CN109886869A - A kind of unreal structure method of face of the non-linear expansion based on contextual information - Google Patents
A kind of unreal structure method of face of the non-linear expansion based on contextual information Download PDFInfo
- Publication number
- CN109886869A CN109886869A CN201811199243.7A CN201811199243A CN109886869A CN 109886869 A CN109886869 A CN 109886869A CN 201811199243 A CN201811199243 A CN 201811199243A CN 109886869 A CN109886869 A CN 109886869A
- Authority
- CN
- China
- Prior art keywords
- resolution
- low
- image
- dictionary
- space
- 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 42
- 238000012360 testing method Methods 0.000 claims abstract description 28
- 230000001815 facial effect Effects 0.000 claims abstract description 3
- 238000012549 training Methods 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000005070 sampling Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 2
- 238000013507 mapping Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 15
- 238000013135 deep learning Methods 0.000 description 5
- 238000013459 approach Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008929 regeneration Effects 0.000 description 1
- 238000011069 regeneration method Methods 0.000 description 1
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The face super-resolution method for the non-linear expansion based on contextual information that the invention discloses a kind of, this method passes through context block first and is sampled to contextual information, to enrich the prior information that facial image indicates, and dimensionality reduction is carried out to context dictionary using setting threshold value in regularization objective function, then nuclear space is converted by initial data using gaussian kernel function, by the non-linear relation for indicating to establish between high-low resolution image that cooperates, testing image is finally reconstructed using the study of context residual error.This method establishes the Nonlinear Mapping between high-low resolution image by gaussian kernel function, and the nonlinear problem in high-dimensional feature space is expressed as linear problem.In addition, it also learns to obtain the prior information that more accurate image indicates using context residual error, the performance of reconstruction is improved.
Description
Technical Field
The invention relates to an image recognition technology, in particular to a non-linear expansion human face illusion method based on context information.
Background
Super-resolution plays an important role in various practical applications, such as remote sensing, medical imaging and video monitoring. The face illusion is a typical super-resolution algorithm, which is to recover a High-resolution (HR) image from a single or multiple Low-resolution (LR) images.
From how the mapping function is modeled, face super-resolution algorithms can be divided into two categories, linear methods and nonlinear methods.
Linear methods assume that each input image can be represented by a linear combination of dictionary atoms, or directly using a linear regression of the LR and HR relationships. Wang et al propose a global linear model to represent LR images in the characteristic face space. While linear methods are simple and effective, linear assumptions limit the expressive power of a priori information in training data. The nonlinear method uses a virtual nonlinear method to model the LR and HR relationships to overcome the limitations of the linear method. Many super-resolution algorithms using non-linear methods achieve good results. Recently, deep learning provides an end-to-end learning model for super resolution tasks. The deep network structure describes the image characteristics through a nonlinear method. Dong et al first proposed a convolutional neural network that uses nonlinear mapping for super resolution. Kim et al accurately represent images with recursive sub-network elements through a deep residual network. Ledig et al take advantage of the fidelity with which confrontation generates network rendered images.
The human face illusion method achieves a good reconstruction effect. These methods, however, have two disadvantages, first, the above method prioritizes the position information in the reconstruction, ignoring the context information in the image and the non-linear nature of the imaging. Second, deep learning based methods have non-linear representation capabilities, but training the network is hardware dependent (GPU) and very time consuming. Based on the enlightenment of the context information block, a simple and effective nonlinear expansion method of the context information is provided to obtain better reconstruction performance. The raw data is extended to a high-dimensional kernel space by a gaussian kernel function, and then the context information is represented using a collaborative expression constraint. Finally, the HR image is reconstructed in the residual domain.
