CN110084745A - Image super-resolution rebuilding method based on dense convolutional neural networks in parallel - Google Patents
Image super-resolution rebuilding method based on dense convolutional neural networks in parallel Download PDFInfo
- Publication number
- CN110084745A CN110084745A CN201910185854.4A CN201910185854A CN110084745A CN 110084745 A CN110084745 A CN 110084745A CN 201910185854 A CN201910185854 A CN 201910185854A CN 110084745 A CN110084745 A CN 110084745A
- Authority
- CN
- China
- Prior art keywords
- dense
- neural networks
- convolutional neural
- characteristic pattern
- resolution
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000004913 activation Effects 0.000 claims description 11
- 239000000284 extract Substances 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 2
- 230000000694 effects Effects 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000004064 recycling Methods 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 210000004209 hair Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000008929 regeneration Effects 0.000 description 1
- 238000011069 regeneration method Methods 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of single-frame image super-resolution reconstruction methods in parallel based on dense convolutional neural networks, it the described method comprises the following steps: the dense convolutional neural networks that building is made of the dense connection structure block of two parallel connections and a skip floor connection structure, for two dense connection structure blocks in parallel as circulation sub-block, every branch includes a sub- block structure and identical mapping；Dense convolutional neural networks are trained, the low-resolution image of arbitrary size is inputted, load trained model, export the high-definition picture of reconstruction.The thought of dense convolutional neural networks structure is applied to the super-resolution rebuilding of single-frame images by the present invention, and improves network structure on the basis of dense convolutional neural networks structure, has been advanced optimized model framework, has been improved algorithm effect.
Description
Technical field
The present invention relates to field of image processing more particularly to a kind of image based on dense convolutional neural networks in parallel are super
Resolution reconstruction method.
Background technique
Image super-resolution rebuilding has extensive theoretical value and practical value, and image super-resolution rebuilding refers to one
Width or several low-resolution images obtain high-definition picture by relevant algorithm process.Mainly will in early stage research
One group there are the different sequence of low resolution pictures of space displacement fog-level to reconstruct high-definition picture, in research side later
It is more biased towards in instructing to rebuild with single low-resolution image additional prior knowledge, referred to as single-frame images super-resolution rebuilding skill
Art.Image after super-resolution rebuilding includes details more abundant, and higher details resolving power, more meeting human eye vision needs
It asks, is applied to the fields such as recognition of face, Car license recognition and Medical Image Processing gradually.
Single-frame images super-resolution rebuilding algorithm is roughly divided into three classes: based on interpolation, based on reconstruction and based on study
Method:
1) method based on interpolation assumes that the gray value of pixel is consecutive variations, and utilizes the gray value meter of neighborhood pixels
The gray value of interpolation pixel is calculated, classical image interpolation method has arest neighbors to insert method, bilinear interpolation and bi-cubic interpolation
Deng.Method based on interpolation is simple and easy, but reconstruction image high-frequency information lacks, and image is excessively fuzzy.
2) based on the method for reconstruction on the basis of the method based on interpolation, some Image Priori Knowledges are artificially added, rebuild
Detail of the high frequency more abundant out.But since Image Priori Knowledge depends on the constraint of practical problem, so based on weight
The method robustness built is not high.
3) it is directed to this problem, the method that the method based on study utilizes machine learning is current by establishing model learning
The priori knowledge of super-resolution problem restores high-frequency information for low sampled data Problems of Reconstruction and provides new approaches, becomes close several
The research hotspot in year.In view of deep learning image domains successful application the experiment has found that convolutional neural networks are in single-frame images
Convolutional neural networks structure also fruitful in super-resolution rebuilding problem, general at present, such as: VGG (Visual
Geometry Group), residual error network (Residual Network, ResNet) etc., it is super all greatly to improve single-frame images
The effect of resolution reconstruction.
But the method based on VGG and residual error network does not all utilize image local feature fully, is difficult by simple
The strategy that ground increases network depth improves reconstructed image quality.
