CN108447020A - A kind of face super-resolution reconstruction method based on profound convolutional neural networks - Google Patents
A kind of face super-resolution reconstruction method based on profound convolutional neural networks Download PDFInfo
- Publication number
- CN108447020A CN108447020A CN201810201306.1A CN201810201306A CN108447020A CN 108447020 A CN108447020 A CN 108447020A CN 201810201306 A CN201810201306 A CN 201810201306A CN 108447020 A CN108447020 A CN 108447020A
- Authority
- CN
- China
- Prior art keywords
- resolution
- face image
- image
- low
- neural networks
- 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
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
- G06T3/4076—Super resolution, i.e. output image resolution higher than sensor resolution by iteratively correcting the provisional high resolution image using the original low-resolution image
-
- 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
- G06T3/4023—Decimation- or insertion-based scaling, e.g. pixel or line decimation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Abstract
The present invention provides a kind of face super-resolution reconstruction methods based on profound convolutional neural networks, include the following steps：One, after carrying out different multiples down-sampling and processing to high-resolution human face image, the training set of low-resolution face image is obtained；Two, after carrying out different multiples down-sampling and processing to another set high-resolution human face image, the test set of low-resolution face image is obtained；Three, the training set that step 1 obtains and the test set that step 2 obtains are put into profound convolutional network to be trained, learn the mapping of residual image, and obtain corresponding convolutional network model；Four, the high-resolution human face image for being rebuild the convolutional network model for needing the low-resolution face image input step three rebuild to learn to obtain.The beneficial effects of the invention are as follows：The face super-resolution reconstruction method based on profound convolutional neural networks can preferably handle the super-resolution rebuilding problem of multiple dimensioned amplification coefficient.
Description
Technical field
The invention belongs to technical field of image information processing, more particularly to a kind of people based on profound convolutional neural networks
Face super-resolution method for reconstructing.
Background technology
Image super-resolution rebuilding is a classical problem of computer vision field, it is intended to from given low resolution
(LR) image infers high-resolution (HR) image with key message, wherein human face super-resolution reconstruction is wherein important point
One of branch.Human face super-resolution is reconstituted in the numerous areas extensive application background such as authentication, intelligent monitoring.Current oversubscription
Resolution method for reconstructing is broadly divided into two classes：(1) it regards oversubscription as ill-posed problem in image procossing, prior information can be introduced
Solve the problems, such as this；(2) method for using machine learning, study obtain the mapping between low resolution and high-definition picture and close
System, to realize image super-resolution rebuilding.
Baker and Kanada describes this point first, and applied to all multi-panels such as face recognition, face's alignment
It applies in portion.More and more concerns are caused due to its actual importance.With the development of machine learning, many is based on study
Method be suggested.
Currently, the algorithm based on study can not only obtain visual effect more better than other super-resolution rebuilding technologies and
And the amplification factor of bigger can be handled.Wherein, the method for field insertion (Neighbor embedding, NB) is mended using interpolation
The strategy of fourth subspace.The method of sparse coding (Sparse Coding, SC) uses the dictionary learning based on Sparse Signal Representation
Mapping relations between LR and HR.Recently, random forest (Random Forest, RF) and convolutional neural networks
(Convolutional Neural Networks, CNN) also be used to promote reconstruction precision.Wherein, Dong et al. is proved
CNN can be effectively applied in the mapping for learning LR to HR end to end, and institute's extracting method is known as SRCNN (Single-Image
Reconstruction Convolutional Neural Networks).It is inspired by the above method, many researchers
Begin to use depth convolutional network come the problem of handling super-resolution rebuilding.Wang et al. mutually ties sparse coding with convolutional network
It closes, to solve the consistency problem in super-resolution method.Cui et al. combine sparse coding speciality and deep learning it is excellent
A kind of point, it is proposed that cascade network of large-scale, arbitrary zoom factor.Multiple convolutional networks are together in series progress by Nie et al.
Human face super-resolution is rebuild.
Although CNN is successfully applied to processing human face super-resolution Problems of Reconstruction and achieves good effect,
There is also limitations following aspects：First, CNN depend critically upon the context of small image-region；Second, with layer
Several increases, receptive field, which becomes conference, causes reconstruction image details easy to be lost；Third, convergence speed are too slow.
Invention content
It is an object of the invention to, provide a kind of to be based on profound convolutional neural networks in view of the drawbacks of the prior art or problem
Face super-resolution reconstruction method.
Technical scheme is as follows：A kind of face super-resolution reconstruction method packet based on profound convolutional neural networks
Include following steps：One, after carrying out different multiples down-sampling and processing to high-resolution human face image, low resolution face figure is obtained
The training set of picture；Two, after carrying out different multiples down-sampling and processing to another set high-resolution human face image, low resolution is obtained
The test set of rate facial image；Three, the training set that step 1 obtains and the test set that step 2 obtains are put into profound convolution net
Network is trained, and learns the mapping of residual image, and obtains corresponding convolutional network model；Four, rebuild needs low point
The high-resolution human face image that the convolutional network model that resolution facial image input step three learns to obtain is rebuild.
