CN109934195A - A kind of anti-spoofing three-dimensional face identification method based on information fusion - Google Patents
A kind of anti-spoofing three-dimensional face identification method based on information fusion Download PDFInfo
- Publication number
- CN109934195A CN109934195A CN201910216754.3A CN201910216754A CN109934195A CN 109934195 A CN109934195 A CN 109934195A CN 201910216754 A CN201910216754 A CN 201910216754A CN 109934195 A CN109934195 A CN 109934195A
- Authority
- CN
- China
- Prior art keywords
- face
- depth
- depth map
- real human
- cromogram
- 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
Abstract
A kind of anti-spoofing three-dimensional face identification method based on information fusion of the invention, comprising: step 1: it acquires the cromogram of multiple real human faces and depth map and handles；Step 2: establishing the Gaussian distribution model of multiple real human faces according to the depth information of key point in every depth map, and determine the threshold range of real human face Gaussian Distribution Parameters；Step 3: establishing the Gaussian distribution model of face to be identified, the threshold range of the Gaussian distribution model parameter of face to be identified and real human face Gaussian Distribution Parameters is compared into judgement, if it is decided that 4 are thened follow the steps for real human face, otherwise without recognition of face；Step 4: building depth convolutional neural networks and training；Step 5: facial image to be identified being input to trained depth convolutional neural networks and is identified, recognition result is exported.By being merged to face depth information progress analysis modeling, and in data terminal, light-weighted network is constructed, the performance of entire face identification system is promoted.
Description
Technical field
The invention belongs to technical field of face recognition, are related to a kind of anti-spoofing three-dimensional face identification side based on information fusion
Method.
Background technique
Currently, face identification system has been widely used in the fields such as access control, identity duplicate removal, video monitoring, however
Most people face recognition method is the differentiation that identity is carried out based on two dimensional image.Two-dimension human face identifies under unfettered environment
It is still faced with huge challenge, such as attitudes vibration, illumination variation, expression shape change, camouflage variation, plastic operation variation, and
For the safeties such as control zone more importance department, face anti-counterfeiting technology, which seems, to be even more important.Three-dimensional face identification
There can be stronger robustness using three-dimensional depth information is anti-fake to above-mentioned variation and face relative to two-dimension human face identification.
Anti-fake for face, current main method has: 1) based on the detection of motion information, such as being acted by blink, mouth
Portion's movement carries out Activity determination；2) gaussian filtering is analyzed based on the analysis of texture, such as Fourier spectrum, this method is vulnerable to light
It is poor for video attack effect according to the influence of, image resolution ratio；3) it is detected based on multispectral reflection case, the party
Method requires strictly acquisition condition, and multispectral image cost is higher than VISIBLE LIGHT SYSTEM；4) it based on the detection of Fusion Features, such as combines
Motion analysis and face unity and coherence in writing are judged that this method verification and measurement ratio is higher but average handling time is longer, and wants to hardware
Ask higher.
These above-mentioned method the high requirements on the equipment and it efficiently can not distinguish that collected photo is derived from simultaneously and take the photograph
As the real human face before head, the photo still printed, or be the video etc. recorded in advance.Currently, preferable three
Tieing up recognition of face is that the Fusion Features mode based on deep learning is realized, but will cause original image information in this way in depth
It is more serious to spend the ratio lost in the communication process of neural network, entire discrimination is caused to be declined.
Summary of the invention
The object of the present invention is to provide a kind of anti-spoofing three-dimensional face identification methods based on information fusion, to promote three-dimensional
Face recognition accuracy rate, while the attack from photos and videos is solved, realize the anti-fake function of face.
