CN109255755A - Image super-resolution rebuilding method based on multiple row convolutional neural networks - Google Patents
Image super-resolution rebuilding method based on multiple row convolutional neural networks Download PDFInfo
- Publication number
- CN109255755A CN109255755A CN201811241002.4A CN201811241002A CN109255755A CN 109255755 A CN109255755 A CN 109255755A CN 201811241002 A CN201811241002 A CN 201811241002A CN 109255755 A CN109255755 A CN 109255755A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- model
- training
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a kind of image super-resolution rebuilding methods based on multiple row convolutional neural networks.Firstly, designing multiple row convolutional neural networks model, including characteristic extraction part and image reconstruction part according to deep learning algorithm.Then, original image is cut into small pieces, and down-sampling is done to these high-resolution fritters, to obtain the fritter of low resolution, using these low resolution and high-resolution fritter to establishing training set.Finally, being trained using stochastic gradient descent algorithm to this model, the model that low-resolution image is reconstructed into high-definition picture is obtained, the low-resolution image reconstruction of input is reverted into corresponding high-definition picture.The method of the present invention is tested on the five general purpose image data libraries Set5, Set14, BSDS100, Urban100 and Manga109, robustness and accuracy all with higher.
Description
Technical field
The present invention relates to a kind of image super-resolution rebuilding methods, are based on multiple row convolutional neural networks more particularly to one kind
The super resolution ratio reconstruction method of image belongs to image procossing, reconstruction technique utilizes.
Background technique
With the development of information technology, image has been widely used for various as wherein main information media
Scene.In various fields, people have higher requirement for the quality of image, so for the information age of high speed development
For, low-quality image has been difficult meet the needs of special scenes.Image resolution ratio is to measure a weight of picture quality
Index is wanted, more high this image that just represents of image resolution ratio includes more detailed information.Image super-resolution (Super-
Resolution, SR) rebuild belong to image processing techniques, obtained from low resolution (Low-Resolution, LR) image reconstruction
High-resolution (High-Resolution, HR) image.The super-resolution rebuilding of image has a wide range of applications, such as face is known
Not, imaging of medical and remote sensing technology.
Currently, convolutional neural networks (Convolutional Neural Networks, CNN) are in target detection, Ren Leihang
To achieve significant progress in the Computer Vision Tasks such as identification and image segmentation.It is based especially on the super of convolutional neural networks
Resolution method has than conventional methods such as dictionary learning, local linear smoothing and random forests and preferably rebuilds effect.2014
Year, Dong [1] etc. realize image super-resolution rebuilding (Super Resolution using convolutional neural networks
Convolutional Neural Network, SRCNN), the pretreated low-resolution image of bicubic interpolation can be input to
End to end in deep layer convolutional neural networks, gradually by low-resolution image to the mapping relations between high-definition picture
Habit processing.Due to using the training method end to end in deep learning, so that this method significantly mentions compared to conventional method
High image super-resolution rebuilding effect.
Although the Super-Resolution of Images Based based on convolutional neural networks proposed solves traditional images super-resolution
That there are robustness is not strong for rate algorithm for reconstructing, calculates the problems such as complicated, but the existing Image Super-resolution based on convolutional neural networks
Rate method will first use the side of bicubic interpolation (Bicubic Interpolation) before extracting low-resolution image feature
Method is amplified to low-resolution image the size for wanting to rebuild obtained high-definition picture, from the image after bicubic interpolation
It goes to extract feature, the information of many redundancies is introduced by the image after bicubic interpolation, this is not help to feature extraction
's.Therefore, existing method still has that there are reconstruction abilities is poor, visual effect is poor etc. asks than more rich image for details
Topic.
