CN107845065A - Super-resolution image reconstruction method and device - Google Patents
Super-resolution image reconstruction method and device Download PDFInfo
- Publication number
- CN107845065A CN107845065A CN201710835137.2A CN201710835137A CN107845065A CN 107845065 A CN107845065 A CN 107845065A CN 201710835137 A CN201710835137 A CN 201710835137A CN 107845065 A CN107845065 A CN 107845065A
- Authority
- CN
- China
- Prior art keywords
- image
- definition picture
- mrow
- levels
- detail
- 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 31
- 238000010276 construction Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 4
- 230000000574 ganglionic effect Effects 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 9
- 238000002595 magnetic resonance imaging Methods 0.000 description 36
- 238000005457 optimization Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 238000009499 grossing Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 230000002490 cerebral effect Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000011524 similarity measure Methods 0.000 description 2
- 206010038743 Restlessness Diseases 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 210000004884 grey matter Anatomy 0.000 description 1
- 210000004885 white matter Anatomy 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2211/00—Image generation
- G06T2211/40—Computed tomography
- G06T2211/416—Exact reconstruction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (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)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention provides super-resolution image reconstruction device, including:Training sample acquiring unit, obtain training sample;Dictionary construction unit, build dictionary;Low-resolution image input block, low-resolution image is converted into high-definition picture initial estimation;Sparse coding unit, sparse coding is carried out to each image sheet in the levels of detail for the high-definition picture currently estimated according to dictionary;Image update unit, update the levels of detail for the high-definition picture currently estimated;Super-resolution image reconstruction unit, when iterative is in convergence state, then the high-definition picture after storage renewal is estimated, otherwise loop iteration performs the sparse coding to each image sheet in the levels of detail of high-definition picture estimation and high-definition picture estimation.The invention also provides corresponding super-resolution image reconstruction method.By technical scheme, MRI resolution ratio can be significantly increased, picture noise is effectively removed and obscures and wait distortion, recover complicated fine structure, there is more preferable subjective and objective effect.
Description
Technical field
The invention belongs to image processing field, and in particular to a kind of super-resolution image reconstruction method and device.
Background technology
With developing rapidly for mr imaging technique, resolution ratio, signal to noise ratio and the sweep speed of MRI have
Larger raising, but neonate it is special MR imaging apparatus it is seldom, neonate's magnetic resonance imaging effect is undesirable
Problem is not well solved still.In neonatal magnetic resonance imaging, single voxel may include several difference
The tissue of type, and to easily cause neonate restless in magnetic resonance imaging environment for external interference, these factors will be further
Reduce neonate's MRI resolution ratio of capture.
In order to improve the diagnosis efficiency of the resolution ratio of neonate's MRI and doctor, clinic generally utilizes interpolation side
Method carries out super-resolution rebuilding to neonate's image.Image interpolation generally using known data point on low resolution raster image come
Estimate the unknown number strong point on high-definition picture grid, common method has:Arest neighbors interpolation, polynomial interopolation, batten are inserted
Value and the interpolation based on edge.Although these methods, which have, calculates the advantages of cost is low, the high-resolution of reconstruction is easily caused
There are the distortion phenomenons such as halation, ring, edge-smoothing, fuzzy and aliasing effect in rate image.
Have in the prior art and propose the image super-resolution method based on sparse expression, it mainly uses convex optimization regular terms
To establish object function, ignore neonatal cerebral white matter and the contrast distribution of grey matter brightness changes, also do not take into full account new
Raw youngster's brain magnetic resonance image noise is big, resolution ratio is low and the distortion phenomenon such as partial volume effect, causes the high-resolution rebuild
Neonate's MRI edge-smoothing and lack of resolution.In addition, the super-resolution method based on sparse expression is generally from interior
Portion's image or exterior view image set train the super complete dictionary for sparse expression.But internal image be from input degrade it is low
Image in different resolution, and external image is the overall situation in these internal images and external image from general Large Scale Graphs image set
May be dissimilar compatible with target high-resolution with local mode, cause target high-resolution image reconstruction errors.It is existing to be based on
The super-resolution method of deep learning is, it is necessary to which substantial amounts of image set and correspondence markings come training convolutional neural networks model or generation
Resist network model, but in reality neonate's MRI be not easy obtain and negligible amounts, do not consider yet it is same individual or
Longitudinal Image Priori Knowledge modeling of Different Individual, so as to constrain the super-resolution performance of training pattern.Therefore, existing oversubscription
The anatomical structure precision and resolution ratio for neonate's MRI that resolution method is rebuild are still not ideal enough.
The content of the invention
In place of solving the deficiencies in the prior art, by using longitudinal magnetic resonance figure of Different Individual
The high frequency detail layer of picture (such as child's image) carrys out training dictionary, to low resolution neonate's MRI of input, establishes
A kind of neonate's MRI super-resolution model based on relict texture sparse expression, picture noise is suppressed with this,
Neonate's MRI super-resolution rebuilding performance is improved, further promotes mr imaging technique in neonate's imaging field
Clinical diagnosis and treatment use.