Summary of the invention
The present invention provides a kind of image super-resolution rebuilding method based on dense convolutional neural networks in parallel, this hairs
The bright thought by dense convolutional neural networks structure (Dense Convolutional Network, DenseNet) is applied to list
The super-resolution rebuilding of frame image, and network structure is improved on the basis of dense convolutional neural networks structure, it is further excellent
Change model framework, improves algorithm effect, described below:
A kind of single-frame image super-resolution reconstruction method in parallel based on dense convolutional neural networks, the method includes with
Lower step:
The dense convolutional neural networks being made of the dense connection structure block of two parallel connections and a skip floor connection structure are constructed,
For two dense connection structure blocks in parallel as circulation sub-block, every branch includes a sub- block structure and identical mapping；
Dense convolutional neural networks are trained, the low-resolution image of arbitrary size is inputted, load trained mould
Type exports the high-definition picture of reconstruction.
Wherein, the dense convolutional neural networks specifically:
The characteristic pattern that convolutional layer exports is input in dense link block in parallel；Dense link block in parallel is stacked, net is increased
Network depth extracts characteristic information；Low-resolution image block is input in skip floor connection structure；
By the output feature of the output characteristic pattern and skip floor connection structure of the last one multi-connection block structure in two branches
Figure is overlapped in channel dimension, shallow-layer feature is combined with further feature, obtained characteristic pattern is as next layer network
Input；
Characteristic pattern is limited to tri- channels RGB by convolutional layer, and uses Tanh activation primitive；
Further, sub-block structure a: characteristic pattern is input to 128 1 × 1 convolutional layers, reduces the dimension of characteristic pattern, so
It is connected in 32 5 × 5 convolutional layers afterwards, uses ReLU as activation primitive；
The characteristic pattern of output is input in 1 × 1 convolutional layer and the structure of sub-pix convolutional layer composition, sub-pix convolutional layer
High-resolution features figure is spliced into using low resolution characteristic pattern in order；
Sub-block structure b: characteristic pattern is input to 32 1 × 1 convolutional layers, reduce characteristic pattern dimension be input to 192 3 ×
3 convolutional layer finally passes through 32 3 × 3 convolution kernels using ReLU as activation primitive.
Wherein, the identical mapping are as follows:
In each multi-connection block structure, all using the output of a upper multi-connection block structure as next multi-connection block structure
Input.
It is preferably, described that dense convolutional neural networks are trained specifically:
Set learning rate and the number of iterations: using L2 norm as loss function, using adaptive matrix Estimation Optimization side
Method.
The beneficial effect of the technical scheme provided by the present invention is that:
(1) present invention is using dense convolutional neural networks for increasing to shallow-layer in single-frame images super-resolution rebuilding
The recycling of local feature strengthens the fusion of the local feature and global characteristics under different feeling open country, extracts more abundant
Detailed information；Optimize the existing super resolution ratio reconstruction method based on convolutional neural networks；
(2) present invention proposes to promote receptive field using multiple small concatenated modes of convolution kernel, improves the non-linear of network
Ability also improves the quality of reconstruction image while increasing receptive field；
(3) in the case where identical parameters, the present invention increases ReLU (Rectified Liner Uints, line rectification letter
Number) before convolution kernel number, under the premise of not increasing computing cost, increase ReLU before characteristic pattern width, extract more images
Characteristic information；
(4) present invention solves the problems, such as feature disperse in propagation, skip floor connection structure is increased, by shallow-layer feature and deep layer
Feature is added connection, further increases the quality of reconstruction image；
(5) present invention devises a network structure end to end, eliminates cumbersome pretreatment and post-treating and other steps,
Method is simple and easy；Since the present invention is only with the structure of convolutional layer, so arbitrary size can be amplified in phase of regeneration
Low-resolution image has preferable generalization ability.
Detailed description of the invention
Fig. 1 is the structural schematic diagram of dense convolutional neural networks.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
Ground detailed description.
Embodiment 1
The embodiment of the present invention proposes a kind of single-frame images super-resolution rebuilding side in parallel based on dense convolutional neural networks
Method, referring to Fig. 1, comprising the following steps:
101: the dense convolutional Neural that building is made of the dense connection structure block of two parallel connections and a skip floor connection structure
Network, as circulation sub-blocks, every branch includes a sub- block structure and identical reflects two dense connection structure blocks in parallel
It penetrates；
102: dense convolutional neural networks being trained, the low-resolution image of arbitrary size is inputted, load trains
Model, export the high-definition picture of reconstruction.