Summary of the invention
The purpose of the invention is to carry out higher-quality reconstruction to low-resolution image, propose a kind of based on multiple row volume
The image super-resolution method of product neural network makes the height rebuild by the extraction to Analysis On Multi-scale Features in low-resolution image
Image in different resolution can restore more image detail informations, and edge is more clear.The method of the present invention can effectively improve oversubscription
The Y-PSNR and structural similarity of resolution reconstruction image, and also have better effect on subjective vision.In addition, this hair
The bright application for convolutional neural networks in image super-resolution also has important reference.
In order to achieve the above objectives, insight of the invention is that
Firstly, designing multiple row convolutional neural networks model, including characteristic extraction part and image according to deep learning algorithm
Rebuild part.Then, original image is cut into small pieces, and down-sampling is done to these high-resolution fritters, to obtain low point
The fritter of resolution, using these low resolution and high-resolution fritter to establishing training set.Finally, using under stochastic gradient
Drop algorithm is trained this model, obtains the model that low-resolution image is reconstructed into high-definition picture, i.e., originally
The image super-resolution rebuilding model of the invention multiple row convolutional neural networks.
According to above-mentioned design, the present invention adopts the following technical scheme:
A kind of image super-resolution method based on multiple row convolutional neural networks, includes the following steps:
Step 1, multiple row convolutional neural networks model foundation: multiple row convolutional neural networks mould is designed according to deep learning algorithm
Type, including characteristic extraction part and image reconstruction part；
Step 2, image augmentation (Image Augmentation): large-scale dataset is successfully using before depth network
It mentioning, image augmentation is that have different training samples by making a series of random changes to training image to generate similar, from
And expand training dataset scale；Increase the scale of training set by image augmentation, reduce dependence of the model to certain attributes,
To improve the generalization ability of model, the image augmentation method used has rotation, scales, mirror image；
Step 3, training set are established: original image being cut on the increased training set of scale obtained according to step 2 small
Block, and down-sampling is done to these high-resolution fritters, to obtain the fritter of low resolution, use these low resolution and height
The fritter of resolution ratio is to establishing training set；
Step 4, multiple row convolutional neural networks model training: the training image super-resolution on the training set that step 3 obtains
Reconstruction model, optimization algorithm use stochastic gradient descent algorithm, obtain one after the completion of training and be reconstructed into low-resolution image
The model of high-definition picture；
Step 5, image super-resolution rebuilding: the low-resolution image of input is rebuild in the model that step 4 training obtains
Revert to corresponding high-definition picture.
The method of the present invention mainly considers the multiple dimensioned characteristic of image, therefore by means of multiple row convolutional neural networks mould
Type can efficiently extract the Analysis On Multi-scale Features in image, and these Analysis On Multi-scale Features are merged.Directly from low resolution figure
Feature is extracted as in, is reduced calculation amount, is improved the reconstruction speed of image.In order to accelerate the convergence of image super-resolution rebuilding model
Speed removes the interpolation image for rebuilding high-definition picture and bicubic interpolation image using the Analysis On Multi-scale Features that extraction obtains, and
It is not directly to go to rebuild high-definition picture from feature, reduces the training difficulty of network, while improving the super-resolution of image
Reconstruction quality.
The present invention compared with prior art, has following obvious prominent substantive distinguishing features and remarkable advantage:
1, the method for the present invention has fully considered the multiple dimensioned feature of image, i.e. the objects in images feelings different there are scale
Condition.Propose a kind of image super-resolution rebuilding model based on multiple row convolutional neural networks.
2, feature is extracted in the directly never pretreated low-resolution image of the method for the present invention, reduces its calculation amount,
To improve the reconstruction speed of model.
3, the method for the present invention goes to rebuild high-definition picture and bicubic interpolation figure using obtained Analysis On Multi-scale Features are extracted
The interpolation image of picture, rather than directly go to rebuild high-definition picture from feature, the training difficulty of model is reduced, figure is improved
The super-resolution rebuilding quality of picture.
Detailed description of the invention
Fig. 1 is that the present invention is based on the network structure block diagrams of the image super-resolution rebuilding method of multiple row convolutional neural networks.