The non-linear method proposed by us is easy to implement and has better performance than some deep learning based methods. Accurate high frequency information is explored by describing a complex relationship between LR and HR images. The non-linear expansion of context information we propose is a non-linear representation method different from deep learning. In this context, compared to the location block based approach, we propose a method that can provide more non-local information using context blocks and also have better representation capability in the residual domain than in the pixel domain.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a face illusion method based on the nonlinear expansion of context information aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a non-linear expansion human face illusion method based on context information comprises the following steps:
s1, obtaining a residual dictionary according to the high-resolution face image in the training set: carrying out fuzzy downsampling on the high-resolution face image in the training set to obtain a corresponding low-resolution face image, then interpolating the low-resolution face image to the same size as the original high-resolution face image, and carrying out interpolation on the high-resolution face image and the high-resolution face imageThe low-resolution face image overlaps and blocks the context information through the context block to form a corresponding context HR dictionary
And context LR dictionaryWherein N represents the number of training samples; we define the size of a context block(Is an integer) is centered in this block at a large window size ω × ω. In this large window we sample multiple blocks using step e, the number of context blocks c can be determined by the window size ω, block sizeAnd step e denotes:
then subtracting the low-resolution dictionary from the high-resolution dictionary to obtain a residual dictionary;
s2, converting the low resolution dictionary to kernel space by using Gaussian kernel function to obtain low resolution dictionary in kernel space(expression dictionary);
s3, interpolating the low-resolution test face image in the test set to the size same as that of the high-resolution face image, then taking the block of the interpolated low-resolution test face image, and converting the block into a kernel space by using a Gaussian kernel function so as to keep the test image and the training sample in the same space;
step S2 refers toIn (1),that is, the training samples are combined by blocks and thenThat is, in the transform space, the transform is performed in the same manner. The kernel space is also referred to as the nonlinear space.
S4, for the interpolated low-resolution test face image, using cooperative expression and setting threshold to solve the optimal expression coefficient matrix in the low-resolution space;
s5, according to the assumption of manifold consistency, keeping the low-resolution collaborative expression coefficient in a high-resolution space, namely keeping the expression coefficients of the high-resolution space and the low-resolution space the same, and obtaining a weight coefficient matrix during reconstruction;
according to the linear divisible condition, the data is divided into a linear space and a nonlinear space (kernel space); according to manifold learning analysis, dividing a picture into a high-resolution space and a low-resolution space;
s6, performing linear combination by using the reconstruction coefficient matrix obtained in the step S5 and the residual dictionary obtained in the step S1, and predicting residual images of the low-resolution test face images in the test set;
and S7, adding the interpolated low-resolution test face image and the residual image obtained in the step S6 to obtain a final reconstructed high-resolution face image.
According to the above scheme, the expression coefficients of the low-resolution image in step S4 are expressed as follows: for an input image block yiExpression coefficient of low-resolution image:
αi＝(G+λI)-1f(·,yi)；
wherein, f (·, y)i)＝[f(l1,yi),…,f(lK,yi)]TRepresenting a low resolution dictionary by kernel function building a non-linear relationship between test samples and an expression dictionaryK represents the number of dictionary atoms, lambda is a balance parameter of nonlinear sparse expression, G represents a Gram matrix, and I is an identity matrix.
According to the above scheme, the LR dictionary is determined by a threshold K in step S4 using the following formulaTo find the expression coefficient of the low resolution image:
wherein,for the context LR dictionary, λ is a balanced parameter of non-linear sparse representation, αi[j]Is the expression coefficient αiThe jth weight coefficient of (C)K(yi) Denotes from yiNeighborhood composed of nearest K dictionary atoms, corresponding reconstructed dictionaryContext HR dictionaries may be indexedTo obtain the compound.
