CN109272450A - A kind of image oversubscription method based on convolutional neural networks - Google Patents
A kind of image oversubscription method based on convolutional neural networks Download PDFInfo
- Publication number
- CN109272450A CN109272450A CN201810959380.XA CN201810959380A CN109272450A CN 109272450 A CN109272450 A CN 109272450A CN 201810959380 A CN201810959380 A CN 201810959380A CN 109272450 A CN109272450 A CN 109272450A
- Authority
- CN
- China
- Prior art keywords
- feature
- image
- unit
- network
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 8
- 230000000007 visual effect Effects 0.000 abstract description 5
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 239000012634 fragment Substances 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012546 transfer 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 belongs to image-editing technology field, specially a kind of image oversubscription method based on convolutional neural networks.Convolutional neural networks include: feature extraction network, feature learning network, image reconstruction network;Method includes: to extract characteristics of image by feature extraction network;By feature learning network, learn high-definition picture feature out;Pass through image reconstruction network reconfiguration image three phases.The network structure of layered characteristic study proposed by the present invention can allow each stage in network to make full use of the characteristic information that all different levels are acquired in network, also largely remain the important feature in network, reduce characteristic loss, maximum characteristic use is realized, while avoiding the repetition and redundancy of feature.The experimental results showed that this method is generated with the high-definition picture for more meeting subjective visual quality, detail textures information true to nature is restored, has also achieved the image oversubscription process of more efficient, there is stronger practical value.
Description
Technical field
The invention belongs to image-editing technology fields, and in particular to a kind of image oversubscription method.
Background technique
Image resolution ratio is to measure an important indicator of picture quality, and resolution ratio is higher, and details is finer, and quality is got over
Good, the information content that image includes is abundanter.Therefore the image with higher resolution is in various Computer Vision Tasks, such as military affairs
Safety, satellite monitoring, traffic monitoring, criminal investigation etc. application in suffer from important application value and Research Prospects.But due to
Cost problem, Image Acquisition usually in a practical situation, storage, transmission process can unavoidably be limited by external condition or
Other interference, leading to it, there are different degrees of quality degradations.
Image oversubscription technology is the research branch as picture quality enhancing technology.It is intended to restore from low-resolution image
Information in high-definition picture out is a science researching value height, the extensive Morden Image Processing Technology of application field.Figure
As oversubscription is not only the size of simple enlarged image, it produces the new image comprising more valuable information.And image is super
The technology of dividing improves image resolution ratio using the method based on signal processing, is that one kind effectively improves image resolution ratio, improves image
The approach of performance, and this method is at low cost, therefore seems more important to the research of the image oversubscription technology of high-effect high-quality.
Image oversubscription can be by algorithm, Case-based Reasoning method and method neural network based based on interpolation come real
It is existing.The oversubscription method of early stage is based on interpolation, such as bicubic interpolation and Lan Suosi resampling[1], since oversubscription is a kind of discomfort
Determine problem, there are many solutions in the mapping from low-resolution image to high-definition picture for each pixel, and such methods only make
With the information of low-resolution image, therefore it is difficult to simulate the visual complexity of true picture, it is complicated for texture, smooth shading
Image, interpolation method are likely to produce false effect.High-definition picture cannot be reconstructed well.
Therefore, oversubscription needs very strong priori to constrain solution space, and nearest most of preferable methods use Case-based Reasoning
Strategy learn powerful priori knowledge.This method is interfragmental right with high-resolution by finding multiple low resolution fragments
It should be related to, find the several fragments most like with the fragment in low-resolution image for each low resolution fragment, and calculate
It sends as an envoy to and reconstructs the smallest weighting parameter of cost, high-resolution piece shape is finally generated using multiple low resolution pieces and weighting parameter
At high-definition picture.The deficiency of this method is the high-frequency content that can be lost in image, additionally due to there are the calculating of superimposed sheets
It will lead to the increase of calculation amount.