Fig. 2 is that super-resolution rebuilding effect when " butterfly " amplification factor is 2 in Set5 test set compares.
Fig. 3 is that super-resolution rebuilding effect when " 21077 " amplification factor is 3 in BSDS100 test set compares.
Fig. 4 is that super-resolution rebuilding effect when " img023 " amplification factor is 4 in Urban100 test set compares.
Fig. 5 is super-resolution rebuilding effect ratio when " UltraEleven " amplification factor is 4 in Manga109 test set
Compared with.
Specific embodiment
Details are as follows for the preferred embodiment of the present invention combination attached drawing:
The multiple row convolutional neural networks structure of the present embodiment is as shown in Figure 1.At Ubuntu 16.04, PyTorch environment
Programming simulation realizes this method.Firstly, designing multiple row convolutional neural networks model, including feature extraction according to deep learning algorithm
Part and image reconstruction part.Then, original image is cut into small pieces, and down-sampling is done to these high-resolution fritters, from
And the fritter of low resolution is obtained, using these low resolution and high-resolution fritter to establishing training set.Finally, using
Stochastic gradient descent algorithm is trained this model, obtains one and low-resolution image is reconstructed into high-definition picture
Model, i.e., the image super-resolution rebuilding model of multiple row convolutional neural networks of the present invention.
This method specifically comprises the following steps:
Step 1, multiple row convolutional neural networks model foundation: multiple row convolutional neural networks mould is designed according to deep learning algorithm
Type, including characteristic extraction part and image reconstruction part；
Step 2, image augmentation (Image Augmentation): large-scale dataset is successfully using before depth network
It mentioning, image augmentation is that have different training samples by making a series of random changes to training image to generate similar, from
And expand training dataset scale；Increase the scale of training set by image augmentation, reduce dependence of the model to certain attributes,
To improve the generalization ability of model, the image augmentation method used has rotation, scales, mirror image；
Step 3, training set are established: original image being cut on the increased training set of scale obtained according to step 2 small
Block, and down-sampling is done to these high-resolution fritters, to obtain the fritter of low resolution, use these low resolution and height
The fritter of resolution ratio is to establishing training set；
Step 4, multiple row convolutional neural networks model training: the training image super-resolution on the training set that step 3 obtains
Reconstruction model, optimization algorithm use stochastic gradient descent algorithm, obtain one after the completion of training and be reconstructed into low-resolution image
The model of high-definition picture；
Step 5, image super-resolution rebuilding: the low-resolution image of input is rebuild in the model that step 4 training obtains
Revert to corresponding high-definition picture.
In the step 1, propose a cascade multiple row convolutional neural networks extracted from low-resolution image it is more
Scale feature, then rebuilds corresponding high-definition picture, and network structure is as shown in Figure 1.The network frame of proposition has used very
More multiple row modules (Multi-Column Block), the convolutional layer group that each multiple row module arranges different convolution kernel sizes by three
At.The model proposed removes prediction image and target high-resolution figure after bicubic interpolation from the low-resolution image of input
Interpolation image as between.The model proposed is divided into two parts, characteristic extraction part and image reconstruction part.
In characteristic extraction part, coarse feature is extracted using a convolutional layer first, which there are 64 3 × 3
Convolution kernel.Then, it goes to extract Analysis On Multi-scale Features using three cascade multiple row modules.In the model, biasing is not used,
So the calculation formula of convolutional layer is as follows:
In above-mentioned formula, WlThe input of the weight and convolutional layer that can learn is respectively indicated with x.σ indicates activation primitive,
In the model, amendment linear unit (Leaky Rectified Linear Unit) is revealed using band.
