CN107730453A - Picture quality method for improving - Google Patents
Picture quality method for improving Download PDFInfo
- Publication number
- CN107730453A CN107730453A CN201711117769.1A CN201711117769A CN107730453A CN 107730453 A CN107730453 A CN 107730453A CN 201711117769 A CN201711117769 A CN 201711117769A CN 107730453 A CN107730453 A CN 107730453A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- improving
- picture quality
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000000694 effects Effects 0.000 claims abstract description 37
- 230000000903 blocking effect Effects 0.000 claims abstract description 33
- 238000013135 deep learning Methods 0.000 claims abstract description 11
- 230000003321 amplification Effects 0.000 claims abstract description 9
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 9
- 230000004927 fusion Effects 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 15
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000006835 compression Effects 0.000 claims description 4
- 238000007906 compression Methods 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 230000001537 neural effect Effects 0.000 claims 2
- 239000007787 solid Substances 0.000 claims 1
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008092 positive effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012545 processing 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 transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- G06T5/73—
-
- 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]
Abstract
The invention discloses a kind of picture quality method for improving, comprise the following steps:S1, establish the image deblocking effect model based on deep learning;S2, establish the image super-resolution model based on deep learning;S3, fusion described image deblocking effect model and image super-resolution model, are handled target image, to lift objective image quality.The quality automatic lifting of fused images deblocking effect model and image super-resolution model realization low-resolution image of the present invention, the problem of image blurring occurred in low-resolution image amplification process and image block effect problem are taken into full account, while the blocking effect of low-resolution image is removed, also the redundancy that image amplification is brought is greatly reduced, ensure that the effect of low-resolution image increased quality.
Description
Technical field
The present invention relates to image processing field, more particularly to a kind of picture quality method for improving.
Background technology
Image contains abundant and intuitively information, currently in fields such as the social activities, shopping and tourism of internet, is required for
Substantial amounts of image gives user's transmission information.Because the source of the Internet images is a lot, causes picture quality uneven, exist
Many low-resolution images, this parts of images can amplify due to size during displaying and occur that image is fuzzy to ask
Topic, while size amplification can also make the blocking effect phenomenon of image more serious, these low-quality images have had a strong impact on user
Experience.With the rapid development of Internet, image provider can utilize all multipaths to obtain great amount of images daily, using artificial
The mode for handling low-resolution image is especially time-consuming, it is necessary to spend substantial amounts of human cost.Therefore, using machine automatically to low point
Resolution image, which carries out increased quality, becomes urgent demand.
Traditional a series of images increased quality method based on interpolation is on the one hand more difficult to eliminate because picture size becomes big
(super-resolution) introduces the problem of image blurring that redundancy is brought, and still further aspect can not solve image blocking effect and deepen to ask
Topic.At this stage, the picture quality method for improving based on deep learning is suggested in succession, but is considered that solution image is fuzzy simultaneously and asked
Topic and image block effect problem method it is less, realize low-resolution image quality be obviously improved be still one have challenge
The task of property.
The content of the invention
The technical problem to be solved in the present invention is that significantly low-resolution image can not be entered in the prior art in order to overcome
The defects of row increased quality, there is provided a kind of picture quality of practical, increased quality positive effect low-resolution image carries
Lifting method.
The present invention is that solve above-mentioned technical problem by following technical proposals:
The invention provides a kind of picture quality method for improving, comprise the following steps:
S1, establish the image deblocking effect model based on deep learning;
S2, establish the image super-resolution model based on deep learning;
S3, fusion described image deblocking effect model and image super-resolution model, are handled target image, to carry
Rise objective image quality.
It is preferred that step S1 includes:
S11, establish the first image data set;
S12, image deblocking effect model of the structure based on the full convolutional neural networks of depth;
S13, the first image data set training step S12 established using step S11 are built refreshing based on the full convolution of depth
Image deblocking effect model through network.
