CN107103266B - The training of two-dimension human face fraud detection classifier and face fraud detection method - Google Patents
The training of two-dimension human face fraud detection classifier and face fraud detection method Download PDFInfo
- Publication number
- CN107103266B CN107103266B CN201610098933.8A CN201610098933A CN107103266B CN 107103266 B CN107103266 B CN 107103266B CN 201610098933 A CN201610098933 A CN 201610098933A CN 107103266 B CN107103266 B CN 107103266B
- Authority
- CN
- China
- Prior art keywords
- image
- feature vector
- dimension
- facial image
- value
- 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.)
- Active
Links
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides the generation methods of two-dimension human face fraud detection model, which comprises and first, all people's face picture in training set is pre-processed, normalization facial image is obtained;Second, LBP feature vector, Gabor wavelet feature vector sum one-dimensional pixel feature vector are extracted from each normalization facial image;Third carries out these three feature vectors to be spliced to form final feature vector;4th, using support vector machines to being trained based on being spliced to form final feature vector, obtain two-dimension human face fraud detection classifier;This method is extracted the characteristic information of the difference of face and photo;Feature extraction is simple and efficient, and is not needed user and is purposely cooperated, and good effect can be obtained in the case where low resolution.In addition, being based on two-dimension human face fraud detection classifier obtained by the above method, the invention also provides face fraud detection method, this method has the advantages that detection accuracy is high, can effectively prevent face fraud.
Description
Technical field
The present invention relates to computer visions and graph and image processing field, in particular to two-dimension human face fraud detection classifier
Training and face fraud detection method.
Background technique
Currently, two-dimentional biological identification technology (identification i.e. based on two-dimension human face biological characteristic) is a critically important research
Field.View transformation, block, complicated outdoor light be always recognition of face difficult point.This is solved although doing a lot of work
A little problems, but the fraud of face identification system attack loophole is but ignored by big multisystem.Face identification system relies on plane
Figure carries out identity detection, and system is highly susceptible to the fraud attack of photograph print or electronic photo.For example, association, Asus
With the laptop of the Windows XP and Vista of Toshiba all built-in IP Camera and biological recognition system, by sweeping
Retouch the face certification user of user.(world-leading technical security meeting), the peace of Ha Noi university in black cap conference in 2009
How easily full loophole research group illustrates is out-tricked the face identification system of notebook using the human face photo of legitimate user
(the Veriface III of association, the SmartLogon V1.0.0005 of Asus, the Face Recognition of Toshiba
2.0.2.32) enter notebook, and the use of these systems is all highest security level.This loophole is now by state, the U.S.
Family's standard and Institute for Research and Technology include into national vulnerability scan.Therefore fraud attack is solved the problems, such as to improve face identification system
Safety and robustness and to be applied to it be a very urgent demand in practice.
Fraud attack refers to that a people attempts to disguise oneself as other people by data falsification to obtain illegal access.Such as
One people cheats face identification system by the photo of a legitimate user, video, mask, 3D model before camera.Although
One people can also use other fraud modes such as makeup, lift face, but photo attack is the most common fraud attack, because of human face photo
Can easily by download on the net, acquisition of taking on the sly.
Currently, fraud attack loophole has attracted more and more people's note that 2011 have also specially held one
" IJCB2011 two-dimension human face cheats attack defending contest ".Although the research in this field is more and more and issues successively
The database of some public openings, but the standard database of relatively objective exploitation test can be provided for fraud detection algorithm simultaneously
It is not very much, this field or not mature enough, up to the present about optimal fraud detection algorithm, there is no unified to be total to
Know.Current existing face fraud detection research method or too complicated, without practicability (it is practical use in require place real-time, quickly
Reason) or some unconventional imaging systems (multiple spectra imaging) and high-resolution camera are used, these do not have reality
The condition of application.
Summary of the invention
It is an object of the invention to overcome drawbacks described above existing for current two-dimension human face fraud detection method, pass through finder
The subtle difference of face and corresponding photo, and a feature space is designed to protrude the difference.In fact, human face photo all contains one
Determine the printing quality defect of degree, can detected well with texture.It is anti-by picture quality, printed matter feature, light
The inspiration of difference is penetrated, the present invention proposes a kind of training method of two-dimension human face fraud detection classifier.This method is by from picture
Middle extraction LBP, Gabor wavelet, corresponding feature vector in three kinds of feature spaces of pixel characteristic, then these three feature vectors are closed
At a final feature vector, this feature vector is finally sent to a nonlinear SVM classifier and carries out classification based training
Obtain a two-dimension human face fraud detection classifier.Based on the two-dimension human face fraud detection classifier, the present invention also provides one
Kind face fraud detection method, it can be determined that going out input picture is face or fraud image.
To achieve the goals above, the present invention provides the training method of two-dimension human face fraud detection classifier, the sides
Method includes: first, is pre-processed to all people's face picture in training set, and normalization facial image is obtained;Second, from each
It normalizes and extracts LBP feature vector, Gabor wavelet feature vector sum one-dimensional pixel feature vector in facial image;Third, by this
Three feature vectors carry out being spliced to form final feature vector;4th, using support vector machines to being spliced to form final spy
Sign vector is trained, and obtains two-dimension human face fraud detection classifier.