Finally, going to up-sample extracted feature using a warp lamination, one 3 × 3 is used after warp lamination
Convolutional layer obtains residual image.The calculation formula of the output picture size of warp lamination is as follows:
Xout=(Xin-1)×λ-2×ρ+κ, (2)
In above-mentioned formula, XinAnd XoutIt is outputting and inputting for warp lamination respectively, λ indicates the step-length of deconvolution, ρ table
Show the line number of the addition 0 in each edge of input, κ indicates the size of deconvolution core.Obviously, it needs λ to be set as and times magnification
Number is the same.Table 1 gives the parameter setting of the warp lamination under different amplification.
Table 1
| Amplification factor | λ | ρ | κ |
| 2 | 2 | 1 | 4 |
| 3 | 3 | 1 | 5 |
| 4 | 4 | 1 | 6 |
In the model proposed, different size of convolution kernel has been used to go to extract feature in each column.Detailed structure
As shown in Figure 1.The calculation formula of the receptive field γ of convolutional layer is as follows:
γ=κ+(κ -1) × (n-1), (3)
In above-mentioned formula, κ indicates the size of convolution kernel, and n indicates the quantity of convolutional layer in each column.According to above formula,
It is 3 × 3 convolutional layer that 6 layers of convolution kernel size have been used in multiple row module, the convolutional layer that 3 layers of convolution kernel size are 5 × 5,2 layers
The convolutional layer that convolution kernel size is 7 × 7, can obtain the receptive field of same size in this way.
In order to extract relatively reliable feature, the feature needs that different lines are extracted carry out feature on the same receptive field
Fusion.The method that Fusion Features are taken is to increase by one 1 × 1 convolutional layer in each column the last layer, then these column
Characteristic pattern, which does element and is added, enters fusion.The benefit of the convolutional layer of increase by 1 × 1 can have Analysis On Multi-scale Features more complicated
Combination.In general, more multiple row modules can have better performance, for the tradeoff of performance and efficiency, the present embodiment is used
Three multiple row modules.
In image reconstruction module, prediction high-definition picture and bicubic interpolation image are removed using one 3 × 3 convolutional layer
Residual image.The residual image of neural network forecast is added with bicubic interpolation image by element, so as to reconstruct phase
The high-definition picture answered.The calculation formula for exporting image is as follows:
In above-mentioned formula,WithRespectively indicate the low-resolution image of the input of model and the high resolution graphics of output
Picture.Indicate bicubic interpolation,Indicate proposed model.
In the step 2, the training set image that uses by Yang [2] 91 pictures and BSDS [3] 200 pictures
Composition.The mode of image augmentation mainly has scaling, rotation, mirror image.The multiple wherein scaled is 1 times, 0.7 times and 0.5 times；Rotation
Angle be 0 °, 90 °, 180 ° and 270 °；Mirror image is horizontal mirror image or holding original image.By image augmentation, in addition to original image with
Outside, 23 additional versions have been obtained.
In the step 3, original image is cut into small pieces on the increased training set of scale obtained according to step 2,
And down-sampling is done to these high-resolution fritters, to obtain the fritter of low resolution, use these low resolution and high score
The fritter of resolution is to establishing training set.When amplification factor is 2, tile size is 82 × 82, step-length 64, and down-sampling is
The inverse of amplification factor, i.e., 1/2 times.Similar, when amplification factor is 3, tile size is 123 × 123, step-length 48,
Down-sampling multiple is 1/3 times；When amplification factor is 4, tile size is 164 × 164, step-length 32, and down-sampling multiple is
1/4 times.The size for inputting low-resolution image block is all 41 × 41.
In the step 4, training image Super-resolution reconstruction established model, optimization algorithm on the training set that step 3 obtains
Using stochastic gradient descent algorithm (Stochastic Gradient Descent), criticizes and be dimensioned to 64, momentum parameter is set as
0.9, weight decaying is set as 10-4, learning rate is set as 0.1, and declines 10 times after every 20 iteration cycles.Due to initial
Learning rate it is bigger, shown to be sliced using gradient to prevent gradient from exploding, gradient slice is set as 0.4, can after the completion of training
The model of high-definition picture is reconstructed into obtain one for low-resolution image.