According to the scheme, the method for keeping the test image and the training sample in the same space in the step S3) is as follows:
each input low resolution image blockAfter interpolation to the same size as the high-resolution image, the image becomesProjection into a low-dimensional embedding space (nonlinear space) using a projection matrix, or using a sample z as a decomposed training sample, by a kernel function:linear low-dimensional spatial low-resolution image block yiBy Gaussian kernel function conversion
The invention has the following beneficial effects: the invention provides a non-linear expansion human face illusion method based on context information, and the non-linear method is easy to operate and superior to some methods based on deep learning in performance. The context residual learning method has proven to have better reconstruction capabilities than the location block based method. The enhanced performance comes from contextual information that always contains more non-local information and residual learning, which is always better expressive than the pixel domain approach.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention;
fig. 2 is a graph comparing the experimental results of the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, a method for a non-linear expansion of human face illusion based on context information includes the following steps:
step 1, a low-resolution training library comprises low-resolution face sample images, and a high-resolution training library comprises high-resolution face sample images. The low-resolution face sample image is a corresponding low-resolution face image obtained by blurring and down-sampling a high-resolution face image, the low-resolution face image is interpolated to the same size as the high-resolution face image, context information of the high-resolution face image and the low-resolution face image is overlapped through context blocks to obtain blocks, a corresponding high-resolution dictionary and a corresponding low-resolution dictionary are formed, and then the low-resolution dictionary is subtracted from the high-resolution dictionary to obtain a residual dictionary.
In this example, the CAS-PEAL-R1 facial database was used, and 1000 pictures were selected as training samples, and the remaining 40 pictures were tested. The size of the HR image is 128 × 112 pixels. An LR image sampled from the corresponding HR image (scale factor t is 4) is formed by blurring (the kernel of blurring is 4 pixels), and thus the size of the LR face image is 32 × 28 pixels.
In an example, inWhen training a sample face image, sampling context information of a low-resolution face image processed by interpolation through a context block, wherein the size of the context block is defined(Is an integer) is centered in this block over a large window size w x w. In this large window we sample multiple blocks using step e, the number of context blocks c can be scaled by window size ω, block sizeAnd step length e:
in the example, the high-resolution image is also overlapped, block-sampling is carried out to obtain a corresponding context information block, then a high-resolution dictionary is formed, and each input image blockWith corresponding HR and LR dictionaries being
And then obtaining a residual dictionary by the following formula:
and 2, converting the low-resolution dictionary into a kernel space by using a Gaussian kernel function to obtain the low-resolution dictionary (expression dictionary) in the kernel space. Defining kernel functionsMapping Euclidean spaceFor Hilbert space F, F is a Hilbert space (RKHS) conforming to a Mercer kernel function F (·), commonly referred to as a regeneration kernel. Given two data pointsWe haveIs the inner product of the kernel feature space F, using the best-known nonlinear kernel function gaussian kernel function:
f(yi,lj)＝exp(-τ||yi-lj||2) (3)
in the above formula, τ is a scalar parameter and ljIs LR dictionary atom in correspondence In d-dimension kernel space F, mapping a database refers to
And 3, blurring the low-resolution test face image, interpolating to the size same as that of the high-resolution face image, taking blocks of the low-resolution test face image, and converting the blocks into a kernel space by using a Gaussian kernel function so as to keep the test image and the training sample in the same space. Each input low resolution image blockAfter interpolation to the same size as the high-resolution image, the image becomesSince d can be very high, it is possible to,we can use the projection matrix to a low-dimensional embedding space. Another approach is to use sample z as the decomposed training sample through a kernel function:linear low-dimensional spatial low-resolution image block yiBy Gaussian kernel function conversionDenote the Gram matrix by G, Gi,j＝f(li,lj)。
And 4, for the test sample with low resolution, calculating the optimal expression coefficient matrix in the low resolution space by using cooperative expression and a set threshold value. For input yiThe LR dictionary is determined by a threshold K using the following formulaTo find the expression coefficient of the low resolution image:
where λ is a parameter balancing the non-linear sparse representation, αi[j]Representative αiThe jth weight coefficient sum CK(yi) Denotes from yiNearest K dictionary atoms and corresponding reconstructed dictionaryCan be found in HR dictionaryIs used to determine the index of (1).
