CN106204489B - The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient - Google Patents
The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient Download PDFInfo
- Publication number
- CN106204489B CN106204489B CN201610545884.8A CN201610545884A CN106204489B CN 106204489 B CN106204489 B CN 106204489B CN 201610545884 A CN201610545884 A CN 201610545884A CN 106204489 B CN106204489 B CN 106204489B
- Authority
- CN
- China
- Prior art keywords
- gradient
- image
- resolution
- deep learning
- conversion
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 238000006243 chemical reaction Methods 0.000 claims abstract description 25
- 230000000694 effects Effects 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims abstract description 12
- 238000005070 sampling Methods 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000011478 gradient descent method Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 10
- 238000002474 experimental method Methods 0.000 description 8
- 230000000052 comparative effect Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 2
- 238000011017 operating method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the single image super resolution ratio reconstruction methods of a kind of combination deep learning and gradient conversion.It mainly comprises the steps that and is up-sampled with low-resolution image of the super-resolution method based on deep learning to input, obtain up-sampling image;Gradient extraction is carried out to up-sampling image with gradient operator;The gradient extracted is converted with depth convolutional neural networks;Using the gradient after the low-resolution image of input and conversion as constraint, establishes and rebuild cost function;Reconstruction cost function is optimized using gradient descent method, obtains the high-definition picture of final output.The image that single image super resolution ratio reconstruction method of the present invention is rebuild has fine structure, in subjective vision effect almost without artifact effect, and objectively evaluates parameter value with very high.Therefore, the present invention is a kind of effective single image super resolution ratio reconstruction method.
Description
Technical field
The present invention relates to image super-resolution rebuilding technologies, and in particular to a kind of list of combination deep learning and gradient conversion
Width image super-resolution rebuilding method, belongs to digital image processing field.
Background technique
In actual life, due to the loss of image information in the limitation and transmission process of imaging device and imaging circumstances,
People obtain image be often low resolution, it is low-quality, it is difficult to meet demand.Image super-resolution rebuilding technology, be
It does not need in the case where increasing hardware cost, by signal processing technology, the low-resolution image of input is reconstructed into high-resolution
A special kind of skill of rate image.Through the image after image super-resolution rebuilding reconstruction not only in spatial resolution better than defeated
Enter image, and is also obviously improved in subjective vision effect.
Image super-resolution rebuilding method can be divided into three classes: method based on interpolation, based on the method for reconstruction be based on
The method of study.In recent years, due to the development of machine learning and deep learning, the super resolution ratio reconstruction method based on study is got
Biggish progress.Super-resolution method based on deep learning compares traditional super-resolution method based on study, has
Structure is simple, fireballing advantage, and since in the training stage, the method based on deep learning optimizes all operations simultaneously,
So the method based on deep learning reconstruct come high-definition picture be better than traditional side based on study in quality
Method.But the used convolutional neural networks of the super-resolution method based on deep learning are obtained according to general structure training
, so rebuilding obtained image would generally be by the influence of ringing effect and sawtooth effect.Remove a kind of side of artifact effect
Method is that the prior information of introducing image constrains reconstruction image.Super-resolution method based on gradient priori can effectively be gone
Except the ringing effect and sawtooth effect of reconstruction image, but such methods cannot play the details of image and fine structure part
Effect is rebuild well.
Summary of the invention
The purpose of the present invention is deep learning is introduced into gradient conversion, and using the gradient information after conversion as constraint
Super-resolution rebuilding is carried out, so that rebuilding obtained image has finer structure, and reduces ringing effect and sawtooth
The influence of effect.The present invention realizes above-mentioned purpose by the technical solution that following operating procedure is constituted.
Combination deep learning proposed by the present invention and gradient conversion single image super resolution ratio reconstruction method, mainly include
Following operating procedure:
(1) it is up-sampled, is obtained with low-resolution image of the super-resolution method based on deep learning to input
Sampled images;
(2) gradient extraction is carried out to up-sampling image with gradient operator;
(3) gradient extracted is converted with depth convolutional neural networks;
(4) gradient for being converted to the low-resolution image of input and step (3) is established as constraint and rebuilds cost letter
Number;
(5) reconstruction cost function is optimized using gradient descent method, obtains the high-definition picture of final output.