In the step 5, the low-resolution image reconstruction of input is reverted in the model that the training of above-mentioned steps 4 obtains
Corresponding high-definition picture.
It is tested in five image data bases of Set5, Set14, BSDS100, Urban100 and Manga109 below
Assess the image super-resolution rebuilding method proposed by the invention based on multiple row convolutional neural networks.Set5, Set14 and
What BSDS100 included is natural image；What Urban100 included is City scenarios image；Scheme in the caricature that Manga109 includes
Picture.The environment of this experiment is the PyTorch platform under 16.04 operating system of Ubuntu, inside saves as 16GB, and GPU is
GeForce1070.Use Y-PSNR (Peak Signal to Noise Ratio, PSNR) and structural similarity coefficient
(Structural Similarity Index, SSIM) is used as super-resolution rebuilding model-evaluation index, and PSNR is bigger, SSIM
Degree of conformity closer to 1 representative model and original image is higher, and accuracy is higher, and the results are shown in Table 2.Fig. 2-Fig. 5 compares difference
Algorithm rebuilds effect on these test sets.
Table 2
Wherein, experimental result best algorithm overstriking font representation, second-best algorithm are indicated with underscore.From table
It can be seen that method of the invention has preferable robustness and accuracy on five databases.By above-mentioned experiment as it can be seen that originally
Inventive method has preferable robustness and accuracy really on image super-resolution rebuilding, and computation complexity is low, can be more
It is suitable for real-time video quality well to monitor.
Bibliography:
1Dong,Chao,et al."Image super-resolution using deep convolutional
networks."IEEE transactions on pattern analysis and machine intelligence 38.2
(2016):295-307.
2Yang,Jianchao,et al."Image super-resolution via sparse
representation."IEEE transactions on image processing 19.11(2010):2861-2873.
3Martin,David,et al."A database of human segmented natural images and
its application to evaluating segmentation algorithms and measuring
ecological statistics."Computer Vision,2001.ICCV 2001.Proceedings.Eighth IEEE
International Conference on.Vol.2.IEEE,2001.
4Timofte,Radu,Vincent De Smet,and Luc Van Gool."A+:Adjusted anchored
neighborhood regression for fast super-resolution."Asian Conference on
Computer Vision.Springer,Cham,2014.
5Huang,Jia-Bin,Abhishek Singh,and Narendra Ahuja."Single image super-
resolution from transformed self-exemplars."Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition.2015.
6Schulter,Samuel,Christian Leistner,and Horst Bischof."Fast and
accurate image upscaling with super-resolution forests."Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition.2015.
7Dong,Chao,Chen Change Loy,and Xiaoou Tang."Accelerating the super-
resolution convolutional neural network."European Conference on Computer
Vision.Springer,Cham,2016.
8Kim,Jiwon,Jung Kwon Lee,and Kyoung Mu Lee."Accurate image super-
resolution using very deep convolutional networks."Proceedings of the IEEE
conference on computer vision and pattern recognition.2016.
9Kim,Jiwon,Jung Kwon Lee,and Kyoung Mu Lee."Deeply-recursive
convolutional network for image super-resolution."Proceedings of the IEEE
conference on computer vision and pattern recognition.2016.
10Lai,Wei-Sheng,et al."Deep laplacian pyramid networks for fast and
accurate superresolution."IEEE Conference on Computer Vision and Pattern
Recognition.Vol.2.No.3.2017.