A kind of anti-spoofing three-dimensional face identification method based on information fusion of the invention, includes the following steps:
Step 1: acquiring the cromogram and depth map of multiple real human faces, and cromogram and depth map are carried out at image
Reason, obtains the cromogram image set and depth map image set of real human face；
Step 2: the Gaussian distribution model of multiple real human faces is established according to the depth information of key point in every depth map,
The threshold range of real human face Gaussian Distribution Parameters is determined according to multiple Gaussian distribution models；
Step 3: the Gaussian distribution model of face to be identified is established, by the Gaussian distribution model parameter of face to be identified and very
The threshold range of real face Gaussian Distribution Parameters compares, and determines the authenticity of face to be identified, if it is decided that is true people
Face thens follow the steps 4 carry out recognition of face steps, otherwise without recognition of face step；
Step 4: building depth convolutional neural networks, and using the cromogram image set and depth map image set of real human face to depth
Degree convolutional neural networks are trained；
Step 5: facial image to be identified is input to trained depth convolutional neural networks and is identified, output identification
As a result.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, the step 1 is specifically included:
Step 1.1: acquiring the cromogram and depth map of multiple real human faces；
Step 1.2: carrying out the detection of real human face characteristic point to cromogram, determine the coordinate information of face frame and multiple
The coordinate information of characteristic point completes face correction alignment and face shear treatment；
Step 1.3: background segment and noise suppression preprocessing are carried out to depth map；
Step 1.4: the space reflection of characteristic point is carried out according to the relationship between cromogram and depth map, according to characteristic point
Mapping carries out face shear treatment to depth map；
Step 1.5: the final cromogram image set and depth map image set for obtaining real human face.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, the step 2 is specifically included:
Step 2.1: dispersion analysis is carried out to the depth information of key point in every depth map；
Step 2.2: determining the maximum difference of depth information between the key point of every depth map, and calculate multiple depth maps
The average value of the maximum difference of depth information between key point；
Step 2.3: according to the minimum value of depth information in depth map and the average value of maximum difference, in certain difference
Operation is normalized in interior depth information；
Step 2.4: according to the depth information after normalization, establishing the height of the nasal area of the depth map of every real human face
This distributed model；
Step 2.5: the corresponding Gaussian distribution model of every depth map being concentrated by depth image, establishes the height of real human face
The threshold range of this distribution parameter.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, the step 3 is specifically included:
Step 3.1: identification facial image being treated according to the image procossing mode in step 1 and is handled, after being handled
Depth map and cromogram；
Step 3.2: dispersion analysis is carried out to the depth information of key point in treated depth map；
Step 3.3: according to the depth information for calculating acquisition in the minimum value and step 2.2 of depth information in depth map
Operation is normalized to the depth information in certain difference in the average value of maximum difference；
Step 3.4: according to the depth information after normalization, establishing the Gaussian Profile mould of the nasal area of depth map to be identified
Type；
Step 3.5: by the Gaussian distribution model parameter of face to be identified and the threshold value of real human face Gaussian Distribution Parameters into
Row comparison, if it is decided that 4 carry out recognition of face steps are thened follow the steps for real human face, otherwise without recognition of face step.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, Gaussian distribution model formula is such as
Under:
Wherein,Indicate depth information,It is the vector that dimension is D,It is multiple vectorsAverage value, Σ indicate institute
Directed quantityCovariance matrix.
Depth convolution mind in the anti-spoofing three-dimensional face identification method of the invention based on information fusion, in step 4
It is that the twin network structure of light-type based on SqueezeNet constructs through network, data input pin is using cromogram and depth
4 dimensional data images made of degree figure fusion, export the feature vector that face is characterized for two；
Network structure is divided into the identical left and right two parts of structure, and every part specifically includes: sequentially connected convolutional layer is criticized
Normalization layer and pond layer, latter linked network structure using lightweight SqueezeNet structure as trunk, lightweight
Regularization is sequentially connected after SqueezeNet structure adds convolutional layer, global draw pond layer and regularization to add full articulamentum；
By after left and right two structures by the feature vector of two characterization faces of output be input to loss layers do it is European
Distance calculates, then compares loss classification.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, real human face is utilized in step 4
Cromogram image set and depth map image set depth convolutional neural networks are trained, specifically:
(1) read real human face cromogram image set and depth image concentrate every real human face cromogram with it is corresponding
Depth map, later by cromogram it is corresponding 3 dimension matrix it is corresponding with depth map 1 dimension matrix carry out dimension fusion, export one
4 dimension matrixes of the matrix of a 4 dimension, cromogram and depth map the fusion output of multiple real human faces constitute the training set of model；
(2) training parameter of depth convolutional neural networks is set, including total the number of iterations, basic studies rate lr, weight
Decaying and Dropout rate；
(3) sample of training set is input to depth convolutional neural networks, while according to the training of depth convolutional neural networks
Parameter constructs Adam optimizer；
(4) ask optimal solution come optimization neural network comparison loss function using the optimization method based on Adam optimizer
Weight parameter, the number of iterations until reaching setting complete the training of depth convolutional neural networks.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, the comparison loss function is as follows
Formula indicates:
Wherein d=| | an-bn||2, represent the Euclidean distances of the feature vector of two characterization faces, y be two samples whether
Matched label, y=1 represent that two samples are similar or matching, y=0 then represent mismatchs, and margin is the threshold value set,
It is set as the quantity that 1, N is sample.
In the anti-spoofing three-dimensional face identification method of the invention based on information fusion, the key point of face is that face is special
The center of nose in sign point, the left and right center of two eyes, mouth five coordinate points of left comer and right corner.
A kind of anti-spoofing three-dimensional face identification method based on information fusion of the invention, at least has below beneficial to effect
Fruit:
1) detection of face authenticity is carried out by using the depth information of facial image, can be used for simultaneously photo with
And the attack detecting of video can carry out the judgement of face authenticity by the foundation of Gaussian distribution model faster.
2) by being merged in data terminal, protection processing greatly is carried out to data in front end, uses light-weighted net
Network greatly reduces storage and space hold promotes entire recognition of face efficiency in the case where guaranteeing accuracy.
