CN106991646A - A kind of image super-resolution method based on intensive connection network - Google Patents
A kind of image super-resolution method based on intensive connection network Download PDFInfo
- Publication number
- CN106991646A CN106991646A CN201710193665.2A CN201710193665A CN106991646A CN 106991646 A CN106991646 A CN 106991646A CN 201710193665 A CN201710193665 A CN 201710193665A CN 106991646 A CN106991646 A CN 106991646A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- resolution
- layer
- super
- 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
Abstract
The present invention discloses a kind of image super-resolution method based on intensive connection network, by the depth for increasing convolutional neural networks, substantial amounts of great-jump-forward connection is introduced in depth network, effectively solve the problems, such as gradient disappearance during depth network backpropagation, flowing of the information on network is optimized, the super-resolution rebuilding ability of convolutional neural networks is improved.Meanwhile, the present invention also effectively combines low-level image feature and higher level of abstraction feature, reduces model parameter, depth network model is have compressed, so as to improve the reconstruction efficiency of image super-resolution.In addition, by introducing depth supervision technology, network different depth can rebuilding super resolution image, not only optimize the training of depth network, and according to the computing capability of test lead appropriate network depth can be selected to rebuild high-definition image in test.Finally, it is trained present invention utilizes the image set of multiple multiplication factors, the model of acquisition can carry out image super-resolution on multiple yardsticks, and without training different models for each multiplication factor.
Description
Technical field
The present invention relates to computer vision field and artificial intelligence technology, more particularly to it is a kind of based on intensive connection network
Image super-resolution method.
Background technology
In computer vision field, most problem all has begun to be solved using deep neural network, and takes
Obtained extensive success.In numerous Computer Vision Tasks, such as recognition of face, object detecting and tracking, image retrieval,
Using the algorithm of deep neural network model, there is great lifting than the performance of traditional algorithm.In the Super-resolution reconstruction of image
Build in task, newest work also has begun to represent ability using the nonlinear characteristic of convolutional neural networks, improves image and surpasses
The reconstruction effect of resolution ratio.Through to the literature search discovery to prior art, a kind of patent name " image super-resolution rebuilding side
Method " (China Patent Publication No. CN105976318A, publication date is 2016.09.28) and patent name are " adaptive based on learning rate
Convolutional neural networks image super-resolution rebuilding method " (China Patent Publication No. CN106228512A, publication date is
2016.12.14 the method for deep learning) has been used to carry out image super-resolution rebuilding, and has obtained than traditional interpolation method more
Good reconstructed results.However, convolutional neural networks structure of this patent only with 3 layers, its nonlinear character representation ability
It is limited in one's ability with Image Reconstruction.Newest neural network model such as AlexNet, VGG and ResNet etc. are main at " width " and " deep
Degree " aspect carries out different degrees of amplification, and the performance of network will be greatly improved.Therefore, study and design one more
Deep network model has for the reconstruction performance of lifting image super-resolution and greatly helped.
The most simple mode for deepening network model is that basic building block (such as convolutional layer and active coating) is stacked.
However, when network becomes increasingly deeper, it is trained and the also corresponding increase therewith of convergence difficulty.In the training process of network model
In, gradient signal needs to propagate backward to the bottom from network top, so as to update network model parameter.For traditional
Have for several layers of neural network models, convergence can be reached in this way.However, there is the network of tens layers for training
Model, when propagating backward to the bottom of network, gradient signal has disappeared almost, and the model parameter of bottom-layer network is not
Effectively it can be updated and be optimized.So, if using it is this directly stack by the way of, can cause on the contrary under algorithm performance
Drop.In order to effectively training depth network, the VDSR algorithms proposed on international conference CVPR for 2016 employ gradient and cut
Cut and the technology such as residual error study so that 20 layers of convolutional neural networks model effectively can optimize and restrain, several layers of than before
The super-resolution rebuilding performance of network model (such as Chinese patent CN105976318A and CN106228512A), which has, greatly to be carried
Rise.But VDSR algorithms are still that convolutional layer and active coating are simply stacked, and are unfavorable for the flowing of gradient information, to excellent
Change deeper Netowrk tape and come difficult.Meanwhile, this simple stack manner can not effectively utilize every layer and train the feature come,
And network model parameter is also very huge.For example, 20 layer networks of VDSR algorithms are accomplished by more than 70 ten thousand model parameter, not only give
Optimization brings difficulty, also increases computation complexity during super-resolution rebuilding.
The methods such as the residual error network structure ResNet and dense network structure DenseNet that are recently proposed pass through in a network
Jump connection is introduced, so as to attempt to solve the problem of profound network is brought in optimization.Connected by largely introducing great-jump-forward,
The interface channel between bottom-layer network and overlay network can effectively be shortened, can so optimize flowing of the information on network,
So as to effectively solve the problems, such as the gradient disappearance of depth network.In addition, dense network structure can be reused with supported feature, it can strengthen
The propagation of feature, and the parameter of model is reduced, reduce the computation complexity of model.The present invention takes full advantage of the excellent of dense network
Gesture, and first Application is in the task of image super-resolution, proposes SRDenseNet algorithms, greatly improves depth network and exist
Reconstruction performance on image super-resolution.Meanwhile, SRDenseNet algorithms proposed by the present invention combine depth supervision technology, make
Each layer parameter of network model can more efficient quick convergence, accelerate training speed, and further lift network model
Super-resolution rebuilding performance.Further it is proposed that algorithm fusion multi-scale information so that the network mould that training is obtained
Type can effectively be rebuild to multiple super-resolution multiplication factors.
The content of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of image oversubscription based on intensive connection network
Resolution method, improves the image super-resolution rebuilding effect of multiple multiplication factors, while also considerably reducing model parameter, has
Compression depth neural network model is imitated, the reconstruction efficiency of image super-resolution is improved.
