CN108230269A - Grid method, device, equipment and storage medium are gone based on depth residual error network - Google Patents
Grid method, device, equipment and storage medium are gone based on depth residual error network Download PDFInfo
- Publication number
- CN108230269A CN108230269A CN201711458971.0A CN201711458971A CN108230269A CN 108230269 A CN108230269 A CN 108230269A CN 201711458971 A CN201711458971 A CN 201711458971A CN 108230269 A CN108230269 A CN 108230269A
- Authority
- CN
- China
- Prior art keywords
- network
- grid
- residual error
- image
- depth residual
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000003475 lamination Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 abstract description 14
- 238000013135 deep learning Methods 0.000 abstract description 10
- 238000010586 diagram Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000008961 swelling Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G06T5/77—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- 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
Grid method, device, equipment and storage medium are gone based on depth residual error network the invention discloses a kind of, this goes grid method to be used as basic network using the full convolutional network based on depth residual error, including：Basic network is trained using training image collection, obtains trained basic network；Grid is carried out to the image of grid to be removed to handle, obtain the image of grid using trained basic network.By using the full convolutional network based on depth residual error as basic network, expand the receptive field of convolution, so that multi-scale information (the more high frequencies of image and low-frequency information) is introduced in grid method is removed to be effectively improved the effect that deep learning goes trellis algorithm, it avoids and existing goes trellis algorithm is computationally intensive to cause treatment effeciency low or information scale is limited the contradiction for causing grid effect to be improved, it enhances deep learning algorithm and goes the application in grid field in image, there is wide popularization and application value.
Description
Technical field
The present invention relates to field of face identification, particularly, it is related to a kind of removing grid method, dress based on depth residual error network
It puts, equipment and storage medium.
Background technology
With the development of deep learning, recognition of face is more and more extensive in the landing popularization of various application scenarios, particularly golden
Melt payment industry, recognition of face is as a kind of application that encipherment is gone without card, and simple and fast with its, fitness is small increasingly to be obtained
To the favor of each bank.But in certain application scenarios, in order to protect the privacy of user, official's network certificate photo that bank takes
Mesh watermarking is added, this can seriously affect the effect of recognition of face, then various that trellis algorithm is gone to come into being.Wherein in addition to passing
The image procossing thinking of system is gone outside trellis algorithm, trellis algorithm is gone also to occur based on deep learning.But these algorithms
Sorter network before being generally basede on is improved, and net is removed as CN107424131A discloses a kind of image based on deep learning
Lattice method and device constructs grid image, generation and the corresponding multi-class grid number of net template online by net template
According to, and sorter network and full convolutional network is respectively trained using multi-class grid data as training data；Use trained point
Class network handles removal grid image carries out classification processing, and according to classification processing as a result, utilizing trained full convolution net
Network carries out grid to the grid image to be removed classified and handles.It is separated by due to going trellis algorithm that can use image current pixel
The information redundancy of some far pixels, existing full convolutional network cause information scale to be limited because its receptive field is smaller, can not
The low-frequency information of comprehensive sampled images, causes it to go grid effect to be improved, if in addition, acquire more image informations, it can
Cause its computationally intensive, influence treatment effeciency.
Invention content
Grid method, device, equipment and storage medium are gone based on depth residual error network the present invention provides a kind of, with solution
It is certainly existing to go trellis algorithm is computationally intensive to cause treatment effeciency low or information scale is limited grid effect is caused to need to be changed
The technical issues of kind.
The technical solution adopted by the present invention is as follows：
According to an aspect of the present invention, provide it is a kind of grid method is gone based on depth residual error network, the present invention removes net
For lattice method using the full convolutional network based on depth residual error as basic network, the present invention goes grid method to include：
Basic network is trained using training image collection, obtains trained basic network；
Grid is carried out to the image of grid to be removed to handle, obtain the figure of grid using trained basic network
Picture.
Further, metanetwork of the basic network for the addition extension convolution on the basis of full convolutional network, basic network
Including a series of metanetwork, often done between two neighboring metanetwork through residual error primary direct-connected.
Further, metanetwork includes first network framework, second network architecture, and any metanetwork is aforementioned two kinds of networks
One of structure；
First network framework includes sequentially connected first convolutional layer, the first relu non-linear layers, the second convolutional layer, second
Relu non-linear layers；
Second network architecture includes sequentially connected mixing convolutional layer, associated for that will mix the output of convolutional layer
Concate layers and the 3rd relu non-linear layers, wherein, mixing convolutional layer presses certain ratio by third convolutional layer and Volume Four lamination
Example composition.