It is preferred that the image that the first view data is concentrated in step S11 Mass production by way of machine automatically processes
To obtain.
It is preferred that step S11 includes:
S111, collect some first images;Wherein, the quality of described first image is higher than the target image;
S112, quality compression is carried out with different compressibility factors to the described first image in step S111, obtain and walk
The size image with blocking effect consistent with content of described first image in rapid S111;
Divide described in S113, the described first image in step S111 and step S112 on the image with blocking effect
The image block of some fixed dimensions is not intercepted;
S114, using the image block with blocking effect obtained in step S113 as the first image pattern, by step S113
The image block of the first image corresponding with the image block with blocking effect of middle acquisition is as described first image sample
Label.
It is preferred that the full convolutional network of the depth in step S12 has 20 layers of network structure, using several residual errors
Unit and batch normalization layer accelerate network convergence, and network uses two norm loss functions.
It is preferred that step S2 includes:
S21, establish the second image data set;
S22, image super-resolution model of the structure based on the full convolutional neural networks of depth;
S23, the second image data set training step S22 established using step S21 are built refreshing based on the full convolution of depth
Image super-resolution model through network.
It is preferred that the image that the second view data is concentrated in step S21 Mass production by way of machine automatically processes
To obtain.
It is preferred that step S21 includes:
S211, collect some second images;Wherein, the high resolution of second image is in the target image;
S212, first carrying out again entering second image in step S211 by way of interpolation after several times are down-sampled
The diminution and amplification of row size, obtain the blurred picture consistent with the size of second image in step S211;
Cut respectively on the corresponding blurred picture in S213, second image and step S212 in step S211
Take the image block of some fixed dimensions;
S214, using the image block of the blurred picture obtained in step S213 as the second image pattern, by step S213
Label of the image block of the second image corresponding with the image block of the blurred picture obtained as second image pattern.
It is preferred that the full convolutional network of the depth in step S22 has 20 layers of network structure, using several residual errors
Unit and batch normalization layer accelerate network convergence, and network uses two norm loss functions.
It is preferred that step S3 includes:
S31, the image deblocking effect model treatment target image established using step S1, complete the blocking effect of target image
Remove;
S32, the image super-resolution model treatment established using step S2 complete the target image that blocking effect removes, and complete
The super-resolution of target image;
S33, repeat step S31 and S32, complete the increased quality of the target image.
It is preferred that in step S31, the input of described image deblocking effect model is target image, is exported to remove block
The target image of effect.
It is preferred that in step s 32, the input of described image super-resolution model is the removal block obtained in step S31
Blurred picture after the interpolated several times of target image of effect, exports the higher quality image for identical size.
The positive effect of the present invention is:
1st, the present invention establishes the image deblocking effect model based on deep learning, trained one kind specifically for image deblocking
The full convolutional neural networks of depth of effect, it can accurately restore the letter of normal picture corresponding to the image containing blocking effect
Breath, preferably solves the problems, such as the blocking effect of image.
2nd, the present invention establishes the image super-resolution model based on deep learning, trained one kind specifically for image oversubscription
The full convolutional neural networks of depth of resolution, it can accurately restore the letter of high-definition picture corresponding to low-resolution image
Breath, preferably solves the problems, such as that low-resolution image introduces bulk redundancy information in amplification process.
3rd, the quality of fused images deblocking effect model and image super-resolution model realization low-resolution image of the present invention
Automatic lifting, the problem of image blurring occurred in low-resolution image amplification process and image block effect problem are taken into full account,
While the blocking effect of low-resolution image is removed, the redundancy that image amplification is brought also is greatly reduced, is ensured
The effect of low-resolution image increased quality.
Brief description of the drawings
Fig. 1 is the flow chart of the picture quality method for improving of presently preferred embodiments of the present invention.
Fig. 2 be presently preferred embodiments of the present invention picture quality method for improving in step 101 flow chart.