The technical solution adopted by the present invention is：
A kind of image super-resolution method based on intensive connection network, it comprises the following steps：
A) according to different interpolation amplification multiples, multiple dimensioned training set of images (I is generatedLR, IHR)；
B) dense network module is built：The dense network module includes the n-layer network knot set gradually along transmission direction
Structure, n is the integer more than 1, and a convolutional layer and an active coating are included per layer network structure, and last layer network structure is rolled up
The feature that product is obtained is superimposed in follow-up each layer network structure, then is per the character representation of the convolutional layer of layer network structure：
Xn=Hn([X0,X1,…,Xn-1]) (1)
Wherein XnFor the feature of the convolutional layer of n-th layer network structure, [X0,X1,...,Xn-1] it is the 1st layer to (n-1)th layer net
The characteristic set of the convolutional layer of network structure；The feature of so bottom-layer network training can be directly added into last layer of module, so that
Effectively combine bottom-layer network feature and the abstract characteristics of top layer；
C) convolutional neural networks model is built, the convolutional neural networks model includes setting gradually along network transmission direction
An input convolutional layer and active coating, and L dense network module；The output end of each dense network module respectively and connects one
Individual convolutional layer as rebuild network,
D) from training set of images (ILR, IHR) it is used as training set, input low-resolution image ILRAnd high-definition picture
IHR, then input picture of the reconstruction image of the reconstruction network of each dense network module respectively with convolutional neural networks model compared
Several loss functions of convolutional neural networks are obtained, are embodied as：
Wherein, fi(w,b,ILR) predicting the outcome for network is rebuild for i-th, w and b are respectively the convolution mould in neutral net
Board parameter and offset parameter；
E) Adam optimized algorithms, convolutional neural networks model parameter w and b that iterative is obtained are utilized；Form low differentiate
Network mapping between rate image and high-definition picture；
F) due to having used multiple dimensioned training set, this network can be used for different multiplication factors.Obtained using training
Convolutional neural networks model parameter w and b, are reconstructed into high-definition picture, and calculate corresponding by the low-resolution image of input
Quantizating index PSNR and SSIM.
Further, the multiplication factor used in the step a) includes 2 times, 3 times and 4 times, forms multiple dimensioned training figure
Image set.
Further, the step a) utilizes the low resolution figure of the different interpolation amplification multiples of ImageNet data set generation
Picture and high resolution graphics image set, and form pairing training set of images (ILR, IHR)。
Further, the low resolution image and high-definition picture in the step a) training set of images can be transformed into
YCbCr space, and carry out Algorithm for Training using Y passages.
Further, the network structure of dense network module is 8 layers in the step b), and convolutional layer in network structure
Convolution kernel size is 3*3, and the activation primitive of active coating is Regularization linear unit function.
Further, the output characteristic of dense network module is controlled in the step b) by introduced feature growth rate k
Figure quantity, n-th layer output has k*n characteristic pattern in dense network module.
Further, the number L of dense network module value is 8 in the step c).
Further, Adam algorithms are dynamically adjusted using the single order moments estimation and second order moments estimation of gradient in the step e)
The learning rate of whole each parameter.
Further, the two indexs of PSNR and SSIM are bigger in the step f), represent reconstruction image and original high score
Difference between resolution image is smaller.
The present invention uses above technical scheme, with the newest convolutional neural networks technology based on intensive connection, effectively
Ground solves the problems, such as the gradient disappearance of profound network, optimizes flowing of the information between each Internet, not only increases figure
As the reconstruction effect of super-resolution, and network model parameter is effectively compressed, has improved the reconstruction efficiency of super-resolution.Tool
The innovative point of body include it is following some：(1) first, super-resolution algorithms of the invention have used multiple dense network moulds first
Block.In each module, each layer network has to be connected with other layer networks in module, so that the information in module
Backpropagation is constantly present direct path, optimizes flowing of the information on profound network, effectively solves the instruction of depth network
Practice problem.(2) secondly, effective utilization of low-level image feature is realized by dense network structure.It is relatively low typically in depth network
Characteristic layer can determine the marginal information of image, and higher layer can train and obtain feature more abstract in image.Intensive net
Network structure passes through the superposition between characteristic layer so that the process of reconstruction of super-resolution can make full use of the edge of bottom-layer network to believe
The abstract characteristics of breath and upper layer network.(3) due to the reuse of characteristic layer so that every layer needs the number of features newly learnt to reduce,
So as to reduce model parameter, the size of network model has been effectively compressed, the calculating speed of test phase is improved.(4) in addition, originally
Invention also introduces depth supervision technology, has accessed reconstruction network outside each dense network module, can so improve
The reconstruction ability of each dense network module, so as to train deeper network model.Further, since network is in different depth
High-definition picture can be reconstructed, so in test lead, can according to the computing capability reasonable selection network depth of test lead,
Export super-resolution reconstruction image.For example, utilizing GPU high performance parallel computation on computers, we can be deep
Dense network module exports reconstructed results, and in mobile phone mobile terminal, because computing capability is limited, we can be selected at first
Or second dense network module output reconstructed results.(5) finally, instructed present invention utilizes the image of multiple yardsticks
Practice, the network that so training is obtained can be amplified the super-resolution rebuilding between 2-4 times, and without for each amplification
Multiple is required for training a depth network model.
Brief description of the drawings
The present invention is described in further details below in conjunction with the drawings and specific embodiments；
Fig. 1 is a kind of schematic flow sheet of the image super-resolution method based on intensive connection network of the present invention；
Fig. 2 is a kind of model structure based on the intensive image super-resolution method for connecting network of the invention；
Fig. 3 is a kind of dense network pie graph based on the intensive image super-resolution method for connecting network of the invention；
Fig. 4 is the low-resolution image of input convolutional neural networks；
Fig. 5 is the super-resolution reconstruction design sketch based on traditional bicubic interpolation algorithm；
Fig. 6 is the super-resolution reconstruction design sketch of the Aplus algorithms based on dictionary learning；
Fig. 7 is the reconstruction design sketch of the SRCNN algorithms based on 3 layers of convolutional neural networks；
Fig. 8 is the reconstruction design sketch of the VDSR algorithms based on 20 layers of convolutional neural networks；
Fig. 9 is a kind of reconstruction design sketch of image super-resolution method based on intensive connection network of the present invention.