Super-resolution image reconstruction device, including:
Training sample acquiring unit, for obtaining training sample, the training sample includes the high score of multiple Different Individuals
Resolution child's MRI;
Dictionary construction unit, for building dictionary, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, feature is carried out to the covariance matrix of each class race
Value, which is decomposed, obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is one
Individual sub- dictionary;
Low-resolution image input block, it is right for inputting any neonate's MRI as low-resolution image
Low-resolution image carries out image interpolation to initialize high-definition picture, the high-definition picture currently estimated;
Sparse coding unit, represent the high-definition picture currently estimated being divided into Primary layer and details using image is double-deck
Layer, all image sheets of current estimation high-definition picture levels of detail are extracted, according to dictionary to current estimation high-definition picture
Levels of detail in each image sheet carry out sparse coding;
Image update unit, for each image sheet in the currently levels of detail of estimation high-definition picture, to estimating high score
The levels of detail of resolution image is updated, i.e., estimation high-definition picture is updated;
Super-resolution image reconstruction unit, for high-definition picture estimation after renewal and current estimation high-resolution
When the sparse coding of each image sheet is in convergence state in image detail layer, the high-definition picture estimation after storage renewal;It is no
Then enter next iteration using the high-definition picture estimation after renewal as current estimation high-definition picture, to continue to height
Image in different resolution is estimated and the current sparse coefficient for estimating each image sheet in high-definition picture levels of detail is updated calculating;
High-definition picture estimation after renewal is super-resolution image.
Further, the high-definition picture currently estimated is divided into Primary layer and levels of detail in sparse coding unit, wrapped
Include:
Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high-definition picture currently estimated with
The difference of its Primary layer is levels of detail.
Further, in sparse coding unit according to dictionary to each in the levels of detail for the high-definition picture currently estimated
Image sheet carries out sparse coding, including:
Step 41, an optional image sheet conduct from all image sheets for the high-definition picture levels of detail currently estimated
Target image piece ei, wherein i=1,2, ..., all image sheets that I, I are the high-definition picture levels of detail currently estimated
Number;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out dilute
Dredge and encode and obtain sparse coefficient αi, wherein k=1,2 ..., K;Complete according to dictionary to the high resolution graphics currently estimated
Each image sheet of the levels of detail of picture carries out sparse coding and obtains sparse coefficient;
It is described that a sub- dictionary D is chosen from K sub- dictionariesk, the sub- dictionary DkThe class race center corresponding to it should be met
With target image piece eiEuclidean distance it is minimum.
Further, the high-resolution of high-definition picture estimation and estimation after being updated in super-resolution image reconstruction unit
When the sparse coding of each image sheet is in convergence state in rate image detail layer, meet formula (1):
In formula (1), i=1,2, ..., I, η are auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Represent target image piece e after the t times iterated revisioniSparse coefficient,For αi (t)Non-local mean, αi (t)Represent the
Target image piece e after t iterated revisioniSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H
For degenerate matrix, Y is the low-resolution image of input, Z(t)Represent that high-resolution neonate's image after the t times iteration renewal is estimated
The Primary layer of meter;
For target image piece e after the t times iterated revisioniRegularization parameter,δ(t)It is t
The noise variance of high-definition picture estimation after secondary iteration renewal;For sparse coefficient after the t times iterationStandard deviation
Difference.
Further, the high-definition picture estimation after being updated in super-resolution image reconstruction unit
Wherein α(t+1)For the sparse coefficient matrix of the levels of detail of the high-definition picture estimation after the t times iterated revision, α(t+1)=
{α1 (t+1),α2 (t+1),...αi (t+1),...,αI (t+1), D is dictionary, D={ D1,D2,...,Dk,..,DK}。
Present invention also offers super-resolution image reconstruction method, including:
Step 1, training sample is obtained, the training sample includes high-resolution child's magnetic resonance figure of multiple Different Individuals
Picture;
Step 2, dictionary is built, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, feature is carried out to the covariance matrix of each class race
Value, which is decomposed, obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is one
Individual sub- dictionary;
Step 3, neonate's MRI of any low resolution is inputted as low-resolution image, to low resolution figure
As initializing high-definition picture using image interpolation, the high-definition picture currently estimated;
Step 4, represent the high-definition picture currently estimated being divided into Primary layer and levels of detail using image is double-deck, extract
All image sheets for the high-definition picture levels of detail currently estimated, according to dictionary to the thin of the high-definition picture currently estimated
Each image sheet carries out sparse coding and obtains corresponding sparse coefficient in ganglionic layer;
Step 5, according to the sparse coefficient of each image sheet in the levels of detail for the high-definition picture currently estimated, to current
The levels of detail of the high-definition picture of estimation is updated, i.e., the high-definition picture currently estimated is updated;
Step 6, it is each in the levels of detail of current estimation high-definition picture and the high-definition picture estimation after renewal
When the sparse coding of image sheet is in convergence state, the high-definition picture estimation after storage renewal;Otherwise with the high score after renewal
Resolution Image estimation enters next iteration as the high-definition picture currently estimated, to continue the high-resolution to currently estimating
The sparse coefficient of each image sheet is updated calculating in image and the high-definition picture levels of detail currently estimated;
High-definition picture estimation after renewal is super-resolution image.
Further, the high-definition picture currently estimated is divided into Primary layer and levels of detail in step 4, including:
Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high-definition picture currently estimated with
The difference of its Primary layer is levels of detail.
Further, in step 4 according to dictionary to each image sheet in the levels of detail for the high-definition picture currently estimated
Carry out sparse coding and obtain corresponding sparse coefficient, including:
Step 41, an optional image sheet conduct from all image sheets for the high-definition picture levels of detail currently estimated
Target image piece ei, wherein i=1,2, ..., all image sheets that I, I are the high-definition picture levels of detail currently estimated
Number;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out dilute
Dredge and encode and obtain sparse coefficient αi, wherein k=1,2 ..., K;Complete according to dictionary to the high resolution graphics currently estimated
Each image sheet of the levels of detail of picture carries out sparse coding.
Further, in the levels of detail that high-definition picture estimation and high-definition picture after being updated in step 6 are estimated
When the sparse coding of each image sheet is in convergence state, meet formula (1):
In formula (1), i=1,2, ..., I, η are auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Represent the target image piece e after the t times iterated revisioniSparse coefficient,For αi (t)Non-local mean, αi (t)Table
Show the target image piece e after the t times iterated revisioniSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H
For degenerate matrix, Y is the low-resolution image of input, Z(t)Represent that high-resolution neonate's image after the t times iteration renewal is estimated
The Primary layer of meter;
For target image piece e after the t times iterated revisioniRegularization parameter,δ(t)It is t
The noise variance of high-definition picture estimation after secondary iteration renewal;For sparse coefficient after the t times iterationStandard deviation
Difference.