Claims (1)
1. a kind of image super-resolution method based on multiple row convolutional neural networks, which comprises the steps of:
Step 1, multiple row convolutional neural networks model foundation: designing multiple row convolutional neural networks model according to deep learning algorithm,
Including characteristic extraction part and image reconstruction part；
Step 2, image augmentation: large-scale dataset is the premise for successfully using depth network, and image augmentation is by training
Image makes a series of random changes, has different training samples to generate similar, to expand training dataset scale；It is logical
Image augmentation is crossed to increase the scale of training set, reduces dependence of the model to certain attributes, so that the generalization ability of model is improved,
The image augmentation method used has rotation, scaling, mirror image；
Step 3, training set are established: original image be cut into small pieces on the increased training set of scale obtained according to step 2, and
Down-sampling is done to these high-resolution fritters, to obtain the fritter of low resolution, uses these low resolution and high-resolution
The fritter of rate is to establishing training set；
Step 4, multiple row convolutional neural networks model training: the training image super-resolution rebuilding on the training set that step 3 obtains
Model, optimization algorithm use stochastic gradient descent algorithm, obtain one after the completion of training and low-resolution image is reconstructed into high score
The model of resolution image；
Step 5, image super-resolution rebuilding: the low-resolution image of input is rebuild in the model that step 4 training obtains and is restored
At corresponding high-definition picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811241002.4A CN109255755B (en) | 2018-10-24 | 2018-10-24 | Image super-resolution reconstruction method based on multi-column convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811241002.4A CN109255755B (en) | 2018-10-24 | 2018-10-24 | Image super-resolution reconstruction method based on multi-column convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109255755A true CN109255755A (en) | 2019-01-22 |
| CN109255755B CN109255755B (en) | 2023-05-23 |
Family
ID=65046172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811241002.4A Active CN109255755B (en) | 2018-10-24 | 2018-10-24 | Image super-resolution reconstruction method based on multi-column convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109255755B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109949224A (en) * | 2019-02-26 | 2019-06-28 | 北京悦图遥感科技发展有限公司 | A kind of method and device of the connection grade super-resolution rebuilding based on deep learning |
| CN110060208A (en) * | 2019-04-22 | 2019-07-26 | 中国科学技术大学 | A method of improving super-resolution algorithms reconstruction property |
| CN110246083A (en) * | 2019-05-10 | 2019-09-17 | 杭州电子科技大学 | A kind of fluorescence microscope images super-resolution imaging method |
| CN110647934A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN110689509A (en) * | 2019-09-10 | 2020-01-14 | 上海大学 | Video super-resolution reconstruction method based on cyclic multi-column 3D convolutional network |
| CN110728219A (en) * | 2019-09-29 | 2020-01-24 | 天津大学 | 3D face generation method based on multi-column multi-scale graph convolution neural network |
| CN110991213A (en) * | 2019-05-14 | 2020-04-10 | 杨春燕 | Automatic big data distribution platform |
| CN111091515A (en) * | 2019-12-24 | 2020-05-01 | 展讯通信(天津)有限公司 | Image restoration method and device, and computer-readable storage medium |
| CN111127317A (en) * | 2019-12-02 | 2020-05-08 | 深圳供电局有限公司 | Image super-resolution reconstruction method and device, storage medium and computer equipment |
| CN111681168A (en) * | 2020-06-05 | 2020-09-18 | 杭州电子科技大学 | Low-resolution cell super-resolution reconstruction method based on parallel residual error network |
| CN111709900A (en) * | 2019-10-21 | 2020-09-25 | 上海大学 | High dynamic range image reconstruction method based on global feature guidance |
| CN111833251A (en) * | 2020-07-13 | 2020-10-27 | 北京安德医智科技有限公司 | Three-dimensional medical image super-resolution reconstruction method and device |