Further, the first convolutional layer is the 2-dilated 3x3 convolution of I=2, and the second convolutional layer is 3x3 convolutional layers, the
Three convolutional layers are the 2-dilated 3x3 convolution of I=2, and Volume Four lamination is 3x3 convolutional layers, and third convolutional layer is accumulated with Volume Four
The ratio of layer is dilate_ratio, and the value range of dilate_ratio is [0,0.5].
Further, before the image of grid to be removed is inputted, the present invention goes grid method to further include：
The image of grid to be removed is pre-processed so that pretreated picture size meets preset dimension requirement.
Further, it is introduced in the step of training basic network using training image collection, obtaining trained basic network
Penalty, penalty are that the grid image for the pre-set dimension that network reconfiguration obtains meets preset dimension requirement with corresponding
Euler's distance that each pixel is subtracted each other on original image is multiplied by the corresponding MASK matrixes of human face region on original image.
According to another aspect of the present invention, also provide it is a kind of mesh device is gone based on depth residual error network, the present invention is gone
Mesh device includes：
Underlay network elements, underlay network elements are using the full convolutional network based on depth residual error as basic network；
Training unit for training basic network using training image collection, obtains trained basic network；
Grid cell is removed, is handled for carrying out grid to the image of grid to be removed using trained basic network,
Obtain the image of grid.
Further, metanetwork of the basic network for the addition extension convolution on the basis of full convolutional network, basic network
Including a series of metanetwork, often done between two neighboring metanetwork through residual error primary direct-connected.
According to another aspect of the present invention, a kind of image based on depth residual error network is also provided and removes grid equipment, including
Processor, processor is for running program, and the execution present invention's removes grid method based on depth residual error network when program is run.
According to another aspect of the present invention, a kind of storage medium is also provided, storage medium includes the program of storage, program fortune
During row control storage medium where equipment perform the present invention grid method is gone based on depth residual error network.
The invention has the advantages that：
Grid method, device, equipment and storage medium are removed the present invention is based on depth residual error network, by using based on depth
The full convolutional network of residual error is spent as basic network, the receptive field of convolution is expanded, so as to introduce more rulers in grid method is removed
Information (the more high frequencies of image and low-frequency information) is spent to be effectively improved the effect that deep learning goes trellis algorithm, is avoided existing
Go trellis algorithm is computationally intensive to cause treatment effeciency low or information scale is limited the lance for causing grid effect to be improved
Shield enhances deep learning algorithm and goes the application in grid field in image, has wide popularization and application value.
Other than objects, features and advantages described above, the present invention also has other objects, features and advantages.
Below with reference to accompanying drawings, the present invention is described in further detail.
Description of the drawings
The attached drawing for forming the part of the application is used to provide further understanding of the present invention, schematic reality of the invention
Example and its explanation are applied for explaining the present invention, is not constituted improper limitations of the present invention.In the accompanying drawings：
Fig. 1 is the step schematic diagram that removes grid method of the preferred embodiment of the present invention based on depth residual error network；
Fig. 2 is the schematic network structure for the basic network that grid is removed in the preferred embodiment of the present invention；
Fig. 3 is the structure diagram of the corresponding first network structure of metanetwork in the preferred embodiment of the present invention；
Fig. 4 is the structure diagram of corresponding second network structure of metanetwork in the preferred embodiment of the present invention；
Fig. 5 is the corresponding receptive field schematic diagram of traditional 3x3 convolution；
Fig. 6 is the corresponding receptive field schematic diagram of 2-dilated 3x3 convolution of I=2 in the preferred embodiment of the present invention；
Fig. 7 is the schematic diagram of original image in the preferred embodiment of the present invention；
Fig. 8 is that image shown in Fig. 7 passes through pretreated schematic diagram；
Fig. 9 is the corresponding MASK matrixes schematic diagram of human face region in image；
Figure 10 is the principle block diagram that goes mesh device of the preferred embodiment of the present invention based on depth residual error network.
Specific embodiment
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the application can phase
Mutually combination.The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
The preferred embodiment of the present invention provide it is a kind of grid method is gone based on depth residual error network, the present embodiment removes net
Lattice method is using the full convolutional network based on depth residual error as basic network, and referring to Fig. 1, the present embodiment removes grid method packet
It includes：
Step S100, basic network is trained using training image collection, obtains trained basic network；
Step S200, grid is carried out to the image of grid to be removed to handle, gone using trained basic network
The image of grid.
