CN108022213A - Video super-resolution algorithm for reconstructing based on generation confrontation network - Google Patents
Video super-resolution algorithm for reconstructing based on generation confrontation network Download PDFInfo
- Publication number
- CN108022213A CN108022213A CN201711224386.4A CN201711224386A CN108022213A CN 108022213 A CN108022213 A CN 108022213A CN 201711224386 A CN201711224386 A CN 201711224386A CN 108022213 A CN108022213 A CN 108022213A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- layers
- resolution
- network
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 15
- 230000004913 activation Effects 0.000 claims description 9
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 6
- 230000003287 optical effect Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000009432 framing Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 description 4
- 238000010606 normalization Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000003412 degenerative effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of video super-resolution algorithm for reconstructing based on generation confrontation network, this method comprises the following steps：Utilize distinguished number modelWith generating algorithm modelIt is trained in an alternating fashion, continues to optimize training result, solves binary problem of game.Compared with prior art, the present invention generates the texture information in image by the generation network containing residual error network, and the problem of robustness is low is rebuild to realize that the prior art is unapproachable.
Description
Technical field
The present invention relates to video reconstruction technical field, more particularly to a kind of video reconstruction side based on generation confrontation network
Method.
Background technology
Video super-resolution reconstruction is an important subject in computer vision field.The super-resolution of video sequence
Rate reconstruction utilizes the offset of the sub-pix between image or high frequency complementary characteristic and specific priori to know from image degradation model
Knowledge solves degradation model.Wherein mainly include iterative backprojection method (Iterative Back Projection, IBP),
Projections onto convex sets (Projection Onto Convex Sets, POCS) and probability theory method three classes.Irani is equal to 1991
The iterative backprojection method of proposition analog image degenerative process by way of back projection gradually reduces reconstruction error, completes oversubscription
Resolution is rebuild；The projections onto convex sets of the propositions such as Stark H can be by being limited in some closures by the feasible solution of reconstruction image
Method in convex set obtains final disaggregation by iteration several times；What chultz was started carries out Super-resolution reconstruction using probability theory
The beginning built, the Optimized model that super-resolution rebuilding is converted into statistics are solved.The it is proposed such as China is grasped by Chinese scholar Soviet Union
《Poisson-ML/MAP methods based on Markov constraints》.In terms of the Study of Registration of sequence image and video, Wang Suyu
Etc. developing《A kind of affine piece of method for registering of multiscale least squares estimation》, for dynamic video superresolution processing etc..
What traditional video super-resolution processing method was generally handled is less amplification factor, when the amplification factor of image
When more than 4 times, it is easy to the result made seems excessively smooth, and lacks the sense of reality in some details, robustness compared with
It is low.Therefore the present invention generates the texture information in image using generation confrontation network, increases the sense of reality in details, and effect
It is superior.
The content of the invention
The present invention proposes a kind of video super-resolution algorithm for reconstructing based on generation confrontation network, utilizes generating algorithm mould
Type can learn to generate the image highly similar to real image, so as to be difficult to classify by D；It is alternately performed and utilizes differentiation
Algorithm model distinguishes super-resolution image and true picture to train.
