image super-resolution rebuilding method based on joint constraint - Google Patents
A kind of image super-resolution rebuilding method based on joint constraint Download PDFInfo
- Publication number
- CN109840888A CN109840888A CN201910062627.2A CN201910062627A CN109840888A CN 109840888 A CN109840888 A CN 109840888A CN 201910062627 A CN201910062627 A CN 201910062627A CN 109840888 A CN109840888 A CN 109840888A
- Authority
- CN
- China
- Prior art keywords
- image block
- image
- dictionary
- coding
- sparse
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of image super-resolution rebuilding methods based on joint constraint, this method extracts image block from natural image first, under conditions of based on low order constraint, dictionary training is carried out using K-SVD algorithm to learn, during dictionary training, the atom in dictionary is updated one by one；Secondly, it is obtained with image block by search with scale and multiple dimensioned similar graph block collection, estimation is weighted to true coding using the sparse coding of similar image block, the difference between true coding and acquired sparse coding is introduced into objective function as bound term；Finally, for the image block that needs are rebuild, it is multiplied using the atom in dictionary with sparse coefficient and estimates to obtain high-definition picture block.The present invention weakens influence of noise by introducing bound term, and at the same time enhancing reconstruct as a result, obtain more image details.
Description
Technical field
The present invention relates to a kind of image super-resolution rebuilding methods based on joint constraint, belong to image processing techniques neck
Domain.
Background technique
In recent years, image super-resolution rebuilding causes the concern of researchers, and has been widely applied to reality
In the every field on border, such as medical image, transmission of video.But image super-resolution rebuilding in actual application also
Many problem needs go to solve, such as: the influence of noise, the edge effect of image, ringing effect etc., and remove the influence of noise
The problem of even more one urgent need is alleviated.
The algorithm for reconstructing of super-resolution image mainly includes the method based on interpolation, the method based on reconstruction and is based on learning
Method.Method based on study has: neighborhood embedding inlay technique, sparse representation method, deep approach of learning.Present most algorithm is all
It is to go from the viewpoint of based on study and step by step to realize, so based on sparse algorithm comparison by scholars
Welcome, but the best algorithm of effect is to be thought deeply in terms of deep learning, but deep learning needs a large amount of data,
Also having time is based on rarefaction representation super-resolution image reconstruction so each thinking direction can have respective strong point and disadvantage
The basic thought of algorithm is: HR image block and obtained LR image block of degenerating being obtained dictionary as data set training, then led to
Cross solution optimization problemRarefaction representation coefficient is obtained, rarefaction representation is finally utilized
Coefficient carries out linear combination to the atom in dictionary.Document (Shang L.Denoising natural images based on
a modified sparse coding algorithm[J].Applied Mathematics&Computation,2008,
205 (2): 883-889) using maximum kurtosis as maximum sparse measurement criterion, it is generated using the constant variance item of sparse coefficient
Fix information capacity simultaneously, in order to improve convergence speed of the algorithm, is obtained using by quick unfixed point Independent Component Analysis Algorithm
Initialization feature basic function of the determination basic function as sparse coding algorithm.Document (Nath A G, Nair M S, Rajan
J.Single Image Super Resolution from Compressive Samples Using Two Level
Sparsity Based Reconstruction[J].Procedia Computer Science,2015,46:1643-1652)
In, a kind of method for reconstructing based on second level degree of rarefication is proposed, this method passes through image interpolation and dictionary learning based on patch
To realize.Document (Peleg T, Elad M.A Statistical Prediction Model Based on Sparse
Representations for Single Image Super-Resolution[J].IEEE Transactions on
Image Processing A Publication of the IEEE Signal Processing Society,2014,23
(6): 2569-82) using the oversubscription for handling single image based on the Statistical Prediction Model of the rarefaction representation of LR and HR image block
Resolution.Estimate to obtain the prediction of HR image block by MMSE, using data clusters and several rudimentary algorithms is cascaded, for gained net
The training pattern of network, it is shown that it computation complexity, in terms of be better than existing method.Document (Xie C, Zeng
W,Jiang S,et al.Multiscale Self-similarity and Sparse Representation Based
Single Image Super-Resolution [J] .Neurocomputing, 2017,260) utilize multi-scale self-similarity structure
It makes by l1Regularization that norm defines compensation is to inhibit sparse noise, to generate more reliable reconstruction effect.
