CN107292821B - A kind of super-resolution image reconstruction method and system - Google Patents
A kind of super-resolution image reconstruction method and system Download PDFInfo
- Publication number
- CN107292821B CN107292821B CN201710488490.8A CN201710488490A CN107292821B CN 107292821 B CN107292821 B CN 107292821B CN 201710488490 A CN201710488490 A CN 201710488490A CN 107292821 B CN107292821 B CN 107292821B
- Authority
- CN
- China
- Prior art keywords
- image
- image block
- block
- dictionary
- classification
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000012360 testing method Methods 0.000 claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 abstract description 7
- 238000011160 research Methods 0.000 description 3
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000009897 systematic effect Effects 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/513—Sparse representations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of super-resolution image reconstruction method and system, carries out including all pixels to training image in training Image DatabaseKMean value classification, records cluster centre；Image block is carried out to each trained image, is classified based on cluster centre according to image block of the minimal distance principle to all trained images, is recorded each image block generic, seek the joint dictionary of each classification respectively；Piecemeal is carried out to low point of testing image to be reconstructed, and calculates each image block at a distance from each cluster centre, is classified according to minimal distance principle, the joint dictionary of adaptively selected respective classes carries out sparse modeling, and solution obtains sparse coefficient；Using joint dictionary and sparse coefficient, each corresponding high-definition picture block of image block in low point of testing image is rebuild, low point of testing image of whole picture optimal reconstructed image accordingly is obtained.The present invention can adaptively train joint dictionary, and suitable dictionary is adaptive selected and carries out sparse modeling, and reconstruction precision is more preferable.
Description
Technical field
The invention belongs to computer vision fields, in particular to a kind of to be built by training joint dictionary and adaptive sparse
Mould carries out the technical solution of super-resolution rebuilding to image.
Background technique
Super-resolution rebuilding technology (SR) refers to using certain method, and a width or several low resolution imageries are reconstructed into
High resolution image with more high pixel density and detailed information.It is sent out in fields such as biomedicine, remote sensing monitoring, public safeties
Wave highly important effect.
The expression of signal sparsity is all the time all of interest for researcher, and such as classical image compression algorithm JPEG's is basic
Principle is exactly that the sparsity using image in DCT domain compresses it.Candes in 2006 et al. proposes compressed sensing reason
After carrying out systematic analysis by and to it, carrying out analysis with application using the sparsity of signal is even more to become research hotspot.
The super-resolution rebuilding algorithm of image is always researchers' concern.High resolution image by deformation,
Degrade after fuzzy, sampling and addition system noise for low resolution image.It is how Converse solved, it restores or constructs more
Add the emphasis that clearly image is studied as everybody.The excessively complete rarefaction representation of image becomes the expression of mainstream image in recent years
It is also therefore constant to carry out image denoising and the research of super-resolution rebuilding using image sparse for model.From 2008
Yang et al. is inspired from compressed sensing, and after carrying out oversubscription reconstruction using sparse coding, the research of this method is persistently risen
Temperature.This method sets up the corresponding relationship of high-resolution and low-resolution image block by dictionary and sparse coefficient, first from image training library
It is middle to obtain corresponding high-resolution and low-resolution image block, then with characteristic symbol searching algorithm (Feature-Sign Search,
FSS) solve to solve l when dictionary and sparse coefficient1Constrained optimization problem, and pass through joint training high-resolution and low-resolution dictionary pair,
So that LR image block and HR the image block sparse coefficient having the same in corresponding dictionary.Pass through LR image block and LR when reconstruct
Dictionary acquires sparse coefficient, and HR dictionary and the sparse coefficient is recycled to acquire corresponding HR image block.It is classical based on sparse table
It is insufficient that the super-resolution rebuilding reached has both sides: 1. using FSS linear programming training dictionary, operand is larger；2. training
Image block is distinguished only with Gradient Features when dictionary, dictionary indicates that ability is limited.
Summary of the invention
For the deficiency of existing method, invent a kind of super based on sparse expression and adaptive multiresolution joint dictionary
Resolution reconstruction method and system, effectively making up original method dictionary ability indicates insufficient disadvantage, improves Image Super-resolution
Rate reconstruction precision.