A kind of video super-resolution algorithm for reconstructing based on generation confrontation network of the present invention, this method include following step
Suddenly：
Utilize distinguished number modelWith generating algorithm modelIt is trained in an alternating fashion, continues to optimize training
As a result, to solve binary problem of game：
Wherein, θG、θDThe training parameter of maker G and arbiter D, I are represented respectivelyLRRepresent low resolution figure to be reconstructed
Picture,I HRRepresent high-definition picture,Represent the probability distribution of true picture,Represent the probability distribution of generation image；
Wherein：
The generating algorithm model specifically includes following processing：
(1) pre-process, video is subjected to framing, takes continuous 5 frame Y successivelyt-2,Yt-1,Yt,Yt+1,Yt+2, to every two field picture
Taken exercises compensation based on intermediate frame, image is had into successional light stream using classical optical flow algorithm Horn-Schunck algorithms
Field E is expressed as follows：
In above formula, it is assumed that a certain pixel P=(x, y) on imageTIt is I (x, y, t) in the gray value of moment t, thenRepresentative image gray scale is along x respectively .y, the partial derivative on t axis,
Represent x, the light stream on y directions,U, partial derivatives of the v on x, y-axis are represented respectively.Section 1 is
Data item；Section 2 is smooth item；α is the smooth item factor, it determines view data and the confidence level of smoothness constraint；
(2) after the image of pretreatment is merged by fused layer, several identical residual blocks is chosen and are trained,
Each residual block includes the convolution kernel of 3*3 and the convolutional network of 64 characteristic patterns, and batch regularization layer, correct linear unit conduct
Activation primitive：
Fi(Y)=max (Wi*Fi-1(Y))+Bi,0)
Wherein, i represents i-th of convolutional layer, and i-1 represents last layer network layer, Fi-1(Y), Fi(Y) i layers are represented, i-1 layers
Output, Wi, BiI layers of weighting parameter and constant parameter is represented respectively；
During decoded, the resolution ratio of input picture is improved using two layers of sub-pixel convolutional layer)；
Distinguished number model includes the convolutional layer of 6 filter kernels, i.e., chosen in VGG networks from 64 to 256 in
If the dried layer characteristic pattern of the 2* factors of core, after two full Dense layers of articulamentums, obtained using Sigmoid shapes activation primitive
Obtain the probability that this image is the super-resolution image from actual high-definition picture or generation.
Compared with prior art, the present invention generates the texture information in image by the generation network containing residual error network,
The texture information of video sequence frame can be more caught, improves robustness.
Brief description of the drawings
Fig. 1 is the video super-resolution algorithm for reconstructing overall flow figure based on generation confrontation network of the present invention；
Fig. 2 is generating algorithm model schematic；
Fig. 3 is distinguished number model schematic.
Embodiment
Embodiments of the present invention are described in further detail below in conjunction with attached drawing.
The video super-resolution algorithm for reconstructing based on generation confrontation network of the present invention is made of i.e. two parts algorithm model
Distinguished number modelWith generating algorithm modelUtilize distinguished number modelWith generating algorithm modelWith alternate
Mode is trained, and continues to optimize training result, to solve binary problem of game：
Wherein, θG、θDThe training parameter of maker G and arbiter D, I are represented respectivelyLRRepresent low resolution figure to be reconstructed
Picture,I HRRepresent high-definition picture,Represent the probability distribution of true picture,Represent the probability distribution of generation image.
Above-mentioned formula represents to allow to train a maker G, its target is to deceive an arbiter D, distinguished number model
Training is used to distinguish super-resolution image and true picture.In this way, generating algorithm model can learn generation with
The highly similar image of real image, so as to be difficult to classify by D.
As shown in Fig. 2, be the present invention generating algorithm model schematic, wherein Yt-2,Yt-1,Yt,Yt+1,Yt+2Represent respectively
Continuous five continuous low-resolution frames by motion compensation (optical flow estimation) centered on time t,
Conv refers to convolutional layer, and concatenate refers to Fusion Features layer, and BN refers to batch regularization layer (batch-normalization
Layers), correct linear unit (Rectified linear unit, ReLU) and be used as activation primitive, Elementwise SUM
Refer to the superposition of different layers, skip connection refer to jump connection, and PixelShuffler represents sub-pixel convolutional layer.The generation
Algorithm model specifically includes following processing：
(1) pre-process, video is subjected to framing, takes continuous 5 frame Y successivelyt-2,Yt-1,Yt,Yt+1,Yt+2, to every two field picture
Taken exercises compensation based on intermediate frame, image is had into successional light stream using classical optical flow algorithm Horn-Schunck algorithms
Field E is expressed as follows：
In above formula, it is assumed that a certain pixel P=(x, y) on imageTIt is I (x, y, t) in the gray value of moment t, then Representative image gray scale is along x respectively, y, the partial derivative on t axis,Generation
Light stream on table x, y directions,U, partial derivatives of the v on x, y-axis are represented respectively.Section 1 is number
According to item；Section 2 is smooth item；α is the smooth item factor, it determines view data and the confidence level of smoothness constraint；
(2) after the image of pretreatment is merged by fused layer (concat layers), several identical residual errors are chosen
Block is trained, and each residual block includes the convolution kernel of 3*3 and the convolutional network of 64 characteristic patterns, batch regularization layer
(batch-normalization layers, BN), correct linear unit (Rectified linear unit, ReLU) conduct
Activation primitive：
Fi(Y)=max (Wi*Fi-1(Y))+Bi,0)
Wherein, i represents i-th of convolutional layer, and i-1 represents last layer network layer, Fi-1(Y), Fi(Y) i layers are represented, i-1 layers
Output, Wi、BiI layers of weighting parameter and constant parameter is represented respectively.