Fig. 3 be presently preferred embodiments of the present invention picture quality method for improving in step 1011 flow chart.
Fig. 4 be presently preferred embodiments of the present invention picture quality method for improving in step 102 flow chart.
Fig. 5 be presently preferred embodiments of the present invention picture quality method for improving in step 1021 flow chart.
Fig. 6 be presently preferred embodiments of the present invention picture quality method for improving in step 103 flow chart.
Embodiment
The present invention is further illustrated below by the mode of embodiment, but does not therefore limit the present invention to described reality
Apply among a scope.
As shown in figure 1, the picture quality method for improving of the present invention comprises the following steps:
Step 101, establish the image deblocking effect model based on deep learning;
Step 102, establish the image super-resolution model based on deep learning;
Step 103, fusion described image deblocking effect model and image super-resolution model, at target image
Reason, to lift objective image quality.
The quality automatic lifting of low-resolution image can be realized by the picture quality method for improving of the present invention, wherein low
Image in different resolution refers to image of the pixel size below 550 × 412, i.e., the pixel size of described target image is 550 × 412
Below.
Wherein, as shown in Fig. 2 step 101 specifically includes following steps:
Step 1011, establish the first image data set;
Step 1012, image deblocking effect model of the structure based on the full convolutional neural networks of depth;
Step 1013, the first image data set training step 1012 established using step 1011 build complete based on depth
The image deblocking effect model of convolutional neural networks.
In the present embodiment, the image that first view data is concentrated in step 1011 is criticized by way of machine automatically processes
Amount is generated to obtain.Specifically, as shown in figure 3, step 1011 comprises the following steps:
Step 10111, the first image of some high quality of collection (specifically can specifically select directly according to visually choosing
What is taken is the image of fine definition);Wherein, the quality of described first image is higher than the target image;In the present embodiment, have
Body collects 2000 the first images altogether;
Step 10112, quality compression is carried out with different compressibility factors to the described first image in step 10111, obtained
Take image consistent with the size of the described first image in step 10111 and content while that there is blocking effect;In the present embodiment
In, compressibility factor includes 10,20,30,40 and 50, and described first image causes blocking effect occur due to compression;
With blocking effect described in step 10113, described first image and step 10112 in step 10111
The image block of some fixed dimensions is intercepted on image respectively;In the present embodiment, it is big to be arranged to 64 × 64 pixels for fixed dimension
Small, it is respectively 200,000 to have the final amt of the image block of blocking effect and the image block of corresponding first image;
Step 10114, using the image block with blocking effect obtained in step 10113 as the first image pattern, will walk
The image block of the first image corresponding with the image block with blocking effect obtained in rapid 10113 is as described first image
The label of sample.
Specifically, the full convolutional network of the depth in step 1012 has 20 layers of network structure, residual using several
Poor unit and batch normalization layer accelerate network convergence, strengthen the deblocking effect ability of network, network uses two norm loss functions.
In the present embodiment, the quantity of residual unit is 5, and each residual unit includes 3 convolutional layers, and the training input of network is 64
The image of × 64 × 1 pixel size.
Specifically, in step 1013, the quantity per a collection of training image is set to 32, and learning rate is set to 0.001, momentum system
Number is set to 0.9, and weights attenuation coefficient is set to 0.0001.All image pattern (image block with blocking effect) head for participating in training
The view data of YCbCr patterns is first converted to, the view data for then extracting Y passages inputs as the training of network, and data are big
Small is 64 × 64 × 1, and the input of the network with being built in step 1012 is consistent.Similarly, all labels (as image pattern
The image block of one image) view data of YCbCr patterns is also converted to, the view data of Y passages is then extracted as corresponding defeated
Enter the label of data.
Wherein, as shown in figure 4, step 102 includes:
Step 1021, establish the second image data set;
Step 1022, image super-resolution model of the structure based on the full convolutional neural networks of depth;
Step 1023, the second image data set training step 1022 established using step 1021 build complete based on depth
The image super-resolution model of convolutional neural networks.