Preferably, is carried out by different multiples down-sampling and is handled for high-resolution human face image in step 1 and step 2 include
Following steps：The down-sampling that different multiples are carried out to high-resolution human face image, obtains different low-resolution face images；It is right
Obtained low-resolution face image is expanded, and carries out overlap sampling to every low-resolution face image, obtains one group
The low-resolution face image block of overlapping.
Preferably, low-resolution face image is carried out adopting under rotationally-varying and different multiples the interpolation of four direction
Sample.
Preferably, further include following steps in step 1：Corresponding high-resolution human face image is similarly weighed
Folded sampling, obtains the high-resolution human face image block of one group of correspondence overlapping, as high-resolution label image.
Preferably, learnt between high-resolution human face image and low-resolution face image using residual error learning method
Detailed information.
Preferably, in step 3, zero filling is identical to keep the size of all characteristic patterns before each convolutional layer.
Preferably, in step 3, the facial image being predicted out is added back in the low-resolution face image of input
To obtain final high-resolution human face image.
Technical solution provided by the invention has the advantages that：
The face super-resolution reconstruction method based on profound convolutional neural networks has the following advantages that：
1, profound convolutional network is used, more accurate image detail information can be extracted；
2, the method for using residual error study, the problem of improving trained speed, avoid loss in detail；
3, the method for using multiple amplification factors while training, reduces the quantity of required model.
Description of the drawings
Fig. 1 is the flow signal of the face super-resolution reconstruction method provided by the invention based on profound convolutional neural networks
Figure；
Fig. 2 is profound in the shown face super-resolution reconstruction method based on profound convolutional neural networks provided by the invention
Convolutional network structure chart；
Fig. 3 is method for reconstructing provided by the invention and bicubic interpolation method and SRCNN methods the case where amplification factor is 4
Under Y-PSNR (PSNR) qualitative assessment figure；
Fig. 4 is the comparison figure of method for reconstructing provided by the invention and other methods in visual effect；
Fig. 5 is the test result figure of method for reconstructing provided by the invention different amplification factors on FEI.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
The description of specific distinct unless the context otherwise, the present invention in element and component, the shape that quantity both can be single
Formula exists, and form that can also be multiple exists, and the present invention is defined not to this.Although step in the present invention with label into
It has gone arrangement, but is not used to limit the precedence of step, unless expressly stated the order of step or holding for certain step
Based on row needs other steps, otherwise the relative rank of step is adjustable.It is appreciated that used herein
Term "and/or" one of is related to and covers associated Listed Items or one or more of any and all possible groups
It closes.
The thinking of the present invention is to obtain big receptive field using profound convolutional network to extract the abundant information in big image,
Secondly, on the other hand the method learnt using residual error, convergence rate when on the one hand can be with training for promotion can also be avoided because of sense
Loss in detail problem caused by being become larger by open country.A kind of method of single mode oversubscription is finally additionally used flexibly to handle amplification coefficient
The problem of.It can also be arbitrary that amplification coefficient can be specified by user.Profound convolutional network designed by the present invention can be preferable
The multiple dimensioned amplification coefficient of processing super-resolution rebuilding problem.
In the present embodiment, the face super-resolution reconstruction method packet provided by the invention based on profound convolutional neural networks
Include following steps：
One, after carrying out different multiples down-sampling and processing to high-resolution human face image, low-resolution face image is obtained
Training set；And corresponding high-resolution human face image is subjected to overlap sampling, obtain the high-resolution human of one group of correspondence overlapping
Face image block, as high-resolution label image.
Specifically, in step 1, high-resolution human face image is carried out different multiples down-sampling and handled to include as follows
Step：
The down-sampling that different multiples are carried out to high-resolution human face image, obtains different low-resolution face images；
Obtained low-resolution face image is expanded, and overlapping is carried out to every low-resolution face image and is adopted
Sample obtains the low-resolution face image block of one group of overlapping.
In fact, in view of super-resolution rebuilding is still an ill-conditioning problem that can not possibly accurately solve, it would be desirable to
Corresponding low-resolution face image is generated, to original high-definition picture with the amplification of different multiples (× 2, × 3, × 4)
The factor carries out bicubic interpolation and obtains low-resolution face image collection.
Moreover, because low-resolution face image sample size is not big enough, need to the obtained low resolution of step 1
Facial image is expanded, first it is carried out 90 degree, 180 degree and 270 degree of rotation transformation obtain the low resolution of different angle
Facial image, then overlap sampling is carried out to the low-resolution face image of different angle, obtain the low resolution people of N × N sizes
Face image block generates final training set.
It should be noted that in step 1, to the overlap sampling method of low-resolution face image with to high-resolution
It is identical that facial image carries out overlap sampling method.
Two, after carrying out different multiples down-sampling and processing to another set high-resolution human face image, low resolution is obtained
The test set of facial image.
Specifically, in step 2, high-resolution human face image is carried out different multiples down-sampling and handled to include as follows
Step：
The down-sampling that different multiples are carried out to high-resolution human face image, obtains different low-resolution face images；
Obtained low-resolution face image is expanded, and overlapping is carried out to every low-resolution face image and is adopted
Sample obtains the low-resolution face image block of one group of overlapping.