In recent years, as convolutional neural networks (CNN) are in the application of computer vision field, occur many based on CNN
Image oversubscription method.These methods realize the development of this technological break-through, wherein with SRCNN[2]And VDSR[3]Method is most
It is representative.By that image oversubscription simply can be expanded to image oversubscription field using these methods to each frame of image.
C.Dong et al. put forward the image oversubscription method (SRCNN) based on convolutional neural networks in 2015, passed through
The mapping relations between low resolution and high-definition picture are practised to rebuild high-definition picture.Mapping shows as a CNN, will be low
Image in different resolution is as input, using high-definition picture as output.This process employs the superiority of neural network, by image
Oversubscription problem is modeled as neural network structure, is simply and effectively increased by the suitable neural network of optimization object function training
The model of strong image resolution ratio.
Neural network is easy to learn to obtain to a large amount of training set datas, furthermore after training the model of oversubscription, to height
The reconstruct of image in different resolution is exactly simple feed forward process, therefore computation complexity is also greatly lowered.C.Dong et al.
SRCNN method is improved again, proposes FSRCNN[4]Method, the structure for improving neural network realize faster oversubscription
Effect.Better effect, while benefit are achieved in image resolution ratio by deepening neural network structure within Kim J et al. 2016
Learn the training speed that raising network efficiency accelerates network with residual error.It realizes with convolutional neural networks in oversubscription field and constantly mentions
The effect risen, more scholars are by continuing to improve network structure in the subjective visual quality and objective value standard of oversubscription result
Constantly broken through.
The method of the present invention is related to image oversubscription technology, is that one parameter of thought building based on deep learning is more efficient
Oversubscription convolutional neural networks.Constructed network structure can be on the basis of existing low-resolution image by utilizing image
The correlation of interior partial structurtes and detailed information learns high frequency detail and texture content in image out, and reconstructing has more preferably
Visual quality clear image.
Summary of the invention
In order to improve the prior art obtain better oversubscription effect, the present invention provides a kind of image spatial resolution that improves
Method promotes oversubscription efficiency to enhance the quality of image.
Image oversubscription method provided by the invention is the image oversubscription method based on convolutional neural networks, the convolution mind
It include: feature extraction network F through networkFE, feature learning network FFL, image reconstruction network FIR, the specific steps are as follows:
(1) extraction of characteristics of image
By low-resolution image ILRIt is input in convolutional neural networks, first passes around feature extraction network FFE, by convolution
Input picture is transformed into feature space from pixel space by operation, generates the feature of input picture:
FLR=FFE(ILR)
(2) study of characteristics of image
This step is by the feature F of the low-resolution image extractedLRAs input, by the feature learning net of a multilayer
Network FFL, from the minutia F learnt in high-definition picture out in original imageHR:
FHR=FFL(FLR)
(3) reconstruct of high-definition picture
The feature F containing rich image detailed information that previous step is predictedHRBy image reconstruction network FIR, recover
High frequency detail content in original image has higher-quality high-definition picture I with reconstructSR:
ISR=FIR(FHR)
Multitiered network structure that the present invention uses (including feature extraction network FFE, feature learning network FFL, image reconstruction
Network FIR) be made of the identical basic unit U of multiple structures;Wherein each unit U includes two convolution blocks and a company
Node is connect, for the structure of convolution block using the GEU unit proposed in paper [5], connecting node is 3 × 3 by a convolution kernel size
Convolutional layer realize;The specific structure of each unit U are as follows:
Firstly, setting finIndicate the input of unit U, first GEU being entered into unit U (is denoted as GEU1) raw in block
At feature f1:
f1=GEU1(fin)
By the feature f of generation1It is input to next GEU and (is denoted as GEU2) unit, obtain feature f2:)
f2=GEU2(f1)