The expression coefficients for the low resolution space can be obtained by solving equation (4):
αi＝(G+λI)-1f(·,yi) (5)
step 5, according to the assumption of manifold consistency, keeping the low-resolution collaborative expression coefficient in the high-resolution space, namely keeping the expression coefficients of the high-resolution space and the low-resolution space the same, and obtaining the weight coefficient matrix during reconstruction
Step 6, reconstructing the residual dictionary and the weight coefficient matrix to obtain a residual face image; reconstructing a residual image by the following formula:
and 7, outputting the target high-resolution face image. Outputting the target high resolution image by the following formula:
whereinIs interpolated yi。
Test example we selected 1000 pictures as training samples in the casual-R1 face database, and the remaining 40 pictures were tested. The size of the high resolution image is 128 × 112 pixels. Each high resolution image is down sampled 4 times to obtain 32 x 28 low resolution images. Here we interpolate the low resolution image to the same size as the high resolution image. We set the image block size to 12 × 12 pixels with 4 pixels overlapping. All experiments were compared under the same conditions.
And (3) preparing data by using formulas (1) and (2), converting the space by using a formula (3), converting the low-resolution dictionary into a kernel space, and obtaining the low-resolution dictionary (expression dictionary) in the kernel space, wherein the dictionary learning is finished, and the reconstruction process is converted into the solution expression coefficient.
Then blurring the low-resolution test face image, interpolating to the size same as that of the high-resolution face image, then taking blocks, converting the blocks into a kernel space by using a Gaussian kernel function, keeping the test image and the training sample in the same space, solving an optimal expression coefficient matrix of the low-resolution space by using a formula (4), keeping the low-resolution collaborative expression coefficient in the high-resolution space according to the manifold consistency assumption, namely keeping the expression coefficients of the high-resolution space and the low-resolution space the same, and obtaining a weight coefficient matrix during reconstruction. And (3) reconstructing a residual image by using the formula (6), and outputting a target high-resolution image by using the formula (7).
The invention is different from other advanced super-resolution algorithms, and experimental comparison is provided below to illustrate the effectiveness of the method.
The image reconstruction performance is evaluated through PSNR and SSIM in the experiment as an algorithm standard. The results of the experiments are shown in the following table:
TABLE average PSNR and SSIM values for different methods
It is evident from the above table that both PSNR and SSIM are higher for the inventive algorithm than for the LSR, WSR, LLE, LCR, CLNE and TLCR, and the depth learning algorithm (SRCNN, VDSR).
Fig. 2 is a graph of experimental results of different algorithms. From left to right in sequence: LR input; LSR; WSR; LLE; LCR; CLNE; TLCR; SRCNN; VDSR; our and HR pictures.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (4)
1. A non-linear expansion human face illusion method based on context information is characterized by comprising the following steps:
s1, obtaining a residual dictionary according to the high-resolution face image in the training set: carrying out fuzzy down-sampling on the high-resolution face image in the training set to obtain a corresponding low-resolution face image, then interpolating the low-resolution face image to the same size as the original high-resolution face image, overlapping the context information of the high-resolution face image and the low-resolution face image through a context block, and taking a block to form a corresponding upper partHR dictionary from below
And context LR dictionaryWherein N represents the number of training samples;
then subtracting the low-resolution dictionary from the high-resolution dictionary to obtain a residual dictionary;
s2, converting the low resolution dictionary to kernel space by using Gaussian kernel function to obtain low resolution dictionary in kernel space
S3, interpolating the low-resolution test face image in the test set to the size same as that of the high-resolution face image, then taking the block of the interpolated low-resolution test face image, and converting the block into a kernel space by using a Gaussian kernel function so as to keep the test image and the training sample in the same space;
s4, for the interpolated low-resolution test face image, using cooperative expression and setting threshold to obtain the optimal expression coefficient matrix in the low-resolution space;
s5, according to the assumption of manifold consistency, keeping the low-resolution collaborative expression coefficient in a high-resolution space, namely keeping the expression coefficients of the high-resolution space and the low-resolution space the same, and obtaining a weight coefficient matrix during reconstruction;
s6, performing linear combination by using the reconstruction coefficient matrix obtained in the step S5 and the residual dictionary obtained in the step S1, and predicting residual images of the low-resolution test face images in the test set;
and S7, adding the interpolated low-resolution test face image and the residual image obtained in the step S6 to obtain a final reconstructed high-resolution face image.