The present embodiment, as basic network, expands the impression of convolution by using the full convolutional network based on depth residual error
Open country, so as to introduce multi-scale information (the more high frequencies of image and low-frequency information) in grid method is removed to be effectively improved depth
Practise and go the effect of trellis algorithm, avoid it is existing go trellis algorithm it is computationally intensive cause treatment effeciency low or information scale by
The contradiction for causing grid effect to be improved is limited, deep learning algorithm is enhanced and goes the application in grid field in image, is had
There is wide popularization and application value.
In the present embodiment, basic network is the addition extension convolution (also known as convolution swelling of nucleus) on the basis of full convolutional network
Metanetwork, so as to expand the receptive field of convolution, introduce multi-scale information to improve the effect that deep learning goes trellis algorithm.Ginseng
See Fig. 2, in the present embodiment, basic network includes a series of mata-network metanetworks, per two neighboring metanetwork between
It is direct-connected that a short-cut is through residual residual errors.
In the present embodiment, metanetwork includes first network framework, second network architecture, and any metanetwork is aforementioned two kinds of nets
One of network structure.
In the present embodiment, first network framework includes sequentially connected first convolutional layer, the first relu non-linear layers, second
Convolutional layer, the 2nd relu non-linear layers.Preferably, referring to Fig. 3, in the present embodiment, the first convolutional layer is the 2-dilated of I=2
3x3 convolution, the second convolutional layer are 3x3 convolutional layers, and input signal first passes through the 2-dilated 3x3 convolution of one layer of I=2, each
The receptive field of convolution is 7x7, then by one layer of relu non-linear layer, the 3x3 convolutional layers (receptive field 3x3) of one layer of routine,
Finally exported by one layer of relu non-linear layer.In the present embodiment, metanetwork is minimal network structural unit, is conducive to realizing
When one network structure, change the depth of network and complexity by simply changing the number of metanetwork, such as
Inception layer inside googlenet.
In the present embodiment, second network architecture includes sequentially connected mixing convolutional layer, for will mix the defeated of convolutional layer
Go out associated concate layers and the 3rd relu non-linear layers, wherein, mixing convolutional layer is pressed by third convolutional layer and Volume Four lamination
Certain ratio composition.Referring to Fig. 4, in the present embodiment, third convolutional layer is the 2-dilated 3x3 convolution of I=2, Volume Four
Lamination is 3x3 convolutional layers, and the ratio of third convolutional layer and Volume Four lamination is dilate_ratio, the value of dilate_ratio
Ranging from [0,0.5].Preferably, for network more toward deep layer, the value of dilate_ratio is smaller.In the present embodiment, dilate_
Feature map numbers and the Volume Four lamination of ratio, that is, third convolutional layer output export feature map number ratios.It is preferred that
Ground, compared with shallow-layer using meta-a (first network framework), the receptive field bigger of the network of entirety deep layer in this way, then deep layer
Web vector graphic meta-b (second network architecture), and the dilate conv3x3 ratios that more deep layer uses are smaller, are conducive to enhancing
Go grid effect.
The present embodiment removes grid method, increases the receptive field of convolution by using expansion convolution, introduces multi-scale information
To improve the effect that deep learning goes trellis algorithm.Fig. 5 shows the corresponding receptive field schematic diagram of traditional 3x3 convolution, and Fig. 6 shows
The corresponding receptive field of 2-dilated 3x3 convolution of I=2 in the preferred embodiment of the present invention is gone out, can have been obtained by Fig. 5 and Fig. 6 comparisons
Know, under the premise of calculation amount is not increased, the receptive field of convolution kernel is become the 3x3 after progress dilate by original 3x3
7x7。
Preferably, before the image of grid to be removed is inputted, the present embodiment goes grid method to further include：
The image of grid to be removed is pre-processed so that pretreated picture size meets preset dimension requirement.
Fig. 7 shows the schematic diagram of original image in the preferred embodiment of the present invention；Fig. 8 is image shown in Fig. 7 by pre- place
Schematic diagram after reason.
The present embodiment is used as basic network, the input picture of network using the full convolutional network based on depth residual error network
Size is 224x224.Since the resolution ratio of grid certificate photo is generally 178x220 or 96x118.In the process of addition grid
In, while very strong ringing effect is generated due to picture compression etc. around grid.The present embodiment goes the process of grid,
Image size is expanded to 224x224 (see Fig. 8) by the way of black surround is mended first, so as to unify the input size of network.