But the above method does not have good rejection ability to noise, so that image reconstruction is ineffective.
Summary of the invention
The technical problems to be solved by the present invention are: providing a kind of image super-resolution rebuilding side based on joint constraint
Method while weakening influence of noise, enhances reconstruction result, obtains more image details by introducing bound term.
The present invention uses following technical scheme to solve above-mentioned technical problem:
A kind of image super-resolution rebuilding method based on joint constraint, includes the following steps:
Step 1, high-definition picture is divided into the identical image block of size, using all image blocks formed image block collection as
Training sample is trained dictionary using K-SVD algorithm, obtains trained dictionary under conditions of based on low order constraint,
Sparse coding is acquired according to trained dictionary；
Step 2, the low-resolution image rebuild for needs, is divided into the identical image block of size for low-resolution image,
The image block rebuild needed for some, search and the image block are with the multiple dimensioned similar image block of scale phase Sihe, using same
The sparse coding of the multiple dimensioned similar image block of scale phase Sihe is weighted estimation to true coding, the true volume estimated
Difference between the sparse coding that the true coding and step 1 of estimation acquire is introduced into objective function as bound term, asks by code
Solution objective function obtains sparse coefficient；
Step 3, the low-resolution image rebuild for needs, utilizes the atom and step 2 in the trained dictionary of step 1
Obtained sparse coefficient is multiplied to obtain high-definition picture.
As a preferred solution of the present invention, detailed process is as follows for the step 1:
(a) dictionary learning constrained optimization problem are as follows:
Wherein, D is dictionary, DoptFor trained dictionary, Λ is sparse coding, and ε indicates that sparsity constraints, Z are for word
The high-definition picture block collection of allusion quotation training, tsFor selected image block Spatial Dimension, tnFor the time point in data, RlowrankIt is low
Order matrix (ts×tn), P representing matrix；
(b) fixed Λ, updates dictionary:
Then in dictionary atom update are as follows:
s.t.P(di,ts,tn)∈Rlowrank
Wherein, E (i) is residual matrix, di、ΛiThe i-th column in D, Λ are respectively indicated, Δ (i) is matrix, Λi,mRespectivelyM column,Indicate trained di。
As a preferred solution of the present invention, detailed process is as follows for the step 2:
(a) search and image block xjWith the multiple dimensioned similar image block of scale phase Sihe, the similar collection Ψ of same scale is obtainedjWith
Multiple dimensioned similar collection Οj, utilize ΨjAnd ΟjThe sparse coding of middle image block is weighted and averaged estimation:
Wherein, ωj,kIt indicates and image block xjThe similar image block x with scalej,kWeight, ωj,qIt indicates and image block xj
Multiple dimensioned similar image block xj,qWeight, Λj,k、Λj,qRespectively Ψj、ΟjMiddle image block xj,kAnd xj,qSparse coding, W
For normalization factor, Respectively Λj、Λj,k、
Λj,qEstimation, ΛjIndicate that the jth column in sparse coding Λ, D are dictionary, h is control constant, ρj、τjRespectively with scale phase
Seemingly, the sparse coding value of multiple dimensioned similar estimation；
(b) difference between the true coding of estimation and obtained sparse coding is introduced objective function as bound term
In, target function model are as follows:
Wherein, subscript (l) indicates the l times iteration, ΛyIndicate that the sparse coding to be solved, y indicate low-resolution image,
η1、η2It is constant；
(c) fixed D, updates Λ, obtains sparse coefficient.
As a preferred solution of the present invention, the constant η1=0.8, η2=0.15.