In fact, the present invention merges the training of convolutional neural networks and test in a network model, rolled up in training
It is tested while product neural network.Moreover, low-resolution face image and high-resolution label image that test set uses
It is generated using method identical with step 1.
Three, the training set that step 1 obtains and the test set that step 2 obtains profound convolutional network is put into be trained,
Learn the mapping of residual image, and obtains corresponding convolutional network model.
Specifically, for example, as shown in Figure 1, devising a profound convolutional network to handle the Super-resolution reconstruction of face
Build problem：20 convolutional layers are used, other than input layer and output layer, other convolutional layers all use 64 3 × 3 × 64
Filter.And linear incentive unit (ReLU) is all employed behind each filter.If input is denoted as Y, need
Learn a kind of details for mapping F (Y) and making it possible to prognostic chart picture, expression formula is as follows：
Fn(Y)=max (0, Wn*Yn-1+Bn), n=1,2 ..., 20 (1)
Wherein FnIndicate the output of n-th of convolutional layer, WnIndicate 64 × 3 × 3 × 64 weight matrix, BnIndicate one
The bias vector of a 64 dimension.
Make by the low-resolution face image of bicubic interpolation (bicubic) as shown in Figure 1, convolutional neural networks are used
To input, and prediction modeling is carried out to image detail.
In fact, image detail is normally used for super resolution ratio reconstruction method, moreover, the deep learning method based on CNN
It can effectively learn the detailed information of image.
But predict that a problem caused by intensive output is using profound convolutional neural networks：Every time with volume
When product operation, characteristic pattern can become smaller.For example, the input picture of one (n+1) × (n+1) sizes, receptive field n
× n, the output after n convolutional layer then become 1 × 1.
Secondly as general super resolution ratio reconstruction method is required for the pixel with surrounding correctly to infer center pixel,
More limitations can be brought to Problems of Reconstruction so as to cause peripheral region.For the information of near image boundaries, due to this pass
System cannot be fully utilized, so many oversubscription methods can cut result images.But if required circular area
Domain is very big cannot then to use the method cut, can not be visually satisfactory because of image meeting very little final after cutting.
In order to solve problem above, before each convolutional layer zero filling (Padding) to keep the size of all characteristic patterns
It is identical.Experiment shows that the result after progress zero filling is very good, therefore can efficiently use image boundary information.
Finally, the image being predicted out is added back in low resolution (LR) image of input to obtain final high-resolution
Rate (HR) image.
The target minimized to find out the optimized parameter of carried model is described below, the low resolution Jing Guo interpolation
Facial image is denoted as Y, and high-resolution human face image is denoted as X.Give a training setThe purpose is to learn one
Model F, and predicted value can be obtainedWhereinIndicate prediction target HR images.In the present embodiment, the present invention adopts
Make to train with mean-square value error is minimized optimal.
Since used convolutional neural networks are due to producing a kind of very long end pair there are many weight layers
The relationship at end.Therefore, it is very easy to lead to the problem of gradient disappearance or gradient explosion.Therefore, present invention employs residual errors
The method of habit solves the problems, such as this.
Since input picture and output image are largely similar, so residual image R=X-Y is defined,
In most of numerical value may be zero or very small.If predicted this residual image, loss function indicates such as
Under：
In loss layer, there are three input parameters：Residual error is estimated, LR images and true HR images are inputted.Costing bio disturbance is
Euclidean distance between reconstruction image and true picture.
Four, the convolutional network model that the low-resolution face image input step three rebuild learns to obtain will be needed to obtain weight
The high-resolution human face image built.
Specifically, a high-resolution human face image different from training set is selected, desired down-sampling times is carried out to it
Number carries out down-sampling and obtains corresponding low-resolution face image, then selects corresponding amplification factor, is trained by step 3
Obtained model can be rebuild after high-resolution human face image.
For the ease of public understanding technical solution of the present invention, a specific example is given below.
Using the face image of CASPEAL-R1 as training set and test set.It is real comprising 1000 anticipatory remarks in CASPEAL-R1
Example face image.It is opened as test set as training set 100 we used 600 therein.Because of the man in CASPEAL
Facial image quantity with Ms is not equal, in order to enable result is more fair, our training set uses 300
Man's facial image and 300 Ms Zhang's facial images.Remaining 50 Ms Zhang facial image and remaining 50 man's face figure
As composition test set.
In this example, network model selects profound convolutional network, in order to verify effectiveness of the invention, reconstructed results difference
It is compared with the reconstructed results based on interpolation method and the method based on convolutional neural networks (CNN)；In order to verify by invented party
The generalization ability of method.200 in FEI face databases face face images are also had chosen to be tested.
The reconstruction process of this example is specific as follows：
1, the acquisition of low-resolution image：
For the high-resolution human face image that given size is 160x160, double the three of 2 times, 3 times and 4 times are carried out to it
Secondary interpolation down-sampling, obtains various sizes of low-resolution face image.