| CN112419150A (en) * | 2020-11-06 | 2021-02-26 | 中国科学技术大学 | Random multiple image super-resolution reconstruction method based on bilateral up-sampling network |
| CN113139899A (en) * | 2021-03-31 | 2021-07-20 | 桂林电子科技大学 | Design method of high-quality light-weight super-resolution reconstruction network model |
| CN113409195A (en) * | 2021-07-06 | 2021-09-17 | 中国标准化研究院 | Image super-resolution reconstruction method based on improved deep convolutional neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101719266A (en) * | 2009-12-25 | 2010-06-02 | 西安交通大学 | Affine transformation-based frontal face image super-resolution reconstruction method |
| CN104463324A (en) * | 2014-11-21 | 2015-03-25 | 长沙马沙电子科技有限公司 | Convolution neural network parallel processing method based on large-scale high-performance cluster |
| CN106650786A (en) * | 2016-11-14 | 2017-05-10 | 沈阳工业大学 | Image recognition method based on multi-column convolutional neural network fuzzy evaluation |
| CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
| US20180137603A1 (en) * | 2016-11-07 | 2018-05-17 | Umbo Cv Inc. | Method and system for providing high resolution image through super-resolution reconstruction |
-
2018
- 2018-10-24 CN CN201811241002.4A patent/CN109255755B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101719266A (en) * | 2009-12-25 | 2010-06-02 | 西安交通大学 | Affine transformation-based frontal face image super-resolution reconstruction method |
| CN104463324A (en) * | 2014-11-21 | 2015-03-25 | 长沙马沙电子科技有限公司 | Convolution neural network parallel processing method based on large-scale high-performance cluster |
| US20180137603A1 (en) * | 2016-11-07 | 2018-05-17 | Umbo Cv Inc. | Method and system for providing high resolution image through super-resolution reconstruction |
| CN106650786A (en) * | 2016-11-14 | 2017-05-10 | 沈阳工业大学 | Image recognition method based on multi-column convolutional neural network fuzzy evaluation |
| CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109949224A (en) * | 2019-02-26 | 2019-06-28 | 北京悦图遥感科技发展有限公司 | A kind of method and device of the connection grade super-resolution rebuilding based on deep learning |
| CN109949224B (en) * | 2019-02-26 | 2023-06-30 | 北京悦图遥感科技发展有限公司 | Deep learning-based cascade super-resolution reconstruction method and device |
| CN110060208A (en) * | 2019-04-22 | 2019-07-26 | 中国科学技术大学 | A method of improving super-resolution algorithms reconstruction property |
| CN110060208B (en) * | 2019-04-22 | 2022-10-28 | 中国科学技术大学 | Method for improving reconstruction performance of super-resolution algorithm |
| CN110246083A (en) * | 2019-05-10 | 2019-09-17 | 杭州电子科技大学 | A kind of fluorescence microscope images super-resolution imaging method |
| CN110246083B (en) * | 2019-05-10 | 2023-02-24 | 杭州电子科技大学 | Fluorescence microscopic image super-resolution imaging method |
| CN110991213A (en) * | 2019-05-14 | 2020-04-10 | 杨春燕 | Automatic big data distribution platform |
| CN110689509A (en) * | 2019-09-10 | 2020-01-14 | 上海大学 | Video super-resolution reconstruction method based on cyclic multi-column 3D convolutional network |
| CN110689509B (en) * | 2019-09-10 | 2024-04-19 | 上海大学 | Video super-resolution reconstruction method based on cyclic multi-column 3D convolution network |
| CN110647934A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN110647934B (en) * | 2019-09-20 | 2022-04-08 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN110728219B (en) * | 2019-09-29 | 2023-09-26 | 天津大学 | 3D face generation method based on multi-column multi-scale graph convolution neural network |
| CN110728219A (en) * | 2019-09-29 | 2020-01-24 | 天津大学 | 3D face generation method based on multi-column multi-scale graph convolution neural network |
| CN111709900A (en) * | 2019-10-21 | 2020-09-25 | 上海大学 | High dynamic range image reconstruction method based on global feature guidance |
| CN111127317A (en) * | 2019-12-02 | 2020-05-08 | 深圳供电局有限公司 | Image super-resolution reconstruction method and device, storage medium and computer equipment |
| CN111127317B (en) * | 2019-12-02 | 2023-07-25 | 深圳供电局有限公司 | Image super-resolution reconstruction method, device, storage medium and computer equipment |