As a preferred solution of the present invention, the reconstruction formula of high-definition picture described in step 3 are as follows:
Wherein, X indicates high-definition picture, and D is dictionary, ΛyFor obtained sparse coefficient, RjTo be extracted from image
Image block xjOperator, J be low-resolution image block number, Λy,jFor image block xjRarefaction representation coefficient.
The invention adopts the above technical scheme compared with prior art, has following technical effect that
The present invention improves method for reconstructing to the rejection ability of noise by introducing bound term, realizes that image is preferably rebuild
Effect.
Detailed description of the invention
Fig. 1 is a kind of basic flow chart of the image super-resolution rebuilding method based on joint constraint of the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings.Below by
The embodiment being described with reference to the drawings is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
As shown in Figure 1, a kind of image super-resolution rebuilding method based on joint constraint, comprising the following steps:
Step 1: under conditions of based on low order constraint, carrying out dictionary training using K-SVD algorithm and learn, in dictionary training
During, the atom in dictionary is updated one by one；
(1) dictionary learning constrained optimization problem:
Wherein, D is dictionary, DoptFor the best dictionary after optimization, Λ is rarefaction representation coefficient, and ε is used to guarantee sparse
Property, Z is the high-definition picture for dictionary training, tsFor selected image block Spatial Dimension, tnFor the time point in data,
RlowrankFor low-rank matrix (ts×tn)。
(2) fixed Λ, updates dictionary.Residual matrix is E (i), due to:
Therefore, the update of dictionary atom are as follows:
Wherein,And Λi,mRespectivelyWithA column therein,Define μiFor
Use diRebuild { xjWhen index set, μi=h | 1≤h≤H, Λi(h) ≠ 0 }, H is total line number of sparse coefficient, Λi(h) it is
ΛiIn h column, Δ (i) be J × | μi| the matrix of size, in (μi(h), h) at value be all 1, at the rest of value be 0, J for figure
As block { xjNumber.
Step 2: estimation is weighted to true coding using the sparse coding of similar image block, will really encode with it is required
The difference obtained between sparse coding is introduced into objective function as bound term, weakens the influence of noise using the bound term, is had
Body process is as follows:
(1) really the difference between coding and obtained Λ is δα=Λy-Λx, wherein ΛxReally to encode.
(2) redundancy between image block, search and image block x are utilizedjWith scale and multiple dimensioned similar image block collection Ψj
And Οj, utilize ΨjAnd ΟjThe Λ of middle image block is weighted and averaged estimation:
Wherein, Λj,kAnd Λj,qRespectively ΨjAnd ΟjMiddle graph block xj,kAnd xj,qSparse Code, ωj,kAnd ωj,qRespectively
The weight of same scale and different scale,W is normalization factor, h
To control constant, ρjAnd τjRespectively with scale and multiple dimensioned sparse estimated value.
(3) final goal function model are as follows:
(4) fixed D, updates Λ.
In the present embodiment, η1=0.8, η2=0.15.
Step 3: utilizingReconstructed image block.
Wherein, RjFor the operator for extracting image block from image.
The above examples only illustrate the technical idea of the present invention, and this does not limit the scope of protection of the present invention, all
According to the technical idea provided by the invention, any changes made on the basis of the technical scheme each falls within the scope of the present invention
Within.
Claims (5)
1. a kind of image super-resolution rebuilding method based on joint constraint, which comprises the steps of:
Step 1, high-definition picture is divided into the identical image block of size, all image blocks is formed into image block collection as training
Sample is trained dictionary using K-SVD algorithm, obtains trained dictionary under conditions of based on low order constraint, according to
Trained dictionary acquires sparse coding；
Step 2, the low-resolution image rebuild for needs, is divided into the identical image block of size for low-resolution image, for
Some needs the image block rebuild, and search, with the multiple dimensioned similar image block of scale phase Sihe, utilizes same scale with the image block
The sparse coding of the multiple dimensioned similar image block of phase Sihe is weighted estimation to true coding, the true coding estimated,
Difference between the sparse coding that the true coding and step 1 of estimation acquire is introduced into objective function as bound term, is solved
Objective function obtains sparse coefficient；
Step 3, for the low-resolution image rebuild of needs, using in the trained dictionary of step 1 atom and step 2 obtain
Sparse coefficient be multiplied to obtain high-definition picture.