In above-mentioned technical proposal, the method is specifically included:
Step S1) i-th of face picture in training set is pre-processed, wherein 1≤i≤L;Obtain 64 × 64 pixels
The normalization facial image z of sizei;
Step S2) from normalization facial image ziMiddle extraction LBP feature vector L (zi);
Step S3) from normalization facial image ziMiddle extraction Gabor wavelet feature vector G (zi);
Step S4) facial image z will be normalizediIt zooms to 8 × 8 sizes, then by two dimensional image thaumatropy is one-dimensional picture
Plain feature vector P (zi);
Step S5) by step S2), step S3), step S4) extract three kinds of textural characteristics be spliced into final feature to
Measure D (zi)=(L (zi),G(zi),P(zi));
Step S6) based on Support vector regression algorithm to all feature vector D (zi) be trained, obtain two-dimentional people
Face fraud detection classifier, wherein 1≤i≤L.
In above-mentioned technical proposal, the step S1) it specifically includes:
Step S1-1) image gray processing processing is carried out to face picture:
Face face picture is traversed, each pixel is handled, the rgb value of each pixel is obtained, passes through operation
Red, blue, green value is extracted respectively, the gray value after calculating each pixel conversion:
Grey=(9798R+19235G+3735B)/32768
Wherein, Grey represents the gray value after conversion, and R, G, B respectively represent the red point of each pixel in image
Amount, green component and blue component;
Step S1-2) gray level image size is adjusted to 64 × 64 using bilinear interpolation;
Step S1-3) image after being sized is enhanced:
The modification that image histogram is carried out using the statistical data of histogram, is gone out by adjusting each grey level pixel of image
Existing probability is equal to change the pixel value of each grey level in image, to realize image enhancement;
Step S1-4) to enhanced image zooming-out image pixel matrix, obtain normalization facial image zi。
In above-mentioned technical proposal, the step S2) it specifically includes:
Step S2-1) to normalization facial image ziWithOperator obtains LBP image, then is by LBP image segmentation
3 × 3 overlapping regions, the statistic histogram for extracting 59 dimensions from each region respectively synthesize the statistic histogram of one 531 dimension
Feature vector;
Step S2-2) to normalization facial image ziWithOperator, extraction obtain the statistic histogram feature of 59 dimensions
Vector;
Step S2-3) to normalization facial image ziWithOperator extracts and obtains the statistic histogram spy of 243 dimensions
Levy vector;
Step S2-4) feature vector that obtains step S2-1), step S2-2), step S2-3) synthesize a feature to
Measure L (zi), the dimension of this feature vector is 59*9+59+243=833 dimension.
In above-mentioned technical proposal, the step S3) it specifically includes:
Step S3-1) facial image z will be normalizedi32 × 32 sizes are zoomed to, it is small to carry out Gabor to the image after scaling
Wave conversion:
The Gabor filter in p different directions and q different scale is taken to handle image after scaling, each picture
Vegetarian refreshments t0It can obtain p × q Gabor amplitude Characteristics, p × q Gabor amplitude Characteristics cascade up a referred to as Jet, write a Chinese character in simplified form
For J, then pixel t in image0Jet are as follows:
J(t0)=(M0,0(t0),...,M0,7(t0),...,M4,0(t0),...,M4,7(t0))
The Gabor amplitude Characteristics of all pixels point are cascaded up to obtain the feature vector F (z of facial imagei):
F(zi)={ J (t0):t0∈zi}
Step S3-2) determine step S3-1) obtained feature vector F (zi) dimensionality reduction dimension, be based on Principal Component Analysis
To feature vector F (zi) carry out dimensionality reduction, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni)。
In above-mentioned technical proposal, the step S3-2) it specifically includes:
Step S3-2-1) the feature vector F (z that obtains step S3-1)i) dimension d be divided into n equal part, determine new dimension
Spend the value range of d';
Original feature vector F (zi) dimension be d, the value of d is divided into n equal portions, value set is as follows:
Wherein,Indicate integer rounding operation;
If the feature that obtains after dimensionality reduction is G (zi), then its dimension d' takes this n value respectively:
Step S3-2-2) d' successively takes each value in set, calculate the corresponding face fraud inspection of all pictures of training set
Survey mean absolute error set { MAEm};
To the L picture in training set, calculate when d' takes each value in set, the corresponding people of all pictures of training set
Face fraud detection mean absolute error is as follows:
Wherein, j indicates that jth picture in training set, k indicate that d' takes k-th of value in set, i.e.,lj
Indicate the corresponding class label of jth picture in training set: 0 indicates fraud image, and 1 indicates real human face image,Indicate training
Concentrate the classification estimated value of jth picture;It may finally obtain different MAE value set { MAEmWherein m ∈ 1,2 ..., n;
Step S3-2-3) take set { MAEmIn minimum M AEmin, with MAEminCorresponding d' is tieed up as final dimensionality reduction
Degree;
Step S3-2-4) it is based on step S3-2-3) obtained d', using Principal Component Analysis to feature vector F (zi) into
Row dimensionality reduction, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni)。
In above-mentioned technical proposal, the step S6) it specifically includes:
Step S6-1) based on Support vector regression algorithm building optimization problem;
Hypothesized model training set sample is { x(i),y(i)(i=1,2 ..., L), x(i)Indicate normalization facial image zi's
Feature vector D (zi), y(i)Indicate the corresponding classification of the image: facial image or fraud image;Assuming that sample dimension is N, thenThe target of Support vector regression algorithm is to solve for two-dimension human face fraud detection classifier f (x), makes f (x(i)) and y(i)Between difference be not more than ε, ε is threshold value, controls the worst error between practical label value and predictive estimation value;So, f
(x) it is defined as follows:
Wherein, " " indicates inner product of vectors;W and b is the parameter solved;
Step S6-2) optimization problem (1) is converted using method of Lagrange multipliers, it is converted to its antithesis
Problem is solved, and the expression formula of two-dimension human face fraud detection classifier f (x) is obtained.
