CN110516616A - A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set - Google Patents
A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set Download PDFInfo
- Publication number
- CN110516616A CN110516616A CN201910809299.8A CN201910809299A CN110516616A CN 110516616 A CN110516616 A CN 110516616A CN 201910809299 A CN201910809299 A CN 201910809299A CN 110516616 A CN110516616 A CN 110516616A
- Authority
- CN
- China
- Prior art keywords
- rgb
- infrared data
- feature
- data set
- extensive
- 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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- 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
-
- 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
-
- 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
- G06V40/40—Spoof detection, e.g. liveness detection
- G06V40/45—Detection of the body part being alive
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Collating Specific Patterns (AREA)
Abstract
The present invention relates to a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set, determine and double judgement including a major punishment, if a major punishment is set to very, carries out the second major punishment and determines, specifically: a major punishment is fixed: 1.1, the acquisition and pretreatment of training dataset；1.2, the textural characteristics of RGB image are extracted using the operator of local binary pattern；1.3, the textural characteristics input classifier of extraction is subjected to an anti-fake judgement of weight；Double judgement: the 2.1, extraction of near-infrared data set global characteristics；2.2, the Local textural feature of extraction and global characteristics are inputted into improved convolutional network and carries out Fusion Features；2.3, the feature of fusion is inputted into improved CNN network classified and determine the true and false of face.This method no user interactivity requirements, without the optional equipment of high-end complexity, the aid of double authentication more makes it be really achieved the detection demand of Face datection.
Description
Technical field
The present invention relates to recognition of face field of anti-counterfeit technology, and in particular to one kind is based on extensive RGB and near-infrared number
According to the double authentication face method for anti-counterfeit of collection.
Background technique
The rapid development of science and technology, the arriving in artificial intelligence big epoch, recognition of face is with its rapidity, validity and user friend
Good property is successfully applied to the various industries such as finance, security protection, education.Recognition of face is by the face in acquired image
It is identified, to verify the validity of its identity.But for collected facial image be true man before camera
Or the false face forged can not but identify.Therefore, system is highly prone to attack.With social media website explosive growth and
The raising of camera resolution, it is easy to the face image or video information for obtaining legitimate user to implement system directly attack,
To disguise oneself as, legitimate user is illegally entered, this brings great threat to the safety of system.Therefore, a complete people
Face identifying system must should have accurate identification function, while will also be from being spoofed and attacking.Face anti-counterfeiting technology is made
For the security protection of recognition of face, guarantee to identify to obtain the possibility of access power by forging face by excluding all
System stablizes the operation of safety.
Face identification system data collection terminal would generally face three kinds of common attack patterns: the photograph print of legitimate user
Attack, video replay attack and the attack of 3D mask.The characteristics of for each attack pattern, it is living that researchers propose different faces
Body detecting method, main thought are by face texture feature, the fortune between manual feature or the true and false face of CNN Characterizations
Dynamic feature, spectral characteristic, physiological property and several how features of 3D, so using these difference characteristics as feature input classifier into
The true and false classification of face two judgement of row.In order to enable defence algorithm is more robust, it will usually while utilizing manual feature and CNN feature
A variety of difference characteristics such as it is described, and merges texture, movement.But the method based on texture analysis needs high-resolution defeated
Enter image and preferable acquisition quality to obtain more accurately grain details；Compared to being only on the defensive by texture, move
State clue has certain effectiveness for printing attack and static video playback attack, but due to needing the height of user to cooperate
To complete verifying, user experience is bad and deception is at low cost；Biopsy method based on texture, motion information under visible light
It is more sensitive to illumination, it is easily affected by it, is according to skin and other materials in spectrum based on multispectral biopsy method
Difference on reflectivity determines true and false face, to illumination-insensitive, can resisting various attacks, but such method is for pick-up slip
Part is stringenter, and equipment requirement is higher；Biopsy method based on multi-feature fusion can often form complementation in mechanism, mention
High detection rate, when performance more single method detection, are significantly increased, but it is longer to be faced with average handling time simultaneously, to hard
The more demanding problem of part.