To achieve the above object, technical solution of the present invention provides a kind of super-resolution image reconstruction method, including following
Step,
Step a carries out the classification of K mean value to all pixels of training image in training Image Database, records cluster centre；To each
Training image carries out image block, is divided based on cluster centre according to image block of the minimal distance principle to all trained images
Class records each image block generic；
Step b, according to the image block classification of step a as a result, seeking each classification C respectivelykJoint dictionary { Dh,Dl},k
=1 ..., K, wherein DhIt is high score dictionary, DlFor low point of dictionary；
Step c carries out piecemeal to low point of testing image y to be reconstructed, and calculates each image block and each cluster centre
Distance is classified according to minimal distance principle, and the joint dictionary of adaptively selected respective classes carries out sparse modeling, is solved
To sparse coefficient α*；
Step d utilizes joint dictionary and sparse coefficient α*, rebuild each image block y in low point of testing image yiIt is corresponding high
Image in different resolution block Obtain low point of testing image y of whole picture optimal reconstructed image accordingly.
Moreover, minimal distance principle is realized using Euclidean distance.
Moreover, in step c, if certain image block yiBelong to classification Ck, select corresponding kth class joint dictionary { Dh,Dl, it presses
Sparse expression is carried out according to following formula,
Wherein, α is sparse coefficient, vectorVectorβ is parameter preset, and matrix P is to extract
The matrix of overlapping region between current image block and the high-definition picture block previously rebuild, ω be current image block and previously
The pixel value of overlapping region on the high-definition picture block of reconstruction, F are Linear feature extraction operators.
The present invention also provides a kind of super-resolution image reconstruction systems, comprise the following modules,
First module carries out the classification of K mean value for all pixels to training image in training Image Database, records in cluster
The heart；Image block is carried out to each trained image, based on cluster centre according to minimal distance principle to the image of all trained images
Block is classified, and each image block generic is recorded；
Second module, for the image block classification according to step a as a result, seeking each classification C respectivelykJoint dictionary
{Dh,Dl, k=1 ..., K, wherein DhIt is high score dictionary, DlFor low point of dictionary；
Third module for carrying out piecemeal to low point of testing image y to be reconstructed, and calculates each image block and each cluster
The distance at center, classifies according to minimal distance principle, and the joint dictionary of adaptively selected respective classes carries out sparse modeling,
Solution obtains sparse coefficient α*；
4th module, for utilizing joint dictionary and sparse coefficient α*, rebuild each image block y in low point of testing image yi
Corresponding high-definition picture block Obtain low point of testing image y of whole picture optimal reconstructed image accordingly.
Moreover, minimal distance principle is realized using Euclidean distance.
Moreover, in third module, if certain image block yiBelong to classification Ck, select corresponding kth class joint dictionary { Dh,Dl,
Sparse expression is carried out according to the following equation,
Wherein, α is sparse coefficient, vectorVectorβ is parameter preset, and matrix P is to extract
The matrix of overlapping region between current image block and the high-definition picture block previously rebuild, ω be current image block and previously
The pixel value of overlapping region on the high-definition picture block of reconstruction, F are Linear feature extraction operators.
The shortcomings that indicating scarce capacity for dictionary in the prior art, the present invention utilize the training image in training Image Database
All pixels classify, then determine each image block classification, the adaptive joint dictionary that constructs is improved, it would be desirable to preferably
Image is indicated.Compared to existing method, the present invention can adaptively train joint dictionary, and can be according to image
The essential characteristic of block is adaptive selected suitable dictionary and carries out sparse modeling, and precision is more preferable, can be in medical image, satellite mapping
The application fields such as picture and video are promoted the use of, and have important market value.
Detailed description of the invention
Fig. 1 is the flow chart of the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions and advantages of the present invention clearer, With reference to embodiment and join
According to attached drawing, the present invention is described in more detail.It should be understood that these descriptions are merely illustrative, and it is not intended to limit this hair
Bright range.
Such as Fig. 1, the embodiment of the present invention proposes to carry out image in the joint dictionary of the adaptive multiresolution based on sparse expression
Super-resolution rebuilding, first training image library is carried out adaptively to combine dictionary training, then sparse build fuzzy image progress
Mould obtains reconstructed image after sparse coefficient and exports.Detailed process is as follows:
Step a carries out the classification of K mean value to all pixels of the training image in training Image Database, records cluster centre, right
Training image carries out piecemeal classification；
In embodiment, there are several high-resolution trained images in training Image Database, read the pixel of all trained images
And K cluster centre is sought using K mean algorithm, and image block is carried out to training image, based on cluster centre according to minimum
Distance principle classifies to the image block of all trained images, records each image block generic of every trained image.
When it is implemented, user can voluntarily preset the value of K, and divide the dimensional parameters of image block.Preferably, image block uses
There is a division of overlapping, i.e., having several pixels between two adjacent pieces is overlapping, such as each image block size is 40 × 40,
5 pixels are overlapped between adjacent image block.