Face location is corresponded in pretreated image labeled as MASK matrixes, white area shown in Figure 9.
Preferably, in the present embodiment, basic network is trained using training image collection, obtains the step of trained basic network
Penalty (Penalty function) is introduced in rapid, penalty is the grid image of pre-set dimension that network reconfiguration obtains
It is multiplied by with Euler's distance L2loss that pixel each on the corresponding original image for meeting preset dimension requirement is subtracted each other original
The corresponding MASK matrixes of human face region on image.The present embodiment penalty is the optimization aim letter of training in step S100
Number.
By the way that grid method is gone to go to grid side with traditional fcn (full convolutional network, whole convolution are 3x3) to the present embodiment
Method is compared, and using the PSNR Y-PSNRs of test image and reconstructed image as judgment criteria, specifically see the table below：
By upper table it is known that the present embodiment goes the PSNR highests of grid method, grid best results are gone.Remove mesh evaluation
Effect judgment criteria is mainly that subjective naked eyes are seen；In addition it is exactly the PSNR that the present embodiment uses, what this example obtained removes grid
Methods of the PSNR compared with baseline of image and original mesh free certificate photo improves 27-24=3dB.
According to another aspect of the present invention, also provide it is a kind of mesh device is gone based on depth residual error network, referring to Figure 10,
The present embodiment goes mesh device to include：
Underlay network elements 100, underlay network elements are using the full convolutional network based on depth residual error as basic network；
Training unit 200 for training basic network using training image collection, obtains trained basic network；
Grid cell 300 is removed, for being carried out at grid to the image of grid to be removed using trained basic network
Reason, obtains the image of grid.
Preferably, metanetwork of the present embodiment basic network for the addition extension convolution on the basis of full convolutional network, base
Plinth network includes a series of metanetwork, is done between every two neighboring metanetwork through residual error primary direct-connected.
It should be noted that the present embodiment goes mesh device for performing above-described embodiment based on depth residual error network
Grid method is removed, specific implementation process is with reference to the description of above-described embodiment method.
According to another aspect of the present invention, a kind of image based on depth residual error network is also provided and removes grid equipment, including
Processor, processor is for running program, and the execution embodiment of the present invention removes net based on depth residual error network when program is run
Lattice method.
According to another aspect of the present invention, a kind of storage medium is also provided, storage medium includes the program of storage, program fortune
Equipment where storage medium is controlled during row performs the embodiment of the present invention and based on depth residual error network removes grid method.
It should be noted that step shown in the flowchart of the accompanying drawings can be in such as a group of computer-executable instructions
It is performed in computer system, although also, show logical order in flow charts, it in some cases, can be with not
The sequence being same as herein performs shown or described step.
If the function described in the present embodiment method is realized in the form of SFU software functional unit and is independent product pin
It sells or in use, can be stored in one or more computing device read/write memory medium.Based on such understanding, this hair
The part or the part of the technical solution that bright embodiment contributes to the prior art can be embodied in the form of software product
Out, which is stored in a storage medium, is used including some instructions so that a computing device (can be
People's computer, server, mobile computing device or network equipment etc.) perform the whole of each embodiment the method for the present invention
Or part steps.And aforementioned storage medium includes：USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD etc. are various can to store journey
The medium of sequence code.
Each embodiment is described by the way of progressive in this specification, the highlights of each of the examples are with it is other
The difference of embodiment, just to refer each other for same or similar part between each embodiment.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, that is made any repaiies
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of remove grid method based on depth residual error network, which is characterized in that described that grid method is gone to use based on depth
The full convolutional network of residual error is described that grid method is gone to include as basic network：
The basic network is trained using training image collection, obtains trained basic network；
Grid is carried out to the image of grid to be removed to handle, obtain the figure of grid using the trained basic network
Picture.
2. according to claim 1 remove grid method based on depth residual error network, which is characterized in that
Metanetwork of the basic network for the addition extension convolution on the basis of full convolutional network, the basic network include one
The metanetwork of series is often done primary direct-connected between two neighboring metanetwork through residual error.
3. according to claim 2 remove grid method based on depth residual error network, which is characterized in that
The metanetwork includes first network framework, second network architecture, and any metanetwork is aforementioned two kinds of network structures
One of；
The first network framework includes sequentially connected first convolutional layer, the first relu non-linear layers, the second convolutional layer, second
Relu non-linear layers；
Second network architecture includes sequentially connected mixing convolutional layer, for the output of the mixing convolutional layer is associated
Concate layers and the 3rd relu non-linear layers, wherein, the mixing convolutional layer is by third convolutional layer and Volume Four lamination by certain
Ratio composition.