3) the detection judgement to face authenticity, while the face proposed can be realized on the basis of relatively low cost
Identifying schemes can largely reduce the loss of face information, reduce model memory space, improve the efficiency of recognition of face,
Promote the performance of entire face identification system.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow charts of the anti-spoofing three-dimensional face identification method of information fusion；
Fig. 2 a is face spatial model figure in the present invention；
Fig. 2 b is face spatial model figure and side curve figure in the present invention；
Fig. 2 c is Gauss two-dimensional spatial model figure in the present invention；
Fig. 3 is the structure chart of depth convolutional neural networks of the invention.
Specific embodiment
As shown in Figure 1, a kind of anti-spoofing three-dimensional face identification method based on information fusion of the invention, including walk as follows
It is rapid:
Step 1: acquiring the cromogram and depth map of multiple real human faces, and cromogram and depth map are carried out at image
Reason, obtains the cromogram image set and depth map image set of real human face, the step 1 is established real human face database and specifically included:
Step 1.1: acquiring the cromogram and depth map of multiple real human faces；
When it is implemented, being adopted using the image that depth camera carries out face to each personnel for participating in database sharing respectively
Collection, obtains everyone the multipair cromogram and depth map under different time, different conditions.
Step 1.2: to cromogram carry out real human face characteristic point detection, determine face frame coordinate information and 68
The coordinate information of characteristic point completes face correction alignment, so that face is in front position as far as possible, and shear to face
Processing；
Step 1.3: background segment and noise suppression preprocessing are carried out to depth map,
When it is implemented, real human face and body part can be used as an entirety for background parts, use
The method of cluster completes the segmentation of background, completes denoising to depth map using the method for bilateral filtering.
Step 1.4: completing the space reflection of 68 characteristic points according to the relationship between cromogram and depth map, obtain face
The position of human face characteristic point in depth map carries out face shear treatment to depth map according to the mapping of characteristic point；
Step 1.5: the final cromogram image set and depth map image set for obtaining real human face.
Step 2: the Gaussian distribution model of multiple real human faces is established according to the depth information of key point in every depth map,
The threshold range of real human face Gaussian Distribution Parameters is determined according to multiple Gaussian distribution models, the step 2 specifically includes:
Step 2.1: dispersion analysis is carried out to the depth information of key point in every depth map；
The key point of face is the center of the nose in 68 characteristic points of face, the left and right center of two eyes, mouth
Five coordinate points of left comer and right corner.
When it is implemented, establishing spatial model according to the depth information in the depth map of obtained real human face to assist pair
The analysis of depth information, the model of real human face in space is shown in Fig. 2 (a), from side it can be seen that face part is
Rough shape.As shown in Fig. 2 (c), look down from front, nose and nose peripheral part and Gauss two-dimensional space mould
Type is similar, and i.e. BC and the curves of two sections of CD formation are two-dimensional side projection lines in Fig. 2 (b).Thus according to collected true
Face depth information carries out the analysis of dispersion and Gauss models.
Step 2.2: determining the maximum difference of depth information between 5 key points of every depth map, and calculate multiple depth
The average value of the maximum difference of depth information between the key point of figure；
Step 2.3: according to the minimum value of depth information in depth map and the average value of maximum difference, in certain difference
Operation is normalized in interior depth information；
Step 2.4: according to the depth information after normalization, establishing the height of the nasal area of the depth map of every real human face
This distributed model；
Step 2.5: the corresponding Gaussian distribution model of every depth map being concentrated by depth image, establishes the height of real human face
The threshold range of this distribution parameter.
Step 3: the Gaussian distribution model of face to be identified is established, by the Gaussian distribution model parameter of face to be identified and very
The threshold range of real face Gaussian Distribution Parameters compares, and determines the authenticity of face to be identified, if it is decided that is true people
Face thens follow the steps 4 carry out recognition of face steps, and otherwise without recognition of face step, the step 3 is specifically included:
Step 3.1: identification facial image being treated according to the image procossing mode in step 1 and is handled, after being handled
Depth map and cromogram；
Step 3.2: dispersion analysis is carried out to the depth information of key point in treated depth map；
Step 3.3: according to the depth information for calculating acquisition in the minimum value and step 2.2 of depth information in depth map
Operation is normalized to the depth information in certain difference in the average value of maximum difference；
Step 3.4: according to the depth information after normalization, establishing the Gaussian Profile mould of the nasal area of depth map to be identified
Type；
Step 3.5: by the Gaussian distribution model parameter of face to be identified and the threshold value of real human face Gaussian Distribution Parameters into
Row comparison, if it is decided that 4 carry out recognition of face steps are thened follow the steps for real human face, otherwise without recognition of face step.
Above-mentioned Gaussian distribution model formula is as follows:
Wherein,Indicate depth information,It is the vector that dimension is D,It is multiple vectorsAverage value, Σ indicates all
VectorCovariance matrix.