For the sake of ease of implementation, it is as follows to provide K mean algorithm:
1) K pixel value is randomly selected as cluster centre from all pixels of N training images；
2) to remaining each pixel, its distance for arriving each mass center is calculated, and it is grouped into the class of nearest mass center；
3) cluster centre of obtained each class is recalculated；
4) iteration 2)~3) up to new cluster centre is equal with former cluster centre or is less than preset specified threshold, algorithm
Terminate.
In the step a of embodiment, if zn, n=1 ..., N is in training image library to training image.It is calculated using K mean value
Method is to pixel x in all high-resolution trained imagesqIt is clustered, q=1 ..., Q, Q is that the pixel of all trained images is total
Number.K classification is generated, if certain classification is denoted asWherein CkFor kth class,For classification CkIn t-th of pixel samples, qkFor classification CkIn number of pixels, meet
Calculate CkClass centerAnd radius
To certain image block on any training image, compare each class center μkDistance, according to minimum value confirm institute
Belong to classification.When it is implemented, distance metric mode, such as Euclidean distance, mahalanobis distance etc. may be selected in user.Embodiment is preferentially adopted
Use Euclidean distance.
Step b, according to the image block classification of step a as a result, carrying out the training of joint dictionary by class；
Every trained image has corresponding low point of image, divides and train image on low point of image as high score image
On the one-to-one low-resolution image block of high-definition picture block.
According to belonging to C of all categorieskHigh-definition picture block and corresponding low-resolution image block, seek each classification respectively
CkJoint dictionary { Dh,Dl, k=1 ..., K.Wherein, DhIt is high score dictionary, DlFor low point of dictionary.
In the step b of embodiment, for different classes of Ck, solve and obtain high-definition picture block and low-resolution image block
Between joint dictionary { Dh,Dl, solution formula is as follows:
In formula, Z indicates sparse coefficient matrix, XhIt is high-definition picture block, YlIt is corresponding low-resolution image block.N and
M is the dimension that high-resolution and low-resolution image block are launched into one-dimensional vector form respectively, and when specific implementation can use upper
Arrive down the modes unfolded image block such as from left to right.Parameter lambda is used for the sparsity of equilibrium solution, desirable empirical value when specific implementation.
Step c carries out piecemeal to low point of testing image y to be reconstructed, and calculates each image block yiWith each cluster centre
Distance, according to minimal distance principle to image block yiClassify, the joint dictionary of respective classes selected to carry out sparse modeling,
Solve sparse coefficient；
In the step c of embodiment, piecemeal is carried out to low point of testing image y, asks image block to each cluster centre μ respectivelyk's
Euclidean distance, and according to minimum distance classification: i.e. for certain image block y in low point of testing image yi, affiliated class (dkRepresent current image block to k-th of cluster centre distance,Expression takes the smallest dk)。
Minimal distance principle is consistent with step a in embodiment, is realized using Euclidean distance.To low point of test shadow to be reconstructed
As the mode of y progress piecemeal, low point of image block mode corresponding with training image is consistent.
Select corresponding kth class joint dictionary { Dh,Dl, sparse expression is carried out according to the following equation:
Wherein, α is sparse coefficient, vectorVectorβ is to keep high-definition picture
Block- matching low resolution inputs the parameter preset the consistency between neighborhood, and β takes empirical value (such as part document order in embodiment
β=1).Matrix P is the matrix for extracting the overlapping region between current image block and the high-definition picture block previously rebuild, ω
Contain current image block and the pixel value of the overlapping region on the high-definition picture block previously rebuild.F is that linear character mentions
Take operator.Thus optimal sparse coefficient is obtained, α is denoted as*。
Step d utilizes joint dictionary and sparse coefficient α*Rebuild each image block y in low point of testing image yiIt is corresponding high
Image in different resolution block Corresponding high-resolution is put into after each image block in low point of testing image y is rebuild respectively
The corresponding position of image takes lap the mode averaged, and thus low point of testing image y of exportable whole picture is corresponding
Optimal reconstructed image.
When it is implemented, method provided by the present invention can realize automatic running process based on software technology, mould can also be used
Block mode realizes corresponding system.
