CN110334588A - Kinship recognition methods and the device of network are paid attention to based on local feature - Google Patents
Kinship recognition methods and the device of network are paid attention to based on local feature Download PDFInfo
- Publication number
- CN110334588A CN110334588A CN201910434461.2A CN201910434461A CN110334588A CN 110334588 A CN110334588 A CN 110334588A CN 201910434461 A CN201910434461 A CN 201910434461A CN 110334588 A CN110334588 A CN 110334588A
- Authority
- CN
- China
- Prior art keywords
- local feature
- result
- feature
- network
- kinship
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- 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
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of kinship recognition methods and devices that network is paid attention to based on local feature.The described method includes: obtaining multiple character images；Notice that network extracts local feature from each character image respectively using local feature trained in advance, and validity feature enhancing processing is carried out to the local feature；According to treated, local feature identifies the kinship of the multiple character image, to improve the recognition accuracy of kinship.
Description
Technical field
The present invention relates to technical field of computer vision, particularly relate to a kind of relatives pass that network is paid attention to based on local feature
It is recognition methods and device.
Background technique
The verifying of kinship has practical use abundant, and if social relationships are investigated, Missing Persons' investigation, discovery is imitated
Etc..Therefore, concern of the technology relevant to kinship increasingly by global researcher.However, this is also a Xiang Fei
Often with challenging task.Different from Generic face identification, kinship verifying faces more difficulties.In one group of relatives' object
Between, their age is often widely different or even gender is also different.In other words, the conventional method of facial similitude is assessed
Relatives' verifying is had little effect.Therefore, kinship verifying needs more explore.
Biologist and psychologist's studies have shown that science of heredity band about face organ information.Two blood relationships are closed
The people of system is entire facial similar usually in certain facials.Therefore, in the prior art using entire facial characteristics into
The identification of row kinship causes recognition accuracy low.
Summary of the invention
In view of this, it is an object of the invention to propose a kind of kinship identification side for paying attention to network based on local feature
Method and device can be improved the recognition accuracy of kinship.
Based on the above-mentioned purpose kinship recognition methods provided by the invention for paying attention to network based on local feature, comprising:
Obtain multiple character images；
Notice that network extracts local feature from each character image respectively using local feature trained in advance, and to institute
It states local feature and carries out validity feature enhancing processing；
According to treated, local feature identifies the kinship of the multiple character image.
Further, the local feature notices that network includes the first convolutional layer set gradually, the first attention network knot
Structure, the first pond layer, the second convolutional layer, the second attention network structure, the second pond layer, third convolutional layer, third pay attention to network
Structure and full articulamentum.
Further, the first attention network structure, the second attention network structure and the third pay attention to network
Structure includes the maximum pond layer set gradually, Volume Four lamination and up-sampling layer.
Further, described that the network extraction office from each character image respectively is paid attention to using local feature trained in advance
Portion's feature, and validity feature enhancing processing is carried out to the local feature, it specifically includes:
The data of the multiple character image are input to first convolutional layer, and export first partial feature；
The first partial feature is input to described first and notices that network structure carries out Feature Mapping, and exports first and reflects
Penetrate result；
First mapping result is sequentially input to first pond layer, second convolutional layer, and exports second
Local feature；
Second local feature is input to described second and notices that network structure carries out Feature Mapping, and exports second and reflects
Penetrate result；
Second mapping result is sequentially input to second pond layer, the third convolutional layer, and exports third
Local feature；
The third local feature is input to the third and notices that network structure carries out Feature Mapping, and exports third and reflects
Penetrate result；
The third mapping result is input to the full articulamentum, and exports the 4th local feature, the 4th part
Feature is validity feature enhancing treated local feature.
Further, each calculation formula for paying attention to network structure and carrying out Feature Mapping are as follows:
P (x)=(1+F (x)) * C (x)；
Wherein, C (x) is the data of convolutional layer output, and F (x) is activation primitive, and P (x) is mapping result.