Embodiment
As shown in figure 1, the present invention discloses a kind of image super-resolution method based on intensive connection network, make full use of close
The advantage of collection connection network, deepens convolutional neural networks model, improves the reconstruction effect of image super-resolution.The method of the present invention
Specifically include following steps：
A) according to different interpolation amplification multiples, multiple dimensioned training set of images (I is generatedLR, IHR)；Further, utilize
ImageNet data set generation different low resolution image and high resolution graphics image set, and form pairing image set (ILR,
IHR)。
The present invention randomly selects 60,000 pictures I in ImageNet databaseHR, by Gaussian Blur and it is interpolated into low
Resolution space, interpolation multiple selects 2 times, 3 times and 4 times respectively, and interpolation method uses bicubic interpolation.Then differentiated again low
Rate image passes through bicubic interpolation to high resolution space, the image I after being handledLR, form image set (ILR, IHR).This hair
It is bright to extract the pairing subgraph image set of 61*61 sizes in image set again, and upset the storage order of subgraph, form final figure
As training set.In addition, given RGB image can be transformed into YCbCr space, all super-resolution computings are entered with Y passages
Row training.
B) dense network module is built：As shown in Fig. 2 the dense network module includes setting successively along network transmission direction
The n-layer network structure put, n is the integer more than 1, includes a convolutional layer and an active coating per layer network structure.It is each
Inside the feature that layer convolution is obtained layer all after being added in the way of superposition, the feature of such bottom-layer network training can be straight
Last layer for adding module is connect, so that bottom-layer network feature and the abstract characteristics of top layer are effectively combined, per layer network structure
The feature of convolutional layer can be specifically expressed as：
Xn=Hn([X0,X1,…,Xn-1]) (1)
Wherein XnFor the feature of n-th layer, [X0,X1,...,Xn-1] it is the 1st layer to (n-1)th layer of characteristic set.Such as Fig. 2 institutes
Show, can be directly from top layer gradient information so when backpropagation because all layers have forward connection in module
Bottom is traveled to, the gradient disappearance that network depth increase is brought is solved the problems, such as.
Specifically, dense network module includes altogether 8 layer network structures in the present embodiment, and each layer is all comprising a convolution
Layer and an active coating, wherein activation primitive are Regularization linear unit function, and the convolution kernel size of all convolutional layers is 3*3.
Further, HnThe quantity of the characteristic pattern of () output is k, that is, is characterized growth rate.Because each layer of input is
The connection of all front layer output, therefore the output of each layer need not be as legacy network as many.Used here as feature increase
Rate k controls the port number of network characterization figure.In dense network module, n-th layer output has k*n characteristic pattern.In the present invention
In, feature growth rate k, which is taken in 16, each dense network module, 8 layers, and so each dense network module exports 128 features
Figure.
C) convolutional neural networks model is built, as shown in figure 3, the convolutional neural networks model is included along network transmission side
To the input convolutional layer and active coating set gradually, and L dense network module；Behind each dense network module
A convolutional layer is respectively connected to as reconstruction network；Specifically, the number L of dense network module value is 8 in the present embodiment.
D) from training set of images (ILR, IHR) it is used as training set, input low-resolution image ILRAnd high-definition picture
IHR is extremelyConvolutional neural networks model, then each dense network module reconstruction network reconstruction image respectively with convolutional neural networks
The input picture of model relatively obtains several loss functions of convolutional neural networks, is embodied as：
Wherein, fi(w,b,ILR) predicting the outcome for network is rebuild for i-th, w and b are respectively the convolution mould in neutral net
Board parameter and offset parameter；In addition, the convergence in order to accelerate depth network, the present invention additionally uses residual image, i.e. network
Information of forecasting is high-definition picture and low-resolution image difference.So neutral net can be lost for low-resolution image
The high-frequency information of mistake is trained, and removes the redundant reconstruction process to low-frequency information in image.
E) Adam optimized algorithms, convolutional neural networks model parameter w and b that iterative is obtained are utilized；Form low differentiate
Network mapping between rate image and high-definition picture；Adam algorithms are the single order moments estimation and second order moments estimation using gradient
The learning rate of each parameter of dynamic adjustment.Adam advantage is essentially consisted in after bias correction, and iterative learning rate is all each time
There is a determination scope so that parameter is more steady.In the present invention, the β among Adam algorithms1It is set to 0.9.Initial learning rate
0.0001 is set to, random during each propagated forward to take 16 samples, algorithm has iteration time altogether 1,000,000 times.
F) due to having used multiple dimensioned training set, this network can be used for different multiplication factors.Obtained using training
Convolutional neural networks model parameter w and b, are reconstructed into high-definition picture, and calculate corresponding by the low-resolution image of input
Quantizating index PSNR and SSIM.Further, the two indexs are bigger, represent between reconstruction image and original high-resolution image
Difference it is smaller.
For the super-resolution rebuilding effect of verification algorithm, the present invention is tested on common test image set Set5, and with
Other several algorithms are compared.Fig. 4 is shown to be compared with the super-resolution rebuilding example of other several algorithms, is respectively：Fig. 4 is
Input the low-resolution image of neutral net；Fig. 5 is the super-resolution reconstruction design sketch based on traditional bicubic interpolation algorithm；
Fig. 6 is the super-resolution reconstruction design sketch of the Aplus algorithms based on dictionary learning；Fig. 7 is based on 3 layers of convolutional neural networks
The reconstruction design sketch of SRCNN algorithms；Fig. 8 is the reconstruction design sketch of the VDSR algorithms based on 20 layers of convolutional neural networks；Fig. 9 is this
The reconstruction design sketch of the proposed SRDenseNet algorithms of invention.As can be seen from the figure SRDenseNet proposed by the present invention is calculated
The details of image also can be reconstructed, show the figure become apparent from by method well in the case of input picture is very unclear
Picture.Meanwhile, the reconstructed results of the SRDenseNet algorithms of the present invention are can be seen that with original high score from the quantizating index in table
Resolution image is more nearly.In addition, although the SRDenseNet algorithms of the present invention employ deeper network, one has 65 layers,
But 20 layer networks of the model parameter but than VDSR are smaller.VDSR 20 layer networks have more than 70 ten thousand model parameters to need optimization,
And the SRDenseNet of the present invention 65 layer networks have 500,000 model parameters, so while network is deepened, it can also press
Contracting model, so as to while the reconstruction effect of super-resolution is improved, it is ensured that the reconstruction efficiency of image.