Based on the two-dimension human face fraud detection classifier that above method training obtains, the present invention also provides face frauds to examine
Survey method, this method comprises:
Step T1) face picture to be detected is pre-processed, obtain the normalization facial image of 64 × 64 pixel sizes
z0;
Step T2) from normalization facial image z0Middle extraction LBP feature vector L (z0);
Step T3) from normalization facial image z0Middle extraction Gabor wavelet feature vector G (z0);
Step T4) facial image z will be normalized0It zooms to 8 × 8 sizes, then by two dimensional image thaumatropy is one-dimensional
Pixel characteristic vector P (z0);
Step T5) by step T2), step T3), step T4) extract three kinds of textural characteristics be spliced into final feature to
Measure D (z0)=(L (z0),G(z0),P(z0));
Step T6) the feature vector D (z that obtains step T5)0)) be input in two-dimension human face fraud detection classifier, it obtains
To testing result: facial image cheats image.
The present invention has the advantages that
1, the training method of two-dimension human face fraud detection classifier proposed by the present invention;By merge a variety of textural characteristics come
Capture the subtle difference of face and corresponding photo;The complementary attribute of LBP, Gabor wavelet two powerful texture operators is absorbed,
In, what LBP included is microtexture information, and what Gabor wavelet included is the information of macroscopic view, and pixel characteristic provides the overall situation
Information, therefore the feature vector of training is fully extracted the information of the difference of face and photo;
2, in the training method of two-dimension human face fraud detection classifier proposed by the present invention, feature extraction is simple and efficient,
It does not need user purposely to cooperate, good effect can be obtained in the case where low resolution;Two-dimension human face feature vector mentions
It takes and combines SVR regression algorithm that lower error can be obtained when realizing face fraud detection, meet practical application scene needs;
3, the textural characteristics that the present invention uses can also be used to carry out recognition of face, be face fraud detection and recognition of face
Provide a unique feature space;
4, face fraud detection method of the invention has the advantages that detection accuracy is high, face can be effectively prevent to cheat
Behavior;
5, method of the invention can be applicable to many occasions such as face verification system, security monitoring.
Detailed description of the invention
Fig. 1 is the flow chart of the training method of two-dimension human face fraud detection classifier of the invention.
Specific embodiment
Technological concept of the present invention is briefly described below.
LBP (Local Binary Pattern, local binary model) is characterized in that face Related Research Domain is more common
Textural characteristics.Most basic LBP image coding is defined as in 3 × 3 window, will be adjacent using window center pixel as threshold value
The gray values of 8 pixels be compared with it, if surrounding pixel values are greater than center pixel value, the position of the pixel is marked
It is denoted as 1, is otherwise 0.In this way, 8 points in 3 × 3 neighborhoods are compared and can produce 8 bits and (be typically converted into the decimal system
Number is LBP codes, totally 256 kinds) to get to the LBP value of the window center pixel, and reflect with this value the texture in the region
Information.In order to adapt to the textural characteristics of different scale, by 3 × 3 neighborhood extendings to any neighborhood, and with circle shaped neighborhood region instead of just
Square Neighborhood calculates the gray value for not entirely falling within the point of location of pixels using bilinear interpolation algorithm.In addition, improved
LBP operator allows to have any number of pixels in the round adjacent, domain that radius is R.LBP equivalent formulations (uniform
Patterns) dimensionality reduction can be carried out to LBP feature vector.The circulation binary number corresponding to some local binary pattern from
0 to 1 or when be up to jumping twice from 1 to 0, binary system corresponding to the local binary pattern is known as an equivalent formulations
Class, such as 00000000,11111111,10001111 be all equivalent formulations class, and the mode in addition to equivalent formulations class is all classified as separately
One kind, referred to as mixed mode class.The type of improvement in this way, local binary patterns greatly reduces.Equivalent formulations class Zhan is total
The overwhelming majority in mode can extract using the histogram of these equivalent formulations classes and mixed mode class more representative of image sheet
The feature of matter characteristic.SymbolIndicate the equivalent formulations operator of the P point in the round field that radius is R.