Further, before the multiple character images of acquisition, further includes:
It establishes local feature and pays attention to network；
Obtain multiple groups character image sample；Every group of character image sample includes the naked image pattern of same character image
With part overlaid image pattern；
The multiple groups character image sample is input to the local feature and pays attention to network, to pay attention to the local feature
Network is trained, and is exported sample and covered result and kinship result；
Result is covered according to the sample and kinship result calculates final loss result, and the final loss is tied
Fruit feeds back to the local feature and pays attention to network.
Further, described that result and the final loss result of kinship result calculating are covered according to the sample, specifically
Include:
The first-loss result that the sample covers result is calculated using cross entropy loss function；
The second loss result of the kinship result is calculated using the cross entropy loss function；
The final loss result is calculated according to the first-loss result and second loss result.
Further, the cross entropy loss function are as follows:
Loss (x, class)=- x [class]+log [∑jexp(x[j])]；
Wherein, x is that sample covers result or kinship as a result, class is the label of lineup's object image sample, Loss
(x, class) is loss result.
Further, the calculation formula of the final loss result are as follows:
Loss=Losslp+λ*Lossks；
Wherein, Loss is final loss result, LosslpFor first-loss as a result, LossksFor the second loss result, λ is
Weight.
The embodiment of the present invention also provides a kind of social relationships identification device based on semantically enhancement network, can be realized above-mentioned
All processes of social relationships recognition methods based on semantically enhancement network, described device include:
Image collection module, for obtaining multiple character images；
Characteristic extracting module, for noticing that network is mentioned from each character image respectively using local feature trained in advance
Local feature is taken, and validity feature enhancing processing is carried out to the local feature；And
Identification module, for local feature to know the kinship of the multiple character image according to treated
Not.
From the above it can be seen that the kinship recognition methods provided by the invention for paying attention to network based on local feature
And device, it can notice that network extracts local feature from multiple character images respectively using local feature trained in advance, into
And validity feature enhancing processing is carried out to local feature, to improve the identification weight of local feature, and then according to treated office
Portion's feature carries out kinship identification, to make full use of the local similarity of relatives' image to promote the recognition performance of kinship,
Effectively improve the recognition accuracy of kinship.
Detailed description of the invention
Fig. 1 is that the process of the kinship recognition methods provided in an embodiment of the present invention that network is paid attention to based on local feature is shown
It is intended to；
Fig. 2 notices that part is special in the kinship recognition methods of network based on local feature to be provided in an embodiment of the present invention
Sign pays attention to the schematic diagram of network；
Fig. 3 is to pay attention to net in the kinship recognition methods provided in an embodiment of the present invention for paying attention to network based on local feature
The schematic diagram of network structure；
Fig. 4 pays attention to Beijing National Sports Training Center in the kinship recognition methods of network based on local feature to be provided in an embodiment of the present invention
Portion's feature pays attention to the schematic diagram of network；
Fig. 5 is that the structure of the kinship identification device provided in an embodiment of the present invention that network is paid attention to based on local feature is shown
It is intended to.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Attached drawing, the present invention is described in more detail.
It is the kinship recognition methods provided in an embodiment of the present invention that network is paid attention to based on local feature referring to Fig. 1
Flow diagram, which comprises
S1, multiple character images are obtained.
In the present embodiment, the image of the multiple personages of multiple character images, that is, kinship to be identified, each character image
Facial image including corresponding personage.
S2, notice that network extracts local feature from each character image respectively using local feature trained in advance, and
Validity feature enhancing processing is carried out to the local feature.
In the present embodiment, local feature refers to the face characteristic in character image, such as five features, may include a left side
Eye, right eye, nose, the left corners of the mouth and right corners of the mouth etc..Enhancing validity feature is referred to the effective enhancing processing of local feature progress and is pressed down
The influence of low-level features processed.