Detailed description of the invention
Fig. 1 is the block diagram that the present invention combines deep learning and the single image super resolution ratio reconstruction method of gradient conversion
Fig. 2 is comparison diagram of the present invention with existing 4 kinds of methods to " Butterfly " image reconstruction result
Fig. 3 is comparison diagram of the present invention with existing 4 kinds of methods to " Foreman " image reconstruction result
Fig. 4 is comparison diagram of the present invention with existing 4 kinds of methods to " Leaves " image reconstruction result
Specific embodiment
The present invention will be further explained below with reference to the attached drawings:
In Fig. 1, in conjunction with the single image super resolution ratio reconstruction method of deep learning and gradient conversion, comprising the following steps:
(1) it is up-sampled, is obtained with low-resolution image of the super-resolution method based on deep learning to input
Sampled images;
(2) gradient extraction is carried out to up-sampling image with gradient operator;
(3) gradient extracted is converted with depth convolutional neural networks;
(4) gradient for being converted to the low-resolution image of input and step (3) is established as constraint and rebuilds cost letter
Number;
(5) reconstruction cost function is optimized using gradient descent method, obtains the high-definition picture of final output.
Specifically, in the step (1), we use low resolution of the super-resolution method based on deep learning to input
Rate image is up-sampled, and up-sampling image is obtained.The specifically used super-resolution method based on deep learning is Dong etc.
The method that people proposes, bibliography " 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,vol.38,no.2,pp.295-307,2016.”。
In the step (2), we use gradient operator [- (1/2), 0, (1/2)] and [- (1/2), 0, (1/2)]TRespectively
Extract the gradient in up-sampling image level direction and vertical direction.
In the step (3), we convert the gradient extracted using convolutional neural networks, so that after conversion
Gradient is closer to original gradient.The conversion method mainly includes two stages, i.e. training stage and conversion stage.
In the training stage, we construct the convolutional neural networks being made of 3 layers of convolutional layer, including Gradient Features first
Extract layer (L1), Gradient Features conversion layer (L2) and Gradient Reconstruction layer (L3)。L1、L2、L3Respectively by the filter of different numbers
Composition.L1Gradient Features extraction is carried out to the gradient of input, obtains character representation f;L2F is mapped as the character representation after conversion
ft, L3Act on ftThe gradient information after conversion to generate final output.Since ReLU can greatly speed up trained convergence speed
Degree, the present invention apply it in the response of the filter in training process.High-resolution natural image is carried out bicubic by us
Down-sampling, and it is up-sampled with the super-resolution method based on deep learning;Then we extract the ladder of up-sampling image
Degree, and it is divided into 36 × 36 block { Gl};For 36 × 36 block of input, the convolutional neural networks that we construct will be defeated
One 20 × 20 block avoids edge effect out, therefore the gradient extracted from primitive nature image is divided into pair by we
Block { the G of 20 × 20 answeredh};We have just obtained for trained training to { G in this wayl,Gh}.We using mean square error as
Loss function trains to obtain the convolutional neural networks for gradient conversion.In the conversion stage, for being mentioned from up-sampling image
The gradient got, we are entered into the convolutional neural networks that training obtains, and the result of final output is after converting
Gradient.In order to make gradient convert effect it is more preferable, to obtain higher-quality reconstruction image, we to horizontal direction with it is vertical
The gradient in direction is respectively trained to obtain the corresponding convolutional neural networks for gradient conversion.
In the step (4), we establish the gradient that low-resolution image is converted to step (3) as constraint
Cost function is rebuild, cost function is rebuild and is defined as
E(H|L,▽Yt)=E1(H|L)+θE2(▽H|▽Yt)
In formula, ▽ YtThe gradient being converted to for step (3);▽ H is the gradient for exporting image H;θ is two about interfasciculars
Weight;E1(H | L) is the constraint of image area, is defined as
E1(H | L)=| H ↓-L |2
E2(▽H|▽Yt) be gradient field constraint, be defined as
E2(▽H|▽Yt)=| ▽ H- ▽ Yt|2
In the step (5), we optimize reconstruction cost function using gradient descent method, obtain the height of final output
Image in different resolution:
Hi+1=Hi-μ((Hi↓-L)↑-θ·(▽2H-▽2Yt))
In formula, HiFor the image of i-th iteration output;μ is iteration step length.