The embodiment of the present invention also provides a kind of super-resolution image reconstruction system, comprises the following modules,
First module carries out the classification of K mean value for all pixels to training image in training Image Database, records in cluster
The heart；Image block is carried out to each trained image, based on cluster centre according to minimal distance principle to the image of all trained images
Block is classified, and each image block generic is recorded；
Second module, for the image block classification according to step a as a result, seeking each classification C respectivelykJoint dictionary
{Dh,Dl, k=1 ..., K, wherein DhIt is high score dictionary, DlFor low point of dictionary；
Third module for carrying out piecemeal to low point of testing image y to be reconstructed, and calculates each image block and each cluster
The distance at center, classifies according to minimal distance principle, and the joint dictionary of adaptively selected respective classes carries out sparse modeling,
Solution obtains sparse coefficient α*；
4th module, for utilizing joint dictionary and sparse coefficient α*, rebuild each image block y in low point of testing image yi
Corresponding high-definition picture block Obtain low point of testing image y of whole picture optimal reconstructed image accordingly.
Each module specific implementation can be found in corresponding steps, and it will not go into details by the present invention.
In conclusion the present invention adaptively constructs multiple joint dictionaries by trained Image Database before this, then to image
Piecemeal processing is carried out, each piece of image block is all adaptive selected corresponding dictionary and carries out sparse reconstruction, exports high score image.It should
Technical solution combines the advantages of joint dictionary and adaptively selected multiword allusion quotation, is the guarantor for reconstructing higher-quality image
Card.
Claims (4)
1. a kind of super-resolution image reconstruction method, it is characterised in that: include the following steps,
Step a carries out the classification of K mean value to all pixels of training image in training Image Database, records cluster centre；To each training
Image carries out image block, is classified based on cluster centre according to image block of the minimal distance principle to all trained images,
Record each image block generic；
Step b, according to the image block classification of step a as a result, seeking each classification C respectivelykJoint dictionary { Dh, Dl, k=
1 ..., K, wherein DhIt is high score dictionary, DlFor low point of dictionary；
Step c carries out piecemeal to low point of testing image y to be reconstructed, and calculates each image block at a distance from each cluster centre,
Classify according to minimal distance principle, the joint dictionary of adaptively selected respective classes carries out sparse modeling, and solution obtains dilute
Sparse coefficient α*, realization is as follows,
If certain image block yiBelong to classification Ck, select corresponding kth class joint dictionary { Dh, Dl, it carries out according to the following equation sparse
Expression,
Wherein, α is sparse coefficient, vectorVectorβ is parameter preset, and matrix P is to extract currently
The matrix of overlapping region between image block and the high-definition picture block previously rebuild, ω are current image block and previously rebuild
High-definition picture block on overlapping region pixel value, F is Linear feature extraction operator, and parameter lambda is dilute for equilibrium solution
Dredge property；
Step d utilizes joint dictionary and sparse coefficient α*, rebuild each image block y in low point of testing image yiCorresponding high-resolution
Rate image block Obtain low point of testing image y of whole picture optimal reconstructed image accordingly.
2. super-resolution image reconstruction method according to claim 1, it is characterised in that: minimal distance principle using it is European away from
From realization.
3. a kind of super-resolution image reconstruction system, it is characterised in that: it comprises the following modules,
First module carries out the classification of K mean value for all pixels to training image in training Image Database, records cluster centre；
Image block is carried out to each trained image, based on cluster centre according to minimal distance principle to the image blocks of all trained images into
Row classification, records each image block generic；
Second module, for the image block classification according to step a as a result, seeking each classification C respectivelykJoint dictionary { Dh,
Dl, k=1 ..., K, wherein DhIt is high score dictionary, DlFor low point of dictionary；
Third module for carrying out piecemeal to low point of testing image y to be reconstructed, and calculates each image block and each cluster centre
Distance, classify according to minimal distance principle, the joint dictionaries of adaptively selected respective classes carries out sparse modeling, solves
Obtain sparse coefficient α*, realization is as follows,
If certain image block yiBelong to classification Ck, select corresponding kth class joint dictionary { Dh, Dl, it carries out according to the following equation sparse
Expression,
Wherein, α is sparse coefficient, vectorVectorβ is parameter preset, and matrix P is to extract currently
The matrix of overlapping region between image block and the high-definition picture block previously rebuild, ω are current image block and previously rebuild
High-definition picture block on overlapping region pixel value, F is Linear feature extraction operator, and parameter lambda is dilute for equilibrium solution
Dredge property；
4th module, for utilizing joint dictionary and sparse coefficient α*, rebuild each image block y in low point of testing image yiAccordingly
High-definition picture block Obtain low point of testing image y of whole picture optimal reconstructed image accordingly.