2, the construction of training sample：
90 degree of low-resolution face image obtained in the previous step progress, 180 degree and 270 degree of rotation transformation are obtained four
Then the facial image of different directions carries out overlap sampling to these images again, obtains the low resolution figure of one group of 41x41 size
As block constitutes the training sample of the present invention；
3, the construction of training label：
High-resolution human face image corresponding with low resolution face is subjected to same overlap sampling, obtains one group of phase
The high-definition picture block of corresponding overlapping, as high-resolution label image；
4, the construction of test set：
Using in CASPEAL-R1 50 man's faces and 50 Ms Zhang's faces as test image, different times are carried out to it
Then several down-samplings carries out the rotation transformation of different angle, then overlap sampling obtains low point of one group of 41x41 size again
Then resolution image block equally carries out overlap sampling to corresponding high-resolution human face image, using obtained picture as mark
Label；
5, the profound convolutional network of training
Data are put into profound convolutional network and are trained, the size of each image block is both configured in this example
41x41, carried network have 20 weight layers, are 64 per batch of picture number when training, momentum and weight decaying are set as 0.9
With 0.0001.Initial learning rate is set as 0.0001.The maximum iteration of this example setting is 400000 times, using under gradient
Drop method optimizes, and then stops iteration when iterations reach maximum times.
6, face image
The low-resolution image rebuild will be needed to carry out super-resolution rebuilding with the convolutional neural networks model learnt to obtain
To the high-resolution human face image of reconstruction.
In order to verify the effect of the method for the present invention, respectively by the present invention method with based on interpolation method and be based on convolution
The method of neural network (SRCNN) is compared.
Fig. 3 is peak value of the method for the present invention with bicubic interpolation method and the method for SRCNN in the case where amplification factor is 4
Signal-to-noise ratio (PSNR) qualitative assessment figure.It can be seen from the figure that for 100 low resolution pictures for rebuilding, the present invention
Method it is higher than the PSNR values of other two methods.This illustrates that the method for the present invention can obtain better reconstructed results.
Fig. 4 is the present invention and comparison figure of the other methods in visual effect, wherein (a) is original high-resolution image；
(b) it is Bicubic interpolation reconstruction images, is (c) SR CNN reconstruction images, is (d) reconstruction image of the present invention.It can be with from figure
Find out, when amplification factor is 4 times, method of the invention compares Bicubic in profile after reconstruction with the method for SRCNN
On it is sharper keen, method of the invention has better details to believe compared to the method for SRCNN at positions such as hair, eyes
Breath.
In order to verify the generalization ability of the method for the present invention, also tested in 200 face face images of FEI.Fig. 5
For the test result figure of the present invention different amplification factors on FEI, wherein first row (a) is original image, and secondary series (b) is to put
The reconstruction image that the big factor is 2, third row (c) are the reconstruction image that amplification factor is 3, and it is 4 that the 4th row (d), which are amplification factor,
Reconstruction image.As can be seen from the figure method of the invention can also obtain preferable visual effect on the test set of FEI faces
And treatment of details.
The method of the present invention copes with the super-resolution rebuilding of the facial image of different amplification factors in summary, and obtains
More better than traditional super-resolution method for reconstructing visual effect.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Profit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent requirements of the claims
Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiment being appreciated that.
Claims (7)
1. a kind of face super-resolution reconstruction method based on profound convolutional neural networks, it is characterised in that：Include the following steps：
One, after carrying out different multiples down-sampling and processing to high-resolution human face image, the instruction of low-resolution face image is obtained
Practice collection；
Two, after carrying out different multiples down-sampling and processing to another set high-resolution human face image, low resolution face is obtained
The test set of image；
Three, the training set that step 1 obtains and the test set that step 2 obtains are put into profound convolutional network to be trained, are learnt
The mapping of residual image, and obtain corresponding convolutional network model；
Four, the convolutional network model for needing the low-resolution face image input step three rebuild to learn to obtain is rebuild
High-resolution human face image.
2. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 1, special
Sign is, carries out different multiples down-sampling and handle to include following step to high-resolution human face image in step 1 and step 2
Suddenly：
The down-sampling that different multiples are carried out to high-resolution human face image, obtains different low-resolution face images；
Obtained low-resolution face image is expanded, and overlap sampling is carried out to every low-resolution face image, is obtained
To the low-resolution face image block of one group of overlapping.
3. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 2, special
Sign is, rotationally-varying and different multiples the interpolation down-sampling of four direction is carried out to low-resolution face image.
4. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 2, special
Sign is, further includes following steps in step 1：
Corresponding high-resolution human face image is subjected to same overlap sampling, obtains the high-resolution human face of one group of correspondence overlapping
Image block, as high-resolution label image.
5. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 1, special
Sign is, learns the letter of the details between high-resolution human face image and low-resolution face image using residual error learning method
Breath.
6. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 1, special
Sign is, in step 3, zero filling is identical to keep the size of all characteristic patterns before each convolutional layer.