Gabor wavelet is widely used in image recognition, field of image processing, in area of pattern recognition, Gabor wavelet
Transformation is also a kind of very effective Feature Descriptor.Gabor wavelet is very sensitive for the edge of image, is capable of providing good
Direction selection and scale selection characteristic, and it is insensitive for illumination variation, it is capable of providing to the good adaptability of illumination variation.
Meanwhile the information of facial critical component (eyes, nose, mouth etc.) can be enhanced in two-dimensional Gabor function, so that retaining
Enhance local characteristics while overall face information to be possibly realized.Gabor filter in airspace, one 2 dimension is a sine
The product of plane wave and gaussian kernel function has and obtains the characteristic of optimal partial simultaneously in spatial domain and frequency domain, with the mankind
Biological vision characteristic is much like, therefore can describe to correspond to spatial frequency (scale), spatial position and direction selection well
The partial structurtes information of property.Substantially, Gabor wavelet transformation is the local message in order to extract signal Fourier transformation, is used
One Gauss function is as window function, because of the Fourier transformation of a Gauss function or Gauss function, institute
It is also local with Fourier inverse transformation.By the selection of frequency parameter and Gaussian function parameter, Gabor transformation can be chosen
The characteristic information at many positions.Facial image is inputted after Gabor wavelet converts, one group of Gabor wavelet response diagram can be obtained
As cluster.
The principle of principal component analysis (Principal Component Analysis, PCA) dimension reduction method is that removal is original
Correlativity in data between each data component removes redundancy, retains most important ingredient.PCA calculates initial data association
Feature vector corresponding to maximum several characteristic values and corresponding subspace is obtained in variance matrix, eigenvector projection is arrived
The subspace is to achieve the purpose that sample space is described with small number of feature.Support vector machines is asked in solution classification
It is that the criterion based on structural risk minimization finds an optimal separating hyper plane, by sample when topic (usually two classification problems)
Originally be divided into two, reach different classes of have maximum class between be spaced.Discrete integer value is commonly used in classification problem indicates sample
This classification, different from classification problem, the label of each sample in regression problem is continuous real number.Therefore support vector machines is returned
Return (Support Vector Regression, SVR) to be intended to find a hyperplane, is capable of the distribution of Accurate Prediction sample, closely
Like sample data.
Now in conjunction with the drawings and specific embodiments, the present invention will be further described.
As shown in Figure 1, a kind of training method of two-dimension human face fraud detection classifier, which comprises
Step S1) successively in training set i-th (1≤i≤L) a face picture is pre-processed, obtain 64 × 64 pixels
The normalization facial image z of sizei(1≤i≤L);It specifically includes:
Step S1-1) image gray processing processing is carried out to a face picture of i-th (1≤i≤L):
The face picture includes: true acquisition face picture and secondary imaging (forgery) face picture.
Face-image is traversed, each pixel is handled, the rgb value of each pixel is obtained, is distinguished by operation
Red, blue, green value is extracted, optimum gradation conversion formula different according to sensitivity of the human eye to three kinds of colors of red blue green are as follows:
Grey=(9798R+19235G+3735B)/32768
Wherein Grey represent conversion after gray value, R, G, B respectively represent the red component of each pixel in image,
Green component and blue component;
Step S1-2) gray level image size is adjusted to 64 × 64 using bilinear interpolation;
Step S1-3) image after being sized is enhanced:
The modification that image histogram is carried out using the statistical data of histogram, is gone out by adjusting each grey level pixel of image
Existing probability is equal to change the pixel value of each grey level in image, to realize image enhancement;
Step S1-4) to enhanced image zooming-out image pixel matrix, obtain normalization facial image zi(1≤i≤
L)。
Step S2) from normalization facial image ziMiddle extraction LBP feature vector L (zi);
The step S2) it specifically includes:
Step S2-1) to normalization facial image ziWithOperator obtains LBP image, then is by LBP image segmentation
3 × 3 overlapping regions, the statistic histogram for extracting 59 dimensions from each region respectively are finally synthesizing a 531 final dimensions
Statistic histogram feature vector;
Wherein,Indicate the equivalent formulations operator of 8 points in the round field that radius is 1.
Step S2-2) to normalization facial image ziWithOperator, extraction obtain the statistic histogram feature of 59 dimensions
Vector;
Wherein,Indicate the equivalent formulations operator of 8 points in the round field that radius is 2.
Step S2-3) to normalization facial image ziWithOperator extracts and obtains the statistic histogram spy of 243 dimensions
Levy vector;
Wherein,Indicate the equivalent formulations operator of 16 points in the round field that radius is 2.
Step S2-4) feature vector that obtains step S2-1), step S2-2), step S2-3) synthesize a feature to
Measure L (zi), feature vector L (zi) dimension be 59*9+59+243=833 dimension.