Further, the high-definition picture estimation after being updated in step 6Wherein α(t+1)
For the sparse coefficient matrix of the levels of detail of the high-definition picture estimation after the t times iterated revision, α(t+1)={ α1 (t+1),
α2 (t+1),...αi (t+1),...,αI (t+1), D is dictionary, D={ D1,D2,...,Dk,..,DK}。
Compared with prior art, the present invention has following technical characteristic:
The present invention can reconstruct high-resolution neonate's MRI, effectively remove picture noise and fuzzy etc.
Distortion, complicated fine structure is recovered, there is more preferable subjective and objective effect, usually more than existing state-of-the-art image super-resolution
Method.
Brief description of the drawings
Fig. 1 is super-resolution image reconstruction schematic device;
Fig. 2 is the flow chart of the present invention;
Fig. 3 (a), the experimental result comparison diagram (one) for being SNR;Fig. 3 (b) is SSIM experimental result comparison diagram (one);Fig. 3
(c) the experimental result comparison diagram (two) for being SNR;Fig. 3 (d) is SSIM experimental result comparison diagram (two).
Embodiment
In order to be more clearly understood that the present invention, the present invention is done with reference to the accompanying drawings and examples further detailed
Explanation.
Embodiment 1
The present embodiment provides super-resolution image reconstruction device, as shown in figure 1, including:
Training sample acquiring unit, for obtaining training sample, the training sample includes the high score of multiple Different Individuals
Resolution child's MRI;
Dictionary construction unit, for building dictionary, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, feature is carried out to the covariance matrix of each class race
Value, which is decomposed, obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is one
Individual sub- dictionary;
The present invention chooses other several bodies for a certain individual low resolution neonate's MRI, the present invention
One group of child's MRI carrys out the dictionary of the sparse modeling of off-line training image super-resolution rebuilding.
The present invention proposes each child's image in training sample being divided into levels of detail and Primary layer, by each child
Image is filtered smoothly, such as Gaussian filter, and filtered smoothed image is referred to as Primary layer, and the image before filtering is with putting down
The difference of sliding image is designated as levels of detail, then the present invention by K mean cluster by the thin of all child's images in training sample
Ganglionic layer is divided into K Ge Lei races and obtains the class race center of each class race.
Specifically, the object function of K mean cluster is defined as follows formula:
Wherein, K is to cluster number, mkIt is k-th of cluster centre or class race SkThe average of middle image sheet, r are represented from one group of children
Youngster image L=(L1,L2,…,LN) extraction levels of detail R in image sheet set, S={ S1,S2,…,SKRepresent class family set.
The present invention solves above-mentioned optimization problem using the iterative refinement method with Fast Convergent local optimum.For each by thin
Save the class race S that image sheet is formedk, the present invention is first using its covariance matrix S of Eigenvalues Decompositionk Again by characteristic vector structure
Into corresponding sub- dictionary Dk, meet formula
Wherein, d is character pair value ρ characteristic vector, and I is unit matrix.By collecting the sub- dictionary of these training, this
Whole super complete dictionary D={ D are constructed in invention1,D2,…,DK}。
Low-resolution image input block, for inputting neonate's MRI of any low resolution as low resolution
Rate image, low-resolution image is initialized into high-definition picture estimation using image interpolation;
Because neonatal cerebral MRI has that noise is big, resolution ratio is low and a distortion phenomenon such as partial volume effect,
Cause high-resolution neonate's MRI edge-smoothing and the lack of resolution estimated using interpolation method, so need to be to working as
The high-definition picture of preceding estimation carries out sparse modeling, to obtain precision height, the preferable super-resolution image of resolution ratio.
Sparse coding unit, for the high-definition picture currently estimated to be divided into Primary layer and levels of detail, extraction is current
All image sheets of the high-definition picture levels of detail of estimation, according to dictionary to the levels of detail for the high-definition picture currently estimated
In each image sheet carry out sparse coding and obtain corresponding sparse coefficient;
In the present invention, Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high score currently estimated
The difference of resolution image and its Primary layer is levels of detail.
The present invention improves the performance of similarity measure using levels of detail come alternate image, and utilizes the sparse coding of levels of detail
To recover the radio-frequency component of high-definition picture, it can effectively remove picture noise and obscure and wait distortion, especially for complexity
Fine structure can be good at recovery effects.
In the present invention, according to dictionary to each in the levels of detail for the high-definition picture currently estimated in sparse coding unit
Image sheet carries out sparse coding and obtains corresponding sparse coefficient, including:
Step 41, an optional image sheet conduct from all image sheets for the high-definition picture levels of detail currently estimated
Target image piece ei, wherein i=1,2, ..., all image sheets that I, I are the high-definition picture levels of detail currently estimated
Number;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out dilute
Dredge and encode and obtain sparse coefficient αi, wherein k=1,2 ..., K;Complete according to dictionary to the high resolution graphics currently estimated
Each image sheet of the levels of detail of picture carries out sparse coding and obtains sparse coefficient;
It is described that a sub- dictionary D is chosen from K sub- dictionariesk, the sub- dictionary DkThe class race center corresponding to it should be met
With target image piece eiEuclidean distance it is minimum.
In the present embodiment, described image piece is the image sheet of 5 × 5 × 5 voxels.