Finally by the output f of two GEU units1、f1The convolutional layer that one convolution kernel size of input is 3 × 3 simultaneously carries out special
Sign fusion, generates the output feature f of the unitout:
fout=Conv (f1,f2)。
In step (1) of the present invention, feature extraction network FEUsing a layer network structure, i.e., it is made of a basic unit U,
The unit specific steps are as follows:
Firstly, low-resolution image is input to feature extraction network FE, low resolution figure is extracted by a unit U
The feature F of pictureLR:
FLR=U (ILR)。
In step (2) of the present invention, feature learning network FFLIt is the network structure of a multilayer, specific feature learning step
Are as follows:
The F that step (1) is exported firstLRInput feature vector learning network FFLIn, by unit U1,1Generate feature f1,1:
f1,1=U1,1(FLR)
By f1,1It is input to the unit U of the second layer2,1In, generate feature f2,1:
f2,1=U2,1(f1,1)
Again by the second layer unit U2,1The feature f of output2,1It is input to next unit U of first layer1,2In, export feature
f1,2:
f1,2=U1,2(f2,1)
By f1,2It is input to the unit U of next second layer2,2In, generate feature f2,2:
f2,2=U2,2(f1,2)
Again by the unit U of the second layer2,2Output feature f2,2It is input to the first layer unit U1,3In, export feature f1,3:
f1,3=U1,3(f2,2)
Finally by the output f of the last one unit of first layer1,3With the output f of the second layer the last one unit2,2Together
Input the unit U of third layer3,1Middle generation feature f3,1, since network structure of the invention is only with three-decker,
The output f of three layers of first unit3,1Minutia F with regard to the high-definition picture acquired as feature learning partSR:
FSR=f1,3
Unit U1,1, U2,1, U1,2, U2,2, U1,3, U3,1, it is basic unit U.
In step (3) of the present invention, image reconstruction network FIRUsing one layer of network structure, i.e., by a basic unit U group
At specific steps are as follows:
Firstly, by the output F of previous stepSRAs input, by a basic unit U1And layer enlarged drawing point that deconvolute
Resolution generates the residual plot I between reconstructed image and true high-definition pictureres:
Ires=Deconv (U1(FSR))
Finally, by use Bicubic method up-sample after low-resolution image and residual plot IresIt is added, obtains network
The high-definition picture I of final outputSR:
ISR=Ires+Bic(ILR)。
Multitiered network structure that the present invention uses learns characteristics of image, by the image acquired between different layers in network spy
Sign is connected by way of layering aggregation, is made full use of each layer in network of structural information, is recovered more accurate height
Frequency picture material effectively reduces information loss of characteristics of image during Internet communication, and it is special to excavate deeper image
Sign, finally obtains better quality reconstruction.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 is the result for carrying out oversubscription reconstruct to low-resolution image using this method.
Specific embodiment
For a low-resolution image, method shown in FIG. 1 can be used, oversubscription processing is carried out.Specific steps are as follows:
(1) low-resolution image is input to network first, through a convolutional layer, a unit U.
The input of unit U is transmitted to first GEU first referring to Fig. 1 by each unit specific steps1Block will generate feature
Input next GEU2Block, the convolution that finally one convolution kernel size of input is 3 × 3 simultaneously again by the output of two GEU units
Layer, generates the output feature f of the unit1;
(2) by f1It is input to the first layer unit U1,1Generate feature f1,1;By f1,1It is input to the second layer unit U2,1It generates special
Levy f2,1;By f2,1It is input to the first layer unit U1,2In, export feature f1,2;By f1,2Input the second layer unit U2,2Generate feature
f2,2;By f2,2Transfer back to the first layer unit U1,3Export feature f1,3;By f1,3And f2,2It is transmitted to third layer unit U3,1Generate feature
f3,1;
(3) by f3,1It is transmitted to a unit U and generates feature f2, by f2Residual plot I is generated by the layer that deconvolutesres;Finally
Former low-resolution image, which is up-sampled, using bicubic interpolation method generates Ibic, by IbicWith residual plot IresIt is added, generates high-resolution
Image IHR。
Fig. 2 is an experimental example.Wherein, (a), (d) figure respectively be input low-resolution image, (b), (e) figure then
Be it is corresponding using the method for the present invention carry out 4 times of oversubscription reconstruct come high-definition picture, (c), (f) be true high-resolution
Rate image.As can be seen that the method for the present invention can effectively recover texture and marginal information in original high-resolution image,
Bring preferable visual effect.