4. according to claim 3 remove grid method based on depth residual error network, which is characterized in that
First convolutional layer is the 2-dilated 3x3 convolution of I=2, and second convolutional layer is 3x3 convolutional layers, described the
Three convolutional layers be I=2 2-dilated 3x3 convolution, the Volume Four lamination be 3x3 convolutional layers, the third convolutional layer with
The ratio of the Volume Four lamination is dilate_ratio, and the value range of dilate_ratio is [0,0.5].
5. according to claim 1 remove grid method based on depth residual error network, which is characterized in that
It is described that grid method is gone to further include before the image of grid to be removed is inputted：
The image of grid to be removed is pre-processed so that pretreated picture size meets preset dimension requirement.
6. according to claim 1 remove grid method based on depth residual error network, which is characterized in that
Punishment letter is introduced in the step of use training image collection trains the basic network, obtains trained basic network
Number, the penalty meet the preset dimension requirement for the grid image of pre-set dimension that network reconfiguration obtains with corresponding
Original image on Euler's distance for subtracting each other of each pixel be multiplied by the corresponding MASK squares of human face region on the original image
Battle array.
7. a kind of go mesh device based on depth residual error network, which is characterized in that including：
Underlay network elements, the underlay network elements are using the full convolutional network based on depth residual error as basic network；
Training unit for training the basic network using training image collection, obtains trained basic network；
Grid cell is removed, is handled for carrying out grid to the image of grid to be removed using the trained basic network,
Obtain the image of grid.
8. according to claim 7 go mesh device based on depth residual error network, which is characterized in that
Metanetwork of the basic network for the addition extension convolution on the basis of full convolutional network, the basic network include one
The metanetwork of series is often done primary direct-connected between two neighboring metanetwork through residual error.
9. a kind of remove grid equipment based on depth residual error network, including processor, the processor is special for running program
Sign is, is performed when described program is run and goes to grid side based on depth residual error network as described in claim 1 to 6 is any
Method.
10. a kind of storage medium, the storage medium includes the program of storage, which is characterized in that described program controls when running
Equipment execution where the storage medium goes to grid side as described in claim 1 to 6 is any based on depth residual error network
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711458971.0A CN108230269B (en) | 2017-12-28 | 2017-12-28 | Grid removing method, device and equipment based on depth residual error network and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711458971.0A CN108230269B (en) | 2017-12-28 | 2017-12-28 | Grid removing method, device and equipment based on depth residual error network and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108230269A true CN108230269A (en) | 2018-06-29 |
| CN108230269B CN108230269B (en) | 2021-02-09 |
Family
ID=62646535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711458971.0A Active CN108230269B (en) | 2017-12-28 | 2017-12-28 | Grid removing method, device and equipment based on depth residual error network and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108230269B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241982A (en) * | 2018-09-06 | 2019-01-18 | 广西师范大学 | Object detection method based on depth layer convolutional neural networks |
| CN109472733A (en) * | 2018-10-22 | 2019-03-15 | 天津大学 | Image latent writing analysis method based on convolutional neural networks |
| CN111062854A (en) * | 2019-12-26 | 2020-04-24 | Oppo广东移动通信有限公司 | Method, device, terminal and storage medium for detecting watermark |
| CN112884666A (en) * | 2021-02-02 | 2021-06-01 | 杭州海康慧影科技有限公司 | Image processing method, image processing device and computer storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107424131A (en) * | 2017-07-14 | 2017-12-01 | 北京智慧眼科技股份有限公司 | Image based on deep learning removes grid method and device |
-
2017
- 2017-12-28 CN CN201711458971.0A patent/CN108230269B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107424131A (en) * | 2017-07-14 | 2017-12-01 | 北京智慧眼科技股份有限公司 | Image based on deep learning removes grid method and device |
Non-Patent Citations (2)
| Title |
|---|
| FISHER YU: "Dilated Residual Networks", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| SHU ZHANG: "Multi-task ConvNet for Blind Face Inpainting with Application to Face Verification", 《2016 INTERNATIONAL CONFERENCE ON BIOMETRICS (ICB)》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241982A (en) * | 2018-09-06 | 2019-01-18 | 广西师范大学 | Object detection method based on depth layer convolutional neural networks |