Image update unit, for according to each image sheet in the levels of detail for obtaining the high-definition picture currently estimated
Sparse coefficient, the levels of detail of the high-definition picture to currently estimating are updated, i.e., high-definition picture estimation are carried out more
Newly;
Super-resolution image reconstruction unit, for the high-definition picture estimation and high-definition picture estimation after renewal
Levels of detail in the sparse coding of each image sheet when be in convergence state, the high-definition picture after storage renewal is estimated;Otherwise
Next iteration as the high-definition picture currently estimated and is entered using the high-definition picture estimation after renewal, to continue to height
The sparse coefficient of each image sheet is updated calculating in the levels of detail that image in different resolution is estimated and high-definition picture is estimated;Most
Eventually, the high-definition picture estimation after renewal is super-resolution image.
In order to obtain the standard of original clean image X sparse coefficient from low resolution neonate's MRI Y of observation
Really estimation, the present invention propose a kind of neonate's magnetic resonance super-resolution image reconstruction model of relict texture sparse expression:
S.t.X=Z+E, E=D α,
Wherein, αYRepresent the sparse coefficient estimated from low resolution neonate's MRI Y of observation, DLIt is from child
The super complete dictionary of MRI collection L levels of detail set study, Z and E are the Primary layer and levels of detail that X is decomposed respectively,
αiIt is that the image sheet e at coordinate i is centrally located in levels of detail EiSparse coefficient, Y be observation low resolution neonate's magnetic resonance
Image, X are original high-definition pictures, and H is the degenerate matrix for describing fuzzy and down-sampling, and λ is punishment parameter.c≥0
It is the parameter for controlling non-convex penalty term Φ.Then, the present invention will select suitable parameter c to ensure that optimization problem is convex letter
Number, and derive the closed solution of image super-resolution problem.The present invention is used as containing parametrization using the secondary penalty in part
Non-convex regular terms:
Wherein, μiIt is image sheet e in neonate's MRI levels of detail EiOne group of similar diagram photo egSparse coefficient
αgNon local weighted average.Specifically, μiIt is defined as follows:
Wherein, weighting functionH is the parameter for controlling rate of decay, and W is to return
One changes parameter and meets
In off-line training dictionary D={ D1,D2,…,DKIt is known under the premise of, can on image super-resolution optimization problem
By be reduced to it is another in the form of:
Theorem:If the λ > 0 and λ of 0≤c < 1/, above-mentioned object function are strictly convex functions.
The present invention goes to solve sparse coding optimization problem using iterative shrinkage algorithm:
High-definition picture estimation and high-definition picture after being updated in super-resolution image reconstruction unit are estimated thin
When the sparse coding of each image sheet is in convergence state in ganglionic layer, meet formula (1):
In formula (1), i=1,2 ..., I, η be auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Represent the target image piece e after the t times iteration renewaliSparse coefficient,For αi (t)Non-local mean, αi (t)Table
Show the target image piece e after the t times iteration renewaliSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H
For degenerate matrix, Y is low resolution neonate's MRI of input, Z(t)Represent the high-resolution after the t times iteration renewal
The Primary layer of neonate's Image estimation;
For target image piece e after the t times iterated revisioniRegularization parameter,δ(t)It is t
The noise variance of the high-definition picture of current estimation after secondary iteration renewal;For sparse coefficient after the t times iteration's
Standard deviation.
As can be seen that the present invention is asked using the non-convex regular terms containing parameter to build image super-resolution from formula (1)
The sparse model of topic, solved using the function of generally use Strict Convex in the prior art, it is proposed that a kind of optimization method.
The high-definition picture of current estimation after being updated in super-resolution image reconstruction unit
Wherein α(t+1)For the sparse coefficient matrix of the levels of detail of the high-definition picture estimation after the t+1 times iterated revision, α(t+1)=
{α1 (t+1),α2 (t+1),...αi (t+1),...,αI (t+1), D is dictionary, D={ D1,D2,...,Dk,..,DK}。
Embodiment 2
The present embodiment additionally provides a kind of super-resolution image reconstruction method, as shown in figure 1, including:
Step 1, training sample is obtained, the training sample includes high-resolution child's magnetic resonance figure of multiple Different Individuals
Picture;
Step 2, dictionary is built, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, feature is carried out to the covariance matrix of each class race
Value, which is decomposed, obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is one
Individual sub- dictionary;
The present invention chooses other several bodies for a certain individual low resolution neonate's MRI, the present invention
One group of child's MRI carrys out the dictionary of training image super-resolution rebuilding sparse model.
The present invention proposes each child's image in training sample being divided into levels of detail and Primary layer, by each child
Image is filtered smoothly, such as Gaussian filter, and filtered smoothed image is taken as Primary layer, and the image before filtering is with putting down
The difference of sliding image is denoted as levels of detail, then the present invention by K mean cluster by the levels of detail of all images in training sample
It is divided into K Ge Lei races and obtains the class race center of each class race.
Specifically, the object function of K mean cluster is defined as follows formula:
Wherein, K is to cluster number, mkIt is k-th of cluster centre or class race SkThe average of middle image sheet, r are represented from one group of children
Youngster image L=(L1,L2,…,LN) extraction levels of detail R in image sheet set, S={ S1,S2,…,SKRepresent class family set.
The present invention solves above-mentioned optimization problem using the iterative refinement method with Fast Convergent local optimum.For each by thin
Save the class race S that image sheet is formedk, the present invention is first using its covariance matrix S of Eigenvalues Decompositionk Again by characteristic vector structure
Into corresponding sub- dictionary Dk, make its satisfaction
Wherein, d is character pair value ρ characteristic vector, and I is unit matrix.By collecting the sub- dictionary of these training, this
One complete super complete dictionary D={ D of invention structure1,D2,…,DK}。
Step 3, neonate's MRI of any one low resolution is inputted as low-resolution image, by low resolution
Imagery exploitation image interpolation is estimated to initialize high-definition picture;
Because neonatal cerebral MRI has that noise is big, resolution ratio is low and a distortion phenomenon such as partial volume effect,
Cause high-resolution neonate's MRI edge-smoothing and the lack of resolution estimated using interpolation method, so need pair
The high-definition picture currently estimated carries out sparse coding to obtain precision height, the preferable super-resolution image of resolution ratio.