2. the image super-resolution rebuilding method according to claim 1 based on joint constraint, which is characterized in that the step
1 detailed process is as follows:
(a) dictionary learning constrained optimization problem are as follows:
Wherein, D is dictionary, DoptFor trained dictionary, Λ is sparse coding, and ε indicates that sparsity constraints, Z are to instruct for dictionary
Experienced high-definition picture block collection, tsFor selected image block Spatial Dimension, tnFor the time point in data, RlowrankFor low-rank square
Battle array (ts×tn), P representing matrix；
(b) fixed Λ, updates dictionary:
Then in dictionary atom update are as follows:
s.t.P(di,ts,tn)∈Rlowrank
Wherein, E (i) is residual matrix, di、ΛiThe i-th column in D, Λ are respectively indicated, Δ (i) is matrix, Λi,mRespectivelyM column,Indicate trained di。
3. the image super-resolution rebuilding method according to claim 1 based on joint constraint, which is characterized in that the step
2 detailed process is as follows:
(a) search and image block xjWith the multiple dimensioned similar image block of scale phase Sihe, the similar collection Ψ of same scale is obtainedjWith more rulers
Spend similar collection Οj, utilize ΨjAnd ΟjThe sparse coding of middle image block is weighted and averaged estimation:
Wherein, ωj,kIt indicates and image block xjThe similar image block x with scalej,kWeight, ωj,qIt indicates and image block xjMore rulers
Spend similar image block xj,qWeight, Λj,k、Λj,qRespectively Ψj、ΟjMiddle image block xj,kAnd xj,qSparse coding, W be return
One changes the factor, Respectively Λj、Λj,k、Λj,q
Estimation, ΛjIndicate that the jth column in sparse coding Λ, D are dictionary, h is control constant, ρj、τjIt is respectively similar, more with scale
The sparse coding value of the similar estimation of scale；
(b) difference between the true coding of estimation and obtained sparse coding is introduced into objective function as bound term,
Target function model are as follows:
Wherein, subscript(l)Indicate the l times iteration, ΛyIndicate that the sparse coding to be solved, y indicate low-resolution image, η1、η2?
For constant；
(c) fixed D, updates Λ, obtains sparse coefficient.
4. the image super-resolution rebuilding method according to claim 3 based on joint constraint, which is characterized in that the constant
η1=0.8, η2=0.15.