In the present embodiment, the image that first view data is concentrated in step 1021 is criticized by way of machine automatically processes
Amount is generated to obtain.Specifically, as shown in figure 5, step 1021 comprises the following steps:
Step 10211, collect some high-resolution second images;Wherein, the high resolution of second image is in institute
Target image is stated, specifically, the pixel size of second image is more than 1200 × 800;
Step 10212, first carry out several times it is down-sampled after again by way of interpolation to described second in step 10211
Image carries out the diminution and amplification of size, obtains the blurred picture consistent with the size of second image in step 10211;
In the present embodiment, the average using 4 times is down-sampled, and interpolation is using bilinear interpolation;
The corresponding blurred picture in step 10213, second image in step 10211 and step 10212
The upper image block for intercepting some fixed dimensions respectively;In the present embodiment, fixed dimension is arranged to 64 × 64 pixel sizes, mould
The final amt for pasting the image block of image and the image block of corresponding second image is respectively 200,000;
Step 10214, using the image block of the blurred picture obtained in step 10213 as the second image pattern, by step
The image block of the second image corresponding with the image block of the blurred picture obtained in 10213 is as second image pattern
Label.
Specifically, the full convolutional network of the depth in step 1022 has 20 layers of network structure, residual using several
Poor unit and batch normalization layer accelerate network convergence, strengthen the deblocking effect ability of network, network uses two norm loss functions.
In the present embodiment, the quantity of residual unit is 5, and each residual unit includes 3 convolutional layers, and the training input of network is 64
The image of × 64 × 3 pixel sizes.
Specifically, in step 1023, the quantity per a collection of training image is set to 32, and learning rate is set to 0.001, momentum system
Number is set to 0.9, and weights attenuation coefficient is set to 0.0001.All image patterns (image block of blurred picture) for participating in training are turned
Inputted for the view data of RGB patterns as the training of network, size of data is 64 × 64 × 3, with building in step 1022
The input of network is consistent.Similarly, all labels (image block of the second image) as image pattern are also converted to RGB patterns
Label of the view data as corresponding input data.
Specifically, as shown in fig. 6, step 103 includes:
Step 1031, the image deblocking effect model treatment target image established using step 101, complete target image
Blocking effect removes;
Step 1032, the image super-resolution model treatment established using step 102 complete the target figure that blocking effect removes
Picture, the super-resolution of target image is completed, obtain final high-definition picture;
Step 1033, repeat step 1031 and 1032, complete the increased quality of the target image of all low resolution.
Specifically, in step 1031, the input of image deblocking effect model is the target image of original low resolution,
The output of model is the target image for the low resolution for eliminating blocking effect.In the present embodiment, for one original low point
Resolution image, YCbCr patterns are converted into first, extract the view data of Y passages, and be inputted step 1013 and train
To network in, it is then that the output data of network and image Cb under YCbCr patterns and Cr two with blocking effect is logical
The view data in road merges, and the view data after merging finally is converted into the storage of RGB patterns, you can obtains removing blocking effect
The target image of low resolution.
Specifically, in step 1032, the input of image super-resolution model is the removal block effect obtained in step 1031
Blurred picture after the interpolated several times of target image for the low resolution answered, is exported as the high quality graphic of identical size.
In the present embodiment, interpolation multiple is 4 times, and interpolation method is bilinear interpolation.Blurred picture after the interpolation is switched into RGB moulds
The view data of formula, and be inputted step 1023 training and obtain in network, the output of network is obtained as final high-resolution
Rate image.
In the picture quality method for improving of the low-resolution image of the present embodiment, image deblocking effect model and image surpass
Resolution model is realized using 20 layers of the full convolutional neural networks of depth residual error, can accurately be restored containing blocking effect
The information of high-definition picture corresponding to the information and blurred picture of normal picture corresponding to image;The fusion of two models makes low
The practicality and lifting effect of image in different resolution increased quality are all preferably ensured.