Step 4, the high-definition picture currently estimated is divided into Primary layer and levels of detail, extracts the high-resolution currently estimated
All image sheets of rate image detail layer, according to dictionary to each image sheet in the levels of detail for the high-definition picture currently estimated
Carry out sparse coding and obtain sparse coefficient;
In the present invention, Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high score currently estimated
The difference of resolution image and its Primary layer is levels of detail.
The present invention substitutes the performance that child's image improves similarity measure using levels of detail, and utilizes the sparse of levels of detail
Coding can effectively remove neonate's picture noise and obscure and wait distortion, especially to recover the radio-frequency component of high-definition picture
It is that can be good at recovery effects for complicated fine structure.
In the present invention, according to dictionary to each image sheet in the levels of detail for the high-definition picture currently estimated in step 4
Carry out sparse coding and obtain sparse coefficient, including:
Step 41, an optional image sheet conduct from all image sheets for the high-definition picture levels of detail currently estimated
Target image piece ei, wherein i=1,2, ..., all image sheets that I, I are the high-definition picture levels of detail currently estimated
Number;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out dilute
Dredge sparse coefficient α corresponding to encoding and obtainingi, wherein k=1,2 ..., K;Complete according to dictionary to the high score currently estimated
The sparse coding of each image sheet of the levels of detail of resolution image;
It is described that a sub- dictionary D is chosen from K sub- dictionariesk, the sub- dictionary DkThe class race center corresponding to it should be met
With target image piece eiEuclidean distance it is minimum.
In the present embodiment, described image piece is the image sheet of 5 × 5 × 5 voxels.
Step 5, according to the sparse coefficient of each image sheet in the levels of detail of the high-definition picture of the current estimation of acquisition,
The levels of detail of high-definition picture to currently estimating is updated, i.e., high-definition picture estimation is updated;
Step 6, each image in the levels of detail of high-definition picture estimation after renewal and high-definition picture estimation
When the sparse coding of piece is in convergence state, the high-definition picture estimation after storage renewal;Otherwise with the high-resolution after renewal
Image estimation is as current estimation high-definition picture and enters next iteration, to continue the high-definition picture to currently estimating
Each image sheet carries out sparse coding in the high-definition picture levels of detail currently estimated;Finally, the high-resolution after renewal
Image estimation is super-resolution image.
In order to obtain the standard of original clean image X sparse coefficient from low resolution neonate's MRI Y of observation
Really estimation, the present invention propose a kind of neonate's magnetic resonance super-resolution image reconstruction model of relict texture sparse expression:
S.t.X=Z+E, E=D α,
Wherein, αYRepresent the sparse coefficient estimated from low resolution neonate's MRI Y of observation, DLIt is from child
The super complete dictionary of MRI collection L levels of detail set study, Z and E are the Primary layer and levels of detail that X is decomposed respectively,
αiIt is that the image sheet e at coordinate i is centrally located in levels of detail EiSparse coefficient, Y be observation low resolution neonate's magnetic resonance
Image, X are original high-definition pictures, and H is the degenerate matrix for describing fuzzy and down-sampling, and λ is punishment parameter.c≥0
It is the parameter for controlling non-convex penalty term Φ.Then, the present invention will select suitable parameter c to ensure that optimization problem is convex letter
Number, and derive the closed solution of image super-resolution problem.The present invention is using the secondary penalty in part as non-convex regular terms:
Wherein, μiIt is image sheet e in neonate's MRI levels of detail EiOne group of similar diagram photo egSparse coefficient
αgNon local weighted average.Specifically, μiIt is defined as follows:
Wherein, weighting functionH is the parameter for controlling rate of decay, and W is to return
One changes parameter and meets
In off-line training dictionary D={ D1,D2,…,DKIt is known under the premise of, can on image super-resolution optimization problem
By be reduced to it is another in the form of:
Theorem:If the λ > 0 and λ of 0≤c < 1/, above-mentioned object function are strictly convex functions.
The present invention goes to solve sparse coding optimization problem using iterative shrinkage algorithm:
In the present invention, in the levels of detail that high-definition picture estimation and high-definition picture after being updated in step 6 are estimated
When the sparse coding of each image sheet is in convergence state, meet formula (1):
In formula (1), i=1,2, ..., I, η are auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Represent target image piece e after the t times iterated revisioniSparse coefficient,For αi (t)Non-local mean, αi (t)Represent the
Target image piece e after t iterated revisioniSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H
For degenerate matrix, Y is the low-resolution image of input, Z(t)Represent that high-resolution neonate's image after the t times iteration renewal is estimated
The Primary layer of meter;
δ(t)It is the noise side of the high-definition picture of the current estimation after the t times iteration updates
Difference;For sparse coefficient after the t times iterationStandard deviation.
As can be seen that the present invention is asked using the non-convex regular terms containing parameter to build image super-resolution from formula (1)
The sparse model of topic, solved using the function of generally use Strict Convex in the prior art, it is proposed that a kind of optimization method.
High-definition picture estimation after being updated in step 6Wherein α(t+1)For the t times repeatedly
For the sparse coefficient matrix of the levels of detail of revised high-definition picture estimation, α(t+1)={ α1 (t+1),α2 (t+1),...