Step 4: building depth convolutional neural networks, and using the cromogram image set and depth map image set of real human face to depth
Degree convolutional neural networks are trained；
Depth convolutional neural networks are that the twin network structure of light-type based on SqueezeNet constructs, data input pin
Using 4 dimensional data images made of cromogram and depth map fusion, the feature vector that face is characterized for two is exported；
SqueezeNet network is that convolutional layer adds pondization to operate first, the fire module, Fire followed by proposed
Module is constituted by two layers, is the squeeze layer and 1 × 1 and 3 × 3 convolution kernels of the convolutional layer composition of 1 × 1 convolution kernel respectively
The expand layer of convolutional layer composition, it is final to replace full articulamentum to be exported using global average pond layer by fire2-9.
As shown in figure 3, depth convolutional neural networks of the invention are divided into the identical left and right two parts of structure, every part tool
Body includes: sequentially connected convolutional layer, batch normalization layer and pond layer, latter linked network structure with lightweight
SqueezeNet structure is trunk, and regularization is sequentially connected after lightweight SqueezeNet structure and adds convolutional layer, global draw
Pond layer and regularization add full articulamentum；By after left and right two structures by the feature of two characterization faces of output to
Amount is input to loss layers and does Euclidean distance calculating, then compares loss classification.
Depth convolutional neural networks are instructed using the cromogram image set of real human face and depth map image set in step 4
Practice, specifically:
(1) read real human face cromogram image set and depth image concentrate every real human face cromogram with it is corresponding
Depth map, later by cromogram it is corresponding 3 dimension matrix it is corresponding with depth map 1 dimension matrix carry out dimension fusion, export one
4 dimension matrixes of the matrix of a 4 dimension, cromogram and depth map the fusion output of multiple real human faces constitute the training set of model；Net
The input of network is two groups of arbitrary 4 dimension matrixes, and the label of network model is 0 and 1 two kind of number, and respectively representing is not the same person
Be the same person；
(2) training parameter of depth convolutional neural networks is set, is including total the number of iterations 300,000, basic studies rate lr
0.001, it is 0.4 that weight, which decays to 0.0003 and Dropout rate,；About 6000 groups of the data of network training, about 100 people for including；
(3) sample of training set is input to depth convolutional neural networks, while according to the training of depth convolutional neural networks
Parameter constructs Adam optimizer；
(4) ask optimal solution come optimization neural network comparison loss function using the optimization method based on Adam optimizer
Weight parameter, the number of iterations until reaching setting complete the training of depth convolutional neural networks.
When it is implemented, comparison loss function such as following formula indicates:
Wherein d=| | an-bn||2, represent the Euclidean distances of the feature vector of two characterization faces, y be two samples whether
Matched label, y=1 represent that two samples are similar or matching, y=0 then represent mismatchs, and margin is the threshold value set,
It is set as the quantity that 1, N is sample.
Step 5: facial image to be identified is input to trained depth convolutional neural networks and is identified, output identification
As a result.
When carrying out face feature vector extraction with corresponding depth map to freshly harvested cromogram using trained model,
Network inputs only need one group of fused 4 dimension matrix.In face recognition process, it is necessary first to carry out the registration of face, lead to
Trained model is crossed to carry out the extraction of face characteristic to the people that needs are registered and save using respective name as ending,
The feature vector of the different people finally saved is as registered set.In the application scenarios of access control system, for the people that enters of needs,
After carrying out man face image acquiring processing and the true sex determination of face by depth camera, if it is decided that for real human face then into
Row recognition of face.The depth map of collected current face and cromogram are fused to the matrix of 14 dimension, Input matrix is arrived
The extraction that trained model carries out face characteristic is found out by doing the comparison of Euclidean distance with feature vector in registered set
The corresponding feature vector of Euclidean distance minimum value, the corresponding name of the feature vector found at this time are this person before depth camera
Name completes recognition of face.
In order to further verify the performance of model, the people not comprising registered set should be had in the middle by needing to know others, be passed through
Model extraction does Euclidean distance calculating with the feature vector of registered set to the feature vector of people to be identified, when distance is less than
Registered set feature vector corresponding people is searched when threshold value 0.7, can not find corresponding people in registered set when greater than 0.7, as
Have neither part nor lot in the personnel of registration.
The present invention constructs lightweight by being merged to face depth information progress analysis modeling, and in data terminal
Network, promote the performance of entire face identification system.