Although the foregoing describing the embodiment of the present invention, it will be appreciated by those of skill in the art that these
It is merely illustrative of, protection scope of the present invention is defined by the appended claims.Those skilled in the art is not carrying on the back
On the premise of principle and essence from the present invention, various changes or modifications can be made to these embodiments, but these are changed
Protection scope of the present invention is each fallen within modification.
Claims (12)
1. a kind of picture quality method for improving, it is characterised in that comprise the following steps:
S1, establish the image deblocking effect model based on deep learning;
S2, establish the image super-resolution model based on deep learning;
S3, fusion described image deblocking effect model and image super-resolution model, are handled target image, to lift mesh
Logo image quality.
2. picture quality method for improving as claimed in claim 1, it is characterised in that step S1 includes:
S11, establish the first image data set;
S12, image deblocking effect model of the structure based on the full convolutional neural networks of depth;
S13, using step S11 establish the first image data set training step S12 build based on the full convolutional Neural net of depth
The image deblocking effect model of network.
3. picture quality method for improving as claimed in claim 2, it is characterised in that the first view data is concentrated in step S11
Image by way of machine automatically processes Mass production obtain.
4. picture quality method for improving as claimed in claim 2, it is characterised in that step S11 includes:
S111, collect some first images;Wherein, the quality of described first image is higher than the target image;
S112, quality compression, acquisition and step are carried out with different compressibility factors to the described first image in step S111
The size of described first image in the S111 image with blocking effect consistent with content;
Cut respectively on the image with blocking effect described in S113, the described first image in step S111 and step S112
Take the image block of some fixed dimensions;
S114, using the image block with blocking effect obtained in step S113 as the first image pattern, will be obtained in step S113
Label of the image block of the first image corresponding with the image block with blocking effect taken as described first image sample.
5. picture quality method for improving as claimed in claim 2, it is characterised in that the full convolution of the depth in step S12
Network has 20 layers of network structure, accelerates network convergence using several residual units and batch normalization layer, network uses two
Norm loss function.
6. picture quality method for improving as claimed in claim 1, it is characterised in that step S2 includes:
S21, establish the second image data set;
S22, image super-resolution model of the structure based on the full convolutional neural networks of depth;
S23, using step S21 establish the second image data set training step S22 build based on the full convolutional Neural net of depth
The image super-resolution model of network.
7. picture quality method for improving as claimed in claim 6, it is characterised in that the second view data is concentrated in step S21
Image by way of machine automatically processes Mass production obtain.
8. picture quality method for improving as claimed in claim 6, it is characterised in that step S21 includes:
S211, collect some second images;Wherein, the high resolution of second image is in the target image;
S212, first carrying out carrying out chi to second image in step S211 by way of interpolation again after several times are down-sampled
Very little diminution and amplification, obtain the blurred picture consistent with the size of second image in step S211;
If intercepted respectively on the corresponding blurred picture in S213, second image and step S212 in step S211
The solid image block being sized;
S214, using the image block of the blurred picture obtained in step S213 as the second image pattern, will be obtained in step S213
The second image corresponding with the image block of the blurred picture label of the image block as second image pattern.
9. picture quality method for improving as claimed in claim 6, it is characterised in that the full convolution of the depth in step S22
Network has 20 layers of network structure, accelerates network convergence using several residual units and batch normalization layer, network uses two
Norm loss function.
10. picture quality method for improving as claimed in claim 1, it is characterised in that step S3 includes:
S31, the image deblocking effect model treatment target image established using step S1, the blocking effect for completing target image are gone
Remove;
S32, the image super-resolution model treatment established using step S2 complete the target image that blocking effect removes, and complete target
The super-resolution of image;
S33, repeat step S31 and S32, complete the increased quality of the target image.