Table 1
Table 1：The quantizating index on Set5 test sets of several algorithms of different.
The present invention uses above technical scheme, with the newest convolutional neural networks technology based on intensive connection, effectively
Ground solves the problems, such as the gradient disappearance of profound network, optimizes flowing of the information between each Internet, not only increases figure
As the reconstruction effect of super-resolution, and network model parameter is effectively compressed, has improved the reconstruction efficiency of super-resolution.Tool
The innovative point of body include it is following some：(1) first, super-resolution algorithms of the invention have used multiple dense network moulds first
Block.In each module, each layer network has to be connected with other layer networks in module, so that the information in module
Backpropagation is constantly present direct path, optimizes flowing of the information on profound network, effectively solves the instruction of depth network
Practice problem.(2) secondly, effective utilization of low-level image feature is realized by dense network structure.It is relatively low typically in depth network
Characteristic layer can determine the marginal information of image, and higher layer can train and obtain feature more abstract in image.Intensive net
Network structure passes through the superposition between characteristic layer so that the process of reconstruction of super-resolution can make full use of the edge of bottom-layer network to believe
The abstract characteristics of breath and upper layer network.(3) due to the reuse of characteristic layer so that every layer needs the number of features newly learnt to reduce,
So as to reduce model parameter, the size of network model has been effectively compressed, the calculating speed of test phase is improved.(4) in addition, originally
Invention also introduces depth supervision technology, has accessed reconstruction network outside each dense network module, can so improve
The reconstruction ability of each dense network module, so as to train deeper network model.Further, since network is in different depth
High-definition picture can be reconstructed, so in test lead, can according to the computing capability reasonable selection network depth of test lead,
Export super-resolution reconstruction image.For example, utilizing GPU high performance parallel computation on computers, we can be deep
Dense network module exports reconstructed results, and in mobile phone mobile terminal, because computing capability is limited, we can be selected at first
Or second dense network module output reconstructed results.(5) finally, instructed present invention utilizes the image of multiple yardsticks
Practice, the network that so training is obtained can be amplified the super-resolution rebuilding between 2-4 times, and without for each amplification
Multiple is required for training a depth network model.
Claims (9)
1. a kind of image super-resolution method based on intensive connection network, it is characterised in that：It comprises the following steps：
A) according to different interpolation amplification multiples, multiple dimensioned training set of images (I is generatedLR, IHR)；
B) dense network module is built：The dense network module includes the n-layer network knot set gradually along network transmission direction
Structure, n is the integer more than 1, and a convolutional layer and an active coating are included per layer network structure, and last layer network structure is rolled up
The feature that product is obtained is superimposed in follow-up each layer network structure, and the character representation per the convolutional layer of layer network structure is：
Xn=Hn([X0,X1,…,Xn-1]) (1)
Wherein XnFor the feature of the convolutional layer of n-th layer network structure, [X0,X1,...,Xn-1] for the 1st layer to the (n-1)th layer network knot
The characteristic set of the convolutional layer of structure；
C) convolutional neural networks model is built, the convolutional neural networks model include setting gradually along transmission direction one is defeated
Enter convolutional layer and active coating, and L dense network module；Each dense network module rear is respectively connected to a convolutional layer work
Attach most importance to establishing network；
D) from training set of images (ILR, IHR) it is used as training set, input low-resolution image ILRWith high-definition picture IHR, then
The reconstruction image of the reconstruction network of each dense network module is compared with the input picture of convolutional neural networks model respectively to be obtained
Several loss functions of convolutional neural networks, are embodied as：
Wherein, fi(w,b,ILR) predicting the outcome for network is rebuild for i-th, w and b are respectively the convolution mask ginseng in neutral net
Number and offset parameter；
E) Adam optimized algorithms are utilized, iterative obtains convolutional neural networks model parameter w and b；Form low-resolution image
The network mapping between high-definition picture；
F) convolutional neural networks the model parameter w and b obtained using training, high-resolution is reconstructed into by the low-resolution image of input
Rate image, and calculate corresponding quantizating index PSNR and SSIM.
2. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
Stating the multiplication factor used in step a) includes 2 times, 3 times and 4 times, forms multiple dimensioned training image collection.
3. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
Low resolution images and high resolution graphics image set of the step a) using the ImageNet different interpolation amplification multiples of data set generation are stated,
And form pairing training set of images (ILR, IHR)。
4. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
YCbCr space can be transformed into by stating low resolution image and high-definition picture in step a) training set of images, and utilize Y passages
Carry out Algorithm for Training.
5. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
The network structure for stating dense network module in step b) is 8 layers, and the convolution kernel size of convolutional layer is 3*3 in network structure, is swashed
The activation primitive of layer living is Regularization linear unit function.
6. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
The output characteristic figure quantity for controlling dense network module in step b) by introduced feature growth rate k is stated, in dense network mould
N-th layer output has k*n characteristic pattern in block.
7. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
The value for stating the number L of dense network module in step c) is 8.
8. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
It is the learning rate that each parameter is dynamically adjusted using the single order moments estimation and second order moments estimation of gradient to state Adam algorithms in step e).