Validity in order to better illustrate the present invention, the present invention will be shown reconstruction effect using the method for comparative experiments
Fruit." Butterfly " image, " Foreman " image and " Leaves " image are 3 width test charts selected by comparative experiments
Picture, respectively as shown in Fig. 2 (a), Fig. 3 (a) and Fig. 4 (a).Comparative experiments, which chooses bicubic interpolation Bicubic and 3, has generation
The single image super resolution ratio reconstruction method of table is compared with experimental result of the invention.This 3 representative lists
Width image super-resolution rebuilding method are as follows:
The method that method 1:Yang et al. is proposed, bibliography " J.Yang, J.Wright, T.S.Huang, and
Y.Ma,"Image super-resolution via sparse representation."IEEE Transactions on
Image Processing,vol.19,no.11,pp.2861-2873,2010.”。
The method that method 2:Timofte et al. is proposed, bibliography " R.Timofte, V.D.Smet, and
L.V.Gool,"A+:Adjusted anchored neighborhood regression for fast super-
resolution."Computer Vision--ACCV 2014.Springer International Publishing,
pp.111-126,2014.”。
The method that method 3:Dong et al. is proposed, bibliography " 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,vol.38,no.2,pp.295-307,2016.”。
The content of comparative experiments is as follows:
Experiment 1, carries out 3 to " Butterfly " image with Bicubic, method 1, method 2, method 3 and the present invention respectively
Times super-resolution rebuilding.Super-resolution Reconstruction result is as shown such as Fig. 2 (b), Fig. 2 (c), Fig. 2 (d), Fig. 2 (e) and Fig. 2 (f) respectively,
It objectively evaluates index as shown in the second row of table one.
Experiment 2, carries out 3 times to " Foreman " image with Bicubic, method 1, method 2, method 3 and the present invention respectively
Super-resolution rebuilding.Super-resolution Reconstruction result is as shown such as Fig. 3 (b), Fig. 3 (c), Fig. 3 (d), Fig. 3 (e) and Fig. 3 (f) respectively,
Index is objectively evaluated as shown in the third line of table one.
Experiment 3, carries out 3 times to " Leaves " image with Bicubic, method 1, method 2, method 3 and the present invention respectively
Super-resolution rebuilding.Super-resolution Reconstruction result is as shown such as Fig. 4 (b), Fig. 4 (c), Fig. 4 (d), Fig. 4 (e) and Fig. 4 (f) respectively,
Index is objectively evaluated as shown in the fourth line of table one.
It can be seen that reconstruction image that Biucbic is obtained is very fuzzy and sawtooth effect is very serious by 3 groups of comparative experimentss;
The reconstruction image ratio Biucbic that method 1 obtains it is slightly clear a bit, but it is still very fuzzy;The reconstruction image that method 2 obtains
Sawtooth effect is avoided well, but it is not clear enough;The reconstruction image that method 3 obtains is than more visible, but edge has saw
Tooth and ringing effect;The reconstruction image that the present invention obtains has fine structure, and almost without sawtooth effect and ringing effect,
With optimal subjective vision effect.
Table one gives two objective parameters of the present invention with the super resolution ratio reconstruction method reconstructed results of 4 kinds of comparisons, point
It Wei not Y-PSNR (PSNR:the Peak Signal to Noise Ratio) and structural similarity (SSIM:the
Structure Similarity Index), objectively to evaluate the quality of reconstruction image.Wherein, PSNR value is bigger, SSIM
Value is closer to 1, then the quality of reconstruction image is better.