Specifically, as shown in Fig. 2, local feature notices that network includes convolutional neural networks and is inserted in convolutional Neural net
Attention network structure between network.Convolutional neural networks generally comprise the first convolutional layer 11 set gradually, the first pond layer 13,
Second convolutional layer 14, the second pond layer 16, third convolutional layer 17 and full articulamentum 19.First convolutional layer 11 and the first pond layer 13
Between inserted with first pay attention to network structure 12, between the second convolutional layer 14 and the second pond layer 16 inserted with second attention network
Structure 15 pays attention to network structure 18 inserted with third between third convolutional layer 17 and full articulamentum 19.Wherein, described first pays attention to
Network structure 12, the second attention network structure 15 and the third notice that network structure 18 includes the maximum set gradually
Pond layer 20, Volume Four lamination 21 and up-sampling layer 22.Network structure is paid attention to for top-down structure from bottom to top, with convolution
Neural network, which is combined, pays attention to network to form local feature.It is special to capture different types of part to use multiple attention network structures
Reference breath.
Specifically, step S2 includes:
The data of the multiple character image are input to first convolutional layer, and export first partial feature；
The first partial feature is input to described first and notices that network structure carries out Feature Mapping, and exports first and reflects
Penetrate result；
First mapping result is sequentially input to first pond layer, second convolutional layer, and exports second
Local feature；
Second local feature is input to described second and notices that network structure carries out Feature Mapping, and exports second and reflects
Penetrate result；
Second mapping result is sequentially input to second pond layer, the third convolutional layer, and exports third
Local feature；
The third local feature is input to the third and notices that network structure carries out Feature Mapping, and exports third and reflects
Penetrate result；
The third mapping result is input to the full articulamentum, and exports the 4th local feature, the 4th part
Feature is validity feature enhancing treated local feature.
It is activated it should be noted that local feature notices that convolutional layer and full articulamentum in network are used as using ReLu function
A pond layer is connect respectively after function, the first convolutional layer and the second convolutional layer, third convolutional layer is followed by a full articulamentum, with
Network performance is improved by addition standardization layer.
In identification process, multiple character images are overlapped, for example, two character images are overlapped, each
Character image is triple channel RGB image, pixel 64*64, to form 6 channel datas of a 64*64 size.As shown in Fig. 2,
The data of character image are input to first convolutional layer 11 of 32 1 step-lengths of convolution kernel, and the size of each convolution kernel is 5*5*6,
Therefore the first partial feature of the first convolutional layer 11 output 60*60*32.First partial feature pays attention to network structure 12 by first
Feature Mapping is carried out, the first mapping result 23 is obtained and is input to the first pond layer 13, the first pond layer 13 exports 30*30*32
Data and input as the second convolutional layer 14.Before second convolutional layer 14 is using 64 convolution kernels filtering that size is 5*5*32
One layer of input obtains the second local feature of 26*26*64.Equally, the second local feature pays attention to network structure 15 by second
Feature Mapping is carried out, the second mapping result 24 is obtained and is input to the second pond layer 16, the second pond layer 16 exports 13*13*64
Data, the input as third convolutional layer 17.Third convolutional layer 17 includes 128 convolution kernels, and each convolution kernel
Size is 5*5*64.Third convolutional layer 17 export 9*9*128 third local feature, and by third pay attention to network structure 18 into
Row Feature Mapping obtains third mapping result 25 and is input to full articulamentum 19.Full articulamentum 19 arrives the data projection of input
In subspace with 512 neurons, then it is special by the 4th part in the child control of data projection to 12 neurons, is obtained
Sign, the 4th local feature are through validity feature enhancing treated local feature.
Wherein, insertion pays attention to network structure after each convolutional layer, imitates fast feedforward and feedback attention process, will pay attention to
Power concentrates on effective local feature.As shown in figure 3, each attention network structure includes the maximum pond layer set gradually
20, Volume Four lamination 21 and up-sampling layer 22.Data C (x) expression of convolutional layer output before each attention network structure, will
C (x) is converted to the Feature Mapping between 0 and 1 by S-shaped activation primitive F (x).But the characteristic pattern phase repeatedly between 0 to 1
Multiply the weight that can reduce further feature, or even destroys the superperformance of local feature network.In order to not reduce original effect
In the case of amplify effective local feature, handled using residual error method, specific formula are as follows:
P (x)=(1+F (x)) * C (x)
Wherein, C (x) is the data of convolutional layer output, and F (x) is activation primitive, and P (x) is mapping result.