Bibliography:
[1]C.E.Duchon.Lanczos filtering in one and two dimensions.Journal of
Applied Meteorology,18(8):1016–1022,1979.
[2]C.Dong,C.C.Loy,K.He,and X.Tang.Image super-resolution using deep
convolutional networks.IEEE Transactions on Pattern Analysis and Machine
Intelligence (TPAMI),38(2):295–307,2015.
[3]Kim J,Lee J K,Lee K M.Accurate Image Super-Resolution Using Very
Deep Convolutional Networks[C]//IEEE Conference on Computer Vision and
Pattern Recognition.IEEE Computer Society,2016:1646-1654.
[4]C.Dong,C.C.Loy,and X.Tang.Accelerating the super-resolution
convolutional neural network.In European Conference on Computer Vision(ECCV),
pages 391–407.Springer International Publishing,2016.
[5]Ke Li,Bahetiyaer Bare,B.Y.B.F.,and Yao,C.2018.Face hallucination
based on key parts enhancement.In IEEE International Conference on Acoustics,
Speech and Signal Processing.。
Claims (4)
1. a kind of image oversubscription method based on convolutional neural networks, the convolutional neural networks include: feature extraction network FFE、
Feature learning network FFL, image reconstruction network FIR, which is characterized in that specific step is as follows:
(1) extraction of characteristics of image
By low-resolution image ILRIt is input in convolutional neural networks, first passes around feature extraction network FFE, by convolution operation
Input picture is transformed into feature space from pixel space, generates the feature of input picture:
FLR=FFE(ILR)
(2) study of characteristics of image
By the feature F of the low-resolution image extractedLRAs input, by feature learning network FFL, from original image
Learn the minutia F in high-definition picture outHR:
FHR=FFL(FLR)
(3) reconstruct of high-definition picture
The feature F containing rich image detailed information that previous step is predictedHRBy image reconstruction network FIR, recover original graph
High frequency detail content as in has higher-quality high-definition picture I with reconstructSR:
ISR=FIR(FHR);
Here, feature extraction network FFE, feature learning network FFL, image reconstruction network FIR, identical by the structure of different number
Basic unit U composition;Wherein each unit U includes two convolution blocks and a connecting node, and the structure of convolution block uses GEU
Unit, connecting node are realized by the convolutional layer that a convolution kernel size is 3 × 3;The process flow of each unit U are as follows:
Firstly, setting finIndicate the input of unit U, first GEU being entered into unit U (is denoted as GEU1) generate in block it is special
Levy f1:
f1=GEU1(fin)
By the feature f of generation1It is input to next GEU and (is denoted as GEU2) unit, obtain feature f2:
f2=GEU2(f1)
Finally, by the output f of two GEU units1、f1The convolutional layer that one convolution kernel size of input is 3 × 3 simultaneously carries out feature
Fusion, generates the output feature f of the unitout:
fout=Conv (f1,f2)。
2. the image oversubscription method according to claim 1 based on convolutional neural networks, which is characterized in that in step (1),
Feature extraction network FEUsing a layer network structure, i.e., it is made of a basic unit U, specific process flow are as follows:
Firstly, low-resolution image is input to feature extraction network FE, low-resolution image is extracted by a unit U
Feature FLR:
FLR=U (ILR)。
3. the image oversubscription method according to claim 1 based on convolutional neural networks, which is characterized in that in step (2),
Feature learning network FFLIt is the network structure of a multilayer, feature learning process are as follows:
Firstly, the F that step (1) is exportedLRInput feature vector learning network FFLIn, by unit U1,1Generate feature f1,1:
f1,1=U1,1(FLR)
By f1,1It is input to the unit U of the second layer2,1In, generate feature f2,1:
f2,1=U2,1(f1,1)
Again by the second layer unit U2,1The feature f of output2,1It is input to next unit U of first layer1,2In, export feature f1,2:
f1,2=U1,2(f2,1)
By f1,2It is input to the unit U of next second layer2,2In, generate feature f2,2:
f2,2=U2,2(f1,2)
Again by the unit U of the second layer2,2Output feature f2,2It is input to the first layer unit U1,3In, export feature f1,3:
f1,3=U1,3(f2,2)
Finally, by the output f of the last one unit of first layer1,3With the output f of the second layer the last one unit2,2It inputs together
The unit U of third layer3,1Middle generation feature f3,1, since network structure of the invention uses three-decker, the of third layer
The output f of one unit3,1Minutia F with regard to the high-definition picture acquired as feature learning partSR:
FSR=f1,3
Wherein, unit U1,1, U2,1, U1,2, U2,2, U1,3, U3,1, it is basic unit U.