As described above, researchers propose a large amount of face In vivo detection technical method, these are to a certain extent all
A possibility that face identification system is invaded by illegal user, but these methods or the height master for needing tester can be reduced
Cooperation is seen to complete to detect, or needs the supply of additional high cost complex device, needed for not being able to satisfy in face In vivo detection
Low cost, the detection demand of high stability.
Summary of the invention
The purpose of the present invention is deficiencies to solve above-mentioned technical problem, provide a kind of based on extensive RGB and close red
The double authentication face method for anti-counterfeit of outer data set, this method no user interactivity requirements, without the optional equipment of high-end complexity,
The aid of double authentication more makes it be really achieved the detection demand of Face datection.
The deficiency of the present invention to solve above-mentioned technical problem, used technical solution is: one kind based on extensive RGB with
And the double authentication face method for anti-counterfeit of near-infrared data set, including a major punishment determines and double judgement, if a major punishment is set to very,
It is fixed to carry out the second major punishment, specifically:
One major punishment is fixed:
1.1, the acquisition and pretreatment of training dataset；
1.2, the textural characteristics of RGB image are extracted using the operator of local binary pattern；
1.3, the textural characteristics input classifier of extraction is subjected to an anti-fake judgement of weight；
Double judgement:
2.1, the global characteristics of near-infrared data set are extracted；
2.2, the Local textural feature of extraction and global characteristics are inputted into improved convolutional network and carries out Fusion Features；
2.3, the feature of fusion is inputted into improved CNN network classified and determine the true and false of face.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: data acquire in the step 1.1 method particularly includes: are acquired using RealSense depth camera device
A large amount of living body and non-live volumetric video RGB and near-infrared data.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: data prediction in the step 1.1 method particularly includes: RGB and near-infrared data are labeled, it is obtained
Coordinate value.Marked content specifically includes 5 key points of face frame and face i.e. left eye, right eye, nose, the left corners of the mouth, the right corners of the mouth.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the step 1.2 method particularly includes:
1) piecemeal processing, is carried out to image；
2) LBP-TOP feature extraction, is carried out respectively to each region after piecemeal.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the step 2) method particularly includes:
1) characteristics extraction, is carried out using equivalent formulations LBP；
Equivalent formulations LBP characteristic value calculates:
Wherein, P indicates the number of field pixel；R is the radius in field；icIt is gray value；ipIt is the gray value of adjacent pixel；s
It is sign function；
2) it, calculates the histogram in each region and is normalized；
3) statistic histogram for, connecting each region synthesizes a feature vector.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the step 2.2 method particularly includes:
1), feature is weighted using improved network structure；
2), by after weighting local feature and global characteristics merge.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the improved network structure includes ＂ Squeeze-and-Excitation ＂ network block, ＂ Squeeze-and-
Excitation ＂ is made of the part Squeeze and the part Excitation.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the part Squeeze is specially the characteristic pattern for polymerizeing characteristic pattern and obtaining that dimension is WxH, is operated by pondization
The spatial information of the global receptive field of each characteristic pattern is placed in characteristic pattern, and subsequent network layer can be according to this
Characteristic pattern obtains the information of global receptive field, is worth the information of spatially all the points all average out to one using following formula；
Wherein, the global characteristics output obtained by previous step convolution is exactly U, or referred to as C size is H × W's
Featuremap, UCIndicate that the C tensor in U, subscript c indicate channel.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: the part Excitation specifically: first reduce characteristic dimension, then pass through one again after Relu is activated
Full articulamentum liter returns to original dimension, then normalized weight between 0-1 is obtained by sigmoid, by the weight of output
The importance for regarding each feature channel after feature selecting as, by multiplication by channel weighting to previous feature,
Complete the recalibration to primitive character on channel dimension, the feature after being weighted；
Fex(z, W)=σ (w2δ(w1z))
Wherein, Squeeze obtain the result is that z, first uses w here1It is exactly a full connection layer operation, w multiplied by z1Dimension be
C/r*C, r is scaling here, and the dimension of z is 1 × 1 × C, w1The result of z is exactly 1 × 1 × C/r；Then using one
Relu layers, the dimension of output is constant, then again and w2Multiplication and w2Be multiplied is also the process of a full articulamentum, w2Dimension be
C*C/r, therefore the dimension exported is exactly 1 × 1 × C finally using sigmoid function, is obtained for portraying in tensor U c
The weight s of featuremap.
As a kind of present invention double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set into
One-step optimization: Fusion Features not only cover the global characteristics of near-infrared data, further include the textural characteristics of RGB data.
Beneficial effect method of the invention is based on RGB and near-infrared data set and constructs big data quantity, multi-modal
Data set, using LBP-TOP Texture descriptor and convolutional network and improved network module, it can be achieved that double authentication face is anti-
Puppet, face method for anti-counterfeit no user interactivity requirements of the invention, without the optional equipment of high-end complexity, the aid of double authentication
It is more set to be really achieved the low cost of Face datection, high-timeliness detects demand.
Detailed description of the invention
Fig. 1 is the flow chart of the double authentication face method for anti-counterfeit of the embodiment of the present invention:
Fig. 2 is that near-infrared data set global characteristics extract convolution used in the double authentication face method for anti-counterfeit of the embodiment of the present invention
The structure chart of network；
Fig. 3 is that improved convolutional network includes newly-increased module ＂ in the double authentication face method for anti-counterfeit of the embodiment of the present invention
The structure chart of Squeeze-and-Excitation ＂ (in dotted line frame).
Specific embodiment
Further technical solution of the present invention is illustrated below in conjunction with specific embodiment.
A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set, includes the following steps
One, which weighs anti-fake judgement, includes:
1.1, the acquisition and pretreatment of training dataset；
1), data acquire: acquired using RealSense depth camera device a large amount of living body and non-live volumetric video RGB with
Near-infrared data.Wherein the collection process of non-live volume data is mainly to the high definition human face photo of printing and the face video of shooting
Reproduction is carried out, every section of video capture is set as continuous 1000 frame picture, and generates its label, and true and false face is respectively set to 0
With 1；
2), data prediction: mainly data are labeled using MTCNN tool, in order to enable MTCNN more accurately to mark
5 key points and face frame of near-infrared face, obtain its coordinate value, we have carried out key to 10000 near-infrared pictures
The manual mark of point and face frame.Then MTCNN network is trained as training set, obtaining can be to near-infrared number
According to the model being labeled.It is last that affine transformation is carried out to image according to the face frame detected and key point, make facial image
Alignment is in same position.
2.2, the textural characteristics of RGB image are extracted using the operator of local binary pattern；
LBP-TOP is extension of the LBP from two-dimensional space to three-dimensional space, and LBP descriptor is often used to state between true and false face
Texture difference is attacked to achieve the purpose that detect, and is mainly used for single frames picture.And LBP-TOP is then from video and image sequence
It sets out, it is also contemplated that temporal characteristics give important dynamic texture information, extract the line of XY, YT, XT tri- orthogonal planes
Information is managed, wherein T is time shaft (frame sequence)；
1) piecemeal processing first, is carried out to image, window is will test and is divided into the zonule 4*4；
2) LBP-TOP feature extraction, is carried out respectively to these regions；Since LBP-TOP is high dimensional feature, in three orthogonal planes
When extracting LBP using equivalent formulations LBP, that is, uniform code coding (the circulation binary number corresponding to some LBP from 0 to
1 or when be up to jumping side twice from 1 to 0, binary system corresponding to the LBP is thus referred to as an equivalent formulations class) dimensionality reduction is carried out,
Each region drops to 59 dimensions in practical application from 256 original dimensions.
Equivalent formulations LBP characteristic value calculates:
Wherein, P indicates the number of field pixel；R is the radius in field；icIt is gray value；ipIt is the gray value of adjacent pixel；s
It is sign function；
3) histogram in each region, i.e., the frequency that each number occurs, are calculated；Then the histogram is normalized
(counting identical number in each section of histogram, be normalized divided by histogram pixel sum)；
4), the statistic histogram in obtained each region is attached as a feature vector, that is, whole picture figure
LBP-TOP texture feature vector.
Step 3: the textural characteristics input classifier of extraction is subjected to an anti-fake judgement of weight；
The normalized feature vector of training set is sent into SVM SVM classifier, by parameter testing come Optimum Classification device
Performance trains data model；
By the judgement before based on LBP-TOP textural characteristics descriptor, we have preliminary judgement to true and false face, when sentencing
When being set to false face, it is classified as false one kind；When being determined as true, double judgement is carried out, steps are as follows:
2.1, the extraction of near-infrared data set global characteristics；
The facial area of cutting is adjusted to 112 × 112 size, and is floated using random, rotation is sized, cut and
Cross-color carries out data enhancing.Pretreated near-infrared image input convolutional network is carried out to the extraction of global characteristics,
By 2,000 repetitive exercises, initial learning rate is 0.1, and learning rate reduces 10 times after every 1000 iteration.Use stochastic gradient
Decline SGD algorithm optimizes, and convolutional network structure is as shown in Figure 2.
Herein using resnet18 as major networks framework, conventional part uses 3 block, and each block includes convolution
Layer, BatchNorm layers, avg-pooling layers, Dropout layers, the structures such as full articulamentum.It is by the stacking of 3*3 convolution come real
The extraction of existing information, obtains global characteristics.
Step 2: the Local textural feature of extraction and global characteristics being inputted into improved convolutional network and carry out Fusion Features；
As shown in figure 3, being added to new network structure elements, referred to as ＂ Squeeze-and- in convolutional network herein
Excitation ＂ network block (in dotted line frame).It is mainly made of the part Squeeze and the part Excitation, can allow net
Network promotes accuracy rate by strengthening important feature from global information.First half in figure is traditional convolution knot
Structure, the part newly increased are the parts after convolutional network output.One GAP (GlobalAveragePooling) is first done to output
That is then the data of output are connected i.e. Excitation process, after output obtains weighting by Squeeze process entirely using two-stage
Feature.This structure makes network carry out the selective valuable feature of amplification from global information and inhibits useless spy
Sign, to make the feature directive property of extraction stronger.Detail section is as follows:
The part Squeeze:
Due to each convolution kernel be carried out in a manner of a local receptor field convolution therefore export each data cell not
The texture information other than data cell can be utilized.It, will be entire in a channel characteristics figure in order to be resolved this problem
The information of position blends in figure, so that assessment is more acurrate.This process is realized by the average pond layer of the overall situation, by characteristic pattern
Polymerization obtains the characteristic pattern that dimension is WxH, is placed in the spatial information of the global receptive field of each characteristic pattern by pondization operation
Into characteristic pattern, and subsequent network layer can obtain the information of global receptive field according to this characteristic pattern.With averaging
Method (formula 1), the information of spatially all the points is all average at a value.
Wherein, the global characteristics output obtained by previous step convolution is exactly U, or referred to as C size is H × W's
Featuremap, UCIndicate that the C tensor in U, subscript c indicate channel.
The part Excitation:
The correlation of modeling interchannel is made of two full articulamentums, and is exported and the same number of weight of input feature vector.It is first
Characteristic dimension is first reduced to 1/16 (experiment show use 16 appropriate as compression factor) of input, is then swashed by Relu
Original dimension is returned to by a full articulamentum liter again after work.In this way than being directly advantageous in that it has with a full articulamentum
Have more non-linear, can preferably be fitted the correlation of interchannel complexity；Additionally greatly reduce parameter amount and meter
Calculation amount.Then normalized weight between 0-1 is obtained by sigmoid, the weight of output is regarded as after feature selecting
The importance in each feature channel complete pair on channel dimension by multiplication by channel weighting to previous feature
The recalibration of primitive character, the feature after being weighted.
Fex(z, W)=σ (w2δ(w1z))
Wherein, Squeeze obtain the result is that z, first uses w here1It is exactly a full connection layer operation, w multiplied by z1Dimension
Degree is C/r*C, and r is scaling here, and the dimension of z is 1 × 1 × C, w1The result of z is exactly 1 × 1 × C/r；Then using
One Relu layers, the dimension of output is constant, then again and w2Multiplication and w2Be multiplied is also the process of a full articulamentum, w2Dimension
Degree is C*C/r, therefore the dimension exported is exactly 1 × 1 × C finally using sigmoid function, is obtained for portraying c in tensor U
The weight s of a feature map.
Step 3: the feature of fusion being inputted into improved CNN network classified and determine the true and false of face.
The feature weighted again is connected with LBP-TOP feature before and carries out Fusion Features, by convolutional layer from fusion
Learn more distinguishing characteristics in characteristic block.It connect the weight that layer operation obtains various modal characteristics entirely with two using GAP,
Last classification is carried out, obtains determining result.
Face In vivo detection is realized by double anti-forge certification as a result, has the function that distinguish true and false face.To sum up institute
It states, present embodiment discloses a kind of double authentication face method for anti-counterfeit based on extensive near-infrared data set.This method can
Face anti-counterfeiting detection recognition accuracy is improved under no user interactivity, without the optional equipment of high-end complexity.
The above described is only a preferred embodiment of the present invention, be not intended to limit the present invention in any form, though
So the present invention has been disclosed as a preferred embodiment, and however, it is not intended to limit the invention, any technology people for being familiar with this profession
Member, without departing from the scope of the present invention, when the technology contents using the disclosure above are modified or are modified
It is right according to the technical essence of the invention for the equivalent embodiment of equivalent variations, but without departing from the technical solutions of the present invention
Any simple modification, equivalent change and modification made by above embodiments, all of which are still within the scope of the technical scheme of the invention.
Claims (10)
1. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set, it is characterised in that: including
One major punishment is fixed and double judgement carries out the second major punishment and determine if a major punishment is set to really, specifically:
One major punishment is fixed:
1.1, the acquisition and pretreatment of training dataset；
1.2, the textural characteristics of RGB image are extracted using the operator of local binary pattern；
1.3, the textural characteristics input classifier of extraction is subjected to an anti-fake judgement of weight；
Double judgement:
2.1, the global characteristics of near-infrared data set are extracted；
2.2, the Local textural feature of extraction and global characteristics are inputted into improved convolutional network and carries out Fusion Features；
2.3, the feature of fusion is inputted into improved CNN network classified and determine the true and false of face.
2. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as described in claim 1,
It is characterized by: data acquisition in the step 1.1 method particularly includes: adopted using RealSense depth camera device
Collect a large amount of living body and non-live volumetric video RGB and near-infrared data.
3. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as described in claim 1,
It is characterized by: data prediction in the step 1.1 method particularly includes: be labeled, obtain to RGB and near-infrared data
Its coordinate value.Marked content specifically includes 5 key points of face frame and face i.e. left eye, right eye, nose, the left corners of the mouth, the right corners of the mouth.
4. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as described in claim 1,
It is characterized by: the step 1.2 method particularly includes:
1) piecemeal processing, is carried out to image；
2) LBP-TOP feature extraction, is carried out respectively to each region after piecemeal.
5. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as claimed in claim 4,
It is characterized by: the step 2) method particularly includes:
1) characteristics extraction, is carried out using equivalent formulations LBP；
Equivalent formulations LBP characteristic value calculates:
Wherein, P indicates the number of field pixel；R is the radius in field；icIt is gray value；ipIt is the gray value of adjacent pixel；S is
Sign function；
2) it, calculates the histogram in each region and is normalized；
3) statistic histogram for, connecting each region synthesizes a feature vector.
6. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as described in claim 1,
It is characterized by: the step 2.2 method particularly includes:
1), feature is weighted using improved network structure；
2), by after weighting local feature and global characteristics merge.
7. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as described in claim 1,
It is characterized by: the improved network structure includes ＂ Squeeze-and-Excitation ＂ network block, ＂ Squeeze-and-
Excitation ＂ is made of the part Squeeze and the part Excitation.
8. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as claimed in claim 7,
It is characterized by: the part Squeeze is specially the characteristic pattern for polymerizeing characteristic pattern and obtaining that dimension is WxH, by Chi Huacao
The spatial information of the global receptive field of each characteristic pattern is placed in characteristic pattern by work, and subsequent network layer can be according to this
A characteristic pattern obtains the information of global receptive field, is worth the information of spatially all the points all average out to one using following formula；
Wherein, the global characteristics output obtained by previous step convolution is exactly U, or the feature that referred to as C size is H × W
Map, UCIndicate that the C tensor in U, subscript c indicate channel.
9. a kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set as claimed in claim 7,
It is characterized by: the part Excitation specifically: first reduce characteristic dimension, then pass through one again after Relu is activated
A full articulamentum liter returns to original dimension, then normalized weight between 0-1 is obtained by sigmoid, by the power of output
The importance for regarding each feature channel after feature selecting as again, by multiplication by channel weighting to previous feature
On, complete the recalibration to primitive character on channel dimension, the feature after being weighted；
Fex(z, W)=σ (w2δ(w1z))
Wherein, Squeeze obtain the result is that z, first uses w here1It is exactly a full connection layer operation, w multiplied by z1Dimension be
C/r*C, r is scaling here, and the dimension of z is 1 × 1 × C, w1The result of z is exactly 1 × 1 × C/r；Then using one
Relu layers, the dimension of output is constant, then again and w2Multiplication and w2Be multiplied is also the process of a full articulamentum, w2Dimension be
C*C/r, therefore the dimension exported is exactly 1 × 1 × C finally using sigmoid function, is obtained for portraying in tensor U c
The weight s of feature map.
10. a kind of anti-fake side of double authentication face based on extensive RGB and near-infrared data set as claimed in claim 6
Method, it is characterised in that: Fusion Features not only cover the global characteristics of near-infrared data, further include the textural characteristics of RGB data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910809299.8A CN110516616A (en) | 2019-08-29 | 2019-08-29 | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910809299.8A CN110516616A (en) | 2019-08-29 | 2019-08-29 | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110516616A true CN110516616A (en) | 2019-11-29 |
Family
ID=68629121
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910809299.8A Pending CN110516616A (en) | 2019-08-29 | 2019-08-29 | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110516616A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259814A (en) * | 2020-01-17 | 2020-06-09 | 杭州涂鸦信息技术有限公司 | Living body detection method and system |
| CN111460419A (en) * | 2020-03-31 | 2020-07-28 | 周亚琴 | Internet of things artificial intelligence face verification method and Internet of things cloud server |
| CN111488856A (en) * | 2020-04-28 | 2020-08-04 | 江西吉为科技有限公司 | Multimode 2D and 3D facial expression recognition based on orthogonal guide learning |
| CN111539311A (en) * | 2020-04-21 | 2020-08-14 | 上海锘科智能科技有限公司 | Living body distinguishing method, device and system based on IR and RGB double photographing |
| CN111666901A (en) * | 2020-06-09 | 2020-09-15 | 创新奇智(北京)科技有限公司 | Living body face detection method and device, electronic equipment and storage medium |
| CN112084946A (en) * | 2020-05-09 | 2020-12-15 | 支付宝(杭州)信息技术有限公司 | Face recognition method and device and electronic equipment |
| CN112200075A (en) * | 2020-10-09 | 2021-01-08 | 西安西图之光智能科技有限公司 | Face anti-counterfeiting method based on anomaly detection |
| CN112434647A (en) * | 2020-12-09 | 2021-03-02 | 浙江光珀智能科技有限公司 | Human face living body detection method |
| CN112464741A (en) * | 2020-11-05 | 2021-03-09 | 马上消费金融股份有限公司 | Face classification method, model training method, electronic device and storage medium |
| CN112528956A (en) * | 2020-12-28 | 2021-03-19 | 佛山科学技术学院 | Face feature extraction method, system, device and medium based on feature recalibration |
| CN112818722A (en) * | 2019-11-15 | 2021-05-18 | 上海大学 | Modular dynamically configurable living body face recognition system |
| CN113542142A (en) * | 2020-04-14 | 2021-10-22 | 中国移动通信集团浙江有限公司 | Portrait anti-counterfeiting detection method and device and computing equipment |
| CN113536843A (en) * | 2020-04-16 | 2021-10-22 | 上海大学 | Anti-counterfeiting face recognition system based on multi-mode fusion convolutional neural network |
| WO2021237968A1 (en) * | 2020-05-28 | 2021-12-02 | 上海依图网络科技有限公司 | Living body detection method and apparatus applied to blocked face scene |
-
2019
- 2019-08-29 CN CN201910809299.8A patent/CN110516616A/en active Pending
Non-Patent Citations (3)
| Title |
|---|
| GUOYING ZHAO等: "Dynamic Texture Recognition Using Local Binary Patterns with an Application to Facial Expressions", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| JIE HU等: "Squeeze-and-Excitation Networks", 《ARCHITECARXIV:1709.01507V4 [CS.CV]》 * |
| 董吉祥: "人脸活体检测算法研究与实现", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112818722A (en) * | 2019-11-15 | 2021-05-18 | 上海大学 | Modular dynamically configurable living body face recognition system |
| CN112818722B (en) * | 2019-11-15 | 2023-08-18 | 上海大学 | Modular dynamic configurable living body face recognition system |
| CN111259814A (en) * | 2020-01-17 | 2020-06-09 | 杭州涂鸦信息技术有限公司 | Living body detection method and system |
| CN111259814B (en) * | 2020-01-17 | 2023-10-31 | 杭州涂鸦信息技术有限公司 | Living body detection method and system |
| CN111460419A (en) * | 2020-03-31 | 2020-07-28 | 周亚琴 | Internet of things artificial intelligence face verification method and Internet of things cloud server |
| CN111460419B (en) * | 2020-03-31 | 2020-11-27 | 深圳市微网力合信息技术有限公司 | Internet of things artificial intelligence face verification method and Internet of things cloud server |
| CN113542142B (en) * | 2020-04-14 | 2024-03-22 | 中国移动通信集团浙江有限公司 | Portrait anti-fake detection method and device and computing equipment |
| CN113542142A (en) * | 2020-04-14 | 2021-10-22 | 中国移动通信集团浙江有限公司 | Portrait anti-counterfeiting detection method and device and computing equipment |
| CN113536843B (en) * | 2020-04-16 | 2023-07-14 | 上海大学 | Anti-fake face recognition system based on multimode fusion convolutional neural network |
| CN113536843A (en) * | 2020-04-16 | 2021-10-22 | 上海大学 | Anti-counterfeiting face recognition system based on multi-mode fusion convolutional neural network |
| CN111539311A (en) * | 2020-04-21 | 2020-08-14 | 上海锘科智能科技有限公司 | Living body distinguishing method, device and system based on IR and RGB double photographing |
| CN111539311B (en) * | 2020-04-21 | 2024-03-01 | 上海锘科智能科技有限公司 | Living body judging method, device and system based on IR and RGB double shooting |
| CN111488856A (en) * | 2020-04-28 | 2020-08-04 | 江西吉为科技有限公司 | Multimode 2D and 3D facial expression recognition based on orthogonal guide learning |
| CN111488856B (en) * | 2020-04-28 | 2023-04-18 | 江西吉为科技有限公司 | Multimodal 2D and 3D facial expression recognition method based on orthogonal guide learning |
| CN112084946A (en) * | 2020-05-09 | 2020-12-15 | 支付宝(杭州)信息技术有限公司 | Face recognition method and device and electronic equipment |
| WO2021237968A1 (en) * | 2020-05-28 | 2021-12-02 | 上海依图网络科技有限公司 | Living body detection method and apparatus applied to blocked face scene |
| CN111666901A (en) * | 2020-06-09 | 2020-09-15 | 创新奇智(北京)科技有限公司 | Living body face detection method and device, electronic equipment and storage medium |
| CN112200075A (en) * | 2020-10-09 | 2021-01-08 | 西安西图之光智能科技有限公司 | Face anti-counterfeiting method based on anomaly detection |
| CN112200075B (en) * | 2020-10-09 | 2024-06-04 | 西安西图之光智能科技有限公司 | Human face anti-counterfeiting method based on anomaly detection |
| CN112464741B (en) * | 2020-11-05 | 2021-11-26 | 马上消费金融股份有限公司 | Face classification method, model training method, electronic device and storage medium |
| CN112464741A (en) * | 2020-11-05 | 2021-03-09 | 马上消费金融股份有限公司 | Face classification method, model training method, electronic device and storage medium |
| CN112434647A (en) * | 2020-12-09 | 2021-03-02 | 浙江光珀智能科技有限公司 | Human face living body detection method |
| CN112528956A (en) * | 2020-12-28 | 2021-03-19 | 佛山科学技术学院 | Face feature extraction method, system, device and medium based on feature recalibration |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110516616A (en) | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set | |
| CN109255322B (en) | A kind of human face in-vivo detection method and device | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| Ramachandra et al. | Towards making morphing attack detection robust using hybrid scale-space colour texture features | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| CN110349136A (en) | A kind of tampered image detection method based on deep learning | |
| CN101923640B (en) | Method for distinguishing false iris images based on robust texture features and machine learning | |
| CN109858439A (en) | A kind of biopsy method and device based on face | |
| CN103605958A (en) | Living body human face detection method based on gray scale symbiosis matrixes and wavelet analysis | |
| JP2020525947A (en) | Manipulated image detection | |
| CN105354554A (en) | Color and singular value feature-based face in-vivo detection method | |
| CN113221655B (en) | Face spoofing detection method based on feature space constraint | |
| CN114067444A (en) | Face spoofing detection method and system based on meta-pseudo label and illumination invariant feature | |
| CN113537027B (en) | Face depth counterfeiting detection method and system based on face division | |
| CN107145841A (en) | A kind of low-rank sparse face identification method and its system based on matrix | |
| Liu et al. | Detecting presentation attacks from 3d face masks under multispectral imaging | |
| Yeh et al. | Face liveness detection based on perceptual image quality assessment features with multi-scale analysis | |
| CN106709458A (en) | Human face living body detection method and device | |
| Zhao et al. | A transferable anti-forensic attack on forensic CNNs using a generative adversarial network | |
| Jingade et al. | DOG-ADTCP: A new feature descriptor for protection of face identification system | |
| CN114842524A (en) | Face false distinguishing method based on irregular significant pixel cluster | |
| Sakthimohan et al. | Detection and Recognition of Face Using Deep Learning | |
| CN113642003A (en) | Safety detection method of face recognition system based on high-robustness confrontation sample generation | |
| CN113468954B (en) | Face counterfeiting detection method based on local area features under multiple channels |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20211210 Address after: 712000 room 10201, building 4a, West Yungu phase II, Fengxi new town, Xixian new area, Xianyang City, Shaanxi Province Applicant after: Shaanxi Xitu Digital Technology Co.,Ltd. Address before: 471000 Room 201, building 1, Chuangzhi Plaza, No. 32, changxiamen street, Luolong District, Luoyang City, Henan Province Applicant before: Henan Zhongyuan big data Research Institute Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191129 |