7. a kind of face super-resolution reconstruction method based on profound convolutional neural networks according to claim 1, special
Sign is, in step 3, the facial image being predicted out is added back in the low-resolution face image of input to obtain most
Whole high-resolution human face image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810201306.1A CN108447020A (en) | 2018-03-12 | 2018-03-12 | A kind of face super-resolution reconstruction method based on profound convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810201306.1A CN108447020A (en) | 2018-03-12 | 2018-03-12 | A kind of face super-resolution reconstruction method based on profound convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108447020A true CN108447020A (en) | 2018-08-24 |
Family
ID=63194156
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810201306.1A Pending CN108447020A (en) | 2018-03-12 | 2018-03-12 | A kind of face super-resolution reconstruction method based on profound convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108447020A (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109191554A (en) * | 2018-09-04 | 2019-01-11 | 清华-伯克利深圳学院筹备办公室 | A kind of super resolution image reconstruction method, device, terminal and storage medium |
| CN109308455A (en) * | 2018-08-28 | 2019-02-05 | 广州图语信息科技有限公司 | A kind of low resolution face identification method, device and processing terminal |
| CN109525859A (en) * | 2018-10-10 | 2019-03-26 | 腾讯科技（深圳）有限公司 | Model training, image transmission, image processing method and relevant apparatus equipment |
| CN109564677A (en) * | 2018-11-09 | 2019-04-02 | 香港应用科技研究院有限公司 | Super-resolution synthesis system and method based on random forest grader weighted results |
| CN109636716A (en) * | 2018-10-29 | 2019-04-16 | 昆明理工大学 | A kind of image super-resolution rebuilding method based on wavelet coefficient study |
| CN109671023A (en) * | 2019-01-24 | 2019-04-23 | 江苏大学 | A kind of secondary method for reconstructing of face image super-resolution |
| CN109741253A (en) * | 2018-12-12 | 2019-05-10 | 深圳慧源创新科技有限公司 | Unmanned plane figure passes video close shot distant view handoff technique |
| CN109859106A (en) * | 2019-01-28 | 2019-06-07 | 桂林电子科技大学 | A kind of image super-resolution rebuilding method based on the high-order converged network from attention |
| CN110070487A (en) * | 2019-04-02 | 2019-07-30 | 清华大学 | Semantics Reconstruction face oversubscription method and device based on deeply study |
| CN110188807A (en) * | 2019-05-21 | 2019-08-30 | 重庆大学 | Tunnel pedestrian target detection method based on cascade super-resolution network and improvement Faster R-CNN |
| CN110211140A (en) * | 2019-06-14 | 2019-09-06 | 重庆大学 | Abdominal vascular dividing method based on 3D residual error U-Net and Weighted Loss Function |
| CN110288697A (en) * | 2019-06-24 | 2019-09-27 | 天津大学 | 3D face representation and method for reconstructing based on multiple dimensioned figure convolutional neural networks |
| CN110490796A (en) * | 2019-04-11 | 2019-11-22 | 福建师范大学 | A kind of human face super-resolution processing method and system of the fusion of low-and high-frequency ingredient |
| CN110503606A (en) * | 2019-08-29 | 2019-11-26 | 广州大学 | A method of improving face clarity |
| CN110532871A (en) * | 2019-07-24 | 2019-12-03 | 华为技术有限公司 | The method and apparatus of image procossing |
| CN110930309A (en) * | 2019-11-20 | 2020-03-27 | 武汉工程大学 | Face super-resolution method and device based on multi-view texture learning |
| CN111127317A (en) * | 2019-12-02 | 2020-05-08 | 深圳供电局有限公司 | Image super-resolution reconstruction method and device, storage medium and computer equipment |
| CN111145096A (en) * | 2019-12-26 | 2020-05-12 | 苏州大学 | Super-resolution image reconstruction method and system based on recursive extremely-deep network |
| CN111259841A (en) * | 2020-01-20 | 2020-06-09 | 深圳云天励飞技术有限公司 | Image processing method and related equipment |
| CN111626937A (en) * | 2020-05-28 | 2020-09-04 | 成都三零凯天通信实业有限公司 | Portrait super-resolution reconstruction method |
| CN111861881A (en) * | 2020-06-09 | 2020-10-30 | 复旦大学 | Image super-resolution reconstruction algorithm for interpolation based on CNN |
| CN112001940A (en) * | 2020-08-21 | 2020-11-27 | Oppo（重庆）智能科技有限公司 | Image processing method and device, terminal and readable storage medium |
| CN112200152A (en) * | 2019-12-06 | 2021-01-08 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back-projection neural network |
| CN112200720A (en) * | 2020-09-29 | 2021-01-08 | 中科方寸知微(南京)科技有限公司 | Super-resolution image reconstruction method and system based on filter fusion |
| CN112950478A (en) * | 2021-03-18 | 2021-06-11 | 武汉工程大学 | Face super-resolution method and system based on dual identity attribute constraint |
| CN113128467A (en) * | 2021-05-11 | 2021-07-16 | 临沂大学 | Low-resolution face super-resolution and recognition method based on face priori knowledge |