αi (t+1),...,αI (t+1), D is dictionary, D={ D1,D2,...,Dk,..,DK}。
Experimental result:
Fig. 3 (a), Fig. 3 (b) respectively pass through spline method (Spline), non-local mean up-samples method (NLMU), low
The full calculus of variations of order (LRTV) puies forward RSSR methods with the present invention and carries out super-resolution to 24 neonate magnetic resonance T1 images respectively
SNR, SSIM of reconstruction experimental result comparison diagram.
Fig. 3 (c), Fig. 3 (d) respectively pass through spline method (Spline), non-local mean up-samples method (NLMU), low
The full calculus of variations of order (LRTV) puies forward RSSR methods with the present invention and carries out super-resolution to 24 neonate magnetic resonance T2 images respectively
SNR, SSIM of reconstruction experimental result comparison diagram.
Claims (10)
1. super-resolution image reconstruction device, it is characterised in that including:
Training sample acquiring unit, for obtaining training sample, the training sample includes the high-resolution of multiple Different Individuals
Child's MRI;
Dictionary construction unit, for building dictionary, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, characteristic value point is carried out to the covariance matrix of each class race
Solution obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is a son
Dictionary;
Low-resolution image input block, for inputting any neonate's MRI as low-resolution image, to low point
Resolution image carries out image interpolation to initialize high-definition picture, obtains currently estimating high-definition picture;
Sparse coding unit, current estimation high-definition picture is divided into Primary layer and levels of detail, extracts current estimation high-resolution
All image sheets of rate image detail layer, each image sheet in the levels of detail of current estimation high-definition picture is entered according to dictionary
Row sparse coding;
Image update unit, for each image sheet in the currently levels of detail of estimation high-definition picture, to estimating high-resolution
The levels of detail of image is updated, i.e., estimation high-definition picture is updated;
Super-resolution image reconstruction unit, for high-definition picture estimation after renewal and current estimation high-definition picture
When the sparse coding of each image sheet is in convergence state in levels of detail, the high-definition picture estimation after storage renewal;Otherwise with
High-definition picture estimation after renewal enters next iteration as current estimation high-definition picture, to continue to high-resolution
The sparse coefficient of each image sheet is updated calculating in rate Image estimation and current estimation high-definition picture levels of detail;
High-definition picture estimation after renewal is super-resolution image.
2. super-resolution image reconstruction device as claimed in claim 1, it is characterised in that will currently estimate in sparse coding unit
The high-definition picture of meter is divided into Primary layer and levels of detail, including:
Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high-definition picture currently estimated and its base
This layer of difference is levels of detail.
3. super-resolution image reconstruction device as claimed in claim 1, it is characterised in that according to dictionary in sparse coding unit
Sparse coding is carried out to each image sheet in the levels of detail for the high-definition picture currently estimated, including:
Step 41, from all image sheets for the high-definition picture levels of detail currently estimated an optional image sheet as target
Image sheet ei, wherein i=1,2, ..., the number for all image sheets that I, I are the high-definition picture levels of detail currently estimated;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out sparse volume
Code simultaneously obtains sparse coefficient αi, wherein k=1,2 ..., K;Complete according to dictionary to the high-definition picture currently estimated
Each image sheet of levels of detail carries out sparse coding and obtains sparse coefficient;
It is described that a sub- dictionary D is chosen from K sub- dictionariesk, the sub- dictionary DkClass race center and the mesh corresponding to it should be met
Mark on a map photo eiEuclidean distance it is minimum.
4. super-resolution image reconstruction device as claimed in claim 1, it is characterised in that in super-resolution image reconstruction unit
The sparse coding of each image sheet is in receipts in the high-definition picture levels of detail that high-definition picture after renewal is estimated and estimated
When holding back state, meet formula (1):
<mrow>
<msubsup>
<mi>&alpha;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>=</mo>
<mi>min</mi>
<mo>{</mo>
<mo>|</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>|</mo>
<mo>,</mo>
<mi>max</mi>
<mrow>
<mo>(</mo>
<mo>|</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>|</mo>
<mo>-</mo>
<msubsup>
<mi>&eta;&lambda;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msubsup>
<mi>c&eta;&lambda;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mn>0</mn>
<mo>}</mo>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula (1), i=1,2, ..., I, η are auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Table
Show target image piece e after the t times iterated revisioniSparse coefficient,For αi (t)Non-local mean, αi (t)Represent the t times repeatedly
Target image piece e after generation amendmentiSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H to move back
Change matrix, Y be input low-resolution image, Z(t)Represent high-resolution neonate's Image estimation after the t times iteration renewal
Primary layer;
For target image piece e after the t times iterated revisioniRegularization parameter,δ(t)Be the t times repeatedly
The noise variance of high-definition picture estimation after generation renewal;For sparse coefficient after the t times iterationStandard deviation.