S3, according to treated, local feature identifies the kinship of the multiple character image.
In the present embodiment, is obtaining after validity feature enhancing treated local feature, local feature is input to one
A classifier (soft-max), for example, two character images export 12 data after local feature pays attention to network processes, classification
This 12 data are classified as 6 channel, two classification results by device, to indicate that relatives' recognition result 44 and image cover result 45.Its
In, first 5 groups in 6 channel, two classification results covering situation (generally five characteristic points for respectively indicating facial five characteristic points
It is not covered), last group indicates that relatives' recognition result, i.e. the two character images whether there is kinship.
Further, before step S1, further includes:
It establishes local feature and pays attention to network；
Obtain multiple groups character image sample；Every group of character image sample includes the naked image pattern of same character image
With part overlaid image pattern；
The multiple groups character image sample is input to the local feature and pays attention to network, to pay attention to the local feature
Network is trained, and is exported sample and covered result and kinship result；
Result is covered according to the sample and kinship result calculates final loss result, and the final loss is tied
Fruit feeds back to the local feature and pays attention to network.
In the present embodiment, as shown in figure 4, obtaining multiple complete personages when noticing that network is trained to local feature
Image pattern (i.e. naked image pattern) 41, and a facial characteristics is randomly choosed in each complete character image sample 41
Point is covered, and covers image pattern 42 with generating portion, and the corresponding part of each completely character image sample 41 hides
Lid image pattern 42 collectively forms lineup's object image sample.Wherein, face feature point includes left eye, right eye, nose, the left corners of the mouth
Or the right corners of the mouth, characteristic point is covered using the pure color square of 9*9, and the color of pure color square and covering area's ambient color phase
Closely.For example, covering in Fig. 4 to the left eye of complete character image sample 41, part overlaid image pattern 42 is obtained.
Multiple groups character image sample is input to local feature and pays attention to network 43, to notice that network 43 carries out to local feature
Training, exports relatives' recognition result 44 and image covers result 45.Wherein, the related data of image covering result 45 is not direct
The identification of kinship is participated in, but a part as loss function feeds back to local feature and pays attention to network 43, using certainly
The mechanism that I supervises enables local feature notice that network 43 is preferably primarily focused on face feature point.
Specifically, described that result and the final loss result of kinship result calculating are covered according to the sample, it is specific to wrap
It includes:
The first-loss result that the sample covers result is calculated using cross entropy loss function；
The second loss result of the kinship result is calculated using the cross entropy loss function；
The final loss result is calculated according to the first-loss result and second loss result.
Specifically, the cross entropy loss function are as follows:
Loss (x, class)=- x [class]+log [∑jexp(x[j])]；
Wherein, x is that sample covers result or kinship as a result, class is the label of lineup's object image sample, Loss
(x, class) is loss result.
Specifically, the calculation formula of the final loss result are as follows:
Loss=Losslp+λ*Lossks；
Wherein, Loss is final loss result, LosslpFor first-loss as a result, LossksFor the second loss result, λ is
Weight.
Intersect entropy function calculating loss it should be noted that relatives' recognition result is passed through respectively with sample covering result,
Different weights is assigned again, and summation obtains final loss result.Wherein, it has been that guidance is made to verifying that sample, which covers result,
With, therefore relatives' recognition result is endowed higher weight λ, it is preferable that λ 10.
The kinship recognition methods provided by the invention that network is paid attention to based on local feature, can be using training in advance
Local feature notices that network extracts local feature from multiple character images respectively, and then carries out validity feature increasing to local feature
Strength reason, to improve the identification weight of local feature, and then local feature carries out kinship identification according to treated, to fill
Divide the recognition performance that kinship is promoted using the local similarity of relatives' image, the identification for effectively improving kinship is accurate
Rate.
Correspondingly, it the present invention also provides a kind of social relationships identification device based on semantically enhancement network, can be realized
State all processes of the social relationships recognition methods based on semantically enhancement network.