9. a kind of image super-resolution method based on intensive connection network according to claim 1, it is characterised in that：Institute
State that the two indexs of PSNR and SSIM in step f) are bigger, represent that the difference between reconstruction image and original high-resolution image is got over
It is small.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710193665.2A CN106991646B (en) | 2017-03-28 | 2017-03-28 | Image super-resolution method based on dense connection network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710193665.2A CN106991646B (en) | 2017-03-28 | 2017-03-28 | Image super-resolution method based on dense connection network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106991646A true CN106991646A (en) | 2017-07-28 |
| CN106991646B CN106991646B (en) | 2020-05-26 |
Family
ID=59413032
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710193665.2A Active CN106991646B (en) | 2017-03-28 | 2017-03-28 | Image super-resolution method based on dense connection network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106991646B (en) |
Cited By (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107767413A (en) * | 2017-09-20 | 2018-03-06 | 华南理工大学 | A kind of image depth estimation method based on convolutional neural networks |
| CN107818302A (en) * | 2017-10-20 | 2018-03-20 | 中国科学院光电技术研究所 | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks |
| CN108022212A (en) * | 2017-11-24 | 2018-05-11 | 腾讯科技（深圳）有限公司 | High-resolution pictures generation method, generating means and storage medium |
| CN108182669A (en) * | 2018-01-02 | 2018-06-19 | 华南理工大学 | A kind of Super-Resolution method of the generation confrontation network based on multiple dimension of pictures |
| CN108257105A (en) * | 2018-01-29 | 2018-07-06 | 南华大学 | A kind of light stream estimation for video image and denoising combination learning depth network model |
| CN108427986A (en) * | 2018-02-26 | 2018-08-21 | 中车青岛四方机车车辆股份有限公司 | A kind of production line electrical fault prediction technique and device |
| CN108536144A (en) * | 2018-04-10 | 2018-09-14 | 上海理工大学 | A kind of paths planning method of fusion dense convolutional network and competition framework |
| CN108615222A (en) * | 2018-04-17 | 2018-10-02 | 中国矿业大学 | A kind of depth convolutional network image super-resolution system based on multipair multi-connection |
| CN108629737A (en) * | 2018-05-09 | 2018-10-09 | 复旦大学 | A method of improving jpeg format image spatial resolution |
| CN108764287A (en) * | 2018-04-24 | 2018-11-06 | 东南大学 | Object detection method and system based on deep learning and grouping convolution |
| CN108765291A (en) * | 2018-05-29 | 2018-11-06 | 天津大学 | Super resolution ratio reconstruction method based on dense neural network and two-parameter loss function |
| CN108765290A (en) * | 2018-05-29 | 2018-11-06 | 天津大学 | A kind of super resolution ratio reconstruction method based on improved dense convolutional neural networks |
| CN108805166A (en) * | 2018-05-03 | 2018-11-13 | 全球能源互联网研究院有限公司 | It is a kind of to establish image classification neural network model and image classification method, device |
| CN108830211A (en) * | 2018-06-11 | 2018-11-16 | 厦门中控智慧信息技术有限公司 | Face identification method and Related product based on deep learning |
| CN108897045A (en) * | 2018-08-28 | 2018-11-27 | 中国石油天然气股份有限公司 | Deep learning model training method and seismic data noise attenuation method, device and equipment |
| CN109035184A (en) * | 2018-06-08 | 2018-12-18 | 西北工业大学 | A kind of intensive connection method based on the deformable convolution of unit |
| CN109064405A (en) * | 2018-08-23 | 2018-12-21 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of multi-scale image super-resolution method based on dual path network |
| CN109064407A (en) * | 2018-09-13 | 2018-12-21 | 武汉大学 | Intensive connection network image super-resolution method based on multi-layer perception (MLP) layer |
| CN109146788A (en) * | 2018-08-16 | 2019-01-04 | 广州视源电子科技股份有限公司 | Super-resolution image reconstruction method and device based on deep learning |
| CN109345476A (en) * | 2018-09-19 | 2019-02-15 | 南昌工程学院 | High spectrum image super resolution ratio reconstruction method and device based on depth residual error network |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
| CN109544457A (en) * | 2018-12-04 | 2019-03-29 | 电子科技大学 | Image super-resolution method, storage medium and terminal based on fine and close link neural network |
| CN109598197A (en) * | 2018-10-31 | 2019-04-09 | 大连大学 | The design method of hourglass model based on intensive link block |
| CN109741260A (en) * | 2018-12-29 | 2019-05-10 | 天津大学 | A kind of efficient super-resolution method based on depth back projection network |
| CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
| CN109816592A (en) * | 2019-01-26 | 2019-05-28 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
| CN109903228A (en) * | 2019-02-28 | 2019-06-18 | 合肥工业大学 | A kind of image super-resolution rebuilding method based on convolutional neural networks |
| CN109919840A (en) * | 2019-01-21 | 2019-06-21 | 南京航空航天大学 | Image super-resolution rebuilding method based on dense feature converged network |
| CN109949332A (en) * | 2017-12-20 | 2019-06-28 | 北京京东尚科信息技术有限公司 | Method and apparatus for handling image |
| CN109949223A (en) * | 2019-02-25 | 2019-06-28 | 天津大学 | Image super-resolution reconstructing method based on the dense connection of deconvolution |
| CN109978003A (en) * | 2019-02-21 | 2019-07-05 | 上海理工大学 | Image classification method based on intensive connection residual error network |
| CN109993109A (en) * | 2019-03-29 | 2019-07-09 | 成都信息工程大学 | Image character recognition method |
| CN110378799A (en) * | 2019-07-16 | 2019-10-25 | 东北大学 | Aluminium oxide comprehensive production index decision-making technique based on multiple dimensioned depth convolutional network |
| CN110415170A (en) * | 2019-06-24 | 2019-11-05 | 武汉大学 | A kind of image super-resolution method based on multiple dimensioned attention convolutional neural networks |
| 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 |
| CN110740350A (en) * | 2019-10-31 | 2020-01-31 | 北京金山云网络技术有限公司 | Image processing method, image processing device, terminal equipment and computer readable storage medium |