Step S3) the extraction Gabor wavelet feature vector G (z from normalization facial imagei);
The step S3) it specifically includes:
Step S3-1) facial image z will be normalizedi32 × 32 sizes are zoomed to, it is small to carry out Gabor to the image after scaling
Wave conversion:
The Gabor filter in p different directions and q different scale is taken to handle image after scaling, each picture
Vegetarian refreshments t0It can obtain p × q Gabor amplitude Characteristics, p × q Gabor amplitude Characteristics cascade up a commonly known as Jet,
It is abbreviated as J, then pixel t in image0Jet are as follows:
J(t0)=(M0,0(t0),...,M0,7(t0),...,M4,0(t0),...,M4,7(t0))
The Gabor amplitude Characteristics of all pixels point are cascaded up to obtain the feature vector F (z of facial imagei):
F(zi)={ J (t0):t0∈zi}
Step S3-2) determine step S3-1) obtained feature vector F (zi) dimensionality reduction dimension, be based on Principal Component Analysis
To feature vector F (zi) carry out dimensionality reduction, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni);It specifically includes:
Step S3-2-1) the feature vector F (z that obtains step S3-1)i) dimension d be divided into n equal part, determine new dimension
Spend the value range of d';
Original feature vector F (zi) dimension be d, the value of d is divided into n equal portions, value set is as follows:
Wherein,Indicate integer rounding operation;
If the feature that obtains after dimensionality reduction is G (zi), then its dimension d' takes this n value respectively:
Step S3-2-2) d' successively takes each value in set, calculate the corresponding face fraud inspection of all pictures of training set
Survey mean absolute error set { MAEm};
To the L picture in training set, calculate when d' takes each value in set, the corresponding people of all pictures of training set
Face fraud detection mean absolute error is as follows:
Wherein, j indicates that jth picture in training set, k indicate that d' takes k-th of value in set, i.e.,lj
Indicate the corresponding class label of jth picture in training set (0 indicates fraud image, and 1 indicates real human face image),Indicate training
Concentrate the classification estimated value of jth picture;It may finally obtain different MAE value set { MAEmWherein m ∈ 1,2 ..., n;
Step S3-2-3) take set { MAEmIn minimum M AEmin, with MAEminCorresponding d' is tieed up as final dimensionality reduction
Degree;
Step S3-2-4) it is based on step S3-2-3) obtained d', using Principal Component Analysis to feature vector F (zi) into
Row dimensionality reduction, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni)。
Step S4) facial image z will be normalizediIt zooms to 8 × 8 sizes, then by two dimensional image thaumatropy is one-dimensional
Pixel characteristic vector P (zi);
Step S5) by step S2), step S3), step S4) extract three kinds of textural characteristics be spliced into final feature to
Measure D (zi)=(L (zi),G(zi),P(zi));
Step S6) based on Support vector regression algorithm to all feature vector D (zi), (1≤i≤L) is trained,
Obtain two-dimension human face fraud detection classifier;Specifically comprise the following steps:
Step S6-1) based on Support vector regression algorithm building optimization problem;
Hypothesized model training set sample is { x(i),y(i)(i=1,2 ..., L), x(i)Indicate normalization facial image zi's
Feature vector D (zi), y(i)Indicate the corresponding classification of the image: facial image or fraud image.Assuming that sample dimension is N, thenThe target of SVR is to solve for two-dimension human face fraud detection classifier f (x), makes f (x(i)) and y(i)Between difference
It is a very small number no more than ε, ε, controls the worst error between practical label value and predictive estimation value.So, f (x)
It is defined as follows:
Wherein, " " indicates inner product of vectors;W and b is the parameter solved;The w of solution should make | | w | |2It is minimum;Commonly referred to as should
Hyperplane model is ε-SVR;So optimization problem of ε-SVR is represented by following formula:
SVR introduces penalty coefficient and slack variable to be adjusted:
Wherein,
Indicate loss function;
In the present embodiment, penalty coefficient C=128, the learning parameter g=0.1 in SVR parameter are set, using RBF (diameter
To gaussian basis) kernel function is trained;
Step S6-2) optimization problem is converted using method of Lagrange multipliers, it is converted to and its antithesis is asked
Topic is solved, and the expression formula of two-dimension human face fraud detection classifier f (x) is obtained;
Introduce following Lagrangian:
WhereinIndicate αiWithIndicate ηiWithAnd αi、ηiAndIt is Lagrange multiplier.It is public
The solution of formula (9) belongs to convex quadratic programming problem scope, and by first solving L (w, α, η, b) function to w, the minimum value of b, ξ turn
Change, then solve and meet L (w, α, η, b) function to w, the local derviation of b, ξ are respectively 0 " saddle point ":
It is availableTaking back in formula (9) can obtain:
In this way, the Optimal solution problem of solution formula (9) is converted into the solution of following dual problem:
In this way, just converting optimization problem only comprising an alpha parameter for problem, phase can be found out after obtaining the value of α
The w answered, final f (x) are as follows:
According to KKT condition, when the dual problem of SVR only meets the following conditions, the solution of dual problem is just equivalent to original and asks
The solution of topic:
It can be seen that and work as αiWhen=C, ξiJust it is not equal to 0, sample point can fall in ε (i.e. outlier) at this time, then:
The value of so b meets:
It is also seen that by KKT condition for | f (x)-y(i)|=ε+ξi (*)Sample point, it is correspondingJust it is not equal to 0;
These sample point, that is, supporting vectors.