The foregoing is merely presently preferred embodiments of the present invention, the thought being not intended to limit the invention, all of the invention
Within spirit and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (9)
1. a kind of anti-spoofing three-dimensional face identification method based on information fusion, which comprises the steps of:
Step 1: acquiring the cromogram and depth map of multiple real human faces, and image procossing is carried out to cromogram and depth map, obtain
Obtain the cromogram image set and depth map image set of real human face；
Step 2: the Gaussian distribution model of multiple real human faces is established according to the depth information of key point in every depth map, according to
Multiple Gaussian distribution models determine the threshold range of real human face Gaussian Distribution Parameters；
Step 3: the Gaussian distribution model of face to be identified is established, by the Gaussian distribution model parameter of face to be identified and true people
The threshold range of face Gaussian Distribution Parameters compares, and determines the authenticity of face to be identified, if it is decided that then for real human face
It executes step 4 and carries out recognition of face step, otherwise without recognition of face step；
Step 4: building depth convolutional neural networks, and depth is rolled up using the cromogram image set of real human face and depth map image set
Product neural network is trained；
Step 5: facial image to be identified is input to trained depth convolutional neural networks and is identified, output identification knot
Fruit.
2. the anti-spoofing three-dimensional face identification method as described in claim 1 based on information fusion, which is characterized in that the step
Rapid 1 specifically includes:
Step 1.1: acquiring the cromogram and depth map of multiple real human faces；
Step 1.2: to cromogram carry out real human face characteristic point detection, determine face frame coordinate information and multiple features
The coordinate information of point completes face correction alignment and face shear treatment；
Step 1.3: background segment and noise suppression preprocessing are carried out to depth map；
Step 1.4: the space reflection of characteristic point is carried out according to the relationship between cromogram and depth map, according to the mapping of characteristic point
Face shear treatment is carried out to depth map；
Step 1.5: the final cromogram image set and depth map image set for obtaining real human face.
3. the anti-spoofing three-dimensional face identification method as described in claim 1 based on information fusion, which is characterized in that the step
Rapid 2 specifically include:
Step 2.1: dispersion analysis is carried out to the depth information of key point in every depth map；
Step 2.2: determining the maximum difference of depth information between the key point of every depth map, and calculate the key of multiple depth maps
The average value of the maximum difference of depth information between point；
Step 2.3: according to the minimum value of depth information in depth map and the average value of maximum difference, in certain difference
Operation is normalized in depth information；
Step 2.4: according to the depth information after normalization, the Gauss for establishing the nasal area of the depth map of every real human face point
Cloth model；
Step 2.5: the corresponding Gaussian distribution model of every depth map being concentrated by depth image, establishes the Gauss point of real human face
The threshold range of cloth parameter.
4. the anti-spoofing three-dimensional face identification method as claimed in claim 3 based on information fusion, which is characterized in that the step
Rapid 3 specifically include:
Step 3.1: identification facial image being treated according to the image procossing mode in step 1 and is handled, treated for acquisition deeply
Degree figure and cromogram；
Step 3.2: dispersion analysis is carried out to the depth information of key point in treated depth map；
Step 3.3: according to the maximum for the depth information for calculating acquisition in the minimum value and step 2.2 of depth information in depth map
Operation is normalized to the depth information in certain difference in the average value of difference；
Step 3.4: according to the depth information after normalization, establishing the Gaussian distribution model of the nasal area of depth map to be identified；
Step 3.5: the threshold value of the Gaussian distribution model parameter of face to be identified and real human face Gaussian Distribution Parameters is carried out pair
Than, if it is decided that 4 carry out recognition of face steps are thened follow the steps for real human face, otherwise without recognition of face step.
5. the anti-spoofing three-dimensional face identification method based on information fusion as described in claim 3 or 4, which is characterized in that high
This distributed model formula is as follows:
Wherein,Indicate depth information,It is the vector that dimension is D,It is multiple vectorsAverage value, Σ indicate institute's directed quantityCovariance matrix.
6. the anti-spoofing three-dimensional face identification method as described in claim 1 based on information fusion, which is characterized in that step 4
In depth convolutional neural networks be that the twin network structure of light-type based on SqueezeNet constructs, data input pin use
Be cromogram and depth map fusion made of 4 dimensional data images, export for two characterize face feature vector；
Network structure is divided into the identical left and right two parts of structure, and every part specifically includes: sequentially connected convolutional layer, batch standard
Change layer and pond layer, latter linked network structure using lightweight SqueezeNet structure as trunk, lightweight SqueezeNet
Regularization is sequentially connected after structure adds convolutional layer, global draw pond layer and regularization to add full articulamentum；
Euclidean distance is done by the feature vector of two characterization faces of output is input to loss layers after left and right two structures
It calculates, then compares loss classification.
7. the anti-spoofing three-dimensional face identification method as described in claim 1 based on information fusion, which is characterized in that step 4
It is middle that depth convolutional neural networks are trained using the cromogram image set and depth map image set of real human face, specifically:
(1) cromogram and the corresponding depth for every real human face that the cromogram image set and depth image for reading real human face are concentrated
The corresponding 3 dimension matrix of cromogram 1 dimension matrix corresponding with depth map is carried out the fusion of dimension later, exports one 4 dimension by degree figure
Matrix, multiple real human faces cromogram and depth map fusion output 4 dimension matrixes constitute models training set；
(2) training parameter of depth convolutional neural networks is set, including total the number of iterations, basic studies rate lr, weight decaying
With Dropout rate；
(3) sample of training set is input to depth convolutional neural networks, while according to depth convolutional neural networks training parameter
Construct Adam optimizer；
(4) ask optimal solution come the weight of optimization neural network comparison loss function using the optimization method based on Adam optimizer
Parameter, the number of iterations until reaching setting complete the training of depth convolutional neural networks.