| CN110738231A (en) * | 2019-07-25 | 2020-01-31 | 太原理工大学 | Method for classifying mammary gland X-ray images by improving S-DNet neural network model |
| NL2022758B1 (en) * | 2018-07-27 | 2020-01-31 | Univ Xuzhou Technology | Image Super-resolution Reconstruction Method Based on Multi-scale Generative Adversarial Network |
| CN110782396A (en) * | 2019-11-25 | 2020-02-11 | 武汉大学 | Light-weight image super-resolution reconstruction network and reconstruction method |
| CN110827963A (en) * | 2019-11-06 | 2020-02-21 | 杭州迪英加科技有限公司 | Semantic segmentation method for pathological image and electronic equipment |
| US10600167B2 (en) | 2017-01-18 | 2020-03-24 | Nvidia Corporation | Performing spatiotemporal filtering |
| CN110956575A (en) * | 2018-09-26 | 2020-04-03 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| CN111052173A (en) * | 2017-07-31 | 2020-04-21 | 巴斯德研究所 | Method, apparatus and computer program for improving reconstruction of dense super-resolution images from diffraction limited images acquired from single molecule positioning microscopy |
| CN111080688A (en) * | 2019-12-25 | 2020-04-28 | 左一帆 | Depth map enhancement method based on depth convolution neural network |
| CN111179314A (en) * | 2019-12-30 | 2020-05-19 | 北京工业大学 | Target tracking method based on residual dense twin network |
| CN111194458A (en) * | 2017-10-11 | 2020-05-22 | 高通股份有限公司 | Image signal processor for processing image |
| CN111402138A (en) * | 2020-03-24 | 2020-07-10 | 天津城建大学 | Image super-resolution reconstruction method of supervised convolutional neural network based on multi-scale feature extraction fusion |
| CN111553861A (en) * | 2020-04-29 | 2020-08-18 | 苏州大学 | Image super-resolution reconstruction method, device and equipment and readable storage medium |
| CN112116526A (en) * | 2019-06-19 | 2020-12-22 | 中国石油化工股份有限公司 | Torch smoke image super-resolution method based on deep convolutional neural network |
| CN112150360A (en) * | 2020-09-16 | 2020-12-29 | 北京工业大学 | IVUS image super-resolution reconstruction method based on dense residual error network |
| CN112866803A (en) * | 2019-11-28 | 2021-05-28 | 三星电子株式会社 | Electronic device and control method thereof |
| CN112868033A (en) * | 2018-10-01 | 2021-05-28 | 谷歌有限责任公司 | System and method for providing machine learning model with adjustable computational requirements |
| CN112907446A (en) * | 2021-02-07 | 2021-06-04 | 电子科技大学 | Image super-resolution reconstruction method based on packet connection network |
| CN113222823A (en) * | 2021-06-02 | 2021-08-06 | 国网湖南省电力有限公司 | Hyperspectral image super-resolution method based on mixed attention network fusion |
| 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 |
| CN113538235A (en) * | 2021-06-30 | 2021-10-22 | 北京百度网讯科技有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
| US11551333B2 (en) * | 2017-12-20 | 2023-01-10 | Huawei Technologies Co., Ltd. | Image reconstruction method and device |
| US11593916B2 (en) | 2018-04-04 | 2023-02-28 | Huawei Technologies Co., Ltd. | Image super-resolution method and apparatus |
| US11836951B2 (en) | 2017-10-11 | 2023-12-05 | Qualcomm Incorporated | Image signal processor for processing images |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016019484A1 (en) * | 2014-08-08 | 2016-02-11 | Xiaoou Tang | An apparatus and a method for providing super-resolution of a low-resolution image |
| CN106204449A (en) * | 2016-07-06 | 2016-12-07 | 安徽工业大学 | A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network |
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
-
2017
- 2017-03-28 CN CN201710193665.2A patent/CN106991646B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016019484A1 (en) * | 2014-08-08 | 2016-02-11 | Xiaoou Tang | An apparatus and a method for providing super-resolution of a low-resolution image |
| CN106204449A (en) * | 2016-07-06 | 2016-12-07 | 安徽工业大学 | A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network |
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
Non-Patent Citations (3)
| Title |
|---|
| 汪慧兰等: "自适应的归一化卷积超分辨率重建算法研究", 《计算机工程与应用》 * |
| 浦剑等: "超分辨率算法研究综述", 《山东大学学报（工学版）》 * |
| 罗鸣威等: "变焦序列图像超分辨路重建算法研究", 《南京大学学报（自然科学）》 * |
Cited By (81)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11113800B2 (en) | 2017-01-18 | 2021-09-07 | Nvidia Corporation | Filtering image data using a neural network |
| US10600167B2 (en) | 2017-01-18 | 2020-03-24 | Nvidia Corporation | Performing spatiotemporal filtering |
| CN111052173A (en) * | 2017-07-31 | 2020-04-21 | 巴斯德研究所 | Method, apparatus and computer program for improving reconstruction of dense super-resolution images from diffraction limited images acquired from single molecule positioning microscopy |
| CN111052173B (en) * | 2017-07-31 | 2023-08-22 | 巴斯德研究所 | Method, apparatus and computer program for improving the reconstruction of dense super resolution images from diffraction limited images acquired by single molecule localization microscopy |
| CN107767413A (en) * | 2017-09-20 | 2018-03-06 | 华南理工大学 | A kind of image depth estimation method based on convolutional neural networks |
| CN107767413B (en) * | 2017-09-20 | 2020-02-18 | 华南理工大学 | Image depth estimation method based on convolutional neural network |
| CN111194458A (en) * | 2017-10-11 | 2020-05-22 | 高通股份有限公司 | Image signal processor for processing image |
| CN111194458B (en) * | 2017-10-11 | 2023-10-13 | 高通股份有限公司 | Image signal processor for processing images |
| US11836951B2 (en) | 2017-10-11 | 2023-12-05 | Qualcomm Incorporated | Image signal processor for processing images |
| CN107818302A (en) * | 2017-10-20 | 2018-03-20 | 中国科学院光电技术研究所 | Non-rigid multiple dimensioned object detecting method based on convolutional neural networks |
| US11373275B2 (en) | 2017-11-24 | 2022-06-28 | Tencent Technology (Shenzhen) Company Limited | Method for generating high-resolution picture, computer device, and storage medium |
| CN108022212A (en) * | 2017-11-24 | 2018-05-11 | 腾讯科技（深圳）有限公司 | High-resolution pictures generation method, generating means and storage medium |
| CN109949332A (en) * | 2017-12-20 | 2019-06-28 | 北京京东尚科信息技术有限公司 | Method and apparatus for handling image |
| US11551333B2 (en) * | 2017-12-20 | 2023-01-10 | Huawei Technologies Co., Ltd. | Image reconstruction method and device |