When kernel function is K (x(i), x) when, detection function f (x) becomes:
After training obtains SVR model, the only corresponding point of the supporting vector predicted value that decides recurrence.
Based on the two-dimension human face fraud detection classifier that above method training obtains, the present invention also provides two-dimension human faces to take advantage of
Cheat detection method, which comprises
Step T1) collected to camera face picture to be detected pre-processes, obtain 64 × 64 pixel sizes
Normalize facial image z0;
Step T2) from normalization facial image z0Middle extraction LBP feature vector L (z0);
Step T3) from normalization facial image z0Middle extraction Gabor wavelet feature vector G (z0);
Step T4) facial image z will be normalized0It zooms to 8 × 8 sizes, then by two dimensional image thaumatropy is one-dimensional
Pixel characteristic vector P (z0);
Step T5) by step T2), step T3), step T4) extract three kinds of textural characteristics be spliced into final feature to
Measure D (z0)=(L (z0),G(z0),P(z0));
Step T6) the feature vector D (z that obtains step T5)0)) be input to step S6) the fraud inspection of obtained two-dimension human face
It surveys classifier f (x), obtains testing result: facial image or fraud image.
It should be noted last that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting.Although ginseng
It is described the invention in detail according to embodiment, those skilled in the art should understand that, to technical side of the invention
Case is modified or replaced equivalently, and without departure from the spirit and scope of technical solution of the present invention, should all be covered in the present invention
Scope of the claims in.
Claims (3)
1. a kind of training method of two-dimension human face fraud detection classifier, which comprises first, to all in training set
Face picture is pre-processed, and normalization facial image is obtained;Second, LBP feature is extracted from each normalization facial image
Vector, Gabor wavelet feature vector sum one-dimensional pixel feature vector;These three feature vectors are spliced to form most by third
Whole feature vector;4th, it is trained using support vector machines to final feature vector is spliced to form, obtains two-dimension human face
Fraud detection classifier;
The method specifically includes:
Step S1) i-th of face picture in training set is pre-processed, wherein 1≤i≤L;Obtain 64 × 64 pixel sizes
Normalization facial image zi;
Step S2) from normalization facial image ziMiddle extraction LBP feature vector L (zi);
Step S3) from normalization facial image ziMiddle extraction Gabor wavelet feature vector G (zi);
Step S4) facial image z will be normalizediIt zooms to 8 × 8 sizes, then by two dimensional image thaumatropy is one-dimensional pixel feature
Vector P (zi);
Step S5) by step S2), step S3), step S4) extract three kinds of textural characteristics be spliced into final feature vector D
(zi)=(L (zi),G(zi),P(zi));
Step S6) based on Support vector regression algorithm to all feature vector D (zi) be trained, it obtains two-dimension human face and takes advantage of
Swindleness detection classifier, wherein 1≤i≤L;
The step S1) it specifically includes:
Step S1-1) image gray processing processing is carried out to face picture:
Face face picture is traversed, each pixel is handled, the rgb value of each pixel is obtained, is distinguished by operation
Red, blue, green value is extracted, the gray value after calculating each pixel conversion:
Grey=(9798R+19235G+3735B)/32768
Wherein, Grey represents the gray value after conversion, and R, G, B respectively represent the red component of each pixel in image, green
Colouring component and blue component;
Step S1-2) gray level image size is adjusted to 64 × 64 using bilinear interpolation;
Step S1-3) image after being sized is enhanced:
The modification of image histogram is carried out using the statistical data of histogram, is occurred by adjusting each grey level pixel of image general
Rate is equal to change the pixel value of each grey level in image, to realize image enhancement;
Step S1-4) to enhanced image zooming-out image pixel matrix, obtain normalization facial image zi;
The step S2) it specifically includes:
Step S2-1) to normalization facial image ziWithOperator obtains LBP image, then by LBP image segmentation is 3 × 3
A overlapping region, the statistic histogram for extracting 59 dimensions from each region respectively synthesize the statistic histogram feature of one 531 dimension
Vector;
Step S2-2) to normalization facial image ziWithOperator, extraction obtain the statistic histogram feature vector of 59 dimensions;
Step S2-3) to normalization facial image ziWithOperator, extract obtain the statistic histogram features of 243 dimensions to
Amount;
Step S2-4) feature vector that obtains step S2-1), step S2-2), step S2-3) synthesizes a feature vector L
(zi), the dimension of this feature vector is 59*9+59+243=833 dimension;
The step S3) it specifically includes:
Step S3-1) facial image z will be normalizedi32 × 32 sizes are zoomed to, Gabor wavelet change is carried out to the image after scaling
It changes:
The Gabor filter in p different directions and q different scale is taken to handle image after scaling, each pixel t0
It can obtain p × q Gabor amplitude Characteristics, p × q Gabor amplitude Characteristics cascade up a referred to as Jet, are abbreviated as J, then
Pixel t in image0Jet are as follows:
J(t0)=(M0,0(t0),...,M0,7(t0),...,M4,0(t0),...,M4,7(t0))
The Gabor amplitude Characteristics of all pixels point are cascaded up to obtain the feature vector F (z of facial imagei):
F(zi)={ J (t0):t0∈zi}
Step S3-2) determine step S3-1) obtained feature vector F (zi) dimensionality reduction dimension, based on Principal Component Analysis to feature
Vector F (zi) carry out dimensionality reduction, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni);
The step S3-2) it specifically includes:
Step S3-2-1) the feature vector F (z that obtains step S3-1)i) dimension d be divided into n equal part, determine new dimension d''s
Value range;
Original feature vector F (zi) dimension be d, the value of d is divided into n equal portions, value set is as follows:
Wherein,Indicate integer rounding operation;
If the feature that obtains after dimensionality reduction is G (zi), then its dimension d' takes this n value respectively:
Step S3-2-2) d' successively takes each value in set, it is flat to calculate the corresponding face fraud detection of all pictures of training set
Equal absolute error set { MAEk};
To the L picture in training set, calculate when d' takes each value in set, the corresponding face of all pictures of training set is taken advantage of
Swindleness detection mean absolute error is as follows:
Wherein, j indicates that jth picture in training set, k indicate that d' takes k-th of value in set, i.e.,ljIt indicates
The corresponding class label of jth picture in training set: 0 indicates fraud image, and 1 indicates real human face image,It indicates in training set
The classification estimated value of jth picture;It may finally obtain different MAE value set { MAEkWherein k ∈ 1,2 ..., n;
Step S3-2-3) take set { MAEkIn minimum M AEmin, with MAEminCorresponding d' is as final dimensionality reduction dimension;
Step S3-2-4) it is based on step S3-2-3) obtained d', using Principal Component Analysis to feature vector F (zi) dropped
Dimension, the Gabor wavelet feature vector G (z after obtaining dimensionality reductioni)。
2. the training method of two-dimension human face fraud detection classifier according to claim 1, which is characterized in that the step
S6 it) specifically includes:
Step S6-1) based on Support vector regression algorithm building optimization problem;
Hypothesized model training set sample is { x(i),y(i), i=1,2 ..., L, x(i)Indicate normalization facial image ziFeature to
Measure D (zi), y(i)Indicate the corresponding classification of the image: facial image or fraud image;Assuming that sample dimension is N, then
The target of Support vector regression algorithm is to solve for two-dimension human face fraud detection classifier f (x), makes f (x(i)) and y(i)Between
Difference is not more than ε, and ε is threshold value, controls the worst error between practical label value and predictive estimation value;So, f (x) determines
Justice is as follows:
Wherein, " " indicates inner product of vectors;W and b is the parameter solved;
Step S6-2) optimization problem (1) is converted using method of Lagrange multipliers, it is converted to its dual problem
It is solved, obtains the expression formula of two-dimension human face fraud detection classifier f (x).
3. a kind of face fraud detection method, the two-dimension human face obtained based on the training of method described in one of claim 1-2 is taken advantage of
Swindleness detection classifier realization, this method comprises:
Step T1) face picture to be detected is pre-processed, obtain the normalization facial image z of 64 × 64 pixel sizes0;
Step T2) from normalization facial image z0Middle extraction LBP feature vector L (z0);
Step T3) from normalization facial image z0Middle extraction Gabor wavelet feature vector G (z0);
Step T4) facial image z will be normalized08 × 8 sizes are zoomed to, then two dimensional image thaumatropy is special for one-dimensional pixel
Levy vector P (z0);
Step T5) by step T2), step T3), step T4) extract three kinds of textural characteristics be spliced into final feature vector D
(z0)=(L (z0),G(z0),P(z0));
Step T6) the feature vector D (z that obtains step T5)0)) be input in two-dimension human face fraud detection classifier, it is examined
Survey result: facial image or fraud image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610098933.8A CN107103266B (en) | 2016-02-23 | 2016-02-23 | The training of two-dimension human face fraud detection classifier and face fraud detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610098933.8A CN107103266B (en) | 2016-02-23 | 2016-02-23 | The training of two-dimension human face fraud detection classifier and face fraud detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107103266A CN107103266A (en) | 2017-08-29 |
CN107103266B true CN107103266B (en) | 2019-08-20 |
Family
ID=59658429
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610098933.8A Active CN107103266B (en) | 2016-02-23 | 2016-02-23 | The training of two-dimension human face fraud detection classifier and face fraud detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107103266B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107704834B (en) * | 2017-10-13 | 2021-03-30 | 深圳壹账通智能科技有限公司 | Micro-surface examination assisting method, device and storage medium |