Wherein, the dense convolutional neural networks in step 101 specifically:
The characteristic pattern that convolutional layer exports is input in dense link block in parallel；Dense link block in parallel is stacked, net is increased
Network depth extracts characteristic information；Low-resolution image block is input in skip floor connection structure；
By the output feature of the output characteristic pattern and skip floor connection structure of the last one multi-connection block structure in two branches
Figure is overlapped in channel dimension, shallow-layer feature is combined with further feature, obtained characteristic pattern is as next layer network
Input；
Characteristic pattern is limited to tri- channels RGB by convolutional layer, and uses Tanh activation primitive.
In conclusion the embodiment of the present invention is used in single-frame images super-resolution rebuilding using dense convolutional neural networks,
The recycling to shallow-layer local feature is increased, the fusion of the local feature and global characteristics under different feeling open country is strengthened, mentions
Take detailed information more abundant；Optimize the existing super-resolution rebuilding based on convolutional neural networks.
Embodiment 2
Below with reference to specific example, Fig. 1, the scheme in embodiment 1 is further introduced, it is as detailed below to retouch
It states:
One, data preparation:
Step 1: dividing data set:
The embodiment of the present invention is tested using NTIRE (New Trends in Image Restoration and
Enhancement workshop) DIV2K data set disclosed in contest, the data set totally 1000 figures, respectively 800 training
Figure, 100 proof diagrams and 100 test charts.Since the test of this data set is no open with figure, it will in experimentation
100 proof diagrams rebuild effect for testing.All images are by one group of high resolution graphics and its corresponding low-resolution image
Composition, wherein low-resolution image is generated by the down-sampled model of bicubic interpolation.
Wherein, the down-sampled model of bicubic interpolation is known to those skilled in the art, and the embodiment of the present invention does not do this
It repeats.
Step 2: 800 trained figures are cut into the image block of 96 × 96 sizes as network inputs.
Two, network structure is built:
The network structure of the embodiment of the present invention is mainly by two dense connection structure blocks in parallel and a skip floor connection knot
Structure composition.Below in conjunction with Fig. 1, explanation is described in detail to the network structure that the embodiment of the present invention is built:
Step 1: every time iteration input 16 96 × 96 sizes cut low-resolution image block, by 64 3 ×
3 convolution kernel.
Step 2: the characteristic pattern that 3 × 3 convolutional layers export is input in dense link block in parallel, based on dense convolution mind
Thought through network structure, as circulation sub-block in the form of two branch circuit parallel connections, every branch wraps the embodiment of the present invention
Containing a sub- block structure and identical mapping, two parallel branches are described in detail as follows:
(a) sub-block structure a: characteristic pattern is input to 128 1 × 1 convolutional layers, reduces the dimension of characteristic pattern, and uses
ReLU is as activation primitive.Then it is connected in 32 5 × 5 convolutional layers, uses ReLU as activation primitive, increase network
Non-thread sexuality.
The characteristic pattern of output is input in 1 × 1 convolutional layer and the structure of sub-pix convolutional layer composition, sub-pix convolutional layer
Low resolution characteristic pattern is amplified to accordingly by the method for being spliced into high-resolution features figure in order using low resolution characteristic pattern
Multiple.For example, being H by height, width W, port number are C × r2Low resolution characteristic pattern (H × W × C × r2) amplification r
Times, obtain height be rH, the high-resolution characteristic pattern (rH × rW × C) that width rW, port number are C.
(b) sub-block structure b: characteristic pattern is input to 32 1 × 1 convolutional layers, carries out dimension-reduction treatment to characteristic pattern, will export
Characteristic pattern be input to 192 3 × 3 convolutional layers, using ReLU as activation primitive, use greater number of convolution herein
Core is the port number in order to increase characteristic pattern before ReLU, finally passes through 32 3 × 3 convolution kernels.