11. picture quality method for improving as claimed in claim 10, it is characterised in that in step S31, described image deblocking
The input of effect model is target image, is exported to remove the target image of blocking effect.
12. picture quality method for improving as claimed in claim 11, it is characterised in that in step s 32, described image oversubscription
The input of resolution model is the blurred picture after the interpolated several times of target image of the removal blocking effect obtained in step S31,
Export the higher quality image for identical size.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711117769.1A CN107730453A (en) | 2017-11-13 | 2017-11-13 | Picture quality method for improving |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711117769.1A CN107730453A (en) | 2017-11-13 | 2017-11-13 | Picture quality method for improving |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107730453A true CN107730453A (en) | 2018-02-23 |
Family
ID=61215181
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711117769.1A Pending CN107730453A (en) | 2017-11-13 | 2017-11-13 | Picture quality method for improving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107730453A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108921785A (en) * | 2018-06-14 | 2018-11-30 | 厦门大学 | Super resolution ratio reconstruction method based on wavelet packet |
CN109345449A (en) * | 2018-07-17 | 2019-02-15 | 西安交通大学 | A kind of image super-resolution based on converged network and remove non-homogeneous blur method |
CN109816592A (en) * | 2019-01-26 | 2019-05-28 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
CN110264415A (en) * | 2019-05-24 | 2019-09-20 | 北京爱诺斯科技有限公司 | It is a kind of to eliminate the fuzzy image processing method of shake |
CN109410158B (en) * | 2018-08-21 | 2020-10-02 | 西安电子科技大学 | Multi-focus image fusion method based on convolutional neural network |
CN113177599A (en) * | 2021-05-10 | 2021-07-27 | 南京信息工程大学 | Enhanced sample generation method based on GAN |
CN113542772A (en) * | 2020-04-21 | 2021-10-22 | 四川大学 | Compressed image deblocking method based on enhanced wide-activation residual error network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140072242A1 (en) * | 2012-09-10 | 2014-03-13 | Hao Wei | Method for increasing image resolution |
CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术(上海)有限公司 | The recognition methods of watermarked image and the system of identification |
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 |
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 |
CN106952228A (en) * | 2017-03-10 | 2017-07-14 | 北京工业大学 | The super resolution ratio reconstruction method of single image based on the non local self-similarity of image |
CN107018422A (en) * | 2017-04-27 | 2017-08-04 | 四川大学 | Still image compression method based on depth convolutional neural networks |
WO2017191461A1 (en) * | 2016-05-05 | 2017-11-09 | Magic Pony Technology Limited | Video encoding using hierarchical algorithms |
-
2017
- 2017-11-13 CN CN201711117769.1A patent/CN107730453A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140072242A1 (en) * | 2012-09-10 | 2014-03-13 | Hao Wei | Method for increasing image resolution |
WO2017191461A1 (en) * | 2016-05-05 | 2017-11-09 | Magic Pony Technology Limited | Video encoding using hierarchical algorithms |
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 |
CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术(上海)有限公司 | The recognition methods of watermarked image and the system of identification |
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 |
CN106952228A (en) * | 2017-03-10 | 2017-07-14 | 北京工业大学 | The super resolution ratio reconstruction method of single image based on the non local self-similarity of image |
CN107018422A (en) * | 2017-04-27 | 2017-08-04 | 四川大学 | Still image compression method based on depth convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
蒋雪等: ""基于卷积神经网络的图像纹理的超分辨率重建"", 《微型机与应用》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108921785A (en) * | 2018-06-14 | 2018-11-30 | 厦门大学 | Super resolution ratio reconstruction method based on wavelet packet |
CN108921785B (en) * | 2018-06-14 | 2020-07-07 | 厦门大学 | Super-resolution reconstruction method based on wavelet packet |