It is the structure of the social relationships identification device provided in an embodiment of the present invention based on semantically enhancement network referring to Fig. 5
Schematic diagram, the device include:
Image collection module 1, for obtaining multiple character images；
Characteristic extracting module 2, for paying attention to network respectively from each character image using local feature trained in advance
Local feature is extracted, and validity feature enhancing processing is carried out to the local feature；And
Identification module 3, for local feature to know the kinship of the multiple character image according to treated
Not.
The kinship identification device provided by the invention that network is paid attention to based on local feature, can be using training in advance
Local feature notices that network extracts local feature from multiple character images respectively, and then carries out validity feature increasing to local feature
Strength reason, to improve the identification weight of local feature, and then local feature carries out kinship identification according to treated, to fill
Divide the recognition performance that kinship is promoted using the local similarity of relatives' image, the identification for effectively improving kinship is accurate
Rate.
It should be understood by those ordinary skilled in the art that: the discussion of any of the above embodiment is exemplary only, not
It is intended to imply that the scope of the present disclosure (including claim) is limited to these examples；Under thinking of the invention, above embodiments
Or can also be combined between the technical characteristic in different embodiments, step can be realized with random order, and be existed such as
Many other variations of the upper different aspect of the invention, for simplicity, they are not provided in details.
In addition, to simplify explanation and discussing, and in order not to obscure the invention, it can in provided attached drawing
It is connect with showing or can not show with the well known power ground of integrated circuit (IC) chip and other components.Furthermore, it is possible to
Device is shown in block diagram form, to avoid obscuring the invention, and this has also contemplated following facts, i.e., about this
The details of the embodiment of a little block diagram arrangements be height depend on will implementing platform of the invention (that is, these details should
It is completely within the scope of the understanding of those skilled in the art).Elaborating that detail (for example, circuit) is of the invention to describe
In the case where exemplary embodiment, it will be apparent to those skilled in the art that can be in these no details
In the case where or implement the present invention in the case that these details change.Therefore, these descriptions should be considered as explanation
Property rather than it is restrictive.
Although having been incorporated with specific embodiments of the present invention, invention has been described, according to retouching for front
It states, many replacements of these embodiments, modifications and variations will be apparent for those of ordinary skills.Example
Such as, discussed embodiment can be used in other memory architectures (for example, dynamic ram (DRAM)).
The embodiment of the present invention be intended to cover fall into all such replacements within the broad range of appended claims,
Modifications and variations.Therefore, all within the spirits and principles of the present invention, any omission, modification, equivalent replacement, the improvement made
Deng should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of kinship recognition methods for paying attention to network based on local feature characterized by comprising
Obtain multiple character images；
Notice that network extracts local feature from each character image respectively using local feature trained in advance, and to the office
Portion's feature carries out validity feature enhancing processing；
According to treated, local feature identifies the kinship of the multiple character image.
2. the kinship recognition methods according to claim 1 for paying attention to network based on local feature, which is characterized in that institute
It states local feature and notices that network includes the first convolutional layer set gradually, the first attention network structure, the first pond layer, volume Two
Lamination, the second attention network structure, the second pond layer, third convolutional layer, third pay attention to network structure and full articulamentum.
3. the kinship recognition methods according to claim 2 for paying attention to network based on local feature, which is characterized in that institute
It states the first attention network structure, the second attention network structure and the third and notices that network structure includes setting gradually
Maximum pond layer, Volume Four lamination and up-sampling layer.
4. the kinship recognition methods according to claim 2 for paying attention to network based on local feature, which is characterized in that institute
It states and notices that network extracts local feature from each character image respectively using local feature trained in advance, and to the part
Feature carries out validity feature enhancing processing, specifically includes:
The data of the multiple character image are input to first convolutional layer, and export first partial feature；
The first partial feature is input to described first and notices that network structure carries out Feature Mapping, and exports the first mapping knot
Fruit；
First mapping result is sequentially input to first pond layer, second convolutional layer, and exports the second part
Feature；
Second local feature is input to described second and notices that network structure carries out Feature Mapping, and exports the second mapping knot
Fruit；
Second mapping result is sequentially input to second pond layer, the third convolutional layer, and exports third part
Feature；
The third local feature is input to the third and notices that network structure carries out Feature Mapping, and exports third mapping knot
Fruit；
The third mapping result is input to the full articulamentum, and exports the 4th local feature, the 4th local feature
As validity feature enhancing treated local feature.