Table one
As can be seen from Table I, the present invention objectively evaluates parameter value with highest.For " Butterfly " image, originally
The PSNR value ratio method 3 of invention is higher by 0.87dB, and SSIM value ratio method 3 is higher by 0.0114;For " Foreman " image, this hair
Bright PSNR value ratio method 2 is higher by 0.43dB, and SSIM value ratio method 3 is higher by 0.0017;It is of the invention for " Leaves " image
PSNR value ratio method 3 is higher by 0.68dB, and SSIM value ratio method 3 is higher by 0.0087.
In conclusion the image rebuild of the present invention has an apparent advantage in subjective vision effect, and for
Other control methods objectively evaluate parameter value with highest.Therefore, the present invention is a kind of effective single image super-resolution
Method for reconstructing.
Claims (4)
1. combining the single image super resolution ratio reconstruction method of deep learning and gradient conversion, it is characterised in that including following step
It is rapid:
Step 1: it is up-sampled, is obtained with low-resolution image of the super-resolution method based on deep learning to input
Sampled images;
Step 2: gradient extraction is carried out to up-sampling image obtained in step 1 with gradient operator;
Step 3: the gradient extracted in step 2 is converted with depth convolutional neural networks;
Step 4: the gradient that the low-resolution image of input and step 3 are converted to is established as constraint and rebuilds cost letter
Number;
Step 5: optimizing reconstruction cost function using gradient descent method, obtains the high-definition picture of final output.
2. the single image super resolution ratio reconstruction method of combination deep learning according to claim 1 and gradient conversion,
It is characterized in that converting the gradient extracted in step 2 with depth convolutional neural networks described in step 3: turn in gradient
It changes the depth convolutional neural networks training stage, constructs the depth convolutional neural networks being made of 3 layers of convolutional layer first, including
Gradient Features extract layer, Gradient Features conversion layer and Gradient Reconstruction layer;Then it will be adopted under high-resolution natural image bicubic
Sample again using the up-sampling image zooming-out gradient obtained based on deep learning super resolution ratio reconstruction method with from high-resolution nature
The gradient extracted in image corresponds to piecemeal composing training pair, and mean square error is trained gradient conversion level as loss function
Convolutional neural networks;The stage is converted in gradient, the depth convolutional neural networks obtained using above-mentioned training in step 2 to extracting
To gradient converted.
3. the single image super resolution ratio reconstruction method of combination deep learning according to claim 1 and gradient conversion,
It is characterized in that converting the gradient extracted in step 2 with depth convolutional neural networks described in step 3: in order to make ladder
The effect for spending conversion is more preferable, to obtain higher-quality reconstruction image, and then to the gradient of horizontal direction and vertical direction point
The depth convolutional neural networks accordingly for gradient conversion Xun Lian not obtained.
4. the single image super resolution ratio reconstruction method of combination deep learning according to claim 1 and gradient conversion,
It is characterized in that establishing described in step 4 using the gradient that the low-resolution image of input is converted to step 3 as constraint
Cost function is rebuild, cost function is rebuild and is defined as
E(H|L,▽Yt)=E1(H|L)+θE2(▽H|▽Yt)
In formula, ▽ YtFor the gradient that step 3 is converted to, ▽ H is the gradient for exporting image H, and θ is the weight of two about interfasciculars, E1
(H | L) is the constraint of image area, E2(▽H|▽Yt) be gradient field constraint.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610545884.8A CN106204489B (en) | 2016-07-12 | 2016-07-12 | The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610545884.8A CN106204489B (en) | 2016-07-12 | 2016-07-12 | The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106204489A CN106204489A (en) | 2016-12-07 |
CN106204489B true CN106204489B (en) | 2019-04-16 |