CN109345449A (en) * | 2018-07-17 | 2019-02-15 | 西安交通大学 | A kind of image super-resolution based on converged network and remove non-homogeneous blur method |
US11928792B2 (en) | 2018-07-17 | 2024-03-12 | Xi'an Jiaotong University | Fusion network-based method for image super-resolution and non-uniform motion deblurring |
CN109345449B (en) * | 2018-07-17 | 2020-11-10 | 西安交通大学 | Image super-resolution and non-uniform blur removing method based on fusion network |
CN109410158B (en) * | 2018-08-21 | 2020-10-02 | 西安电子科技大学 | Multi-focus image fusion method based on convolutional neural network |
CN109816592B (en) * | 2019-01-26 | 2022-05-13 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
CN109816592A (en) * | 2019-01-26 | 2019-05-28 | 福州大学 | Single-frame image continuous scale super-resolution method based on convolutional neural network |
CN110264415A (en) * | 2019-05-24 | 2019-09-20 | 北京爱诺斯科技有限公司 | It is a kind of to eliminate the fuzzy image processing method of shake |
CN110264415B (en) * | 2019-05-24 | 2020-06-12 | 北京爱诺斯科技有限公司 | Image processing method for eliminating jitter blur |
CN113542772A (en) * | 2020-04-21 | 2021-10-22 | 四川大学 | Compressed image deblocking method based on enhanced wide-activation residual error network |
CN113177599A (en) * | 2021-05-10 | 2021-07-27 | 南京信息工程大学 | Enhanced sample generation method based on GAN |
CN113177599B (en) * | 2021-05-10 | 2023-11-21 | 南京信息工程大学 | Reinforced sample generation method based on GAN |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107730453A (en) | Picture quality method for improving | |
CN110120011B (en) | Video super-resolution method based on convolutional neural network and mixed resolution | |
CN111861961B (en) | Single image super-resolution multi-scale residual error fusion model and restoration method thereof | |
CN110033410B (en) | Image reconstruction model training method, image super-resolution reconstruction method and device | |
CN110782399B (en) | Image deblurring method based on multitasking CNN | |
CN106683067A (en) | Deep learning super-resolution reconstruction method based on residual sub-images | |
CN109345449A (en) | A kind of image super-resolution based on converged network and remove non-homogeneous blur method | |
CN107492070A (en) | A kind of single image super-resolution computational methods of binary channels convolutional neural networks | |
CN106709875A (en) | Compressed low-resolution image restoration method based on combined deep network | |
CN109389556A (en) | The multiple dimensioned empty convolutional neural networks ultra-resolution ratio reconstructing method of one kind and device | |
CN108921786A (en) | Image super-resolution reconstructing method based on residual error convolutional neural networks | |
CN109118432A (en) | A kind of image super-resolution rebuilding method based on Rapid Circulation convolutional network | |
CN107563965A (en) | Jpeg compressed image super resolution ratio reconstruction method based on convolutional neural networks | |
CN110070489A (en) | Binocular image super-resolution method based on parallax attention mechanism | |
CN109389552A (en) | A kind of Image Super-resolution based on context-sensitive multitask deep learning | |
CN108961186A (en) | A kind of old film reparation recasting method based on deep learning | |
CN108428212A (en) | A kind of image magnification method based on double laplacian pyramid convolutional neural networks | |
CN110276354A (en) | A kind of training of high-resolution Streetscape picture semantic segmentation and real time method for segmenting | |
CN108805839A (en) | Combined estimator image defogging method based on convolutional neural networks | |
CN110163801A (en) | A kind of Image Super-resolution and color method, system and electronic equipment | |
CN109272447A (en) | A kind of depth map super-resolution method | |
CN109919840A (en) | Image super-resolution rebuilding method based on dense feature converged network | |
CN107833182A (en) | The infrared image super resolution ratio reconstruction method of feature based extraction | |
CN102243711A (en) | Neighbor embedding-based image super-resolution reconstruction method | |
CN110363704B (en) | Construction and reconstruction method of image super-resolution reconstruction model fusing morphology and color |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180223 |