2. The method for facial illusion of non-linear expansion based on context information as claimed in claim 1, wherein the expression coefficients of the low-resolution image in the step S4 are expressed as follows: for an input image block yiExpression coefficient of low-resolution image:
αi＝(G+λI)-1f(·,yi)；
wherein, f (·, y)i)＝[f(l1,yi),…,f(lK,yi)]TRepresenting a non-linear relationship between test samples and an expression dictionary established by a kernel function, a low resolution dictionaryK represents the number of dictionary atoms, lambda is a balance parameter of nonlinear sparse expression, G represents a Gram matrix,and I is an identity matrix.
3. The method of claim 1, wherein the LR dictionary determination in step S4 is determined by a threshold K according to the following formulaTo find the expression coefficient of the low resolution image:
wherein,for the context LR dictionary, λ is a balanced parameter of non-linear sparse representation, αi[j]Is the expression coefficient αiThe jth weight coefficient of (C)K(yi) Denotes from yiNeighborhood composed of nearest K dictionary atoms, corresponding reconstructed dictionaryContext HR dictionaries may be indexedTo obtain the compound.
4. The method for the non-linear expansion of the human face illusion based on the contextual information as claimed in claim 1, wherein the method for keeping the test image and the training sample in the same space in the step S3) is specifically as follows:
each input low resolution image blockAfter interpolation to the same size as the high-resolution image, the image becomesUsing the projection matrix to project into a low-dimensional embedding space, or using the sample z as a decomposed training sample, by a kernel function:linear low-dimensional spatial low-resolution image block yiBy Gaussian kernel function conversion
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811199243.7A CN109886869B (en) | 2018-10-15 | 2018-10-15 | Non-linear expansion face illusion method based on context information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811199243.7A CN109886869B (en) | 2018-10-15 | 2018-10-15 | Non-linear expansion face illusion method based on context information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886869A true CN109886869A (en) | 2019-06-14 |
| CN109886869B CN109886869B (en) | 2022-12-20 |
Family
ID=66924877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811199243.7A Active CN109886869B (en) | 2018-10-15 | 2018-10-15 | Non-linear expansion face illusion method based on context information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886869B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860356A (en) * | 2020-07-23 | 2020-10-30 | 中国电子科技集团公司第五十四研究所 | Polarization SAR image classification method based on nonlinear projection dictionary pair learning |
| CN111915693A (en) * | 2020-05-22 | 2020-11-10 | 中国科学院计算技术研究所 | Sketch-based face image generation method and system |
| CN112966554A (en) * | 2021-02-02 | 2021-06-15 | 重庆邮电大学 | Robust face recognition method and system based on local continuity |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550649A (en) * | 2015-12-09 | 2016-05-04 | 武汉工程大学 | Extremely low resolution human face recognition method and system based on unity coupling local constraint expression |
| CN107169928A (en) * | 2017-05-12 | 2017-09-15 | 武汉华大联创智能科技有限公司 | A kind of human face super-resolution algorithm for reconstructing learnt based on deep layer Linear Mapping |
| US9865036B1 (en) * | 2015-02-05 | 2018-01-09 | Pixelworks, Inc. | Image super resolution via spare representation of multi-class sequential and joint dictionaries |
| WO2018120329A1 (en) * | 2016-12-28 | 2018-07-05 | 深圳市华星光电技术有限公司 | Single-frame super-resolution reconstruction method and device based on sparse domain reconstruction |
-
2018
- 2018-10-15 CN CN201811199243.7A patent/CN109886869B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9865036B1 (en) * | 2015-02-05 | 2018-01-09 | Pixelworks, Inc. | Image super resolution via spare representation of multi-class sequential and joint dictionaries |
| CN105550649A (en) * | 2015-12-09 | 2016-05-04 | 武汉工程大学 | Extremely low resolution human face recognition method and system based on unity coupling local constraint expression |