5. super-resolution image reconstruction device as claimed in claim 1, it is characterised in that in super-resolution image reconstruction unit
High-definition picture estimation after renewalWherein α(t+1)For the high-resolution after the t times iterated revision
The sparse coefficient matrix of the levels of detail of rate Image estimation, α(t+1)={ α1 (t+1),α2 (t+1),...αi (t+1),...,αI (t+1), D is word
Allusion quotation, D={ D1,D2,...,Dk,..,DK}。
6. super-resolution image reconstruction method, it is characterised in that including:
Step 1, training sample is obtained, the training sample includes high-resolution child's MRI of multiple Different Individuals;
Step 2, dictionary is built, the dictionary includes K sub- dictionaries;
The acquisition of sub- dictionary includes:
The levels of detail of training sample is clustered to obtain K Ge Lei races, characteristic value point is carried out to the covariance matrix of each class race
Solution obtains the unitary matrice that the characteristic vector of each class race is formed, and the unitary matrice that the characteristic vector of each class race is formed is a son
Dictionary;
Step 3, neonate's MRI of any low resolution is inputted as low-resolution image, to low-resolution image profit
High-definition picture is initialized with image interpolation, the high-definition picture currently estimated;
Step 4, represent the high-definition picture currently estimated being divided into Primary layer and levels of detail using image is double-deck, extraction is current
All image sheets of the high-definition picture levels of detail of estimation, according to dictionary to the levels of detail for the high-definition picture currently estimated
In each image sheet carry out sparse coding and obtain corresponding sparse coefficient;
Step 5, according to the sparse coefficient of each image sheet in the levels of detail for the high-definition picture currently estimated, to current estimation
The levels of detail of high-definition picture be updated, i.e., the high-definition picture currently estimated is updated;
Step 6, each image in the levels of detail of current estimation high-definition picture and the high-definition picture estimation after renewal
When the sparse coding of piece is in convergence state, the high-definition picture estimation after storage renewal;Otherwise with the high-resolution after renewal
Image estimation enters next iteration as the high-definition picture currently estimated, to continue the high-definition picture to currently estimating
The sparse coefficient of each image sheet is updated calculating in the high-definition picture levels of detail currently estimated;
High-definition picture estimation after renewal is super-resolution image.
7. super-resolution image reconstruction method as claimed in claim 6, it is characterised in that the height that will currently estimate in step 4
Image in different resolution is divided into Primary layer and levels of detail, including:
Primary layer is obtained after being filtered to the high-definition picture currently estimated, the high-definition picture currently estimated and its base
This layer of difference is levels of detail.
8. super-resolution image reconstruction method as claimed in claim 6, it is characterised in that according to dictionary to current in step 4
Each image sheet carries out sparse coding and obtains corresponding sparse coefficient in the levels of detail of the high-definition picture of estimation, including:
Step 41, from all image sheets for the high-definition picture levels of detail currently estimated an optional image sheet as target
Image sheet ei, wherein i=1,2, ..., the number for all image sheets that I, I are the high-definition picture levels of detail currently estimated;
Step 42, a sub- dictionary D is chosen from K sub- dictionariesk, according to sub- dictionary DkTo target image piece eiCarry out sparse volume
Code simultaneously obtains sparse coefficient αi, wherein k=1,2 ..., K;Complete according to dictionary to the high-definition picture currently estimated
Each image sheet of levels of detail carries out sparse coding.
9. super-resolution image reconstruction method as claimed in claim 6, it is characterised in that the high-resolution after being updated in step 6
When the sparse coding of each image sheet is in convergence state in rate Image estimation and the levels of detail of high-definition picture estimation, meet formula
(1):
<mrow>
<msubsup>
<mi>&alpha;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>=</mo>
<mi>min</mi>
<mo>{</mo>
<mo>|</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>|</mo>
<mo>,</mo>
<mi>max</mi>
<mrow>
<mo>(</mo>
<mo>|</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>|</mo>
<mo>-</mo>
<msubsup>
<mi>&eta;&lambda;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msubsup>
<mi>c&eta;&lambda;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mn>0</mn>
<mo>}</mo>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&gamma;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula (1), i=1,2, ..., I, η are auxiliary parameter, c >=0, c represent to parameterize the parameter of non-convex penalty term,Table
Show the target image piece e after the t times iterated revisioniSparse coefficient,For αi (t)Non-local mean, αi (t)Represent the t times
Target image piece e after iterated revisioniSparse coefficient;
γi (t)=αi (t)-μi (t)+ηDk THT(Y-H(Z(t)+Dkαi (t))), DkFor k-th of sub- dictionary, k=1,2 ..., K, H to move back
Change matrix, Y be input low-resolution image, Z (t) represent high-resolution neonate's Image estimation after the t times iteration renewal
Primary layer;
For target image piece e after the t times iterated revisioniRegularization parameter,δ(t)Be the t times repeatedly
The noise variance of high-definition picture estimation after generation renewal;For sparse coefficient after the t times iterationStandard deviation.
10. super-resolution image reconstruction method as claimed in claim 6, it is characterised in that the high-resolution after being updated in step 6
Rate Image estimationWherein α(t+1)For the thin of the high-definition picture estimation after the t times iterated revision
The sparse coefficient matrix of ganglionic layer,
α(t+1)={ α1 (t+1),α2 (t+1),...αi (t+1),...,αI (t+1), D is dictionary, D={ D1,D2,...,Dk,..,DK}。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710835137.2A CN107845065B (en) | 2017-09-15 | 2017-09-15 | Super-resolution image reconstruction method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710835137.2A CN107845065B (en) | 2017-09-15 | 2017-09-15 | Super-resolution image reconstruction method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107845065A true CN107845065A (en) | 2018-03-27 |
CN107845065B CN107845065B (en) | 2021-06-15 |
Family
ID=61683290