During decoded, using two layers sub-pixel convolutional layer (i.e. step-length be fraction convolutional layer) improve input figure
The resolution ratio of picture.
As shown in figure 3, for the distinguished number model schematic of the present invention, wherein Input is input picture, and Conv is convolution
Layer, for Leakly ReLU to correct activation primitive, BN refers to batch regularization layer (batch-normalization layers),
Dense is full articulamentum, and Sigmoid is grader.Its details is as follows：
The distinguished number model of the present invention is to distinguish actual high-definition picture and the super-resolution image of generation, abide by
Follow the architecture guide of Radford et al. summaries.And use amending unit (Leaky Rectified linear unit,
LeakyReLU) activate, to avoid the maximum pond (maxpooling) in whole model.This distinguished number includes 6 characteristic patterns
The convolutional layer (n64, n64, n128, n128, n256, n256, n512, n512) of more and more filter kernels, i.e., in VGG nets
If choosing the dried layer characteristic pattern of the 2* factors of a kernel from 64 to 256 in network, by Dense layers (1024,1 of two full articulamentums
Characteristic pattern number is represented respectively) after, using Sigmoid shapes activation primitive come to obtain this image be to come from actual high-definition picture
Or the probability of the super-resolution image of generation.
The training loss processing details of the present invention is as follows：
Optimized by training maker G and arbiter D and perceive loss function lSR, work as lSRDuring loss reduction, network training
Effect is best, so defining lSRIt is most important for the performance of generation network.Although lSRMSE is typically based on to model, but
Improved on the basis of Johnson et al. and Bruna et al., and devise a new loss function, perceived for assessing
The loss of correlated characteristic, by lSRIt is established as content loss componentComponent is lost with antagonismWeighted sum：
Wherein, low-resolution image ILR(W*H) the r times of high-definition picture I up-sampled is passed throughHR(rW*rH) picture
Plain level MSE costing bio disturbances are as follows：
This is to obtain the most widely used optimization aim of super-resolution image, and the state-of-the-art method of many of which depends on
It.Although however, realizing extra high PSNR, the solution of MSE optimization problems often lacks high-frequency content, has
The texture information of excess smoothness.Then, in 19 layers of VGG networks based on the pre-training described in Simonyan and Zisserman
ReLU active coatings come define VGG loss, useBefore representing in VGG19 networks by i-th of maximum pond layer, jth is secondary
The characteristic pattern that convolutional layer (after activation) obtains.So VGG losses are defined as reconstruction image by inventionAnd reference picture
IHRCharacter representation between Euclidean distance：
Wherein, Wi,jAnd Hi,jDescribe the size of each characteristic pattern in VGG networks.
In addition to the content loss described so far, the generating portion of GAN is also added to perception loss by the present invention
In, this encourages generating algorithm model by attempting to cheat arbiter network to support the solution party in many aspects of natural image
Case.Based on arbiter on all training samplesProbability define generational loss
It is reconstruction imageBe nature HR images probability it is special in order to obtain more preferable gradient
Property, minimizeRather than
Preferred forms：
The bicubic down-sampling that decimation factor r=4 is carried out to original image obtains low-resolution image.In optimization process,
Using Stochastic Optimization Algorithms, β=0.9.During training SRGAN, in order to avoid generation confrontation network is absorbed in local optimum, adopt
Initialization by the use of the trained residual error network block based on MSE as generation model, wherein residual error network block need to be with 10-4 and 10-
6 learning rates update iterative process.The variable of all SRGAN should all be trained to carry out 105 iteration with the learning rate of 10-4.