(c) identical mapping:, directly will be upper all comprising one from the connection for being input to output in each multi-connection block structure
Input of the output of one multi-connection block structure as next multi-connection block structure, to strengthen the sharp again of shallow-layer feature
With, increase extract feature complexity and diversity.
Step 3: stacking dense link block in parallel, increase network depth, extracts more abundant, complicated characteristic information.
Step 4: the low-resolution image block for 96 × 96 sizes that the first step is cut is input in skip floor connection structure,
Skip floor connection structure is made of 256 3 × 3 convolutional layers and 1 sub-pix convolutional layer.
Step 5: by two branches output characteristic pattern and the skip floor connection structure of the last one multi-connection block structure it is defeated
Characteristic pattern is overlapped in channel dimension out, and shallow-layer feature is combined with further feature, under 192 obtained characteristic pattern is used as
The input of one layer network.
Step 6: characteristic pattern is limited to tri- channels RGB using 31 × 1 convolutional layers, and use Tanh activation primitive.
Three, model training:
Step 1: the learning rate in the embodiment of the present invention is set as 10-3, and every 75000 iterative learning rates reduce half,
Until error no longer reduce or 250000 iteration after, deconditioning.
Step 2: using L2 norm as loss function.
Step 3: using ADAM (Adaptive Moment Estimation, adaptive matrix estimation) optimization method, β1
=0.9, β2=0.999.
Wherein, the step of above-mentioned model training is known to those skilled in the art, and the embodiment of the present invention does not do this superfluous
It states.
Four, image reconstruction, the method is as follows:
The low-resolution image for inputting arbitrary size, loads trained model, exports the high-definition picture of reconstruction.
In conclusion the embodiment of the present invention proposes to promote receptive field using multiple small concatenated modes of convolution kernel, improve
The non-thread sexuality of network also improves the quality of reconstruction image while increasing receptive field.
Embodiment 3
It is verified below with reference to feasibility of the table 1 to the scheme in Examples 1 and 2, described below:
Experiment of the embodiment of the present invention is measured using Y-PSNR (Peak Signal to Noise Ratio, PSNR) calculates
Method effect, PSNR is defined as:
Wherein, f (x, y) and f'(x, y) original image and reconstruction figure are respectively represented, M × N represents spatial resolution.Higher
PSNR value represents better reconstruction effect.On DIV2K data set, the specific PSNR value of the experiment of the embodiment of the present invention such as 1 institute of table
Show.
1 present invention of table compares (unit: dB) using the PSNR value of model
In conclusion method provided in an embodiment of the present invention improves on the basis of conventional method Bicubic algorithm
3.91dB improves reconstructed image quality.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (5)
1. a kind of single-frame image super-resolution reconstruction method in parallel based on dense convolutional neural networks, which is characterized in that described
Method the following steps are included:
The dense convolutional neural networks that building is made of the dense connection structure block of two parallel connections and a skip floor connection structure, two
For dense connection structure block in parallel as circulation sub-block, every branch includes a sub- block structure and identical mapping；
Dense convolutional neural networks are trained, the low-resolution image of arbitrary size is inputted, load trained model, it is defeated
The high-definition picture rebuild out.
2. a kind of single-frame images super-resolution rebuilding side in parallel based on dense convolutional neural networks according to claim 1
Method, which is characterized in that the dense convolutional neural networks specifically:
The characteristic pattern that convolutional layer exports is input in dense link block in parallel；Dense link block in parallel is stacked, it is deep to increase network
Degree extracts characteristic information；Low-resolution image block is input in skip floor connection structure；
The output characteristic pattern of the output characteristic pattern of the last one multi-connection block structure in two branches and skip floor connection structure is existed
Channel dimension is overlapped, and shallow-layer feature is combined with further feature, input of the obtained characteristic pattern as next layer network；
Characteristic pattern is limited to tri- channels RGB by convolutional layer, and uses Tanh activation primitive.