Family
ID=57477466
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610545884.8A Active CN106204489B (en) | 2016-07-12 | 2016-07-12 | The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106204489B (en) |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106846463B (en) * | 2017-01-13 | 2020-02-18 | 清华大学 | Microscopic image three-dimensional reconstruction method and system based on deep learning neural network |
US11113800B2 (en) | 2017-01-18 | 2021-09-07 | Nvidia Corporation | Filtering image data using a neural network |
CN106910161B (en) * | 2017-01-24 | 2020-06-19 | 华南理工大学 | Single image super-resolution reconstruction method based on deep convolutional neural network |
CN106981046B (en) * | 2017-03-21 | 2019-10-11 | 四川大学 | Single image super resolution ratio reconstruction method based on multi-gradient constrained regression |
CN107018422B (en) * | 2017-04-27 | 2019-11-05 | 四川大学 | Still image compression method based on depth convolutional neural networks |
WO2018212599A1 (en) | 2017-05-17 | 2018-11-22 | Samsung Electronics Co., Ltd. | Super-resolution processing method for moving image and image processing apparatus therefor |
US10474161B2 (en) * | 2017-07-03 | 2019-11-12 | Baidu Usa Llc | High resolution 3D point clouds generation from upsampled low resolution lidar 3D point clouds and camera images |
CN107480772B (en) * | 2017-08-08 | 2020-08-11 | 浙江大学 | License plate super-resolution processing method and system based on deep learning |
CN107481192B (en) * | 2017-08-11 | 2021-08-24 | 北京市商汤科技开发有限公司 | Image processing method, image processing apparatus, storage medium, computer program, and electronic device |
CN107464217B (en) * | 2017-08-16 | 2020-12-29 | 清华-伯克利深圳学院筹备办公室 | Image processing method and device |
CN108053368B (en) * | 2017-12-18 | 2020-11-03 | 清华大学 | Cross-scale resolution light field image super-resolution and depth estimation method and device |
US10540749B2 (en) | 2018-03-29 | 2020-01-21 | Mitsubishi Electric Research Laboratories, Inc. | System and method for learning-based image super-resolution |
CN108550115B (en) * | 2018-04-25 | 2020-06-16 | 中国矿业大学 | Image super-resolution reconstruction method |
CN108898560B (en) * | 2018-06-21 | 2023-04-28 | 四川大学 | Core CT image super-resolution reconstruction method based on three-dimensional convolutional neural network |
CN109214989B (en) * | 2018-09-04 | 2019-08-13 | 四川大学 | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori |
CN110517210B (en) * | 2019-07-08 | 2021-09-03 | 河北工业大学 | Multi-exposure welding area image fusion method based on Haar wavelet gradient reconstruction |
CN110677649B (en) * | 2019-10-16 | 2021-09-28 | 腾讯科技(深圳)有限公司 | Artifact removing method based on machine learning, artifact removing model training method and device |
CN111161146B (en) * | 2019-12-25 | 2022-10-14 | 大连理工大学 | Coarse-to-fine single-image super-resolution reconstruction method |
CN111754405B (en) * | 2020-06-22 | 2023-08-08 | 北京大学深圳研究生院 | Image resolution reduction and restoration method, device and readable storage medium |
CN111932463B (en) * | 2020-08-26 | 2023-05-30 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment and storage medium |
CN112116527B (en) * | 2020-09-09 | 2024-02-23 | 北京航空航天大学杭州创新研究院 | Image super-resolution method based on cascade network frame and cascade network |
CN112184555B (en) * | 2020-10-21 | 2022-07-19 | 天津大学 | Stereo image super-resolution reconstruction method based on deep interactive learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246594A (en) * | 2008-02-22 | 2008-08-20 | 华南师范大学 | Optimized amalgamation remote sensing image processing method based on gradient field |
CN101950415A (en) * | 2010-09-14 | 2011-01-19 | 武汉大学 | Shape semantic model constraint-based face super-resolution processing method |
CN103854257A (en) * | 2012-12-07 | 2014-06-11 | 山东财经大学 | Depth image enhancement method based on self-adaptation trilateral filtering |
CN103985107A (en) * | 2014-05-27 | 2014-08-13 | 武汉科技大学 | Self-adaptive image super-resolution reconstruction method based on visual perception |
CN104778659A (en) * | 2015-04-15 | 2015-07-15 | 杭州电子科技大学 | Single-frame image super-resolution reconstruction method on basis of deep learning |
US9225889B1 (en) * | 2014-08-18 | 2015-12-29 | Entropix, Inc. | Photographic image acquisition device and method |
-
2016