More newly-generated network and differentiation network successively, it is equivalent to k=1.There are 4 identical (B=16) residual blocks in maker network.
Table 1, experiment algorithm comparing result
The present invention is used to train and test using the Myanmar video sequences in a disclosed video database.Burma
Video sequence is used to training and testing inventive algorithm.Experimental result is surveyed using the average PSNR and SSIM values of test frame as performance
Amount.Experiment is as shown in table 1, is demonstrated by the comparing result with algorithms of different, which improves experimental result, and can more catch
The texture information of video sequence frame is caught, improves robustness.
Claims (1)
1. a kind of video super-resolution algorithm for reconstructing based on generation confrontation network, it is characterised in that this method includes following step
Suddenly：
Utilize distinguished number modelWith generating algorithm modelIt is trained in an alternating fashion, continues to optimize training result,
To solve binary problem of game：
Wherein, θG、θDThe training parameter of maker G and arbiter D, I are represented respectivelyLRRepresent low-resolution image to be reconstructed,
IHRRepresent high-definition picture,Represent the probability distribution of true picture,Represent the probability distribution of generation image；Its
In：
The generating algorithm model specifically includes following processing：
(1) pre-process, video is subjected to framing, takes continuous 5 frame Y successivelyt-2,Yt-1,Yt,Yt+1,Yt+2, every two field picture is based on
Intermediate frame is taken exercises compensation, and image is had successional optical flow field E using classical optical flow algorithm Horn-Schunck algorithms
It is expressed as follows：
In above formula, it is assumed that a certain pixel P=(x, y) on imageTIt is I (x, y, t) in the gray value of moment t, thenRepresentative image gray scale is along x respectively .y, the partial derivative on t axis,
Represent x, the light stream on y directions,U, partial derivatives of the v on x, y-axis are represented respectively.Section 1 is
Data item；Section 2 is smooth item；α is the smooth item factor, it determines view data and the confidence level of smoothness constraint；
(2) after the image of pretreatment is merged by fused layer, several identical residual blocks is chosen and are trained, it is each
Residual block includes the convolution kernel of 3*3 and the convolutional network of 64 characteristic patterns, and batch regularization layer, correct linear unit as activation
Function：
Fi(Y)=max (Wi*Fi-1(Y))+Bi,0)
Wherein, i represents i-th of convolutional layer, and i-1 represents last layer network layer, Fi-1(Y), Fi(Y) i layers of expression, i-1 layers of output,
Wi, BiI layers of weighting parameter and constant parameter is represented respectively；
During decoded, the resolution ratio of input picture is improved using two layers of sub-pixel convolutional layer)；
Distinguished number model includes the convolutional layer of 6 filter kernels, i.e., a kernel from 64 to 256 is chosen in VGG networks
If the dried layer characteristic pattern of the 2* factors, after two full Dense layers of articulamentums, this is obtained using Sigmoid shapes activation primitive
Image is the probability of the super-resolution image from actual high-definition picture or generation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711224386.4A CN108022213A (en) | 2017-11-29 | 2017-11-29 | Video super-resolution algorithm for reconstructing based on generation confrontation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711224386.4A CN108022213A (en) | 2017-11-29 | 2017-11-29 | Video super-resolution algorithm for reconstructing based on generation confrontation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108022213A true CN108022213A (en) | 2018-05-11 |
Family
ID=62077392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711224386.4A Pending CN108022213A (en) | 2017-11-29 | 2017-11-29 | Video super-resolution algorithm for reconstructing based on generation confrontation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108022213A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765295A (en) * | 2018-06-12 | 2018-11-06 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus and storage medium |