3. a kind of single-frame images Super-resolution reconstruction in parallel based on dense convolutional neural networks according to claim 1 or 2
Construction method, which is characterized in that
Sub-block structure a: characteristic pattern is input to 128 1 × 1 convolutional layers, reduces the dimension of characteristic pattern, is then connected to 32
In 5 × 5 convolutional layer, use ReLU as activation primitive；
The characteristic pattern of output is input in 1 × 1 convolutional layer and the structure of sub-pix convolutional layer composition, and sub-pix convolutional layer uses
Low resolution characteristic pattern is spliced into high-resolution features figure in order；
Sub-block structure b: characteristic pattern is input to 32 1 × 1 convolutional layers, and the dimension for reducing characteristic pattern is input to 192 3 × 3
Convolutional layer finally passes through 32 3 × 3 convolution kernels using ReLU as activation primitive.
4. a kind of single-frame images Super-resolution reconstruction in parallel based on dense convolutional neural networks according to claim 1 or 2
Construction method, which is characterized in that the identical mapping are as follows:
In each multi-connection block structure, all by the output of a upper multi-connection block structure as the defeated of next multi-connection block structure
Enter.
5. a kind of single-frame images super-resolution rebuilding side in parallel based on dense convolutional neural networks according to claim 1
Method, which is characterized in that described that dense convolutional neural networks are trained specifically:
Set learning rate and the number of iterations: using L2 norm as loss function, using adaptive matrix Estimation Optimization method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185854.4A CN110084745A (en) | 2019-03-12 | 2019-03-12 | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185854.4A CN110084745A (en) | 2019-03-12 | 2019-03-12 | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110084745A true CN110084745A (en) | 2019-08-02 |
Family
ID=67413160
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910185854.4A Pending CN110084745A (en) | 2019-03-12 | 2019-03-12 | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084745A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647934A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN111080688A (en) * | 2019-12-25 | 2020-04-28 | 左一帆 | Depth map enhancement method based on depth convolution neural network |
| CN112288626A (en) * | 2020-10-10 | 2021-01-29 | 武汉大学 | Face illusion method and system based on dual-path depth fusion |
| CN112489103A (en) * | 2020-11-19 | 2021-03-12 | 北京的卢深视科技有限公司 | High-resolution depth map acquisition method and system |
| CN113139904A (en) * | 2021-04-29 | 2021-07-20 | 厦门大学 | Image blind super-resolution method and system |
| CN114066726A (en) * | 2020-08-04 | 2022-02-18 | 深圳市中兴微电子技术有限公司 | Image processing method, device, equipment and storage medium |
| TWI826184B (en) * | 2022-12-14 | 2023-12-11 | 瑞昱半導體股份有限公司 | Super resolution image generating device and super resolution image generating method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
| CN107240066A (en) * | 2017-04-28 | 2017-10-10 | 天津大学 | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks |
| EP3319039A1 (en) * | 2016-11-07 | 2018-05-09 | UMBO CV Inc. | A method and system for providing high resolution image through super-resolution reconstruction |
| CN109242849A (en) * | 2018-09-26 | 2019-01-18 | 上海联影智能医疗科技有限公司 | Medical image processing method, device, system and storage medium |
| WO2019025298A1 (en) * | 2017-07-31 | 2019-02-07 | Institut Pasteur | Method, device, and computer program for improving the reconstruction of dense super-resolution images from diffraction-limited images acquired by single molecule localization microscopy |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
-
2019
- 2019-03-12 CN CN201910185854.4A patent/CN110084745A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
| EP3319039A1 (en) * | 2016-11-07 | 2018-05-09 | UMBO CV Inc. | A method and system for providing high resolution image through super-resolution reconstruction |
| CN107240066A (en) * | 2017-04-28 | 2017-10-10 | 天津大学 | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks |
| WO2019025298A1 (en) * | 2017-07-31 | 2019-02-07 | Institut Pasteur | Method, device, and computer program for improving the reconstruction of dense super-resolution images from diffraction-limited images acquired by single molecule localization microscopy |