| CN113269765A (en) * | 2021-06-04 | 2021-08-17 | 重庆大学 | Expandable convolutional neural network training method and CT image segmentation model construction method |
| CN113538242A (en) * | 2021-07-22 | 2021-10-22 | 苏州大学 | Image interpolation model training method based on residual error guide strategy |
| CN113628107A (en) * | 2021-07-02 | 2021-11-09 | 上海交通大学 | Face image super-resolution method and system |
| CN116205806A (en) * | 2023-01-28 | 2023-06-02 | 荣耀终端有限公司 | Image enhancement method and electronic equipment |
| CN112200152B (en) * | 2019-12-06 | 2024-04-26 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back projection neural network |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101872472A (en) * | 2010-06-02 | 2010-10-27 | 中国科学院自动化研究所 | Method for super-resolution reconstruction of facial image on basis of sample learning |
| EP3028246A1 (en) * | 2013-07-31 | 2016-06-08 | MBDA UK Limited | Method and apparatus for tracking an object |
| CN106096547A (en) * | 2016-06-11 | 2016-11-09 | 北京工业大学 | A kind of towards the low-resolution face image feature super resolution ratio reconstruction method identified |
| EP3166068A1 (en) * | 2015-11-06 | 2017-05-10 | Thomson Licensing | Method for upscaling an image and apparatus for upscaling an image |
| CN106910161A (en) * | 2017-01-24 | 2017-06-30 | 华南理工大学 | A kind of single image super resolution ratio reconstruction method based on depth convolutional neural networks |
| CN106991648A (en) * | 2017-04-10 | 2017-07-28 | 中国石油大学(华东) | A kind of image super-resolution rebuilding method based on wavelet transformation and convolutional neural networks |
| CN107358575A (en) * | 2017-06-08 | 2017-11-17 | 清华大学 | A kind of single image super resolution ratio reconstruction method based on depth residual error network |
| CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
| CN107464216A (en) * | 2017-08-03 | 2017-12-12 | 济南大学 | A kind of medical image ultra-resolution ratio reconstructing method based on multilayer convolutional neural networks |
| CN107481188A (en) * | 2017-06-23 | 2017-12-15 | 珠海经济特区远宏科技有限公司 | A kind of image super-resolution reconstructing method |
| CN107507134A (en) * | 2017-09-21 | 2017-12-22 | 大连理工大学 | Super-resolution method based on convolutional neural networks |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
| CN107680037A (en) * | 2017-09-12 | 2018-02-09 | 河南大学 | The improved face super-resolution reconstruction method based on nearest feature line manifold learning |
-
2018
- 2018-03-12 CN CN201810201306.1A patent/CN108447020A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101872472A (en) * | 2010-06-02 | 2010-10-27 | 中国科学院自动化研究所 | Method for super-resolution reconstruction of facial image on basis of sample learning |
| EP3028246A1 (en) * | 2013-07-31 | 2016-06-08 | MBDA UK Limited | Method and apparatus for tracking an object |
| EP3166068A1 (en) * | 2015-11-06 | 2017-05-10 | Thomson Licensing | Method for upscaling an image and apparatus for upscaling an image |
| CN106096547A (en) * | 2016-06-11 | 2016-11-09 | 北京工业大学 | A kind of towards the low-resolution face image feature super resolution ratio reconstruction method identified |
| CN106910161A (en) * | 2017-01-24 | 2017-06-30 | 华南理工大学 | A kind of single image super resolution ratio reconstruction method based on depth convolutional neural networks |
| CN106991648A (en) * | 2017-04-10 | 2017-07-28 | 中国石油大学(华东) | A kind of image super-resolution rebuilding method based on wavelet transformation and convolutional neural networks |
| CN107358575A (en) * | 2017-06-08 | 2017-11-17 | 清华大学 | A kind of single image super resolution ratio reconstruction method based on depth residual error network |
| CN107481188A (en) * | 2017-06-23 | 2017-12-15 | 珠海经济特区远宏科技有限公司 | A kind of image super-resolution reconstructing method |
| CN107358576A (en) * | 2017-06-24 | 2017-11-17 | 天津大学 | Depth map super resolution ratio reconstruction method based on convolutional neural networks |
| CN107464216A (en) * | 2017-08-03 | 2017-12-12 | 济南大学 | A kind of medical image ultra-resolution ratio reconstructing method based on multilayer convolutional neural networks |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
| CN107680037A (en) * | 2017-09-12 | 2018-02-09 | 河南大学 | The improved face super-resolution reconstruction method based on nearest feature line manifold learning |
| CN107507134A (en) * | 2017-09-21 | 2017-12-22 | 大连理工大学 | Super-resolution method based on convolutional neural networks |
Non-Patent Citations (3)
| Title |
|---|
| CHAO DONG;CHEN CHANGE LOY;KAIMING HE;XIAOOU TANG: "Image Super-Resolution Using Deep Convolutional Networks", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 王一宁，秦品乐，李传朋，崔雨豪: "基于残差神经网络的图像超分辨率改进算法", 《计算机应用》 * |