| CN108829855A (en) * | 2018-06-21 | 2018-11-16 | 山东大学 | It is worn based on the clothing that condition generates confrontation network and takes recommended method, system and medium |
| CN108898066A (en) * | 2018-06-06 | 2018-11-27 | 天津大学 | A kind of body movement detection method based on production confrontation network |
| CN109151443A (en) * | 2018-10-15 | 2019-01-04 | Oppo广东移动通信有限公司 | High degree of comfort three-dimensional video-frequency generation method, system and terminal device |
| CN109363632A (en) * | 2018-09-26 | 2019-02-22 | 北京三医智慧科技有限公司 | The deciphering method of pulse profile data and the solution read apparatus of pulse profile data |
| CN109447936A (en) * | 2018-12-21 | 2019-03-08 | 江苏师范大学 | A kind of infrared and visible light image fusion method |
| CN109544450A (en) * | 2018-11-09 | 2019-03-29 | 武汉烽火众智数字技术有限责任公司 | A kind of confrontation generates network establishing method and device, image reconstructing method and device |
| CN109785270A (en) * | 2019-01-18 | 2019-05-21 | 四川长虹电器股份有限公司 | A kind of image super-resolution method based on GAN |
| CN109859107A (en) * | 2019-02-12 | 2019-06-07 | 广东工业大学 | Remote Sensing Image Super Resolution method, apparatus, equipment and readable storage medium storing program for executing |
| CN109919058A (en) * | 2019-02-26 | 2019-06-21 | 武汉大学 | A kind of multisource video image highest priority rapid detection method based on Yolo V3 |
| CN109978804A (en) * | 2019-03-08 | 2019-07-05 | 清华大学 | Human eye sight antidote and system based on deep learning |
| CN109993702A (en) * | 2019-04-10 | 2019-07-09 | 大连民族大学 | Based on the language of the Manchus image super-resolution rebuilding method for generating confrontation network |
| CN110070935A (en) * | 2019-03-20 | 2019-07-30 | 中国科学院自动化研究所 | Medical image synthetic method, classification method and device based on confrontation neural network |
| CN110580682A (en) * | 2019-09-16 | 2019-12-17 | 电子科技大学 | Countermeasure network seismic data super-resolution reconstruction method based on optimization generation |
| CN110599421A (en) * | 2019-09-12 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Model training method, video fuzzy frame conversion method, device and storage medium |
| CN111221118A (en) * | 2020-02-26 | 2020-06-02 | 南京理工大学 | Microscopic imaging method based on phase coding single lens |
| CN111340768A (en) * | 2020-02-21 | 2020-06-26 | 之江实验室 | Multi-center effect compensation method based on PET/CT intelligent diagnosis system |
| CN112101204A (en) * | 2020-09-14 | 2020-12-18 | 北京百度网讯科技有限公司 | Training method of generative countermeasure network, image processing method, device and equipment |
| CN112711995A (en) * | 2020-12-22 | 2021-04-27 | 中南大学 | Image-based marine target identification method |
| CN113487503A (en) * | 2021-07-01 | 2021-10-08 | 安徽大学 | PET (positron emission tomography) super-resolution method for generating antagonistic network based on channel attention |
| CN109087243B (en) * | 2018-06-29 | 2023-05-05 | 中山大学 | Video super-resolution generation method for generating countermeasure network based on depth convolution |
| CN113487503B (en) * | 2021-07-01 | 2024-07-09 | 安徽大学 | PET super-resolution method for generating countermeasure network based on channel attention |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170019615A1 (en) * | 2015-07-13 | 2017-01-19 | Asustek Computer Inc. | Image processing method, non-transitory computer-readable storage medium and electrical device thereof |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN106991648A (en) * | 2017-04-10 | 2017-07-28 | 中国石油大学(华东) | A kind of image super-resolution rebuilding method based on wavelet transformation and convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107274347A (en) * | 2017-07-11 | 2017-10-20 | 福建帝视信息科技有限公司 | A kind of video super-resolution method for reconstructing based on depth residual error network |
-
2017
- 2017-11-29 CN CN201711224386.4A patent/CN108022213A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170019615A1 (en) * | 2015-07-13 | 2017-01-19 | Asustek Computer Inc. | Image processing method, non-transitory computer-readable storage medium and electrical device thereof |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN106991648A (en) * | 2017-04-10 | 2017-07-28 | 中国石油大学(华东) | A kind of image super-resolution rebuilding method based on wavelet transformation and convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107274347A (en) * | 2017-07-11 | 2017-10-20 | 福建帝视信息科技有限公司 | A kind of video super-resolution method for reconstructing based on depth residual error network |