8. the anti-spoofing three-dimensional face identification method as claimed in claim 6 based on information fusion, which is characterized in that described right
For example than loss function following formula indicates:
Wherein d=| | an-bn||2, the Euclidean distance of the feature vector of two characterization faces is represented, y is whether two samples match
Label, y=1 represents that two samples are similar or matching, y=0 then represent mismatchs, and margin be the threshold value set, is arranged
It is the quantity of sample for 1, N.
9. the anti-spoofing three-dimensional face identification method as claimed in claim 3 based on information fusion, which is characterized in that face
Key point be the center of nose in human face characteristic point, the left and right center of two eyes, mouth five coordinate points of left comer and right corner.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910216754.3A CN109934195A (en) | 2019-03-21 | 2019-03-21 | A kind of anti-spoofing three-dimensional face identification method based on information fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910216754.3A CN109934195A (en) | 2019-03-21 | 2019-03-21 | A kind of anti-spoofing three-dimensional face identification method based on information fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109934195A true CN109934195A (en) | 2019-06-25 |
Family
ID=66987889
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910216754.3A Pending CN109934195A (en) | 2019-03-21 | 2019-03-21 | A kind of anti-spoofing three-dimensional face identification method based on information fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109934195A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110517482A (en) * | 2019-07-29 | 2019-11-29 | 杭州电子科技大学 | A kind of Short-time Traffic Flow Forecasting Methods based on 3D convolutional neural networks |
| CN110602476A (en) * | 2019-08-08 | 2019-12-20 | 南京航空航天大学 | Hole filling method of Gaussian mixture model based on depth information assistance |
| CN110674677A (en) * | 2019-08-06 | 2020-01-10 | 厦门大学 | Multi-mode multi-layer fusion deep neural network for anti-spoofing of human face |
| CN110956080A (en) * | 2019-10-14 | 2020-04-03 | 北京海益同展信息科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN111027394A (en) * | 2019-11-12 | 2020-04-17 | 天津大学 | Behavior classification method based on twin three-dimensional convolution neural network |
| CN111091089A (en) * | 2019-12-12 | 2020-05-01 | 新华三大数据技术有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN111144298A (en) * | 2019-12-26 | 2020-05-12 | 北京华捷艾米科技有限公司 | Pedestrian identification method and device |
| CN111476222A (en) * | 2020-06-11 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Image processing method, image processing device, computer equipment and computer readable storage medium |
| CN111739168A (en) * | 2020-06-30 | 2020-10-02 | 华东交通大学 | Large-scale three-dimensional face synthesis method with suppressed sample similarity |
| CN111881746A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
| CN111881706A (en) * | 2019-11-27 | 2020-11-03 | 马上消费金融股份有限公司 | Living body detection, image classification and model training method, device, equipment and medium |
| CN111898746A (en) * | 2020-08-11 | 2020-11-06 | 中国人民解放军海军航空大学 | Deep learning method for association of interrupted flight path continuation |
| CN112200125A (en) * | 2020-10-26 | 2021-01-08 | 东北大学 | Three-dimensional face recognition method |
| CN112446345A (en) * | 2020-12-09 | 2021-03-05 | 西安交通大学 | Low-quality three-dimensional face recognition method, system, equipment and storage medium |
| CN112800941A (en) * | 2021-01-26 | 2021-05-14 | 中科人工智能创新技术研究院(青岛)有限公司 | Face anti-fraud method and system based on asymmetric auxiliary information embedded network |
| CN112861586A (en) * | 2019-11-27 | 2021-05-28 | 马上消费金融股份有限公司 | Living body detection, image classification and model training method, device, equipment and medium |
| CN113743379A (en) * | 2021-11-03 | 2021-12-03 | 杭州魔点科技有限公司 | Light-weight living body identification method, system, device and medium for multi-modal characteristics |
| CN114267068A (en) * | 2021-12-24 | 2022-04-01 | 北京的卢深视科技有限公司 | Face recognition method based on continuous frame information, electronic equipment and storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012071677A1 (en) * | 2010-11-29 | 2012-06-07 | Technicolor (China) Technology Co., Ltd. | Method and system for face recognition |
| CN104573743A (en) * | 2015-01-14 | 2015-04-29 | 南京烽火星空通信发展有限公司 | Filtering method for facial image detection |
| US20150125049A1 (en) * | 2013-11-04 | 2015-05-07 | Facebook, Inc. | Systems and methods for facial representation |
| CN107424153A (en) * | 2017-04-18 | 2017-12-01 | 辽宁科技大学 | Face cutting techniques based on deep learning and Level Set Method |