| 陈敏: "《认知计算导论》", 30 April 2017, 华中科技大学出版社 * |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109308455A (en) * | 2018-08-28 | 2019-02-05 | 广州图语信息科技有限公司 | A kind of low resolution face identification method, device and processing terminal |
| CN109191554A (en) * | 2018-09-04 | 2019-01-11 | 清华-伯克利深圳学院筹备办公室 | A kind of super resolution image reconstruction method, device, terminal and storage medium |
| CN109525859B (en) * | 2018-10-10 | 2021-01-15 | 腾讯科技（深圳）有限公司 | Model training method, image sending method, image processing method and related device equipment |
| CN109525859A (en) * | 2018-10-10 | 2019-03-26 | 腾讯科技（深圳）有限公司 | Model training, image transmission, image processing method and relevant apparatus equipment |
| CN109636716A (en) * | 2018-10-29 | 2019-04-16 | 昆明理工大学 | A kind of image super-resolution rebuilding method based on wavelet coefficient study |
| CN109636716B (en) * | 2018-10-29 | 2022-07-22 | 昆明理工大学 | Image super-resolution reconstruction method based on wavelet coefficient learning |
| CN109564677A (en) * | 2018-11-09 | 2019-04-02 | 香港应用科技研究院有限公司 | Super-resolution synthesis system and method based on random forest grader weighted results |
| CN109564677B (en) * | 2018-11-09 | 2022-09-27 | 香港应用科技研究院有限公司 | Super-resolution synthesis system and method based on random forest classifier weighting result |
| CN109741253A (en) * | 2018-12-12 | 2019-05-10 | 深圳慧源创新科技有限公司 | Unmanned plane figure passes video close shot distant view handoff technique |
| CN109671023A (en) * | 2019-01-24 | 2019-04-23 | 江苏大学 | A kind of secondary method for reconstructing of face image super-resolution |
| CN109671023B (en) * | 2019-01-24 | 2023-07-21 | 江苏大学 | Face image super-resolution secondary reconstruction method |
| CN109859106A (en) * | 2019-01-28 | 2019-06-07 | 桂林电子科技大学 | A kind of image super-resolution rebuilding method based on the high-order converged network from attention |
| CN110070487B (en) * | 2019-04-02 | 2021-05-11 | 清华大学 | Semantic reconstruction face hyper-segmentation method and device based on deep reinforcement learning |
| CN110070487A (en) * | 2019-04-02 | 2019-07-30 | 清华大学 | Semantics Reconstruction face oversubscription method and device based on deeply study |
| CN110490796B (en) * | 2019-04-11 | 2023-02-14 | 福建师范大学 | High-low frequency component fused face super-resolution processing method and system |
| CN110490796A (en) * | 2019-04-11 | 2019-11-22 | 福建师范大学 | A kind of human face super-resolution processing method and system of the fusion of low-and high-frequency ingredient |
| CN110188807A (en) * | 2019-05-21 | 2019-08-30 | 重庆大学 | Tunnel pedestrian target detection method based on cascade super-resolution network and improvement Faster R-CNN |
| CN110211140A (en) * | 2019-06-14 | 2019-09-06 | 重庆大学 | Abdominal vascular dividing method based on 3D residual error U-Net and Weighted Loss Function |
| CN110288697A (en) * | 2019-06-24 | 2019-09-27 | 天津大学 | 3D face representation and method for reconstructing based on multiple dimensioned figure convolutional neural networks |
| CN110532871A (en) * | 2019-07-24 | 2019-12-03 | 华为技术有限公司 | The method and apparatus of image procossing |
| CN110532871B (en) * | 2019-07-24 | 2022-05-10 | 华为技术有限公司 | Image processing method and device |
| CN110503606A (en) * | 2019-08-29 | 2019-11-26 | 广州大学 | A method of improving face clarity |
| CN110503606B (en) * | 2019-08-29 | 2023-06-20 | 广州大学 | Method for improving face definition |
| CN110930309A (en) * | 2019-11-20 | 2020-03-27 | 武汉工程大学 | Face super-resolution method and device based on multi-view texture learning |
| CN110930309B (en) * | 2019-11-20 | 2023-04-18 | 武汉工程大学 | Face super-resolution method and device based on multi-view texture learning |
| CN111127317B (en) * | 2019-12-02 | 2023-07-25 | 深圳供电局有限公司 | Image super-resolution reconstruction method, device, storage medium and computer equipment |
| CN111127317A (en) * | 2019-12-02 | 2020-05-08 | 深圳供电局有限公司 | Image super-resolution reconstruction method and device, storage medium and computer equipment |
| CN112200152A (en) * | 2019-12-06 | 2021-01-08 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back-projection neural network |
| CN112200152B (en) * | 2019-12-06 | 2024-04-26 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back projection neural network |
| CN111145096A (en) * | 2019-12-26 | 2020-05-12 | 苏州大学 | Super-resolution image reconstruction method and system based on recursive extremely-deep network |
| CN111259841A (en) * | 2020-01-20 | 2020-06-09 | 深圳云天励飞技术有限公司 | Image processing method and related equipment |
| CN111626937A (en) * | 2020-05-28 | 2020-09-04 | 成都三零凯天通信实业有限公司 | Portrait super-resolution reconstruction method |
| CN111861881A (en) * | 2020-06-09 | 2020-10-30 | 复旦大学 | Image super-resolution reconstruction algorithm for interpolation based on CNN |
| CN111861881B (en) * | 2020-06-09 | 2022-05-20 | 复旦大学 | Image super-resolution reconstruction method based on CNN interpolation |