Non-Patent Citations (2)
| Title |
|---|
| 浦煜 等: "基于互信息最大化正则的生成式对抗网络", 《现代计算机(专业版)》 * |
| 薛翠红: "图像及视频序列中人脸超分辨率重建技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108898066B (en) * | 2018-06-06 | 2022-01-04 | 天津大学 | Human motion detection method based on generating type countermeasure network |
| CN108898066A (en) * | 2018-06-06 | 2018-11-27 | 天津大学 | A kind of body movement detection method based on production confrontation network |
| CN108765295A (en) * | 2018-06-12 | 2018-11-06 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus and storage medium |
| CN108765295B (en) * | 2018-06-12 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus and storage medium |
| CN108829855A (en) * | 2018-06-21 | 2018-11-16 | 山东大学 | It is worn based on the clothing that condition generates confrontation network and takes recommended method, system and medium |
| CN109087243B (en) * | 2018-06-29 | 2023-05-05 | 中山大学 | Video super-resolution generation method for generating countermeasure network based on depth convolution |
| CN109363632A (en) * | 2018-09-26 | 2019-02-22 | 北京三医智慧科技有限公司 | The deciphering method of pulse profile data and the solution read apparatus of pulse profile data |
| CN109151443A (en) * | 2018-10-15 | 2019-01-04 | Oppo广东移动通信有限公司 | High degree of comfort three-dimensional video-frequency generation method, system and terminal device |
| CN109544450A (en) * | 2018-11-09 | 2019-03-29 | 武汉烽火众智数字技术有限责任公司 | A kind of confrontation generates network establishing method and device, image reconstructing method and device |
| CN109544450B (en) * | 2018-11-09 | 2022-08-19 | 武汉烽火众智数字技术有限责任公司 | Method and device for constructing confrontation generation network and method and device for reconstructing image |
| CN109447936A (en) * | 2018-12-21 | 2019-03-08 | 江苏师范大学 | A kind of infrared and visible light image fusion method |
| CN109785270A (en) * | 2019-01-18 | 2019-05-21 | 四川长虹电器股份有限公司 | A kind of image super-resolution method based on GAN |
| CN109859107A (en) * | 2019-02-12 | 2019-06-07 | 广东工业大学 | Remote Sensing Image Super Resolution method, apparatus, equipment and readable storage medium storing program for executing |
| CN109859107B (en) * | 2019-02-12 | 2023-04-07 | 广东工业大学 | Remote sensing image super-resolution method, device, equipment and readable storage medium |
| CN109919058A (en) * | 2019-02-26 | 2019-06-21 | 武汉大学 | A kind of multisource video image highest priority rapid detection method based on Yolo V3 |
| CN109978804A (en) * | 2019-03-08 | 2019-07-05 | 清华大学 | Human eye sight antidote and system based on deep learning |
| CN109978804B (en) * | 2019-03-08 | 2021-02-26 | 清华大学 | Human eye sight line correction method and system based on deep learning |
| CN110070935A (en) * | 2019-03-20 | 2019-07-30 | 中国科学院自动化研究所 | Medical image synthetic method, classification method and device based on confrontation neural network |
| CN110070935B (en) * | 2019-03-20 | 2021-04-30 | 中国科学院自动化研究所 | Medical image synthesis method, classification method and device based on antagonistic neural network |
| CN109993702B (en) * | 2019-04-10 | 2023-09-26 | 大连民族大学 | Full-text image super-resolution reconstruction method based on generation countermeasure network |
| CN109993702A (en) * | 2019-04-10 | 2019-07-09 | 大连民族大学 | Based on the language of the Manchus image super-resolution rebuilding method for generating confrontation network |
| CN110599421A (en) * | 2019-09-12 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Model training method, video fuzzy frame conversion method, device and storage medium |
| CN110599421B (en) * | 2019-09-12 | 2023-06-09 | 腾讯科技（深圳）有限公司 | Model training method, video fuzzy frame conversion method, device and storage medium |
| CN110580682A (en) * | 2019-09-16 | 2019-12-17 | 电子科技大学 | Countermeasure network seismic data super-resolution reconstruction method based on optimization generation |
| WO2021164495A1 (en) * | 2020-02-21 | 2021-08-26 | 之江实验室 | Multi-center effect compensation method based on pet/ct intelligent diagnosis system |