4. the image oversubscription method according to claim 1 based on convolutional neural networks, which is characterized in that in step (3),
Image reconstruction network FIRUsing one layer of network structure, i.e., it is made of a basic unit U, the detailed process of processing are as follows:
Firstly, by the output F of previous stepSRAs input, by a basic unit U (U1) and a layer enlarged drawing resolution of deconvoluting
Rate generates the residual plot I between reconstructed image and true high-definition pictureres
Ires=Deconv (U1(FSR))
Finally, by use Bicubic method up-sample after low-resolution image and residual plot IresIt is added, it is final to obtain network
The high-definition picture I of outputSR:
ISR=Ires+Bic(ILR)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810959380.XA CN109272450B (en) | 2018-08-22 | 2018-08-22 | Image super-resolution method based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810959380.XA CN109272450B (en) | 2018-08-22 | 2018-08-22 | Image super-resolution method based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109272450A true CN109272450A (en) | 2019-01-25 |
CN109272450B CN109272450B (en) | 2023-01-06 |
Family
ID=65153956
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810959380.XA Active CN109272450B (en) | 2018-08-22 | 2018-08-22 | Image super-resolution method based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109272450B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109978785A (en) * | 2019-03-22 | 2019-07-05 | 中南民族大学 | The image super-resolution reconfiguration system and its method of multiple recurrence Fusion Features |
CN111182254A (en) * | 2020-01-03 | 2020-05-19 | 北京百度网讯科技有限公司 | Video processing method, device, equipment and storage medium |
CN113556496A (en) * | 2020-04-23 | 2021-10-26 | 京东方科技集团股份有限公司 | Video resolution improving method and device, storage medium and electronic equipment |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN106683067A (en) * | 2017-01-20 | 2017-05-17 | 福建帝视信息科技有限公司 | Deep learning super-resolution reconstruction method based on residual sub-images |
CN106709875A (en) * | 2016-12-30 | 2017-05-24 | 北京工业大学 | Compressed low-resolution image restoration method based on combined deep network |
CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
CN107392855A (en) * | 2017-07-19 | 2017-11-24 | 苏州闻捷传感技术有限公司 | Image Super-resolution Reconstruction method based on sparse autoencoder network Yu very fast study |
CN107492070A (en) * | 2017-07-10 | 2017-12-19 | 华北电力大学 | A kind of single image super-resolution computational methods of binary channels convolutional neural networks |
CN108259994A (en) * | 2018-01-15 | 2018-07-06 | 复旦大学 | A kind of method for improving video spatial resolution |
-
2018
- 2018-08-22 CN CN201810959380.XA patent/CN109272450B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN106709875A (en) * | 2016-12-30 | 2017-05-24 | 北京工业大学 | Compressed low-resolution image restoration method based on combined deep network |
CN106683067A (en) * | 2017-01-20 | 2017-05-17 | 福建帝视信息科技有限公司 | Deep learning super-resolution reconstruction method based on residual sub-images |
CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
CN107492070A (en) * | 2017-07-10 | 2017-12-19 | 华北电力大学 | A kind of single image super-resolution computational methods of binary channels convolutional neural networks |
CN107392855A (en) * | 2017-07-19 | 2017-11-24 | 苏州闻捷传感技术有限公司 | Image Super-resolution Reconstruction method based on sparse autoencoder network Yu very fast study |
CN108259994A (en) * | 2018-01-15 | 2018-07-06 | 复旦大学 | A kind of method for improving video spatial resolution |
Non-Patent Citations (2)
Title |
---|
KE LI; BAHETIYAER BARE; BO YAN; BAILAN FENG; CHUNFENG YAO: "HNSR: Highway Networks Based Deep Convolutional Neural Networks Model for Single Image Super-Resolution", 《2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)》 * |