| WO2018120329A1 (en) * | 2016-12-28 | 2018-07-05 | 深圳市华星光电技术有限公司 | Single-frame super-resolution reconstruction method and device based on sparse domain reconstruction |
| CN107169928A (en) * | 2017-05-12 | 2017-09-15 | 武汉华大联创智能科技有限公司 | A kind of human face super-resolution algorithm for reconstructing learnt based on deep layer Linear Mapping |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111915693A (en) * | 2020-05-22 | 2020-11-10 | 中国科学院计算技术研究所 | Sketch-based face image generation method and system |
| CN111915693B (en) * | 2020-05-22 | 2023-10-24 | 中国科学院计算技术研究所 | Sketch-based face image generation method and sketch-based face image generation system |
| CN111860356A (en) * | 2020-07-23 | 2020-10-30 | 中国电子科技集团公司第五十四研究所 | Polarization SAR image classification method based on nonlinear projection dictionary pair learning |
| CN111860356B (en) * | 2020-07-23 | 2022-07-01 | 中国电子科技集团公司第五十四研究所 | Polarization SAR image classification method based on nonlinear projection dictionary pair learning |
| CN112966554A (en) * | 2021-02-02 | 2021-06-15 | 重庆邮电大学 | Robust face recognition method and system based on local continuity |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886869B (en) | 2022-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109064396B (en) | Single image super-resolution reconstruction method based on deep component learning network | |
| US10593021B1 (en) | Motion deblurring using neural network architectures | |
| CN112801877B (en) | Super-resolution reconstruction method of video frame | |
| CN106204447A (en) | The super resolution ratio reconstruction method with convolutional neural networks is divided based on total variance | |
| CN103413286B (en) | United reestablishing method of high dynamic range and high-definition pictures based on learning | |
| CN105631807B (en) | The single-frame image super-resolution reconstruction method chosen based on sparse domain | |
| CN109886869B (en) | Non-linear expansion face illusion method based on context information | |
| CN106127688B (en) | A kind of super-resolution image reconstruction method and its system | |
| CN107067380B (en) | High-resolution image reconstruction method based on low-rank tensor and hierarchical dictionary learning | |
| Chen et al. | Low-rank neighbor embedding for single image super-resolution | |
| CN110675321A (en) | Super-resolution image reconstruction method based on progressive depth residual error network | |
| CN107341776B (en) | Single-frame super-resolution reconstruction method based on sparse coding and combined mapping | |
| CN107194873B (en) | Low-rank nuclear norm regular face image super-resolution method based on coupled dictionary learning | |
| CN112529776B (en) | Training method of image processing model, image processing method and device | |
| CN110942424A (en) | Composite network single image super-resolution reconstruction method based on deep learning | |
| CN105513033B (en) | A kind of super resolution ratio reconstruction method that non local joint sparse indicates | |
| CN111861886B (en) | Image super-resolution reconstruction method based on multi-scale feedback network | |
| CN109360147A (en) | Multispectral image super resolution ratio reconstruction method based on Color Image Fusion | |
| CN111986105A (en) | Video time sequence consistency enhancing method based on time domain denoising mask | |
| CN110675318A (en) | Main structure separation-based sparse representation image super-resolution reconstruction method | |
| CN108460723B (en) | Bilateral total variation image super-resolution reconstruction method based on neighborhood similarity | |
| CN115578255A (en) | Super-resolution reconstruction method based on inter-frame sub-pixel block matching | |
| CN111476748A (en) | MR image fusion method based on MCP constraint convolution sparse representation | |
| Lu et al. | Two-stage self-supervised cycle-consistency network for reconstruction of thin-slice MR images | |
| CN111767679B (en) | Method and device for processing time-varying vector field data |
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 |