| US11715562B2 (en) | 2020-02-21 | 2023-08-01 | Zhejiang Lab | Method for multi-center effect compensation based on PET/CT intelligent diagnosis system |
| CN111340768B (en) * | 2020-02-21 | 2021-03-09 | 之江实验室 | Multi-center effect compensation method based on PET/CT intelligent diagnosis system |
| CN111340768A (en) * | 2020-02-21 | 2020-06-26 | 之江实验室 | Multi-center effect compensation method based on PET/CT intelligent diagnosis system |
| CN111221118A (en) * | 2020-02-26 | 2020-06-02 | 南京理工大学 | Microscopic imaging method based on phase coding single lens |
| CN112101204A (en) * | 2020-09-14 | 2020-12-18 | 北京百度网讯科技有限公司 | Training method of generative countermeasure network, image processing method, device and equipment |
| CN112101204B (en) * | 2020-09-14 | 2024-01-23 | 北京百度网讯科技有限公司 | Training method, image processing method, device and equipment for generating type countermeasure network |
| CN112711995A (en) * | 2020-12-22 | 2021-04-27 | 中南大学 | Image-based marine target identification method |
| CN113487503A (en) * | 2021-07-01 | 2021-10-08 | 安徽大学 | PET (positron emission tomography) super-resolution method for generating antagonistic network based on channel attention |
| CN113487503B (en) * | 2021-07-01 | 2024-07-09 | 安徽大学 | PET super-resolution method for generating countermeasure network based on channel attention |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108022213A (en) | Video super-resolution algorithm for reconstructing based on generation confrontation network | |
| Zamir et al. | Restormer: Efficient transformer for high-resolution image restoration | |
| Wang et al. | Esrgan: Enhanced super-resolution generative adversarial networks | |
| Li et al. | MDCN: Multi-scale dense cross network for image super-resolution | |
| Zhang et al. | Image restoration: From sparse and low-rank priors to deep priors [lecture notes] | |
| CN109829855B (en) | Super-resolution reconstruction method based on fusion of multi-level feature images | |
| WO2020037965A1 (en) | Method for multi-motion flow deep convolutional network model for video prediction | |
| CN109559276B (en) | Image super-resolution reconstruction method based on quality evaluation and feature statistics | |
| CN107977932A (en) | It is a kind of based on can differentiate attribute constraint generation confrontation network face image super-resolution reconstruction method | |
| CN109671023A (en) | A kind of secondary method for reconstructing of face image super-resolution | |
| CN110675321A (en) | Super-resolution image reconstruction method based on progressive depth residual error network | |
| CN111091503A (en) | Image out-of-focus blur removing method based on deep learning | |
| Fu et al. | Residual scale attention network for arbitrary scale image super-resolution | |
| CN111080591A (en) | Medical image segmentation method based on combination of coding and decoding structure and residual error module | |
| CN106157249A (en) | Based on the embedded single image super-resolution rebuilding algorithm of optical flow method and sparse neighborhood | |
| CN105513033B (en) | A kind of super resolution ratio reconstruction method that non local joint sparse indicates | |
| CN107610049A (en) | Image super-resolution method based on sparse Regularization Technique and weighting guiding filtering | |
| Guan et al. | Srdgan: learning the noise prior for super resolution with dual generative adversarial networks | |
| CN114170088A (en) | Relational reinforcement learning system and method based on graph structure data | |
| CN112184549B (en) | Super-resolution image reconstruction method based on space-time transformation technology | |
| CN116664397B (en) | TransSR-Net structured image super-resolution reconstruction method | |
| CN112163998A (en) | Single-image super-resolution analysis method matched with natural degradation conditions | |
| CN115222592A (en) | Underwater image enhancement method based on super-resolution network and U-Net network and training method of network model | |
| CN112598604A (en) | Blind face restoration method and system | |
| US7684635B2 (en) | Signal processing device, and signal processing method, and program, and recording medium |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180511 |
|
| WD01 | Invention patent application deemed withdrawn after publication |