| CN108182669A (en) * | 2018-01-02 | 2018-06-19 | 华南理工大学 | A kind of Super-Resolution method of the generation confrontation network based on multiple dimension of pictures |
| CN108257105A (en) * | 2018-01-29 | 2018-07-06 | 南华大学 | A kind of light stream estimation for video image and denoising combination learning depth network model |
| CN108427986A (en) * | 2018-02-26 | 2018-08-21 | 中车青岛四方机车车辆股份有限公司 | A kind of production line electrical fault prediction technique and device |
| US11593916B2 (en) | 2018-04-04 | 2023-02-28 | Huawei Technologies Co., Ltd. | Image super-resolution method and apparatus |
| CN108536144A (en) * | 2018-04-10 | 2018-09-14 | 上海理工大学 | A kind of paths planning method of fusion dense convolutional network and competition framework |
| CN108615222A (en) * | 2018-04-17 | 2018-10-02 | 中国矿业大学 | A kind of depth convolutional network image super-resolution system based on multipair multi-connection |
| CN108764287B (en) * | 2018-04-24 | 2021-11-16 | 东南大学 | Target detection method and system based on deep learning and packet convolution |
| CN108764287A (en) * | 2018-04-24 | 2018-11-06 | 东南大学 | Object detection method and system based on deep learning and grouping convolution |
| CN108805166A (en) * | 2018-05-03 | 2018-11-13 | 全球能源互联网研究院有限公司 | It is a kind of to establish image classification neural network model and image classification method, device |
| CN108629737A (en) * | 2018-05-09 | 2018-10-09 | 复旦大学 | A method of improving jpeg format image spatial resolution |
| CN108629737B (en) * | 2018-05-09 | 2022-11-18 | 复旦大学 | Method for improving JPEG format image space resolution |
| CN108765291A (en) * | 2018-05-29 | 2018-11-06 | 天津大学 | Super resolution ratio reconstruction method based on dense neural network and two-parameter loss function |
| CN108765290A (en) * | 2018-05-29 | 2018-11-06 | 天津大学 | A kind of super resolution ratio reconstruction method based on improved dense convolutional neural networks |
| CN109035184A (en) * | 2018-06-08 | 2018-12-18 | 西北工业大学 | A kind of intensive connection method based on the deformable convolution of unit |
| CN108830211A (en) * | 2018-06-11 | 2018-11-16 | 厦门中控智慧信息技术有限公司 | Face identification method and Related product based on deep learning |
| NL2022758B1 (en) * | 2018-07-27 | 2020-01-31 | Univ Xuzhou Technology | Image Super-resolution Reconstruction Method Based on Multi-scale Generative Adversarial Network |
| CN109146788A (en) * | 2018-08-16 | 2019-01-04 | 广州视源电子科技股份有限公司 | Super-resolution image reconstruction method and device based on deep learning |
| CN109064405A (en) * | 2018-08-23 | 2018-12-21 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of multi-scale image super-resolution method based on dual path network |
| CN108897045A (en) * | 2018-08-28 | 2018-11-27 | 中国石油天然气股份有限公司 | Deep learning model training method and seismic data noise attenuation method, device and equipment |
| CN109064407B (en) * | 2018-09-13 | 2023-05-05 | 武汉大学 | Dense connection network image super-resolution method based on multi-layer perceptron layers |
| CN109064407A (en) * | 2018-09-13 | 2018-12-21 | 武汉大学 | Intensive connection network image super-resolution method based on multi-layer perception (MLP) layer |
| CN109345476A (en) * | 2018-09-19 | 2019-02-15 | 南昌工程学院 | High spectrum image super resolution ratio reconstruction method and device based on depth residual error network |
| CN110956575B (en) * | 2018-09-26 | 2022-04-12 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| US11842267B2 (en) | 2018-09-26 | 2023-12-12 | Boe Technology Group Co., Ltd. | Computer-implemented method using a neural network comprising down-sampling segment, densely connected segment, and up-sampling segment, apparatus, and computer-program product |
| CN110956575A (en) * | 2018-09-26 | 2020-04-03 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| CN112868033A (en) * | 2018-10-01 | 2021-05-28 | 谷歌有限责任公司 | System and method for providing machine learning model with adjustable computational requirements |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
| CN109598197A (en) * | 2018-10-31 | 2019-04-09 | 大连大学 | The design method of hourglass model based on intensive link block |
| CN109544457A (en) * | 2018-12-04 | 2019-03-29 | 电子科技大学 | Image super-resolution method, storage medium and terminal based on fine and close link neural network |
| CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
| CN109741260A (en) * | 2018-12-29 | 2019-05-10 | 天津大学 | A kind of efficient super-resolution method based on depth back projection network |
| CN109741260B (en) * | 2018-12-29 | 2023-05-12 | 天津大学 | Efficient super-resolution method based on depth back projection network |
| CN109919840A (en) * | 2019-01-21 | 2019-06-21 | 南京航空航天大学 | Image super-resolution rebuilding method based on dense feature converged network |
| CN109816592A (en) * | 2019-01-26 | 2019-05-28 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
| CN109816592B (en) * | 2019-01-26 | 2022-05-13 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
| CN109978003A (en) * | 2019-02-21 | 2019-07-05 | 上海理工大学 | Image classification method based on intensive connection residual error network |
| CN109949223A (en) * | 2019-02-25 | 2019-06-28 | 天津大学 | Image super-resolution reconstructing method based on the dense connection of deconvolution |
| CN109949223B (en) * | 2019-02-25 | 2023-06-20 | 天津大学 | Image super-resolution reconstruction method based on deconvolution dense connection |
| CN109903228B (en) * | 2019-02-28 | 2023-03-24 | 合肥工业大学 | Image super-resolution reconstruction method based on convolutional neural network |
| CN109903228A (en) * | 2019-02-28 | 2019-06-18 | 合肥工业大学 | A kind of image super-resolution rebuilding method based on convolutional neural networks |
| CN109993109A (en) * | 2019-03-29 | 2019-07-09 | 成都信息工程大学 | Image character recognition method |
| CN112116526A (en) * | 2019-06-19 | 2020-12-22 | 中国石油化工股份有限公司 | Torch smoke image super-resolution method based on deep convolutional neural network |
| CN110415170A (en) * | 2019-06-24 | 2019-11-05 | 武汉大学 | A kind of image super-resolution method based on multiple dimensioned attention convolutional neural networks |