| CN109472733A (en) * | 2018-10-22 | 2019-03-15 | 天津大学 | Image latent writing analysis method based on convolutional neural networks |
| CN111062854A (en) * | 2019-12-26 | 2020-04-24 | Oppo广东移动通信有限公司 | Method, device, terminal and storage medium for detecting watermark |
| CN111062854B (en) * | 2019-12-26 | 2023-08-25 | Oppo广东移动通信有限公司 | Method, device, terminal and storage medium for detecting watermark |
| CN112884666A (en) * | 2021-02-02 | 2021-06-01 | 杭州海康慧影科技有限公司 | Image processing method, image processing device and computer storage medium |
| CN112884666B (en) * | 2021-02-02 | 2024-03-19 | 杭州海康慧影科技有限公司 | Image processing method, device and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108230269B (en) | 2021-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230269A (en) | Grid method, device, equipment and storage medium are gone based on depth residual error network | |
| CN109492627B (en) | Scene text erasing method based on depth model of full convolution network | |
| CN109816612A (en) | Image enchancing method and device, computer readable storage medium | |
| CN107992891A (en) | Based on spectrum vector analysis multi-spectral remote sensing image change detecting method | |
| CN107958247A (en) | Method and apparatus for facial image identification | |
| CN111179196B (en) | Multi-resolution depth network image highlight removing method based on divide-and-conquer | |
| CN112232914B (en) | Four-stage virtual fitting method and device based on 2D image | |
| CN108765282B (en) | Real-time super-resolution method and system based on FPGA | |
| Kim et al. | Deep edge-aware interactive colorization against color-bleeding effects | |
| CN104881684A (en) | Stereo image quality objective evaluate method | |
| CN108921162A (en) | Licence plate recognition method and Related product based on deep learning | |
| CN108616471A (en) | A kind of signal modulate method and apparatus based on convolutional neural networks | |
| CN104021567B (en) | Based on the fuzzy altering detecting method of image Gauss of first numeral law | |
| CN103366175B (en) | Based on the scene image classification method that potential Di Li Cray distributes | |
| CN115393698A (en) | Digital image tampering detection method based on improved DPN network | |
| CN107492082A (en) | A kind of MRF sample block image repair methods using edge statistics feature | |
| US10521918B2 (en) | Method and device for filtering texture, using patch shift | |
| CN113362255A (en) | Character image restoration method and system based on improved DCGAN | |
| CN108664906A (en) | The detection method of content in a kind of fire scenario based on convolutional network | |
| CN116844052A (en) | Method, device and equipment for resisting attack by deployable patches facing remote sensing image | |
| CN114820674B (en) | Arc profile extraction method, device, computer equipment and storage medium | |
| CN110176021A (en) | In conjunction with the level set image segmentation method and system of the conspicuousness information of gamma correction | |
| CN116310323A (en) | Aircraft target instance segmentation method, system and readable storage medium | |
| CN107862316A (en) | Convolution algorithm method and device | |
| CN115512258A (en) | Desensitization method and device for video image, terminal equipment and storage medium |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 100097 Beijing Haidian District Kunming Hunan Road 51 C block two floor 207. Applicant after: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. Address before: 100193 4, 403, block A, 14 building, 10 East North Road, Haidian District, Beijing. Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 410205 14 Changsha Zhongdian Software Park Phase I, 39 Jianshan Road, Changsha High-tech Development Zone, Yuelu District, Changsha City, Hunan Province Applicant after: Wisdom Eye Technology Co.,Ltd. Address before: 100097 2nd Floor 207, Block C, 51 Hunan Road, Kunming, Haidian District, Beijing Applicant before: BEIJING ATHENA EYES SCIENCE & TECHNOLOGY CO.,LTD. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Method, device, equipment and storage medium for grid removal based on deep residual network Effective date of registration: 20221205 Granted publication date: 20210209 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20231220 Granted publication date: 20210209 Pledgee: Agricultural Bank of China Limited Hunan Xiangjiang New Area Branch Pledgor: Wisdom Eye Technology Co.,Ltd. Registration number: Y2022430000107 |
|
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: No. 205, Building B1, Huigu Science and Technology Industrial Park, No. 336 Bachelor Road, Bachelor Street, Yuelu District, Changsha City, Hunan Province, 410000 Patentee after: Wisdom Eye Technology Co.,Ltd. Address before: 410205 building 14, phase I, Changsha Zhongdian Software Park, No. 39, Jianshan Road, Changsha high tech Development Zone, Yuelu District, Changsha City, Hunan Province Patentee before: Wisdom Eye Technology Co.,Ltd. |