| US20180005018A1 (en) * | 2016-06-30 | 2018-01-04 | U.S. Army Research Laboratory Attn: Rdrl-Loc-I | System and method for face recognition using three dimensions |
| CN107832677A (en) * | 2017-10-19 | 2018-03-23 | 深圳奥比中光科技有限公司 | Face identification method and system based on In vivo detection |
| CN108197587A (en) * | 2018-01-18 | 2018-06-22 | 中科视拓（北京）科技有限公司 | A kind of method that multi-modal recognition of face is carried out by face depth prediction |
| CN108416291A (en) * | 2018-03-06 | 2018-08-17 | 广州逗号智能零售有限公司 | Face datection recognition methods, device and system |
| CN108520204A (en) * | 2018-03-16 | 2018-09-11 | 西北大学 | A kind of face identification method |
| CN108764091A (en) * | 2018-05-18 | 2018-11-06 | 北京市商汤科技开发有限公司 | Biopsy method and device, electronic equipment and storage medium |
| CN109034102A (en) * | 2018-08-14 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Human face in-vivo detection method, device, equipment and storage medium |
-
2019
- 2019-03-21 CN CN201910216754.3A patent/CN109934195A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012071677A1 (en) * | 2010-11-29 | 2012-06-07 | Technicolor (China) Technology Co., Ltd. | Method and system for face recognition |
| US20150125049A1 (en) * | 2013-11-04 | 2015-05-07 | Facebook, Inc. | Systems and methods for facial representation |
| CN104573743A (en) * | 2015-01-14 | 2015-04-29 | 南京烽火星空通信发展有限公司 | Filtering method for facial image detection |
| US20180005018A1 (en) * | 2016-06-30 | 2018-01-04 | U.S. Army Research Laboratory Attn: Rdrl-Loc-I | System and method for face recognition using three dimensions |
| CN107424153A (en) * | 2017-04-18 | 2017-12-01 | 辽宁科技大学 | Face cutting techniques based on deep learning and Level Set Method |
| CN107832677A (en) * | 2017-10-19 | 2018-03-23 | 深圳奥比中光科技有限公司 | Face identification method and system based on In vivo detection |
| CN108197587A (en) * | 2018-01-18 | 2018-06-22 | 中科视拓（北京）科技有限公司 | A kind of method that multi-modal recognition of face is carried out by face depth prediction |
| CN108416291A (en) * | 2018-03-06 | 2018-08-17 | 广州逗号智能零售有限公司 | Face datection recognition methods, device and system |
| CN108520204A (en) * | 2018-03-16 | 2018-09-11 | 西北大学 | A kind of face identification method |
| CN108764091A (en) * | 2018-05-18 | 2018-11-06 | 北京市商汤科技开发有限公司 | Biopsy method and device, electronic equipment and storage medium |
| CN109034102A (en) * | 2018-08-14 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Human face in-vivo detection method, device, equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| JIAN-GANG WANG等: "Fusion of Appearance and Depth Information for Face Recognition", 《IEEE EXPLORE》 * |
| WUMING ZHANG等: "Improving Heterogeneous Face Recognition with Conditional Adversarial Networks", 《ARXIV》 * |
| 凌仁兵等: "基于人脸面部特征的三维人脸预处理方法", 《信息通信》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110517482A (en) * | 2019-07-29 | 2019-11-29 | 杭州电子科技大学 | A kind of Short-time Traffic Flow Forecasting Methods based on 3D convolutional neural networks |
| CN110674677A (en) * | 2019-08-06 | 2020-01-10 | 厦门大学 | Multi-mode multi-layer fusion deep neural network for anti-spoofing of human face |
| CN110602476A (en) * | 2019-08-08 | 2019-12-20 | 南京航空航天大学 | Hole filling method of Gaussian mixture model based on depth information assistance |
| CN110602476B (en) * | 2019-08-08 | 2021-08-06 | 南京航空航天大学 | Hole filling method of Gaussian mixture model based on depth information assistance |
| CN110956080A (en) * | 2019-10-14 | 2020-04-03 | 北京海益同展信息科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN110956080B (en) * | 2019-10-14 | 2023-11-03 | 京东科技信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN111027394B (en) * | 2019-11-12 | 2023-07-07 | 天津大学 | Behavior classification method based on twin three-dimensional convolutional neural network |
| CN111027394A (en) * | 2019-11-12 | 2020-04-17 | 天津大学 | Behavior classification method based on twin three-dimensional convolution neural network |
| CN112861586A (en) * | 2019-11-27 | 2021-05-28 | 马上消费金融股份有限公司 | Living body detection, image classification and model training method, device, equipment and medium |
| CN111881706A (en) * | 2019-11-27 | 2020-11-03 | 马上消费金融股份有限公司 | Living body detection, image classification and model training method, device, equipment and medium |
| CN113642466A (en) * | 2019-11-27 | 2021-11-12 | 马上消费金融股份有限公司 | Living body detection and model training method, apparatus and medium |
| CN111881706B (en) * | 2019-11-27 | 2021-09-03 | 马上消费金融股份有限公司 | Living body detection, image classification and model training method, device, equipment and medium |