| CN110415170B (en) * | 2019-06-24 | 2022-12-16 | 武汉大学 | Image super-resolution method based on multi-scale attention convolution neural network |
| CN110378799B (en) * | 2019-07-16 | 2022-07-12 | 东北大学 | Alumina comprehensive production index decision method based on multi-scale deep convolution network |
| CN110378799A (en) * | 2019-07-16 | 2019-10-25 | 东北大学 | Aluminium oxide comprehensive production index decision-making technique based on multiple dimensioned depth convolutional network |
| CN110738231A (en) * | 2019-07-25 | 2020-01-31 | 太原理工大学 | Method for classifying mammary gland X-ray images by improving S-DNet neural network model |
| 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 |
| CN110740350A (en) * | 2019-10-31 | 2020-01-31 | 北京金山云网络技术有限公司 | Image processing method, image processing device, terminal equipment and computer readable storage medium |
| CN110827963A (en) * | 2019-11-06 | 2020-02-21 | 杭州迪英加科技有限公司 | Semantic segmentation method for pathological image and electronic equipment |
| CN110782396A (en) * | 2019-11-25 | 2020-02-11 | 武汉大学 | Light-weight image super-resolution reconstruction network and reconstruction method |
| CN110782396B (en) * | 2019-11-25 | 2023-03-28 | 武汉大学 | Light-weight image super-resolution reconstruction network and reconstruction method |
| CN112866803A (en) * | 2019-11-28 | 2021-05-28 | 三星电子株式会社 | Electronic device and control method thereof |
| CN112866803B (en) * | 2019-11-28 | 2023-07-21 | 三星电子株式会社 | Electronic device and control method thereof |
| CN111080688A (en) * | 2019-12-25 | 2020-04-28 | 左一帆 | Depth map enhancement method based on depth convolution neural network |
| CN111179314A (en) * | 2019-12-30 | 2020-05-19 | 北京工业大学 | Target tracking method based on residual dense twin network |
| CN111179314B (en) * | 2019-12-30 | 2023-05-02 | 北京工业大学 | Target tracking method based on residual intensive twin network |
| CN111402138A (en) * | 2020-03-24 | 2020-07-10 | 天津城建大学 | Image super-resolution reconstruction method of supervised convolutional neural network based on multi-scale feature extraction fusion |
| CN111553861A (en) * | 2020-04-29 | 2020-08-18 | 苏州大学 | Image super-resolution reconstruction method, device and equipment and readable storage medium |
| CN111553861B (en) * | 2020-04-29 | 2023-11-24 | 苏州大学 | Image super-resolution reconstruction method, device, equipment and readable storage medium |
| CN112150360A (en) * | 2020-09-16 | 2020-12-29 | 北京工业大学 | IVUS image super-resolution reconstruction method based on dense residual error network |
| CN112907446A (en) * | 2021-02-07 | 2021-06-04 | 电子科技大学 | Image super-resolution reconstruction method based on packet connection network |
| CN113222823A (en) * | 2021-06-02 | 2021-08-06 | 国网湖南省电力有限公司 | Hyperspectral image super-resolution method based on mixed attention network fusion |
| CN113538235A (en) * | 2021-06-30 | 2021-10-22 | 北京百度网讯科技有限公司 | Training method and device of image processing model, electronic equipment and storage medium |
| CN113538235B (en) * | 2021-06-30 | 2024-01-09 | 北京百度网讯科技有限公司 | Training method and device for image processing model, electronic equipment and storage medium |
| 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 |
|---|---|
| CN106991646B (en) | 2020-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106991646A (en) | A kind of image super-resolution method based on intensive connection network | |
| CN111161150B (en) | Image super-resolution reconstruction method based on multi-scale attention cascade network | |
| CN110458750B (en) | Unsupervised image style migration method based on dual learning | |
| CN109903223B (en) | Image super-resolution method based on dense connection network and generation type countermeasure network | |
| CN108510485B (en) | Non-reference image quality evaluation method based on convolutional neural network | |
| CN110197468A (en) | A kind of single image Super-resolution Reconstruction algorithm based on multiple dimensioned residual error learning network | |
| CN106683067A (en) | Deep learning super-resolution reconstruction method based on residual sub-images | |
| CN109767386A (en) | A kind of rapid image super resolution ratio reconstruction method based on deep learning | |
| CN109389556A (en) | The multiple dimensioned empty convolutional neural networks ultra-resolution ratio reconstructing method of one kind and device | |
| CN107507134A (en) | Super-resolution method based on convolutional neural networks | |
| CN108765296A (en) | A kind of image super-resolution rebuilding method based on recurrence residual error attention network | |
| CN107578377A (en) | A kind of super-resolution image reconstruction method and system based on deep learning | |
| CN108537733A (en) | Super resolution ratio reconstruction method based on multipath depth convolutional neural networks | |
| CN107092960A (en) | A kind of improved parallel channel convolutional neural networks training method | |
| CN111429347A (en) | Image super-resolution reconstruction method and device and computer-readable storage medium | |
| CN106228512A (en) | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method | |
| CN108830790A (en) | It is a kind of based on the fast video super resolution ratio reconstruction method for simplifying convolutional neural networks | |
| CN109544457A (en) | Image super-resolution method, storage medium and terminal based on fine and close link neural network | |
| CN111861906A (en) | Pavement crack image virtual augmentation model establishment and image virtual augmentation method | |
| CN113837946B (en) | Lightweight image super-resolution reconstruction method based on progressive distillation network | |
| CN112215755A (en) | Image super-resolution reconstruction method based on back projection attention network | |
| CN113538234A (en) | Remote sensing image super-resolution reconstruction method based on lightweight generation model | |
| CN112270366B (en) | Micro target detection method based on self-adaptive multi-feature fusion | |
| CN111861886A (en) | Image super-resolution reconstruction method based on multi-scale feedback network | |
| CN112017116B (en) | Image super-resolution reconstruction network based on asymmetric convolution and construction method thereof |
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 |