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710835137.2A Expired - Fee Related CN107845065B (en) | 2017-09-15 | 2017-09-15 | Super-resolution image reconstruction method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107845065B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108532700A (en) * | 2018-03-02 | 2018-09-14 | 李军 | The hospital's self-stripping system avoided infection |
CN108764368A (en) * | 2018-06-07 | 2018-11-06 | 西安邮电大学 | A kind of image super-resolution rebuilding method based on matrix mapping |
CN109145738A (en) * | 2018-07-18 | 2019-01-04 | 浙江工业大学 | The dynamic video dividing method of beam low-rank representation is weighed about based on the non-convex regularization of weighting and iteration |
CN109166093A (en) * | 2018-07-09 | 2019-01-08 | 西北大学 | A kind of detection method for image salient region |
CN109447905A (en) * | 2018-11-06 | 2019-03-08 | 大连海事大学 | Based on the marine image super-resolution rebuilding method for differentiating dictionary |
CN110619603A (en) * | 2019-08-29 | 2019-12-27 | 浙江师范大学 | Single image super-resolution method for optimizing sparse coefficient |
CN113344829A (en) * | 2021-04-21 | 2021-09-03 | 复旦大学 | Portable ultrasonic video optimization reconstruction method for multi-channel generation countermeasure network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105550988A (en) * | 2015-12-07 | 2016-05-04 | 天津大学 | Super-resolution reconstruction algorithm based on improved neighborhood embedding and structure self-similarity |
CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
CN107067367A (en) * | 2016-09-08 | 2017-08-18 | 南京工程学院 | A kind of Image Super-resolution Reconstruction processing method |
-
2017
- 2017-09-15 CN CN201710835137.2A patent/CN107845065B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105550988A (en) * | 2015-12-07 | 2016-05-04 | 天津大学 | Super-resolution reconstruction algorithm based on improved neighborhood embedding and structure self-similarity |
CN107067367A (en) * | 2016-09-08 | 2017-08-18 | 南京工程学院 | A kind of Image Super-resolution Reconstruction processing method |
CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
Non-Patent Citations (3)
Title |
---|
徐国明: "基于稀疏表示的图像超分辨率重建方法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
陈佳文: "基于稀疏高频双字典的磁共振成像超分辨率重建法", 《电子技术与软件工程》 * |
韩玉兵 等: "基于MG-GMRES算法的图像超分辨率重建", 《计算机学报》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108532700B (en) * | 2018-03-02 | 2019-06-21 | 广州医谷生物医疗科技有限公司 | The hospital's self-stripping system avoided infection |
CN108532700A (en) * | 2018-03-02 | 2018-09-14 | 李军 | The hospital's self-stripping system avoided infection |
CN108764368B (en) * | 2018-06-07 | 2021-11-30 | 西安邮电大学 | Image super-resolution reconstruction method based on matrix mapping |
CN108764368A (en) * | 2018-06-07 | 2018-11-06 | 西安邮电大学 | A kind of image super-resolution rebuilding method based on matrix mapping |
CN109166093A (en) * | 2018-07-09 | 2019-01-08 | 西北大学 | A kind of detection method for image salient region |
CN109166093B (en) * | 2018-07-09 | 2020-09-01 | 西北大学 | Image salient region detection method |
CN109145738A (en) * | 2018-07-18 | 2019-01-04 | 浙江工业大学 | The dynamic video dividing method of beam low-rank representation is weighed about based on the non-convex regularization of weighting and iteration |
CN109145738B (en) * | 2018-07-18 | 2021-07-27 | 浙江工业大学 | Dynamic video segmentation method based on weighted non-convex regularization and iterative re-constrained low-rank representation |
CN109447905A (en) * | 2018-11-06 | 2019-03-08 | 大连海事大学 | Based on the marine image super-resolution rebuilding method for differentiating dictionary |
CN109447905B (en) * | 2018-11-06 | 2022-11-18 | 大连海事大学 | Maritime image super-resolution reconstruction method based on discrimination dictionary |
CN110619603A (en) * | 2019-08-29 | 2019-12-27 | 浙江师范大学 | Single image super-resolution method for optimizing sparse coefficient |
CN110619603B (en) * | 2019-08-29 | 2023-11-10 | 浙江师范大学 | Single image super-resolution method for optimizing sparse coefficient |
CN113344829A (en) * | 2021-04-21 | 2021-09-03 | 复旦大学 | Portable ultrasonic video optimization reconstruction method for multi-channel generation countermeasure network |
Also Published As
Publication number | Publication date |
---|---|
CN107845065B (en) | 2021-06-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107845065A (en) | Super-resolution image reconstruction method and device | |
Armanious et al. | Unsupervised medical image translation using cycle-MedGAN | |
CN107369160B (en) | Choroid neogenesis blood vessel segmentation algorithm in OCT image | |
CN108734659B (en) | Sub-pixel convolution image super-resolution reconstruction method based on multi-scale label | |
CN105046672B (en) | A kind of image super-resolution rebuilding method | |
CN112465827A (en) | Contour perception multi-organ segmentation network construction method based on class-by-class convolution operation | |
CN106204447A (en) | The super resolution ratio reconstruction method with convolutional neural networks is divided based on total variance | |
CN109035172B (en) | Non-local mean ultrasonic image denoising method based on deep learning | |
CN106952228A (en) | The super resolution ratio reconstruction method of single image based on the non local self-similarity of image | |
CN108932536A (en) | Human face posture method for reconstructing based on deep neural network | |
CN107766874B (en) | Measuring method and measuring system for ultrasonic volume biological parameters | |
CN107146228A (en) | A kind of super voxel generation method of brain magnetic resonance image based on priori | |
CN109214989A (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
CN102915527A (en) | Face image super-resolution reconstruction method based on morphological component analysis | |
CN110363719A (en) | A kind of cell layered image processing method and system | |
CN108280804A (en) | A kind of multi-frame image super-resolution reconstruction method | |
CN111899165A (en) | Multi-task image reconstruction convolution network model based on functional module | |
CN116682120A (en) | Multilingual mosaic image text recognition method based on deep learning | |
CN110853048A (en) | MRI image segmentation method, device and storage medium based on rough training and fine training | |
CN114897780A (en) | MIP sequence-based mesenteric artery blood vessel reconstruction method | |
CN106169174A (en) | A kind of image magnification method | |
CN114266939A (en) | Brain extraction method based on ResTLU-Net model | |
CN115147600A (en) | GBM multi-mode MR image segmentation method based on classifier weight converter | |
CN110211193A (en) | Three dimensional CT interlayer image interpolation reparation and super-resolution processing method and device | |
CN115661459A (en) | 2D mean teacher model using difference information |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210615 Termination date: 20210915 |