5. the kinship recognition methods according to claim 4 for paying attention to network based on local feature, which is characterized in that every
One notices that network structure carries out the calculation formula of Feature Mapping are as follows:
P (x)=(1+F (x)) * C (x)；
Wherein, C (x) is the data of convolutional layer output, and F (x) is activation primitive, and P (x) is mapping result.
6. the kinship recognition methods according to claim 1 for paying attention to network based on local feature, which is characterized in that
Before the multiple character images of acquisition, further includes:
It establishes local feature and pays attention to network；
Obtain multiple groups character image sample；Every group of character image sample includes naked image pattern and the portion of same character image
Divide and covers image pattern；
The multiple groups character image sample is input to the local feature and pays attention to network, to pay attention to network to the local feature
It is trained, and exports sample and cover result and kinship result；
Result is covered according to the sample and kinship result calculates final loss result, and the final loss result is anti-
The local feature of feeding pays attention to network.
7. the kinship recognition methods according to claim 6 for paying attention to network based on local feature, which is characterized in that institute
It states and result and the final loss result of kinship result calculating is covered according to the sample, specifically include:
The first-loss result that the sample covers result is calculated using cross entropy loss function；
The second loss result of the kinship result is calculated using the cross entropy loss function；
The final loss result is calculated according to the first-loss result and second loss result.
8. the kinship recognition methods according to claim 7 for paying attention to network based on local feature, which is characterized in that institute
State cross entropy loss function are as follows:
Loss (x, class)=- x [class]+log [∑jexp(x[j])]；
Wherein, x is that sample covers result or kinship as a result, class is the label of lineup's object image sample, Loss (x,
It class) is loss result.
9. the kinship recognition methods according to claim 1 for paying attention to network based on local feature, which is characterized in that institute
State the calculation formula of final loss result are as follows:
Loss=Losslp+λ*Lossks；
Wherein, Loss is final loss result, LosslpFor first-loss as a result, LossksFor the second loss result, λ is weight.
10. a kind of social relationships identification device based on semantically enhancement network can be realized such as any one of claim 1 to 9 institute
The social relationships recognition methods based on semantically enhancement network stated, which is characterized in that described device includes:
Image collection module, for obtaining multiple character images；
Characteristic extracting module, for paying attention to the network extraction office from each character image respectively using local feature trained in advance
Portion's feature, and validity feature enhancing processing is carried out to the local feature；And
Identification module, for local feature to identify the kinship of the multiple character image according to treated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910434461.2A CN110334588A (en) | 2019-05-23 | 2019-05-23 | Kinship recognition methods and the device of network are paid attention to based on local feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910434461.2A CN110334588A (en) | 2019-05-23 | 2019-05-23 | Kinship recognition methods and the device of network are paid attention to based on local feature |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110334588A true CN110334588A (en) | 2019-10-15 |
Family
ID=68139183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910434461.2A Pending CN110334588A (en) | 2019-05-23 | 2019-05-23 | Kinship recognition methods and the device of network are paid attention to based on local feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110334588A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112668509A (en) * | 2020-12-31 | 2021-04-16 | 深圳云天励飞技术股份有限公司 | Training method and recognition method of social relationship recognition model and related equipment |
| CN113920573A (en) * | 2021-11-22 | 2022-01-11 | 河海大学 | Face change decoupling relativity relationship verification method based on counterstudy |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105005774A (en) * | 2015-07-28 | 2015-10-28 | 中国科学院自动化研究所 | Face relative relation recognition method based on convolutional neural network and device thereof |