- 2016-07-12 CN CN201610545884.8A patent/CN106204489B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246594A (en) * | 2008-02-22 | 2008-08-20 | 华南师范大学 | Optimized amalgamation remote sensing image processing method based on gradient field |
CN101950415A (en) * | 2010-09-14 | 2011-01-19 | 武汉大学 | Shape semantic model constraint-based face super-resolution processing method |
CN103854257A (en) * | 2012-12-07 | 2014-06-11 | 山东财经大学 | Depth image enhancement method based on self-adaptation trilateral filtering |
CN103985107A (en) * | 2014-05-27 | 2014-08-13 | 武汉科技大学 | Self-adaptive image super-resolution reconstruction method based on visual perception |
US9225889B1 (en) * | 2014-08-18 | 2015-12-29 | Entropix, Inc. | Photographic image acquisition device and method |
CN104778659A (en) * | 2015-04-15 | 2015-07-15 | 杭州电子科技大学 | Single-frame image super-resolution reconstruction method on basis of deep learning |
Non-Patent Citations (4)
Title |
---|
Gradient Profile Prior and Its Applications in Image Super-Resolution and Enhancement;Jian Sun et al.;《IEEE Transactions on Image Processing》;20110630;第1529-1542页 |
Image Super-Resolution using Gradient Profile Prior;Jian Sun et al.;《Computer Vision and Pattern Rcongtion 2008》;20081231;第1-8页 |
Learning a Deep Convolutional Network for Image Super-Resolution;Chao Dong et al.;《European Conference Computer Vision 2014》;20141231;第184-199页 |
基于深度学习的图像超分辨率算法研究;胡传平 等;《铁道警察学院学报》;20160505;第26卷(第1期);第5-10页 |
Also Published As
Publication number | Publication date |
---|---|
CN106204489A (en) | 2016-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106204489B (en) | The single image super resolution ratio reconstruction method converted in conjunction with deep learning and gradient | |
CN108765296B (en) | Image super-resolution reconstruction method based on recursive residual attention network | |
CN108537733B (en) | Super-resolution reconstruction method based on multi-path deep convolutional neural network | |
CN111340744B (en) | Attention double-flow depth network-based low-quality image down-sampling method and system | |
CN111681166B (en) | Image super-resolution reconstruction method of stacked attention mechanism coding and decoding unit | |
CN110033410A (en) | Image reconstruction model training method, image super-resolution rebuilding method and device | |
CN110889895B (en) | Face video super-resolution reconstruction method fusing single-frame reconstruction network | |
CN113362223A (en) | Image super-resolution reconstruction method based on attention mechanism and two-channel network | |
CN109214989B (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
CN107240066A (en) | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks | |
CN109118432A (en) | A kind of image super-resolution rebuilding method based on Rapid Circulation convolutional network | |
CN106910161A (en) | A kind of single image super resolution ratio reconstruction method based on depth convolutional neural networks | |
CN109584161A (en) | The Remote sensed image super-resolution reconstruction method of convolutional neural networks based on channel attention | |
CN110675321A (en) | Super-resolution image reconstruction method based on progressive depth residual error network | |
CN109949222B (en) | Image super-resolution reconstruction method based on semantic graph | |
CN109272452A (en) | Learn the method for super-resolution network in wavelet field jointly based on bloc framework subband | |
CN103606136B (en) | Based on the video super resolution of key frame and non local constraint | |
CN114463183A (en) | Image super-resolution method based on frequency domain and spatial domain | |
CN111768340A (en) | Super-resolution image reconstruction method and system based on dense multi-path network | |
Yeh et al. | HEVC intra frame coding based on convolutional neural network | |
Chen et al. | Single image super-resolution based on deep learning and gradient transformation | |
Chen et al. | Hierarchical generative adversarial networks for single image super-resolution | |
CN117274059A (en) | Low-resolution image reconstruction method and system based on image coding-decoding | |
CN110288529A (en) | A kind of single image super resolution ratio reconstruction method being locally synthesized network based on recurrence | |
Wu et al. | Lightweight asymmetric convolutional distillation network for single image super-resolution |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |