CN110334698A - Glasses detection system and method - Google Patents
Glasses detection system and method Download PDFInfo
- Publication number
- CN110334698A CN110334698A CN201910811078.4A CN201910811078A CN110334698A CN 110334698 A CN110334698 A CN 110334698A CN 201910811078 A CN201910811078 A CN 201910811078A CN 110334698 A CN110334698 A CN 110334698A
- Authority
- CN
- China
- Prior art keywords
- face
- glasses
- image
- module
- face image
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/18—Eye characteristics, e.g. of the iris
- G06V40/19—Sensors therefor
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Geometry (AREA)
- Probability & Statistics with Applications (AREA)
- Ophthalmology & Optometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Glasses detection system and methods, and for detecting whether user wears glasses, wherein eyeglass detection method is the following steps are included: a. image capture module acquires the facial image of user;B. half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half-face image, and half-face image is respectively stored into half face raw data base of glasses-free and half face raw data base of wearing glasses;C. half face rebuilds module and is based on the half face raw data base of glasses-free progress principal component analysis, obtain half face reconstruction model, and half-face image is rebuild by the half face reconstruction model, the half-face image after reconstruction is respectively stored into half face of glasses-free and rebuilds database and half face reconstruction database of wearing glasses;D. glasses discrimination module is based on confrontation sorter network and carries out network training to database, obtains glasses discrimination model;And differentiated by the glasses discrimination model to what whether the half-face image of input wore glasses.
Description
Technical field
The present invention relates to technical field of biometric identification, and in particular to a kind of Glasses detection system and method.
Background technique
Due to myopia population substantial amounts, and most of all wearing spectacles, in recognition of face, blocking for glasses be will affect
The precision of recognition of face;And in iris recognition, influence more very, therefore face recognition technology and iris recognition sound at present
Technology will usually be related to Glasses detection technology.Iris identification equipment mostly illuminates iris region using infrared polishing, then leads to
Cross the infrared imaging that acquisition is returned from iris reflex.If user matches wear a pair of spectacles, eyeglass can generate strong reflection to infrared light, that
The iris image captured serious will be interfered by hot spot, shown as iris texture and watered down or invisible.In addition iris
Identification equipment is all made of high definition probe, can clearly capture the stain on glasses, stain can be superimposed upon on iris texture
Iris image is formed and is polluted.Once contaminated iris image is added to register list, it may appear that two kinds of risks: risk
One, registration user is rejected when identifying next time because that cannot register due to iris templates match with it;Risk two, due to registration
Hot spot and stain are superimposed in iris templates, it is more likely that unlock is attacked by the iris under the conditions of other same, shows as registering
User A is with the identity logs or nonregistered user of another one registration user B to register user identity login.Above risk is raw
Strict demand is evaded in object identification technology.The method for evading the above risk is to notify user to cooperate by Glasses detection technology
It removes glasses completion registration and the user's head portrait or iris image for not allowing to wear glasses is mixed into user registry database.In face
Or during iris recognition, many developers allow user to wear glasses identification, but still have to cooperate the Experience Degree of user
By the possibility of rejection, user is reminded to remove glasses by Glasses detection technology at this time, Experience Degree can be more preferable for users.
At present Glasses detection technology be mainly pass through principal component analysis method rebuild facial image, then obtain original image with
The error image of reconstruction image, and frame is obtained by the method for binaryzation error image.The difficulty of this method be first to
The facial image of detection has the difference worn glasses and do not worn glasses, regardless of whether wear glasses, facial image and its reconstruction image
Between certainly exist difference, and this species diversity is derived from the otherness of glasses or face also or the error of image reconstruction is difficult really
It is fixed, thus judge whether that there are many disturbing factor worn glasses by error image;Secondly glasses have color difference (to show on image
For the difference of gray scale), black surround glasses, it is greatly poor that white edge glasses have with the grey value profile that transparent glasses show on the image
It is different, and then the great difference of its error image will be caused, then the threshold value of stable binaryzation error image can be very difficult to
It determines, is likely to not divide spectacle-frame and Glasses detection is caused to fail if the threshold value of binaryzation is excessively high, if binaryzation
Too low other textures for being likely to be partitioned on face of threshold value and accidentally know for glasses;Thus inaccurate threshold value will directly affect
To the precision of Glasses detection.To which the threshold value that this method is difficult to obtain stalwartness carrys out binaryzation spectacle-frame, the presence of spectacle-frame is judged.
Summary of the invention
It, can be with technical problem to be solved by the invention is to provide a kind of quickly and effectively Glasses detection system and method
Accurate detection user whether wearing spectacles, improve recognition of face and the registration of iris recognition and recognition efficiency.In view of many rainbows
Film identifying system may not be able to capture complete facial image in the application, in order to expand Glasses detection systematic difference model
It enclosing, the present invention uses face half-face image, i.e., and it is left to left eyebrow tail down toward nose tabula lower edge up to eyebrow peak in face, it is right to right eyebrow
The region of tail.
In order to realize the above technical effect, the invention discloses a kind of Glasses detection systems, for detecting whether user wears
Glasses comprising:
Image capture module, for acquiring facial image;
Half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half-face image,
Wherein, the half-face image that do not wear glasses stores the half-face image worn glasses to half face raw data base of glasses-free and stores to hyperphoria with fixed eyeballs
Half face raw data base of mirror;
Half face rebuilds module, carries out principal component based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
It analyzes (PCA), obtains half face reconstruction model, and rebuild to half-face image by half face reconstruction model, wherein do not wear glasses
Half-face image rebuild after storage to half face of glasses-free rebuild database, storage is to wearing glasses after the half-face image worn glasses is rebuild
Half face rebuilds database;
Glasses discrimination module, based on confrontation sorter network to the half face raw data base of glasses-free, half face original number of wearing glasses
The half-face image progress network training that database, half face of wearing glasses reconstruction database are rebuild according to library, half face of glasses-free, obtains glasses
Discrimination model;And the differentiation worn glasses is made whether to half-face image by glasses discrimination model.
The improvement of Glasses detection system of the present invention is, further includes a data update module, the data update module root
Half face raw data base of glasses-free, half face raw data base of wearing glasses are updated according to the differentiation result of glasses discrimination module.
Invention additionally discloses there is a kind of eyeglass detection method, for detecting whether user wears glasses comprising following step
It is rapid:
A. the facial image of image capture module acquisition user;
B. half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half face figure
Picture, and the half-face image that do not wear glasses is stored to half face raw data base of glasses-free, the half-face image storage worn glasses is arrived
It wears glasses half face raw data base;
C. half face rebuild module based on the half-face image that do not wear glasses in the half face raw data base of glasses-free carry out it is main at
Analysis (PCA), obtains half face reconstruction model, and rebuild to half-face image by the half face reconstruction model, will not hyperphoria with fixed eyeballs
The half-face image of mirror stores to half face of glasses-free after rebuilding and rebuilds database, and hyperphoria with fixed eyeballs is arrived in storage after the half-face image worn glasses is rebuild
Half face of mirror rebuilds database;
D. it is original to the half face raw data base of glasses-free, half face of wearing glasses to be based on confrontation sorter network for glasses discrimination module
Database, half face of glasses-free rebuild database, the half-face image of half face of wearing glasses reconstruction database carries out network training, obtain eye
Mirror discrimination model;And differentiated by the glasses discrimination model to whether the half-face image of input wears glasses.
The improvement of eyeglass detection method of the present invention is that it includes the first study module and reconstruction mould that half face, which rebuilds module,
Block, step c further comprises:
First study module carries out principal component based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
It analyzes (PCA), obtains half face reconstruction model, and the half face reconstruction model is published to and is rebuild in module;
It rebuilds module to rebuild half-face image by the half face reconstruction model, and the half-face image that do not wear glasses is rebuild
Storage rebuilds database to half face of glasses-free afterwards, and storage rebuilds data to half face of wearing glasses after the half-face image worn glasses is rebuild
Library.
Eyeglass detection method of the present invention further improvement lies in that, the glasses discrimination module include the second study module and
Discrimination module, step d further comprises:
Second study module is based on confrontation sorter network to the half face raw data base of glasses-free, half face initial data of wearing glasses
Library, half face of glasses-free rebuild database, the half-face image of half face of wearing glasses reconstruction database carries out network training, obtain glasses and sentence
Other model, and the glasses discrimination model is published in discrimination module;
Discrimination module differentiates by the way that whether half-face image of the glasses discrimination model to input wears glasses.
Eyeglass detection method of the present invention further improvement lies in that, further include step e:
Data update module is according to the differentiation result of glasses discrimination module to half face raw data base of glasses-free, half face original of wearing glasses
Beginning database is updated.
Eyeglass detection method of the present invention further improvement lies in that, the half-face image extraction module include detection module,
Rectification module and normalization module, step b further comprises:
Detection module detects face in the facial image of input, and detects the position where face key point;
Rectification module will test the face key point that module detects and compare with normalized standard faces model, based on pair
The face in facial image is corrected than use of information perspective transform, forms standard faces image, and will test module inspection
The face key point measured is mapped on the standard faces image corrected;
Module is normalized according to the position coordinates of the face key point being mapped on the standard faces image corrected by half face
The image cut of position comes out and normalizes to fixed dimension, forms half-face image, wherein the half-face image storage that do not wear glasses is arrived
Half face raw data base of glasses-free, the half-face image storage worn glasses to half face raw data base of wearing glasses.
Eyeglass detection method of the present invention further improvement lies in that, half-face image is in facial image up to eyebrow peak, down toward
Nose tabula lower edge, it is left to left eyebrow tail, it is right to the region of right eyebrow tail.
Eyeglass detection method of the present invention further improvement lies in that, the second study module is classified net for training confrontation
The classification of the complete paired samples of network, confrontation sorter network are formed based on half-face image and the half-face image through more than half face reconstruction models
The difference characteristic for rebuilding half-face image is completed to half face raw data base of glasses-free, wear glasses half face raw data base, glasses-free
Half face rebuilds database, half face of wearing glasses rebuilds the classification of half-face image in database.
Glasses detection system and method for the present invention obtains half face reconstruction model by principal component analysis (PCA), then based on original
The difference for the reconstruction half-face image that beginning half-face image and the original half-face image are rebuild through more than half face reconstruction models, in conjunction with depth
The method of study is spent to judge whether user wears glasses.Glasses detection system and method prediction accuracy of the present invention is higher, and
Dynamic data update module can dynamic more new database, so that glasses discrimination model is constantly updated, generalization ability is stronger.
Detailed description of the invention
Fig. 1 is the schematic diagram of Glasses detection system of the present invention;
Fig. 2 is the schematic diagram of database module in Glasses detection system of the present invention;
Fig. 3 is the flow diagram that half-face image extraction module of the present invention extracts half-face image;
Fig. 4 is the schematic diagram of normalized standard faces model in the present invention;
Fig. 5-7 is the schematic diagram of facial image in a preferred embodiment of the present invention half-face image extraction module extraction process;
Fig. 8 is the schematic diagram that half face of the invention rebuilds the first study module in module;
Fig. 9 is the flow diagram that half-face image of the present invention is rebuild;
Figure 10 is that a preferred embodiment of the present invention is worn glasses the schematic diagram that half-face image rebuilds through half face reconstruction model;
Figure 11 is the schematic diagram that a preferred embodiment of the present invention glasses-free half-face image is rebuild through half face reconstruction model;
Figure 12 is the schematic diagram of the second study module in glasses discrimination module of the present invention;
Figure 13 is the flow diagram that glasses differentiate in glasses discrimination module of the present invention;
Figure 14 is the class categories that sorter network is fought in glasses discrimination module of the present invention;
Figure 15 is that sorter network structural schematic diagram is fought in glasses discrimination module of the present invention;
Figure 16 is the schematic diagram of data update module of the present invention.
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention is described in further detail.
As shown in Figure 1, a kind of Glasses detection system of the present invention, for detecting whether user wears glasses comprising:
Image capture module, for acquiring facial image, which can directly be passed to the image capture module
Half-face image extraction module can also first be stored in face image database according to the actual situation, to be extracted again by half-face image later
Module handles the facial image in the face image database.Preferably, in the present embodiment, described image acquires mould
Block includes iris camera and infrared lamp.
In conjunction with Fig. 2 to Fig. 7, half-face image extraction module carries out the facial image for inputting the half-face image extraction module
Processing, i.e., half-face image extraction module is corrected the facial image of input, then extracts and normalizes to obtain half-face image,
Wherein, the half-face image that do not wear glasses stores the half-face image worn glasses to half face raw data base of glasses-free and stores to hyperphoria with fixed eyeballs
Half face raw data base of mirror.In the present embodiment, the facial image for inputting half-face image extraction module can be Image Acquisition mould
The directly incoming facial image of block, the facial image being also possible in face image database, this can according to the actual situation voluntarily
Selection, i.e., in the present invention, as long as a facial image of input half-face image extraction module, either image is adopted
Collect that module is collected in real time, is also possible to extract from database.
In the present embodiment, half face refers in face, and up to eyebrow peak, down toward nose tabula lower edge, left to left eyebrow tail, the right side is extremely
The region of right eyebrow tail.Firstly, the ratio of the partial region each organ distribution in different crowds is relatively stable, and will not be because
For expression variation and generate very big denaturation, thus the region is more advantageous to image normalization;Secondly, using the partial region
Can meet the needs of Face datection and iris detection simultaneously.It under normal circumstances, is the clear iris texture of bat, the view of iris capturing system
Wild range is smaller, it may not be possible to obtain complete facial image, can preferably be compatible with iris capturing system, i.e. base using half-face image
Can be not only used for face identification system in the Glasses detection system of half-face image can be used for iris recognition again.
Further, half-face image extraction module includes detection module, rectification module and normalization module.Wherein, it detects
Module detects the position where face key point for detecting face in the facial image of input, as shown in Figure 5.It rectifys
The face key point and normalized standard faces model that positive module is detected by Determination module extrapolate the face figure
When as acquisition, position coordinates and user's head corner in three-dimensional space of user's face relative to image capture device, base
The face rotated in the facial image is corrected into standard faces in information above and using perspective transform, while detection module
The face key point detected is mapped on the standard faces image corrected, as shown in Figure 6.Normalization module is according to being mapped to
The image cut of half face position is come out and is normalized to solid by the position coordinates of the face key point on the facial image corrected
Scale cun forms half-face image, as shown in Figure 7, wherein the half-face image storage that do not wear glasses to half face initial data of glasses-free
Library, the half-face image storage worn glasses to half face raw data base of wearing glasses.
As shown in figure 4, in the present embodiment, face key point refers to the key extracted in facial image based on human face five-sense-organ
Point information, and the image for extracting face key point when standard faces model is then head irrotationality gyration, and being formed through normalization.
As shown in Figs. 8 to 11, half face rebuilds module, based on not wearing glasses in the half face raw data base of glasses-free
Half-face image carry out principal component analysis (PCA), obtain half face reconstruction model, and by half face reconstruction model to half-face image into
Row is rebuild, wherein storage rebuilds database, the half face figure worn glasses to half face of glasses-free after the half-face image that do not wear glasses is rebuild
Database is rebuild as being stored after rebuilding to half face of wearing glasses.In the present embodiment, it half face raw data base of glasses-free and wears glasses
The half-face image of half face raw data base is real human face image, and half face of glasses-free rebuilds database and half face of wearing glasses rebuilds number
Half-face image according to library is to rebuild facial image.The half face raw data base of glasses-free, wear glasses half face raw data base, nothing
Half face of glasses rebuilds database and half face of wearing glasses rebuilds database and constitutes database module.
Further, it includes the first study module and reconstruction module, in the present embodiment, described first that half face, which rebuilds module,
Study module drops the face characteristic of half-face image in half face raw data base of glasses-free using the method for principal component analysis (PCA)
Dimension, extracts the main common feature of different faces, and generate eigenface, these eigenfaces i.e. half face reconstruction model, should
The half face reconstruction model that study is completed, which is published to, to be rebuild in module.The half face reconstruction model does not include any information of glasses.
The reconstruction module rebuilds half-face image by the half face reconstruction model, wherein module is rebuild in input
Half-face image can be by image capture module and half-face image extraction module acquisition extract half-face image, be also possible to
Directly read half face raw data base of glasses-free and the half-face image in half face raw data base of wearing glasses.Half face that do not wear glasses
Image after the reconstruction of half face reconstruction model to half face of glasses-free reconstruction database, rebuild through half face for storage by the half-face image worn glasses
Storage rebuilds database to half face of wearing glasses after Model Reconstruction.
As shown in Figure 12 to Figure 15, glasses discrimination module, based on confrontation sorter network to the half face original number of glasses-free
Database is rebuild according to library, half face raw data base of wearing glasses, half face of glasses-free, half face of wearing glasses rebuilds the half-face image of database
Network training is carried out, glasses discrimination model is obtained;And sentenced by glasses discrimination model to what half-face image was made whether to wear glasses
Not.
Further, glasses discrimination module includes the second study module and discrimination module, wherein the second study module is used for
The classification of one confrontation complete paired samples of sorter network of training.In the present embodiment, confrontation network refers to that the network can be fought
Half face algorithm for reconstructing of PCA distinguishes the half-face image and true half-face image of reconstruction, that is, distinguishes and rebuild half-face image and original
Beginning half-face image;Confrontation sorter network refers to that the network can rebuild half-face image and while original half-face image pair distinguishing
Whether original half-face image, which wears glasses, is completed classification.Purpose using confrontation network is powerful by dual training production one
Classifier can be distinguished to the greatest extent rebuilds half-face image and original half-face image.Rebuild half-face image and original half-face image
It is primarily present following two difference: (1) half-face image rebuild not wear a pair of spectacles;(2) half-face image rebuild using PCA method
Lack the order of information of real human face.Therefore the half-face image rebuild in confrontation network has reference role, for extracting half
The difference characteristic of real human face and reconstruction face in face image, i.e. confrontation network are extracted same based on the half-face image in database
The original half-face image of user and the difference characteristic for rebuilding half-face image.Sorter network is fought based on above difference characteristic completion pair
The classification of half-face image, the classification of classification include wearing glasses, glasses-free and be to rebuild face.Database used in dual training
For half-face image all in database module, i.e. half face raw data base of glasses-free, wear glasses half face raw data base, anophthalmia
Half face of mirror rebuilds database and half face of wearing glasses rebuilds database.Half-face image is corresponding in half face raw data base of glasses-free
Label is 2, is desired for [0,1,0];The corresponding label of half-face image in half face raw data base of wearing glasses is 1, be desired for [1,0,
0];It is 3 that half face of glasses-free, which rebuilds database label corresponding with half-face image in half face reconstruction database of wearing glasses, is desired for
[0,0,1].Loss function is as follows:
WhereinDifferentiate that the loss function of network, y indicate classification for glasses, y=1 indicates the original half-face image worn glasses, y=2
Indicate the original half-face image that do not wear glasses, y=3 indicates to rebuild half-face image.First item in formulaIndicate defeated
Enter image x to meet the distribution of the i.e. original half-face image of true picture and belong to the expectation of y classification,
Indicate that input picture is predicted to be the probability of y classification, y=1 therein or 2 are the original half-face image worn glasses or not hyperphoria with fixed eyeballs
Any one classification in the original half-face image of mirror, first item integrally indicate to allow discrimination model to authentic specimen generic
Probability output maximize, that is, maximize a kind of other true picture of one Zhang of input and be the category by network Accurate Prediction
Possibility.Section 2 in formulaIndicate that input picture x meets the phase of the i.e. original half-face image distribution of true picture
It hopes,Indicate that input picture belongs to the probability for rebuilding half-face image classification by neural network forecast,Indicate that input picture belongs to the non-probability for rebuilding half-face image classification, Section 2 by neural network forecast
It is whole to indicate that discrimination model is allowed to maximize a possibility that true picture is predicted to be true picture;Section 3 in formulaTable
Show that input picture x is the expectation for rebuilding half-face image,Indicate that input picture belongs to by neural network forecast
The probability of half-face image classification is rebuild, Section 3 integrally indicates that allowing discrimination model to maximize reconstruction image is predicted to be reconstruction image
A possibility that.The loss function purpose of design is the difference for maximizing the i.e. original half-face image of true picture and rebuilding half-face image with it
It is different and simultaneously to wearing glasses half-face image and half-face image of not wearing glasses is classified.Learn the publication of obtained glasses discrimination model
Into discrimination module.
Such as Figure 15, in the present embodiment, confrontation sorter network structure includes:
Input is that a height is the normalization facial image that 64 pixel wides are 128 pixels.
First layer, convolutional layer
The input of first layer is original image, and image size is 64*128.First layer convolutional layer filter size is 3*3, depth
It is 64, is filled using full 0, step-length 1.This layer is that output is 64*128*64.
The second layer, pond layer
The input of the second layer is the output of first layer, is the matrix of 64*128*64.The size of this layer of filter is 2*2, and step-length is
2.This layer is that output is 32*64*64.
Third layer, convolutional layer
This layer of input is the output of the second layer, is the matrix of 32*64*64.This layer of filter size is 3*3, depth 128,
It is filled using full 0, step-length 1.This layer is that output is 32*64*128.
4th layer, pond layer
This layer of input is the output of third layer, is the matrix of 32*64*128.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 16*32*128.
Layer 5, convolutional layer
The output that this layer of input is the 4th layer is the matrix of 16*32*128.This layer of filter size is 3*3, depth 256,
It is filled using full 0, step-length 1.This layer is that output is 16*32*256.
Layer 6, pond layer
This layer of input is the output of layer 5, is the matrix of 16*32*256.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 8*16*256.
Layer 7, convolutional layer
The output that this layer of input is the six or seven layer is the matrix of 8*16*256.This layer of filter size is 3*3, and depth is
512, it is filled using full 0, step-length 1.This layer is that output is 8*16*512.
8th layer, pond layer
This layer of input is the output of layer 7, is the matrix of 8*16*512.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 4*8*256.
9th layer, full articulamentum
This layer of input is the matrix of 4*8*256, and output node number is 128.
Tenth layer, full articulamentum
This layer of input number of nodes is 128, and output node number is 64.
Eleventh floor, full articulamentum
This layer of input number of nodes is 64, and output node number is 3.
Discrimination module differentiates by the way that whether half-face image of the glasses discrimination model to input wears glasses.Wherein,
The half-face image of the input is the half-face image extracted by image capture module and the acquisition of half-face image extraction module.At this
In embodiment, the output of the discrimination module is the result is that whether the half-face image of input wears glasses, do not wear glasses, being to rebuild half face
Confidence level, that is, confidence level that the half-face image inputted is worn glasses, the confidence level that do not wear glasses are the confidence levels for rebuilding half face.
Preferably, confidence level is the numerical value between 0 to 1, numerical value is bigger, and confidence level is higher.
Since the data volume of initial data base is limited, make the entire extensive energy of Glasses detection system to more collect data
Power is stronger, and in conjunction with Figure 16, invention introduces database update modules.Database update module is sentenced according to glasses discrimination module
Other result more new database, when output result is not bespectacled confidence level very high (being greater than a biggish threshold value) and is weight
Build image confidence level it is very low when (less than one lesser threshold value), which can be added to half face original number of glasses-free
According to library;When output result be bespectacled confidence level very high (be greater than a biggish threshold value) and be reconstruction image confidence level
When very low (less than one lesser threshold value), which can be added to half face raw data base of wearing glasses.Based on aforementioned
The update of database, half face reconstruction model update, and half face raw data base enlarged meeting of glasses-free makes half face reconstruction model increasingly
Accurately.Half face raw data base of wearing glasses expands the diversity that will increase Glasses detection.Preferably, in database update module also
It may include manual intervention module, can manually add and safeguard database with Rejection of samples.
As shown in Figure 1, the present invention proposes a kind of eyeglass detection method, for detecting whether user wears glasses comprising
Following steps:
A. the facial image of image capture module acquisition user, and the collected facial image is directly passed to half-face image
Extraction module, or the collected facial image is first stored in face image database according to the actual situation, to later again by
Half-face image extraction module handles the facial image in the face image database.Preferably, in the present embodiment, institute
Stating image capture module includes iris camera and infrared lamp.
B. half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half
Face image, i.e. half-face image extraction module are corrected the facial image of input, then extract and normalize to obtain half face figure
By the half-face image that do not wear glasses storage to half face raw data base of glasses-free, the half-face image worn glasses is stored later for picture
To half face raw data base of wearing glasses.In the present embodiment, the facial image for inputting half-face image extraction module can be image
The directly incoming facial image of acquisition module, the facial image being also possible in face image database, this can be according to practical feelings
Condition voluntarily selects, i.e., in the present invention, as long as a facial image of input half-face image extraction module, either
Image capture module is collected in real time, is also possible to extract from database.
In the present embodiment, half face is referred to up to eyebrow peak in face, and down toward nose tabula lower edge, left to left eyebrow tail, the right side is extremely
The region of right eyebrow tail.Firstly, the ratio of the partial region each organ distribution in different crowds is relatively stable, and will not be because
For expression variation and generate very big denaturation, thus the region is more advantageous to image normalization;Secondly, using the partial region
Can meet the needs of Face datection and iris detection simultaneously.It under normal circumstances, is the clear iris texture of bat, the view of iris capturing system
Wild range is smaller, it may not be possible to obtain complete facial image, can preferably be compatible with iris capturing system, i.e. base using half-face image
Can be not only used for face identification system in the Glasses detection system of half-face image can be used for iris authentication system again.
In conjunction with Fig. 2 to Fig. 7, half-face image extraction module includes detection module, rectification module and normalization module, step
B further comprises:
Detection module detects face in the facial image of input, and detects the position where face key point.
Rectification module will test the face key point that module detects and compare with normalized standard faces model, push away
Calculate the facial image acquisition when, user's face relative to image capture device position coordinates and user's head in three-dimensional
The corner in space is corrected the face rotated in the facial image at standard faces, simultaneously based on this use of information perspective transform
It will test the face key point that module detects to be mapped on the standard faces image corrected.
Module is normalized according to the position coordinates of the key point being mapped on the facial image corrected by half face
Image cut comes out and normalizes to fixed dimension, forms half-face image, wherein the half-face image storage that do not wear glasses to anophthalmia
Half face raw data base of mirror, the half-face image storage worn glasses to half face raw data base of wearing glasses.
As shown in figure 4, in the present embodiment, face key point refers to the key extracted in facial image based on human face five-sense-organ
Point information, and the image for extracting face key point when standard faces model is then head irrotationality gyration, and being formed through normalization.
C. half face is rebuild module and is carried out based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
Principal component analysis (PCA) obtains half face reconstruction model, and is rebuild by the half face reconstruction model to half-face image, will not
Storage rebuilds database to half face of glasses-free after the half-face image worn glasses is rebuild, and the half-face image worn glasses is stored after rebuilding and arrived
Half face of wearing glasses rebuilds database.In the present embodiment, half face raw data base of glasses-free and half face raw data base of wearing glasses
Half-face image be real human face image, half face of glasses-free rebuild database and wear glasses half face rebuild database half-face image
To rebuild facial image.The half face raw data base of glasses-free, half face raw data base of wearing glasses, half face of glasses-free rebuild number
Database, which is rebuild, according to library and half face of wearing glasses constitutes database module.
As shown in Figs. 8 to 11, half face rebuilds module and includes the first study module and rebuild module, and step c is further wrapped
It includes:
First study module carries out principal component based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
It analyzes (PCA), obtains half face reconstruction model, and the half face reconstruction model is published to and is rebuild in module.In the present embodiment, institute
Stating the first study module utilizes the method for principal component analysis (PCA) to the face of half-face image in half face raw data base of glasses-free
Feature Dimension Reduction, extracts the main common feature of different faces, and generates eigenface, and i.e. half face of these eigenfaces rebuilds mould
Type, the half face reconstruction model for later completing the study, which is published to, to be rebuild in module.By being then based on the half face figure that do not wear glasses
As the principal component analysis carried out, therefore half face reconstruction model in the present invention does not include any information of glasses.
Rebuild the half-face image that module rebuilds half-face image, and will not worn glasses by the half face reconstruction model
Storage rebuilds database to half face of glasses-free after the reconstruction of half face reconstruction model, and the half-face image worn glasses is through half face reconstruction model
Storage rebuilds database to half face of wearing glasses after reconstruction.In the present embodiment, the half-face image that module is rebuild in input can be logical
The half-face image for crossing image capture module and the acquisition extraction of half-face image extraction module, is also possible to directly read half face of glasses-free
Half-face image in raw data base and half face raw data base of wearing glasses.
D. glasses discrimination module is based on confrontation sorter network to the half face raw data base of glasses-free, half face of wearing glasses
Raw data base, half face of glasses-free rebuild database, the half-face image of half face of wearing glasses reconstruction database carries out dual training, obtain
To glasses discrimination model;And differentiated by the glasses discrimination model to whether the half-face image of input wears glasses.
As shown in Figure 12 to Figure 15, the glasses discrimination module includes the second study module and discrimination module, and step d is into one
Step includes:
Second study module is used to train the classification of the confrontation complete paired samples of sorter network.In the present embodiment, net is fought
Network refers to that the network can fight half-face image and true half-face image that half face algorithm for reconstructing of PCA distinguishes reconstruction, i.e. area
It separates and rebuilds half-face image and original half-face image;Confrontation sorter network refer to the network can distinguish rebuild half-face image and
Classification is completed to whether original half-face image wears glasses while original half-face image.Purpose using confrontation network is by right
One powerful classifier of anti-training production can be distinguished to the greatest extent rebuilds half-face image and original half-face image.Rebuild half
Face image and original half-face image are primarily present following two difference: (1) half-face image rebuild not wear a pair of spectacles;(2) it uses
The half-face image that PCA method is rebuild lacks the order of information of real human face.Therefore the half-face image tool rebuild in confrontation network
There is reference role, for extracting the difference characteristic of real human face and reconstruction face in half-face image, i.e. confrontation network is based on data
Half-face image in library extracts the original half-face image of same user and rebuilds the difference characteristic of half-face image.Fight sorter network base
The classification to half-face image is completed in above difference characteristic, the classification of classification includes wearing glasses, glasses-free and be to rebuild people
Face.Database used in dual training is all half-face image in database module, i.e., half face raw data base of glasses-free, wear
Half face raw data base of glasses, half face of glasses-free rebuild database and half face of wearing glasses rebuilds database.Half face of glasses-free is former
The corresponding label of half-face image is 2 in beginning database, is desired for [0,1,0];It wears glasses half-face image in half face raw data base
Corresponding label is 1, is desired for [1,0,0];Half face of glasses-free rebuilds database and half face of wearing glasses rebuilds half face in database
The corresponding label of image is 3, is desired for [0,0,1].Loss function is as follows:
WhereinDifferentiate that the loss function of network, y indicate classification for glasses, y=1 indicates the original half-face image worn glasses, y=2
Indicate the original half-face image that do not wear glasses, y=3 indicates to rebuild half-face image.First item in formulaIt indicates
Input picture x meets the distribution of the i.e. original half-face image of true picture and belongs to the expectation of y classification,Table
Show that input picture is predicted to be the probability of y classification, y=1 therein or 2 are the original half-face image worn glasses or do not wear glasses
Original half-face image in any one classification, first item integrally indicates to allow discrimination model to authentic specimen generic
Probability output maximizes, that is, maximize a kind of other true picture of one Zhang of input and by network Accurate Prediction be the category can
It can property.Section 2 in formulaIndicate that input picture x meets the expectation of the i.e. original half-face image distribution of true picture,Indicate that input picture belongs to the probability for rebuilding half-face image classification by neural network forecast,Indicate that input picture belongs to the non-probability for rebuilding half-face image classification, Section 2 by neural network forecast
It is whole to indicate that discrimination model is allowed to maximize a possibility that true picture is predicted to be true picture;Section 3 in formulaTable
Show that input picture x is the expectation for rebuilding half-face image,Indicate that input picture belongs to weight by neural network forecast
The probability of half-face image classification is built, Section 3 integrally indicates that allowing discrimination model to maximize reconstruction image is predicted to be reconstruction image
Possibility.The loss function purpose of design is the difference for maximizing the i.e. original half-face image of true picture and rebuilding half-face image with it
And simultaneously to wearing glasses half-face image and half-face image of not wearing glasses is classified.The glasses discrimination model for learning to obtain is published to
In discrimination module.
Such as Figure 15, in the present embodiment, confrontation sorter network structure includes:
Input is that a height is the normalization facial image that 64 pixel wides are 128 pixels.
First layer, convolutional layer
The input of first layer is original image, and image size is 64*128.First layer convolutional layer filter size is 3*3, depth
It is 64, is filled using full 0, step-length 1.This layer is that output is 64*128*64.
The second layer, pond layer
The input of the second layer is the output of first layer, is the matrix of 64*128*64.The size of this layer of filter is 2*2, and step-length is
2.This layer is that output is 32*64*64.
Third layer, convolutional layer
This layer of input is the output of the second layer, is the matrix of 32*64*64.This layer of filter size is 3*3, depth 128,
It is filled using full 0, step-length 1.This layer is that output is 32*64*128.
4th layer, pond layer
This layer of input is the output of third layer, is the matrix of 32*64*128.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 16*32*128.
Layer 5, convolutional layer
The output that this layer of input is the 4th layer is the matrix of 16*32*128.This layer of filter size is 3*3, depth 256,
It is filled using full 0, step-length 1.This layer is that output is 16*32*256.
Layer 6, pond layer
This layer of input is the output of layer 5, is the matrix of 16*32*256.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 8*16*256.
Layer 7, convolutional layer
The output that this layer of input is the six or seven layer is the matrix of 8*16*256.This layer of filter size is 3*3, and depth is
512, it is filled using full 0, step-length 1.This layer is that output is 8*16*512.
8th layer, pond layer
This layer of input is the output of layer 7, is the matrix of 8*16*512.The size of this layer of filter is 2*2, step-length 2.
This layer is that output is 4*8*256.
9th layer, full articulamentum
This layer of input is the matrix of 4*8*256, and output node number is 128.
Tenth layer, full articulamentum
This layer of input number of nodes is 128, and output node number is 64.
Eleventh floor, full articulamentum
This layer of input number of nodes is 64, and output node number is 3.
Discrimination module differentiates by the way that whether half-face image of the glasses discrimination model to input wears glasses.Wherein,
The half-face image of the input is the half-face image extracted by image capture module and the acquisition of half-face image extraction module.At this
In embodiment, the output of the discrimination module is the result is that whether the half-face image of input wears glasses, do not wear glasses, being to rebuild half face
Confidence level, that is, confidence level that the half-face image inputted is worn glasses, the confidence level that do not wear glasses are the confidence levels for rebuilding half face.
Preferably, confidence level is the numerical value between 0 to 1, numerical value is bigger, and confidence level is higher.
Preferably, the data volume due to initial data base is limited, make entire Glasses detection system to more collect data
Uniting, generalization ability is stronger, and in conjunction with Figure 16, invention introduces database update modules.Database update module differentiates according to glasses
The differentiation result more new database of module, when output result is that not bespectacled confidence level is very high (being greater than a biggish threshold value)
It and is (less than one lesser threshold value) when the confidence level of reconstruction image is very low, which can be added to glasses-free half
Face raw data base;When output result is bespectacled confidence level very high (being greater than a biggish threshold value) and is reconstruction image
Confidence level it is very low when (less than one lesser threshold value), which can be added to half face raw data base of wearing glasses.
Based on the update in aforementioned data library, half face reconstruction model updates, and half face raw data base enlarged meeting of glasses-free makes half face rebuild mould
Type is more and more accurate.Half face raw data base of wearing glasses expands the diversity that will increase Glasses detection.Preferably, database update
It can also include manual intervention module in module, can manually add and safeguard database with Rejection of samples.
Glasses detection system and method for the present invention, detecting user in such a way that principal component analysis and deep learning combine is
No wearing spectacles, relatively reliable stabilization;It goes to learn wear glasses half-face image and its reconstructed shape present invention introduces deep learning
At half-face image of not wearing glasses difference come to user, whether wearing spectacles provide judgement, when avoiding existing Threshold segmentation glasses
Threshold value cannot accurately set and cause error in judgement.Database update module of the invention can be with automatic updating data library, it is ensured that sample
This diversity, and improve the precision of half face reconstruction model and glasses discrimination model.
It is described the invention in detail above in conjunction with accompanying drawings and embodiments, those skilled in the art can basis
Above description makes many variations example to the present invention.Thus, certain details in embodiment should not constitute limitation of the invention,
The present invention will be using the range that the appended claims define as protection scope of the present invention.
Claims (9)
1. a kind of Glasses detection system, for detecting whether user wears glasses, characterized by comprising:
Image capture module, for acquiring facial image;
Half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half-face image,
Wherein, the half-face image that do not wear glasses stores the half-face image worn glasses to half face raw data base of glasses-free and stores to hyperphoria with fixed eyeballs
Half face raw data base of mirror;
Half face rebuilds module, carries out principal component based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
It analyzes (PCA), obtains half face reconstruction model, and rebuild to half-face image by half face reconstruction model, wherein do not wear glasses
Half-face image rebuild after storage to half face of glasses-free rebuild database, storage is to wearing glasses after the half-face image worn glasses is rebuild
Half face rebuilds database;
Glasses discrimination module, based on confrontation sorter network to the half face raw data base of glasses-free, half face original number of wearing glasses
The half-face image progress network training that database, half face of wearing glasses reconstruction database are rebuild according to library, half face of glasses-free, obtains glasses
Discrimination model;And the differentiation worn glasses is made whether to half-face image by glasses discrimination model.
2. a kind of Glasses detection system according to claim 1, it is characterised in that: further include a data update module, institute
It is original to half face raw data base of glasses-free, half face of wearing glasses according to the differentiation result of glasses discrimination module to state data update module
Database is updated.
3. a kind of eyeglass detection method, for detecting whether user wears glasses, it is characterised in that the following steps are included:
A. the facial image of image capture module acquisition user;
B. half-face image extraction module handles the facial image for inputting the half-face image extraction module, obtains half face figure
Picture, and the half-face image that do not wear glasses is stored to half face raw data base of glasses-free, the half-face image storage worn glasses is arrived
It wears glasses half face raw data base;
C. half face rebuild module based on the half-face image that do not wear glasses in the half face raw data base of glasses-free carry out it is main at
Analysis (PCA), obtains half face reconstruction model, and rebuild to half-face image by the half face reconstruction model, will not hyperphoria with fixed eyeballs
The half-face image of mirror stores to half face of glasses-free after rebuilding and rebuilds database, and hyperphoria with fixed eyeballs is arrived in storage after the half-face image worn glasses is rebuild
Half face of mirror rebuilds database;
D. it is original to the half face raw data base of glasses-free, half face of wearing glasses to be based on confrontation sorter network for glasses discrimination module
Database, half face of glasses-free rebuild database, the half-face image of half face of wearing glasses reconstruction database carries out network training, obtain eye
Mirror discrimination model;And differentiated by the glasses discrimination model to what whether the half-face image of input wore glasses.
4. eyeglass detection method according to claim 3, which is characterized in that it includes the first study that half face, which rebuilds module,
Module and reconstruction module, step c further comprises:
First study module carries out principal component based on the half-face image that do not wear glasses in the half face raw data base of glasses-free
It analyzes (PCA), obtains half face reconstruction model, and the half face reconstruction model is published to and is rebuild in module;
It rebuilds module to rebuild half-face image by the half face reconstruction model, and the half-face image that do not wear glasses is rebuild
Storage rebuilds database to half face of glasses-free afterwards, and storage rebuilds data to half face of wearing glasses after the half-face image worn glasses is rebuild
Library.
5. eyeglass detection method according to claim 3, which is characterized in that the glasses discrimination module includes the second study
Module and discrimination module, step d further comprises:
Second study module is based on confrontation sorter network to the half face raw data base of glasses-free, half face initial data of wearing glasses
Library, half face of glasses-free rebuild database, the half-face image of half face of wearing glasses reconstruction database carries out network training, obtain glasses and sentence
Other model, and the glasses discrimination model is published in discrimination module;
Discrimination module differentiates by the way that whether half-face image of the glasses discrimination model to input wears glasses.
6. eyeglass detection method according to claim 3, which is characterized in that further include step e:
Data update module is according to the differentiation result of glasses discrimination module to half face raw data base of glasses-free, half face original of wearing glasses
Beginning database is updated.
7. eyeglass detection method according to claim 3, which is characterized in that the half-face image extraction module includes detection
Module, rectification module and normalization module, step b further comprises:
Detection module detects face in the facial image of input, and detects the position where face key point;
Rectification module will test the face key point that module detects and compare with normalized standard faces model, based on pair
The face in facial image is corrected than use of information perspective transform, forms standard faces image, and will test module inspection
The face key point measured is mapped on the standard faces image corrected;
Module is normalized according to the position coordinates of the face key point being mapped on the standard faces image corrected by half face
The image cut of position comes out and normalizes to fixed dimension, forms half-face image, wherein the half-face image storage that do not wear glasses is arrived
Half face raw data base of glasses-free, the half-face image storage worn glasses to half face raw data base of wearing glasses.
8. eyeglass detection method according to claim 3, it is characterised in that: half-face image is in facial image up to eyebrow
Peak, it is left to left eyebrow tail down toward nose tabula lower edge, it is right to the region of right eyebrow tail.
9. eyeglass detection method according to claim 5, it is characterised in that: the second study module is for training a confrontation
The classification of the complete paired samples of sorter network, confrontation sorter network are based on half-face image and the half-face image through more than half face reconstruction models
The difference characteristic of the reconstruction image of formation is completed to half face raw data base of glasses-free, wear glasses half face raw data base, anophthalmia
Half face of mirror rebuilds database, half face of wearing glasses rebuilds the classification of half-face image in database.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910811078.4A CN110334698A (en) | 2019-08-30 | 2019-08-30 | Glasses detection system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910811078.4A CN110334698A (en) | 2019-08-30 | 2019-08-30 | Glasses detection system and method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110334698A true CN110334698A (en) | 2019-10-15 |
Family
ID=68150200
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910811078.4A Pending CN110334698A (en) | 2019-08-30 | 2019-08-30 | Glasses detection system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110334698A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113052068A (en) * | 2021-03-24 | 2021-06-29 | 深圳威富云数科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260731A (en) * | 2015-11-25 | 2016-01-20 | 商汤集团有限公司 | Human face living body detection system and method based on optical pulses |
CN107016370A (en) * | 2017-04-10 | 2017-08-04 | 电子科技大学 | One kind is based on the enhanced partial occlusion face identification method of data |
CN107341435A (en) * | 2016-08-19 | 2017-11-10 | 北京市商汤科技开发有限公司 | Processing method, device and the terminal device of video image |
CN107808120A (en) * | 2017-09-30 | 2018-03-16 | 平安科技(深圳)有限公司 | Glasses localization method, device and storage medium |
CN107808142A (en) * | 2017-11-09 | 2018-03-16 | 北京小米移动软件有限公司 | Eyeglass detection method and device |
CN108229348A (en) * | 2017-12-21 | 2018-06-29 | 中国科学院自动化研究所 | Block the identification device of facial image |
CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
CN109977841A (en) * | 2019-03-20 | 2019-07-05 | 中南大学 | A kind of face identification method based on confrontation deep learning network |
-
2019
- 2019-08-30 CN CN201910811078.4A patent/CN110334698A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260731A (en) * | 2015-11-25 | 2016-01-20 | 商汤集团有限公司 | Human face living body detection system and method based on optical pulses |
CN107341435A (en) * | 2016-08-19 | 2017-11-10 | 北京市商汤科技开发有限公司 | Processing method, device and the terminal device of video image |
CN107016370A (en) * | 2017-04-10 | 2017-08-04 | 电子科技大学 | One kind is based on the enhanced partial occlusion face identification method of data |
CN107808120A (en) * | 2017-09-30 | 2018-03-16 | 平安科技(深圳)有限公司 | Glasses localization method, device and storage medium |
CN107808142A (en) * | 2017-11-09 | 2018-03-16 | 北京小米移动软件有限公司 | Eyeglass detection method and device |
CN108229348A (en) * | 2017-12-21 | 2018-06-29 | 中国科学院自动化研究所 | Block the identification device of facial image |
CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
CN109977841A (en) * | 2019-03-20 | 2019-07-05 | 中南大学 | A kind of face identification method based on confrontation deep learning network |
Non-Patent Citations (1)
Title |
---|
万里鹏: "基于生成对抗网络的多属性人脸生成及辅助识别研究", 《万方学位论文全文数据库》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113052068A (en) * | 2021-03-24 | 2021-06-29 | 深圳威富云数科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
CN113052068B (en) * | 2021-03-24 | 2024-04-30 | 深圳威富云数科技有限公司 | Image processing method, device, computer equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102844766B (en) | Human eyes images based multi-feature fusion identification method | |
CN103902961B (en) | Face recognition method and device | |
CN109840565A (en) | A kind of blink detection method based on eye contour feature point aspect ratio | |
CN107766786A (en) | Activity test method and active testing computing device | |
KR102554391B1 (en) | Iris recognition based user authentication apparatus and method thereof | |
Boehnen et al. | A fast multi-modal approach to facial feature detection | |
KR20050025927A (en) | The pupil detection method and shape descriptor extraction method for a iris recognition, iris feature extraction apparatus and method, and iris recognition system and method using its | |
CN104504362A (en) | Face detection method based on convolutional neural network | |
TW201137768A (en) | Face recognition apparatus and methods | |
CN111178208A (en) | Pedestrian detection method, device and medium based on deep learning | |
CN106203391A (en) | Face identification method based on intelligent glasses | |
CN106599785B (en) | Method and equipment for establishing human body 3D characteristic identity information base | |
CN109034133A (en) | A kind of face identification method and device | |
RU2316051C2 (en) | Method and system for automatically checking presence of a living human face in biometric safety systems | |
Pauca et al. | Challenging ocular image recognition | |
Bastias et al. | A method for 3D iris reconstruction from multiple 2D near-infrared images | |
Mirabet-Herranz et al. | Impact of digital face beautification in biometrics | |
CN110135391A (en) | System is matched using the program and spectacle-frame of computer apolegamy spectacle-frame | |
CN110334698A (en) | Glasses detection system and method | |
CN108319935A (en) | Based on the face group recognizer that region is sparse | |
Borkar et al. | IRIS recognition system | |
Durga et al. | Face Mask Detection using MobileNetV2 | |
CN113657197A (en) | Image recognition method, training method of image recognition model and related device | |
CN113553890A (en) | Multi-modal biological feature fusion method and device, storage medium and equipment | |
Zhou et al. | Eye localization based on face alignment |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191015 |