CN108038413A (en) * | 2017-11-02 | 2018-05-15 | 平安科技(深圳)有限公司 | Cheat probability analysis method, apparatus and storage medium |
CN107886070A (en) * | 2017-11-10 | 2018-04-06 | 北京小米移动软件有限公司 | Verification method, device and the equipment of facial image |
CN108009531B (en) * | 2017-12-28 | 2022-01-07 | 北京工业大学 | Multi-strategy anti-fraud face recognition method |
CN109558794A (en) * | 2018-10-17 | 2019-04-02 | 平安科技(深圳)有限公司 | Image-recognizing method, device, equipment and storage medium based on moire fringes |
WO2020097834A1 (en) * | 2018-11-14 | 2020-05-22 | 北京比特大陆科技有限公司 | Feature processing method and apparatus, storage medium and program product |
CN111428666A (en) * | 2020-03-31 | 2020-07-17 | 齐鲁工业大学 | Intelligent family accompanying robot system and method based on rapid face detection |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101021900A (en) * | 2007-03-15 | 2007-08-22 | 上海交通大学 | Method for making human face posture estimation utilizing dimension reduction method |
CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
CN105117688A (en) * | 2015-07-29 | 2015-12-02 | 重庆电子工程职业学院 | Face identification method based on texture feature fusion and SVM |
-
2016
- 2016-02-23 CN CN201610098933.8A patent/CN107103266B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101021900A (en) * | 2007-03-15 | 2007-08-22 | 上海交通大学 | Method for making human face posture estimation utilizing dimension reduction method |
CN100492399C (en) * | 2007-03-15 | 2009-05-27 | 上海交通大学 | Method for making human face posture estimation utilizing dimension reduction method |
CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
CN105117688A (en) * | 2015-07-29 | 2015-12-02 | 重庆电子工程职业学院 | Face identification method based on texture feature fusion and SVM |
Non-Patent Citations (5)
Title |
---|
"Face Spoofing Detection From Single Images Using Micro-Texture Analysis";J.Maatta等;《2011 International Joint Conference on Biometrics》;20111013;第4页第3.1节 * |
"Face spoofing detection from single images using texture and local shape analysis";J. Maatta 等;《IET Biometrics》;20120331;第1卷(第1期);第5-6页第3节,第7页第4.1节 * |
"基于人脸图像特征表达的年龄估计模型算法研究与实现";吴仰波;《中国优秀硕士学位论文全文数据库 信息科技辑》;20150315(第3期);第12-16页第3.1-3.2节,第28-34页第4.1.1-4.2.3节 * |
"基于人脸图像的性别识别与年龄估计研究";陆丽;《中国博士学位论文全文数据库 信息科技辑》;20101015(第10期);第47-50页第3.2.1-3.2.4节 * |
J. Maatta 等."Face spoofing detection from single images using texture and local shape analysis".《IET Biometrics》.2012,第1卷(第1期),第5-6页第3节,第7页第4.1节. * |
Also Published As
Publication number | Publication date |
---|---|
CN107103266A (en) | 2017-08-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107103266B (en) | The training of two-dimension human face fraud detection classifier and face fraud detection method | |
Jourabloo et al. | Face de-spoofing: Anti-spoofing via noise modeling | |
Hu et al. | Iris liveness detection using regional features | |
Xiong et al. | Unknown presentation attack detection with face rgb images | |
CN101142584B (en) | Method for facial features detection | |
CN109858439A (en) | A kind of biopsy method and device based on face | |
Rusia et al. | A comprehensive survey on techniques to handle face identity threats: challenges and opportunities | |
CN106529378B (en) | A kind of the age characteristics model generating method and age estimation method of asian ancestry's face | |
Zhong et al. | Change detection based on pulse-coupled neural networks and the NMI feature for high spatial resolution remote sensing imagery | |
Wang et al. | Robust head pose estimation via supervised manifold learning | |
Nguyen et al. | Face presentation attack detection based on a statistical model of image noise | |
Barni et al. | Iris deidentification with high visual realism for privacy protection on websites and social networks | |
Rahman et al. | An automatic face detection and gender classification from color images using support vector machine | |
Mohamed et al. | Automated face recogntion system: Multi-input databases | |
Majumdar et al. | Pseudo-Fisherface method for single image per person face recognition | |
Ribarić et al. | Personal recognition based on the Gabor features of colour palmprint images | |
Gowda | Age estimation by LS-SVM regression on facial images | |
Bevilacqua et al. | Face detection by means of skin detection | |
Mitra et al. | Machine learning approach for signature recognition by harris and surf features detector | |
Singh et al. | Development of accurate face recognition process flow for authentication | |
Jiang et al. | Face recognition by combining wavelet transform and K-nearest neighbor | |
Singh et al. | Rough set based pose invariant face recognition with mug shot images | |
Navya | SMART ATTENDANCE USING IMAGE PROCESSING TECHNIQUES | |
Huang et al. | Real time face detection based on skin tone detector | |
Daniel et al. | Analysing texture, color and spatial features for face spoof detection with hybrid classification model |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220729 Address after: 100190, No. 21 West Fourth Ring Road, Beijing, Haidian District Patentee after: INSTITUTE OF ACOUSTICS, CHINESE ACADEMY OF SCIENCES Patentee after: NANHAI RESEARCH STATION, INSTITUTE OF ACOUSTICS, CHINESE ACADEMY OF SCIENCES Address before: 100190, No. 21 West Fourth Ring Road, Beijing, Haidian District Patentee before: INSTITUTE OF ACOUSTICS, CHINESE ACADEMY OF SCIENCES |