4. super-resolution image reconstruction system according to claim 3, it is characterised in that: minimal distance principle using it is European away from
From realization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710488490.8A CN107292821B (en) | 2017-06-23 | 2017-06-23 | A kind of super-resolution image reconstruction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710488490.8A CN107292821B (en) | 2017-06-23 | 2017-06-23 | A kind of super-resolution image reconstruction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107292821A CN107292821A (en) | 2017-10-24 |
| CN107292821B true CN107292821B (en) | 2019-09-10 |
Family
ID=60098769
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710488490.8A Active CN107292821B (en) | 2017-06-23 | 2017-06-23 | A kind of super-resolution image reconstruction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107292821B (en) |
Families Citing this family (4)
| 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 |
| CN109886900B (en) * | 2019-03-15 | 2023-04-28 | 西北大学 | Synthetic rain map rain removing method based on dictionary training and sparse representation |
| CN110826467B (en) * | 2019-11-22 | 2023-09-29 | 中南大学湘雅三医院 | Electron microscope image reconstruction system and method thereof |
| CN111260558B (en) * | 2020-01-22 | 2021-10-22 | 武汉大学 | Image super-resolution network model with variable magnification |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902961A (en) * | 2012-09-21 | 2013-01-30 | 武汉大学 | Face super-resolution processing method based on K neighbor sparse coding average value constraint |
| CN103150713A (en) * | 2013-01-29 | 2013-06-12 | 南京理工大学 | Image super-resolution method of utilizing image block classification sparse representation and self-adaptive aggregation |
| CN106204667A (en) * | 2016-07-01 | 2016-12-07 | 山东大学 | A kind of similarity solved in image super-resolution rebuilding retains the sparse coding method of problem |
-
2017
- 2017-06-23 CN CN201710488490.8A patent/CN107292821B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102902961A (en) * | 2012-09-21 | 2013-01-30 | 武汉大学 | Face super-resolution processing method based on K neighbor sparse coding average value constraint |
| CN103150713A (en) * | 2013-01-29 | 2013-06-12 | 南京理工大学 | Image super-resolution method of utilizing image block classification sparse representation and self-adaptive aggregation |
| CN106204667A (en) * | 2016-07-01 | 2016-12-07 | 山东大学 | A kind of similarity solved in image super-resolution rebuilding retains the sparse coding method of problem |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107292821A (en) | 2017-10-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292821B (en) | A kind of super-resolution image reconstruction method and system | |
| CN109255755A (en) | Image super-resolution rebuilding method based on multiple row convolutional neural networks | |
| CN101630405B (en) | Multi-focusing image fusion method utilizing core Fisher classification and redundant wavelet transformation | |
| Aghamaleki et al. | Multi-stream CNN for facial expression recognition in limited training data | |
| Hara et al. | Towards good practice for action recognition with spatiotemporal 3d convolutions | |
| CN106097253B (en) | A kind of single image super resolution ratio reconstruction method based on block rotation and clarity | |
| CN104077742B (en) | Human face sketch synthetic method and system based on Gabor characteristic | |
| Sheng et al. | Second-order ResU-Net for automatic MRI brain tumor segmentation | |
| Shi et al. | Exploiting multi-scale parallel self-attention and local variation via dual-branch transformer-CNN structure for face super-resolution | |
| Gu et al. | Fast image super resolution via local regression | |
| CN112950480A (en) | Super-resolution reconstruction method integrating multiple receptive fields and dense residual attention | |
| CN103020940B (en) | Local feature transformation based face super-resolution reconstruction method | |
| Zhao et al. | Deep equilibrium models for video snapshot compressive imaging | |
| CN108062779B (en) | Lossless compression method for breast tomography radiographic image | |
| Ma et al. | Hyper-convolution networks for biomedical image segmentation | |
| CN104636711A (en) | Facial emotion recognition method based on local sparse representation classifier | |
| Zhang et al. | Hepatocellular carcinoma histopathological images grading with a novel attention-sharing hybrid network based on multi-feature fusion | |
| Razali et al. | High-level features in deeper deep learning layers for breast cancer classification | |
| CN112634308A (en) | Nasopharyngeal carcinoma target area and endangered organ delineation method based on different receptive fields | |
| CN116469172A (en) | Bone behavior recognition video frame extraction method and system under multiple time scales | |
| CN104574320B (en) | A kind of image super-resolution restored method based on sparse coding coefficients match | |
| Yang et al. | A breast ultrasound tumor detection framework using convolutional neural networks | |
| CN115272739A (en) | Weakly supervised cervical carcinoma pre-lesion grading method based on self-attention mechanism | |
| CN110633385B (en) | Medical image retrieval and compression method | |
| Wang et al. | Lightweight non-local network for image super-resolution |
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 |