| CN112001940B (en) * | 2020-08-21 | 2023-04-07 | Oppo（重庆）智能科技有限公司 | Image processing method and device, terminal and readable storage medium |
| CN112001940A (en) * | 2020-08-21 | 2020-11-27 | Oppo（重庆）智能科技有限公司 | Image processing method and device, terminal and readable storage medium |
| CN112200720B (en) * | 2020-09-29 | 2023-08-08 | 中科方寸知微(南京)科技有限公司 | Super-resolution image reconstruction method and system based on filter fusion |
| CN112200720A (en) * | 2020-09-29 | 2021-01-08 | 中科方寸知微(南京)科技有限公司 | Super-resolution image reconstruction method and system based on filter fusion |
| CN112950478A (en) * | 2021-03-18 | 2021-06-11 | 武汉工程大学 | Face super-resolution method and system based on dual identity attribute constraint |
| CN112950478B (en) * | 2021-03-18 | 2022-07-19 | 武汉工程大学 | Face super-resolution method and system based on dual identity attribute constraint |
| CN113128467A (en) * | 2021-05-11 | 2021-07-16 | 临沂大学 | Low-resolution face super-resolution and recognition method based on face priori knowledge |
| CN113128467B (en) * | 2021-05-11 | 2022-03-29 | 临沂大学 | Low-resolution face super-resolution and recognition method based on face priori knowledge |
| CN113269765A (en) * | 2021-06-04 | 2021-08-17 | 重庆大学 | Expandable convolutional neural network training method and CT image segmentation model construction method |
| CN113269765B (en) * | 2021-06-04 | 2022-10-28 | 重庆大学 | Expandable convolutional neural network training method and CT image segmentation model construction method |
| CN113628107A (en) * | 2021-07-02 | 2021-11-09 | 上海交通大学 | Face image super-resolution method and system |
| CN113628107B (en) * | 2021-07-02 | 2023-10-27 | 上海交通大学 | Face image super-resolution method and system |
| CN113538242A (en) * | 2021-07-22 | 2021-10-22 | 苏州大学 | Image interpolation model training method based on residual error guide strategy |
| WO2023000526A1 (en) * | 2021-07-22 | 2023-01-26 | 苏州大学 | Image interpolation model training method based on residual-guided policy |
| CN116205806A (en) * | 2023-01-28 | 2023-06-02 | 荣耀终端有限公司 | Image enhancement method and electronic equipment |
| CN116205806B (en) * | 2023-01-28 | 2023-09-19 | 荣耀终端有限公司 | Image enhancement method and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108447020A (en) | A kind of face super-resolution reconstruction method based on profound convolutional neural networks | |
| CN107154023B (en) | Based on the face super-resolution reconstruction method for generating confrontation network and sub-pix convolution | |
| CN110097550B (en) | Medical image segmentation method and system based on deep learning | |
| CN110659727B (en) | Sketch-based image generation method | |
| CN110084863A (en) | A kind of multiple domain image conversion method and system based on generation confrontation network | |
| CN109727195B (en) | Image super-resolution reconstruction method | |
| CN108734659A (en) | A kind of sub-pix convolved image super resolution ratio reconstruction method based on multiple dimensioned label | |
| CN110211045A (en) | Super-resolution face image method based on SRGAN network | |
| CN105488472B (en) | A kind of digital cosmetic method based on sample form | |
| CN107609638A (en) | A kind of method based on line decoder and interpolation sampling optimization convolutional neural networks | |
| CN107729819A (en) | A kind of face mask method based on sparse full convolutional neural networks | |
| CN105069825A (en) | Image super resolution reconstruction method based on deep belief network | |
| CN108921789A (en) | Super-resolution image reconstruction method based on recurrence residual error network | |
| CN109214989A (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| CN110060204A (en) | A kind of single image super-resolution method based on reciprocal networks | |
| CN107590775B (en) | Image super-resolution amplification method using regression tree field | |
| CN109886881A (en) | Face dressing minimizing technology | |
| CN106169174A (en) | A kind of image magnification method | |
| CN110188667B (en) | Face rectification method based on three-party confrontation generation network | |
| CN113298736B (en) | Face image restoration method based on face pattern | |
| CN112767252B (en) | Image super-resolution reconstruction method based on convolutional neural network | |
| CN109801225A (en) | Face reticulate pattern stain minimizing technology based on the full convolutional neural networks of multitask | |
| Dogan et al. | Semi-supervised image attribute editing using generative adversarial networks | |
| CN109977989A (en) | A kind of processing method of image tensor data | |
| Qu et al. | Perceptual-DualGAN: perceptual losses for image to image translation with generative adversarial nets |
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 |
Application publication date: 20180824 |
|
| RJ01 | Rejection of invention patent application after publication |