| CN111091089A (en) * | 2019-12-12 | 2020-05-01 | 新华三大数据技术有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN111091089B (en) * | 2019-12-12 | 2022-07-29 | 新华三大数据技术有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN111144298A (en) * | 2019-12-26 | 2020-05-12 | 北京华捷艾米科技有限公司 | Pedestrian identification method and device |
| CN111476222A (en) * | 2020-06-11 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Image processing method, image processing device, computer equipment and computer readable storage medium |
| CN111881746B (en) * | 2020-06-23 | 2024-04-02 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
| CN111881746A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
| CN111739168A (en) * | 2020-06-30 | 2020-10-02 | 华东交通大学 | Large-scale three-dimensional face synthesis method with suppressed sample similarity |
| CN111898746B (en) * | 2020-08-11 | 2024-01-02 | 中国人民解放军海军航空大学 | Deep learning method for continuous relevance of broken flight path |
| CN111898746A (en) * | 2020-08-11 | 2020-11-06 | 中国人民解放军海军航空大学 | Deep learning method for association of interrupted flight path continuation |
| CN112200125A (en) * | 2020-10-26 | 2021-01-08 | 东北大学 | Three-dimensional face recognition method |
| CN112446345A (en) * | 2020-12-09 | 2021-03-05 | 西安交通大学 | Low-quality three-dimensional face recognition method, system, equipment and storage medium |
| CN112446345B (en) * | 2020-12-09 | 2024-04-05 | 西安交通大学 | Low-quality three-dimensional face recognition method, system, equipment and storage medium |
| CN112800941A (en) * | 2021-01-26 | 2021-05-14 | 中科人工智能创新技术研究院(青岛)有限公司 | Face anti-fraud method and system based on asymmetric auxiliary information embedded network |
| CN112800941B (en) * | 2021-01-26 | 2023-06-06 | 中科人工智能创新技术研究院(青岛)有限公司 | Face anti-fraud method and system based on asymmetric auxiliary information embedded network |
| CN113743379A (en) * | 2021-11-03 | 2021-12-03 | 杭州魔点科技有限公司 | Light-weight living body identification method, system, device and medium for multi-modal characteristics |
| CN114267068B (en) * | 2021-12-24 | 2022-11-01 | 合肥的卢深视科技有限公司 | Face recognition method based on continuous frame information, electronic equipment and storage medium |
| CN114267068A (en) * | 2021-12-24 | 2022-04-01 | 北京的卢深视科技有限公司 | Face recognition method based on continuous frame information, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109934195A (en) | A kind of anti-spoofing three-dimensional face identification method based on information fusion | |
| CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
| CN106815566B (en) | Face retrieval method based on multitask convolutional neural network | |
| CN104008370B (en) | A kind of video face identification method | |
| Turk et al. | Eigenfaces for recognition | |
| CN102194131B (en) | Fast human face recognition method based on geometric proportion characteristic of five sense organs | |
| CN105447441B (en) | Face authentication method and device | |
| CN109409297B (en) | Identity recognition method based on dual-channel convolutional neural network | |
| CN104933414B (en) | A kind of living body faces detection method based on WLD-TOP | |
| CN111274916B (en) | Face recognition method and face recognition device | |
| CN105138954A (en) | Image automatic screening, query and identification system | |
| CN111652827A (en) | Front face synthesis method and system based on generation countermeasure network | |
| CN112801015B (en) | Multi-mode face recognition method based on attention mechanism | |
| CN106295600A (en) | Driver status real-time detection method and device | |
| Wang et al. | Head pose estimation with combined 2D SIFT and 3D HOG features | |
| CN111476222B (en) | Image processing method, image processing device, computer equipment and computer readable storage medium | |
| CN106650574A (en) | Face identification method based on PCANet | |
| CN113205002B (en) | Low-definition face recognition method, device, equipment and medium for unlimited video monitoring | |
| CN113963032A (en) | Twin network structure target tracking method fusing target re-identification | |
| CN111914643A (en) | Human body action recognition method based on skeleton key point detection | |
| CN112613480A (en) | Face recognition method, face recognition system, electronic equipment and storage medium | |
| Wu et al. | Kinship verification from faces: Methods, databases and challenges | |
| CN110969101A (en) | Face detection and tracking method based on HOG and feature descriptor | |
| CN114239754A (en) | Pedestrian attribute identification method and system based on attribute feature learning decoupling | |
| CN112800979B (en) | Dynamic expression recognition method and system based on characterization flow embedded network |
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 | ||
| AD01 | Patent right deemed abandoned | ||
| AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20230915 |