| US20160202756A1 (en) * | 2015-01-09 | 2016-07-14 | Microsoft Technology Licensing, Llc | Gaze tracking via eye gaze model |
| CN108596211A (en) * | 2018-03-29 | 2018-09-28 | 中山大学 | It is a kind of that pedestrian's recognition methods again is blocked based on focusing study and depth e-learning |
| CN109543606A (en) * | 2018-11-22 | 2019-03-29 | 中山大学 | A kind of face identification method that attention mechanism is added |
| CN109784144A (en) * | 2018-11-29 | 2019-05-21 | 北京邮电大学 | A kind of kinship recognition methods and system |
-
2019
- 2019-05-23 CN CN201910434461.2A patent/CN110334588A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160202756A1 (en) * | 2015-01-09 | 2016-07-14 | Microsoft Technology Licensing, Llc | Gaze tracking via eye gaze model |
| CN105005774A (en) * | 2015-07-28 | 2015-10-28 | 中国科学院自动化研究所 | Face relative relation recognition method based on convolutional neural network and device thereof |
| CN108596211A (en) * | 2018-03-29 | 2018-09-28 | 中山大学 | It is a kind of that pedestrian's recognition methods again is blocked based on focusing study and depth e-learning |
| CN109543606A (en) * | 2018-11-22 | 2019-03-29 | 中山大学 | A kind of face identification method that attention mechanism is added |
| CN109784144A (en) * | 2018-11-29 | 2019-05-21 | 北京邮电大学 | A kind of kinship recognition methods and system |
Non-Patent Citations (1)
| Title |
|---|
| FEI WANG等: "Residual Attention Network for Image Classification", 《 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112668509A (en) * | 2020-12-31 | 2021-04-16 | 深圳云天励飞技术股份有限公司 | Training method and recognition method of social relationship recognition model and related equipment |
| CN112668509B (en) * | 2020-12-31 | 2024-04-02 | 深圳云天励飞技术股份有限公司 | Training method and recognition method of social relation recognition model and related equipment |
| CN113920573A (en) * | 2021-11-22 | 2022-01-11 | 河海大学 | Face change decoupling relativity relationship verification method based on counterstudy |
| CN113920573B (en) * | 2021-11-22 | 2022-05-13 | 河海大学 | Face change decoupling relativity relationship verification method based on counterstudy |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108229490B (en) | Key point detection method, neural network training method, device and electronic equipment | |
| CN109409435B (en) | Depth perception significance detection method based on convolutional neural network | |
| Nogueira et al. | Exploiting convnet diversity for flooding identification | |
| CN110353675B (en) | Electroencephalogram signal emotion recognition method and device based on picture generation | |
| CN108345818B (en) | Face living body detection method and device | |
| Chen et al. | Detection evolution with multi-order contextual co-occurrence | |
| CN111814574B (en) | Face living body detection system, terminal and storage medium applying double-branch three-dimensional convolution model | |
| CN108710847A (en) | Scene recognition method, device and electronic equipment | |
| CN112800894A (en) | Dynamic expression recognition method and system based on attention mechanism between space and time streams | |
| TW200842733A (en) | Object image detection method | |
| WO2013063765A1 (en) | Object detection using extended surf features | |
| CN110334588A (en) | Kinship recognition methods and the device of network are paid attention to based on local feature | |
| Xiao et al. | Attention-based deep neural network for driver behavior recognition | |
| CN115082698B (en) | Distraction driving behavior detection method based on multi-scale attention module | |
| CN106971161A (en) | Face In vivo detection system based on color and singular value features | |
| CN113283338A (en) | Method, device and equipment for identifying driving behavior of driver and readable storage medium | |
| Agarwal et al. | Presentation attack detection system for fake Iris: a review | |
| Shen et al. | A competitive method to vipriors object detection challenge | |
| CN107239827A (en) | A kind of spatial information learning method based on artificial neural network | |
| Rahim et al. | Dynamic hand gesture based sign word recognition using convolutional neural network with feature fusion | |
| CN110309832A (en) | A kind of object classification method based on image, system and electronic equipment | |
| US20100268301A1 (en) | Image processing algorithm for cueing salient regions | |
| She et al. | Micro-expression recognition based on multiple aggregation networks | |
| CN115984919A (en) | Micro-expression recognition method and system | |
| US20220207261A1 (en) | Method and apparatus for detecting associated objects |
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: 20191015 |