| CN111091515A (en) * | 2019-12-24 | 2020-05-01 | 展讯通信(天津)有限公司 | Image restoration method and device, and computer-readable storage medium |
| CN111091515B (en) * | 2019-12-24 | 2022-08-09 | 展讯通信(天津)有限公司 | Image restoration method and device, and computer-readable storage medium |
| CN111681168B (en) * | 2020-06-05 | 2023-03-21 | 杭州电子科技大学 | Low-resolution cell super-resolution reconstruction method based on parallel residual error network |
| CN111681168A (en) * | 2020-06-05 | 2020-09-18 | 杭州电子科技大学 | Low-resolution cell super-resolution reconstruction method based on parallel residual error network |
| CN111833251A (en) * | 2020-07-13 | 2020-10-27 | 北京安德医智科技有限公司 | Three-dimensional medical image super-resolution reconstruction method and device |
| CN112419150A (en) * | 2020-11-06 | 2021-02-26 | 中国科学技术大学 | Random multiple image super-resolution reconstruction method based on bilateral up-sampling network |
| CN112419150B (en) * | 2020-11-06 | 2024-04-02 | 中国科学技术大学 | Image super-resolution reconstruction method of arbitrary multiple based on bilateral upsampling network |
| CN113139899A (en) * | 2021-03-31 | 2021-07-20 | 桂林电子科技大学 | Design method of high-quality light-weight super-resolution reconstruction network model |
| CN113409195A (en) * | 2021-07-06 | 2021-09-17 | 中国标准化研究院 | Image super-resolution reconstruction method based on improved deep convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109255755B (en) | 2023-05-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109255755A (en) | Image super-resolution rebuilding method based on multiple row convolutional neural networks | |
| WO2023185243A1 (en) | Expression recognition method based on attention-modulated contextual spatial information | |
| CN110111256B (en) | Image super-resolution reconstruction method based on residual distillation network | |
| CN103279933B (en) | A kind of single image super resolution ratio reconstruction method based on bilayer model | |
| Fu et al. | Image super-resolution based on generative adversarial networks: a brief review | |
| CN105488759B (en) | A kind of image super-resolution rebuilding method based on local regression model | |
| CN103020940B (en) | Local feature transformation based face super-resolution reconstruction method | |
| CN110533591A (en) | Super resolution image reconstruction method based on codec structure | |
| Shi et al. | Exploiting multi-scale parallel self-attention and local variation via dual-branch transformer-cnn structure for face super-resolution | |
| Chen et al. | Single image super-resolution based on deep learning and gradient transformation | |
| CN107292821B (en) | A kind of super-resolution image reconstruction method and system | |
| Luo et al. | Bi-GANs-ST for perceptual image super-resolution | |
| CN103413351B (en) | Three-dimensional face fast reconstructing method based on compressive sensing theory | |
| CN117575915A (en) | Image super-resolution reconstruction method, terminal equipment and storage medium | |
| Zeng et al. | Densely connected transformer with linear self-attention for lightweight image super-resolution | |
| CN108090873B (en) | Pyramid face image super-resolution reconstruction method based on regression model | |
| CN113096015A (en) | Image super-resolution reconstruction method based on progressive sensing and ultra-lightweight network | |
| CN116071239B (en) | CT image super-resolution method and device based on mixed attention model | |
| CN111696167A (en) | Single image super-resolution reconstruction method guided by self-example learning | |
| CN108846797B (en) | Image super-resolution method based on two training sets | |
| Shao et al. | SRWGANTV: image super-resolution through wasserstein generative adversarial networks with total variational regularization | |
| CN113344110B (en) | Fuzzy image classification method based on super-resolution reconstruction | |
| CN104574320B (en) | A kind of image super-resolution restored method based on sparse coding coefficients match | |
| CN103632358B (en) | For the method that low-resolution image is converted to high-definition picture | |
| Shao et al. | Two-stream coupling network with bidirectional interaction between structure and texture for image inpainting |
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 |