5. the image super-resolution rebuilding method according to claim 1 based on joint constraint, which is characterized in that step 3 institute
State the reconstruction formula of high-definition picture are as follows:
Wherein, X indicates high-definition picture, and D is dictionary, ΛyFor obtained sparse coefficient, RjTo extract image from image
Block xjOperator, J be low-resolution image block number, Λy,jFor image block xjRarefaction representation coefficient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910062627.2A CN109840888B (en) | 2019-01-23 | 2019-01-23 | Image super-resolution reconstruction method based on joint constraint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910062627.2A CN109840888B (en) | 2019-01-23 | 2019-01-23 | Image super-resolution reconstruction method based on joint constraint |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109840888A true CN109840888A (en) | 2019-06-04 |
| CN109840888B CN109840888B (en) | 2023-08-01 |
Family
ID=66884004
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910062627.2A Active CN109840888B (en) | 2019-01-23 | 2019-01-23 | Image super-resolution reconstruction method based on joint constraint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109840888B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112449140A (en) * | 2019-08-29 | 2021-03-05 | 华为技术有限公司 | Video super-resolution processing method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563968A (en) * | 2017-07-26 | 2018-01-09 | 昆明理工大学 | A kind of method based on the group medicine image co-registration denoising for differentiating dictionary learning |
| CN109064406A (en) * | 2018-08-26 | 2018-12-21 | 东南大学 | A kind of rarefaction representation image rebuilding method that regularization parameter is adaptive |
-
2019
- 2019-01-23 CN CN201910062627.2A patent/CN109840888B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563968A (en) * | 2017-07-26 | 2018-01-09 | 昆明理工大学 | A kind of method based on the group medicine image co-registration denoising for differentiating dictionary learning |
| CN109064406A (en) * | 2018-08-26 | 2018-12-21 | 东南大学 | A kind of rarefaction representation image rebuilding method that regularization parameter is adaptive |
Non-Patent Citations (1)
| Title |
|---|
| 刘冬雪 等: "一种改进的在线字典学习图像超分辨率重建算法", 《云南民族大学学报(自然科学版)》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112449140A (en) * | 2019-08-29 | 2021-03-05 | 华为技术有限公司 | Video super-resolution processing method and device |
| CN112449140B (en) * | 2019-08-29 | 2021-09-14 | 华为技术有限公司 | Video super-resolution processing method and device |
| US11922599B2 (en) | 2019-08-29 | 2024-03-05 | Huawei Technologies Co., Ltd. | Video super-resolution processing method and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109840888B (en) | 2023-08-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107610194B (en) | Magnetic resonance image super-resolution reconstruction method based on multi-scale fusion CNN | |
| CN111145181B (en) | Skeleton CT image three-dimensional segmentation method based on multi-view separation convolutional neural network | |
| CN110992270A (en) | Multi-scale residual attention network image super-resolution reconstruction method based on attention | |
| CN103295196B (en) | Based on the image super-resolution rebuilding method of non local dictionary learning and biregular item | |
| CN109389585B (en) | Brain tissue extraction method based on full convolution neural network | |
| CN109214989B (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| CN109614991A (en) | A kind of segmentation and classification method of the multiple dimensioned dilatancy cardiac muscle based on Attention | |
| CN107993204B (en) | MRI image reconstruction method based on image block enhanced sparse representation | |
| CN102156875A (en) | Image super-resolution reconstruction method based on multitask KSVD (K singular value decomposition) dictionary learning | |
| CN107301630B (en) | CS-MRI image reconstruction method based on ordering structure group non-convex constraint | |
| CN112132959A (en) | Digital rock core image processing method and device, computer equipment and storage medium | |
| Couturier et al. | Image denoising using a deep encoder-decoder network with skip connections | |
| CN106157244A (en) | A kind of QR Code Image Super-resolution Reconstruction method based on rarefaction representation | |
| CN106056647A (en) | Magnetic resonance rapid imaging method based on convolution sparse dual-layer iteration learning | |
| CN109522971A (en) | A kind of CS-MRI image reconstructing method based on classification image block rarefaction representation | |
| CN112686830B (en) | Super-resolution method of single depth map based on image decomposition | |
| CN116665065B (en) | Cross attention-based high-resolution remote sensing image change detection method | |
| CN113066025A (en) | Image defogging method based on incremental learning and feature and attention transfer | |
| Manimala et al. | Sparse MR image reconstruction considering Rician noise models: A CNN approach | |
| CN113208641B (en) | Auxiliary diagnosis method for lung nodule based on three-dimensional multi-resolution attention capsule network | |
| CN117333750A (en) | Spatial registration and local global multi-scale multi-modal medical image fusion method | |
| CN109840888A (en) | A kind of image super-resolution rebuilding method based on joint constraint | |
| CN106056554B (en) | A kind of magnetic resonance fast imaging method of gradient field convolution sparse coding | |
| CN116258730B (en) | Semi-supervised medical image segmentation method based on consistency loss function | |
| CN117593199A (en) | Double-flow remote sensing image fusion method based on Gaussian prior distribution self-attention |
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 |