兰妙萍: "基于卷积神经网络的图像复原方法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109978785A (en) * | 2019-03-22 | 2019-07-05 | 中南民族大学 | The image super-resolution reconfiguration system and its method of multiple recurrence Fusion Features |
CN111182254A (en) * | 2020-01-03 | 2020-05-19 | 北京百度网讯科技有限公司 | Video processing method, device, equipment and storage medium |
CN111182254B (en) * | 2020-01-03 | 2022-06-24 | 北京百度网讯科技有限公司 | Video processing method, device, equipment and storage medium |
US11416967B2 (en) | 2020-01-03 | 2022-08-16 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Video processing method, apparatus, device and storage medium |
CN113556496A (en) * | 2020-04-23 | 2021-10-26 | 京东方科技集团股份有限公司 | Video resolution improving method and device, storage medium and electronic equipment |
CN113556496B (en) * | 2020-04-23 | 2022-08-09 | 京东方科技集团股份有限公司 | Video resolution improving method and device, storage medium and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN109272450B (en) | 2023-01-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111047515B (en) | Attention mechanism-based cavity convolutional neural network image super-resolution reconstruction method | |
CN111105352B (en) | Super-resolution image reconstruction method, system, computer equipment and storage medium | |
CN111932461B (en) | Self-learning image super-resolution reconstruction method and system based on convolutional neural network | |
CN106204449A (en) | A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network | |
CN108259994B (en) | Method for improving video spatial resolution | |
CN111861961A (en) | Multi-scale residual error fusion model for single image super-resolution and restoration method thereof | |
CN110889895B (en) | Face video super-resolution reconstruction method fusing single-frame reconstruction network | |
CN109035146A (en) | A kind of low-quality image oversubscription method based on deep learning | |
CN111768340B (en) | Super-resolution image reconstruction method and system based on dense multipath network | |
CN107590775B (en) | Image super-resolution amplification method using regression tree field | |
CN113205096B (en) | Attention-based combined image and feature self-adaptive semantic segmentation method | |
CN109272452A (en) | Learn the method for super-resolution network in wavelet field jointly based on bloc framework subband | |
CN109949222A (en) | Image super-resolution rebuilding method based on grapheme | |
CN109272450A (en) | A kind of image oversubscription method based on convolutional neural networks | |
Yang et al. | Image super-resolution based on deep neural network of multiple attention mechanism | |
CN114418850A (en) | Super-resolution reconstruction method with reference image and fusion image convolution | |
CN112200720A (en) | Super-resolution image reconstruction method and system based on filter fusion | |
CN114463183A (en) | Image super-resolution method based on frequency domain and spatial domain | |
CN115953294A (en) | Single-image super-resolution reconstruction method based on shallow channel separation and aggregation | |
CN116563100A (en) | Blind super-resolution reconstruction method based on kernel guided network | |
Zang et al. | Cascaded dense-UNet for image super-resolution | |
CN109087247A (en) | The method that a kind of pair of stereo-picture carries out oversubscription | |
CN116206105A (en) | Collaborative learning enhanced colon polyp segmentation method integrating deep learning | |
CN115375537A (en) | Nonlinear sensing multi-scale super-resolution image generation system and method | |
CN117274064B (en) | Image super-resolution method |
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 |