| CN109242849A (en) * | 2018-09-26 | 2019-01-18 | 上海联影智能医疗科技有限公司 | Medical image processing method, device, system and storage medium |
| CN109360152A (en) * | 2018-10-15 | 2019-02-19 | 天津大学 | 3 d medical images super resolution ratio reconstruction method based on dense convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| FU ZHANG 等: "Image super-resolution via a novel cascaded convolutional neural network framework", 《SIGNAL PROCESSING: IMAGE COMMUNICATION》 * |
| 姚琴娟 等: "基于双通道CNN的单幅图像超分辨率重建", 《华东理工大学学报（自然科学版）》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647934A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN110647934B (en) * | 2019-09-20 | 2022-04-08 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN111080688A (en) * | 2019-12-25 | 2020-04-28 | 左一帆 | Depth map enhancement method based on depth convolution neural network |
| CN114066726A (en) * | 2020-08-04 | 2022-02-18 | 深圳市中兴微电子技术有限公司 | Image processing method, device, equipment and storage medium |
| CN112288626A (en) * | 2020-10-10 | 2021-01-29 | 武汉大学 | Face illusion method and system based on dual-path depth fusion |
| CN112489103A (en) * | 2020-11-19 | 2021-03-12 | 北京的卢深视科技有限公司 | High-resolution depth map acquisition method and system |
| CN112489103B (en) * | 2020-11-19 | 2022-03-08 | 北京的卢深视科技有限公司 | High-resolution depth map acquisition method and system |
| CN113139904A (en) * | 2021-04-29 | 2021-07-20 | 厦门大学 | Image blind super-resolution method and system |
| CN113139904B (en) * | 2021-04-29 | 2022-05-27 | 厦门大学 | Image blind super-resolution method and system |
| TWI826184B (en) * | 2022-12-14 | 2023-12-11 | 瑞昱半導體股份有限公司 | Super resolution image generating device and super resolution image generating method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084745A (en) | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel | |
| CN109903228B (en) | Image super-resolution reconstruction method based on convolutional neural network | |
| CN107240066A (en) | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks | |
| CN106683067B (en) | Deep learning super-resolution reconstruction method based on residual sub-images | |
| CN108765290A (en) | A kind of super resolution ratio reconstruction method based on improved dense convolutional neural networks | |
| CN111861961B (en) | Single image super-resolution multi-scale residual error fusion model and restoration method thereof | |
| CN110276721A (en) | Image super-resolution rebuilding method based on cascade residual error convolutional neural networks | |
| CN110197468A (en) | A kind of single image Super-resolution Reconstruction algorithm based on multiple dimensioned residual error learning network | |
| CN108765291A (en) | Super resolution ratio reconstruction method based on dense neural network and two-parameter loss function | |
| CN107123089A (en) | Remote sensing images super-resolution reconstruction method and system based on depth convolutional network | |
| CN108805808A (en) | A method of improving video resolution using convolutional neural networks | |
| CN107563965A (en) | Jpeg compressed image super resolution ratio reconstruction method based on convolutional neural networks | |
| CN109727195B (en) | Image super-resolution reconstruction method | |
| CN110751597B (en) | Video super-resolution method based on coding damage repair | |
| CN110211038A (en) | Super resolution ratio reconstruction method based on dirac residual error deep neural network | |
| CN111028150A (en) | Rapid space-time residual attention video super-resolution reconstruction method | |
| CN111798377B (en) | Magnetic resonance image super-resolution reconstruction method based on multi-resolution learning strategy | |
| CN109934771B (en) | Unsupervised remote sensing image super-resolution reconstruction method based on recurrent neural network | |
| CN109949224A (en) | A kind of method and device of the connection grade super-resolution rebuilding based on deep learning | |
| CN112767243B (en) | Method and system for realizing super-resolution of hyperspectral image | |
| CN110163803A (en) | A kind of image super-resolution rebuilding method and system based on convolutional neural networks | |
| CN104299193B (en) | Image super-resolution reconstruction method based on high-frequency information and medium-frequency information | |
| CN114331831A (en) | Light-weight single-image super-resolution reconstruction method | |
| CN116091313A (en) | Image super-resolution network model and reconstruction method | |
| CN113781308A (en) | Image super-resolution reconstruction method and device, storage medium and electronic equipment |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190802 |