CN104615997A - Human face anti-fake method based on multiple cameras - Google Patents

Abstract

The invention relates to the field of human face recognition, in particular to a human face anti-fake method based on multiple cameras. A plurality of cameras are arranged to conduct camera shooting from different angles, pictures shot by different cameras are utilized to conduct human face characteristic angle comparison to further achieve human face recognition and fake resistance, hardware configuration is simple, demarcation is not required, the method can be easily integrated with the existing human face recognition system, meanwhile, judgment can be conducted by mixing multiple human face characteristics to prevent fakeness, and the anti-fake accuracy is effectively improved.

Description

A kind of face method for anti-counterfeit based on multiple-camera

Technical field

The present invention relates to field of face identification, particularly a kind of face method for anti-counterfeit based on multiple-camera.

Background technology

Current two-dimension human face recognition technology obtains a large amount of uses in the application such as work attendance and gate inhibition.But two-dimension human face recognition system faces some difficult problems in actual applications, cheat face identification system as used photo or video.In order to address this problem, face identification system method for anti-counterfeit arises at the historic moment.

Conventional face identification system method for anti-counterfeit comprises use additional hardware and uses the large class of software algorithm two.Wherein using the method for additional hardware generally to adopt increases extra near infrared sensor to judge whether object to be identified is real human face,

Use the method for software algorithm then to distinguish real human face and photo etc. by the feature (as detected nictation, expression shape change or behavior etc.) of the image or video of analyzing object to be identified and forge face.These methods or introduce diverse extras thus add complexity and the cost of system, or depend on the feature of the image sequence of single image or shooting continuously, even require that user performs certain action according to instruction, thus cause the inconvenience in use, also have impact on false proof precision.

Summary of the invention

The object of the invention is to overcome and rely on increase hardware based on the face method for anti-counterfeit of multiple-camera or rely on the problem that user performs instruction action at present, there is provided a kind of and can further improve the anti-counterfeit capability of face identification system and the face method for anti-counterfeit based on multiple-camera of ease for use, comprise following steps:

(1) be symmetrical arranged and take facial image by n platform video camera, n is more than 2 natural numbers.

(2) respectively face feature point location is carried out to the facial image of different cameras shooting, obtain the coordinate of each facial feature points.

(3) select any three not facial feature points on the same line, wherein fisrt feature point coordinate is (x ₁, y ₁), second feature point coordinate is (x ₂, y ₂), third feature point coordinate is (x ₃, y ₃), according to the face visual angle characteristic d β of each image of following formulae discovery _i:

β ₁＝arctan(y ₂-y ₁,x ₂-x ₁)×180/π；

β ₂＝arctan(y ₃-(y ₁+y ₂)/2,x ₃-(x ₁+x ₂)/2)×180/π；

dβ _i＝|β ₂-β ₁|。

(4) the face visual angle characteristic of the facial image taken by symmetrical camera compares, and show that face visual angle characteristic is poor

(5) by face visual angle characteristic difference Δ and predetermined threshold value T _Δcompare, as Δ > T _Δtime, be then judged as true, otherwise be false.

Further, described facial feature points comprises nose, the corners of the mouth, pupil center, canthus.

Preferably, in step (3), three facial feature points selected are respectively Liang Ge pupil center and nose.

Further, also comprise and judge true and false step by face characteristic similarity S as follows:

(6) from the facial image that each video camera is taken, textural characteristics value T is extracted _i∈ R ^m, i is camera number, and m is intrinsic dimensionality.

(7) calculate each shot by camera facial image and and its facial image being in the shot by camera of symmetrical seat in the plane between human face similarity degree S _k, computing formula is:

$S_k=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{i,h}-T_{j,h})}^2}{T_{i,h}+T_{j,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{i,h}-T_{j,h})}^p}\end{array}\right.$

Wherein, T _i, T _jbe respectively the textural characteristics value of two mutual symmetrical video cameras in position, the span of k be 1 to the value of p is 1 or 2, P value when being 1, represents Euclidean distance, when P value is 2, and expression L1 distance.

(8) each face image similarity S calculated _imean value, be this shooting human face similarity degree S, its computing formula is: wherein for rounding under n/2.

(9) as S > T _s, be then judged as vacation, otherwise be true, wherein TS is for presetting face characteristic similarity threshold.

Further, in step (2-1), textural characteristics value obtains initial value for taking local binary patterns (LBP), gradient orientation histogram pattern (HOG) or Gabor filter patterns, and the initial value of acquisition is carried out dimensionality reduction or conversion process obtains in some embodiment, as Δ > T through principal component analysis (PCA) or linear discriminant analysis _Δand S≤T _stime, be judged to be real human face, otherwise be judged to forge face.

In other embodiment, as α (Δ-T _Δ)+β (T _s-S) > T time, be judged to be real human face, otherwise be judged to be forge face, wherein α, β are constant, and meet alpha+beta=1; The threshold value that T pre-sets.

Further, the predetermined angle in n platform video camera between adjacent camera is 0-45 degree.

In some embodiments, a described n video camera is in same level height.

In other embodiment, when the quantity of described video camera is even number, symmetrical between two and form symmetrical group relative to same axis between video camera, the video camera of different symmetrical group is in differentiated levels.

Further, the angle between two video cameras in different symmetrical group is identical or not identical.

In sum, owing to have employed technique scheme, the invention has the beneficial effects as follows: the present invention adopts configuration multi-cam to make a video recording from different perspectives, and it is false proof to realize recognition of face to the method that the image that different camera is taken carries out comparing at face characteristic visual angle, hardware configuration is simple, without the need to demarcating, be easy to mutually integrated with existing face identification system; Also being undertaken false proof by merging the differentiation of multiple face characteristic similarity simultaneously, effectively improving anti-spurious accuracy.

Accompanying drawing explanation

Fig. 1 is face visual angle characteristic method for anti-counterfeit process flow diagram provided by the invention.

The Double anti-counterfeit method integrated use process flow diagram of Fig. 2 for providing in the embodiment of the present invention 2.

Fig. 3 is human face similarity degree method for anti-counterfeit process flow diagram in the Double anti-counterfeit method that provides in the embodiment of the present invention 2.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in detail.

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

Embodiment 1: rely on increase hardware or rely on the problem that user performs instruction action based on the face method for anti-counterfeit of multiple-camera as Fig. 1 the object of the invention is to overcome at present, there is provided a kind of and can further improve the anti-counterfeit capability of face identification system and the face method for anti-counterfeit based on multiple-camera of ease for use, when video camera is even number, for 2 video cameras, 2 video cameras are arranged (preferably between 0-45 degree at same level height according to predetermined angle, predetermined angle in the present embodiment between two video cameras is 30 degree), and take facial image (should symmetrically take), and comprise the step of being carried out true and false judgement as follows by face visual angle characteristic difference Δ:

S101: respectively face feature point location is carried out to the facial image of different cameras shooting, has both determined each facial feature points, as the coordinate position at nose, the corners of the mouth, pupil center, canthus.

S102: select Liang Ge pupil center and nose as three facial feature points, wherein the first center coordinate of eye pupil is (x ₁, y ₁), the second center coordinate of eye pupil is (x ₂, y ₂), nose coordinate is (x ₃, y ₃), according to following formulae discovery face visual angle characteristic d β _i:

Calculate the angle of Liang Ge pupil center line and transverse axis: β ₁=arctan (y ₂-y ₁, x ₂-x ₁) × 180/ π;

Calculate nose and the line of interpupillary line central point and the angle of transverse axis:

β ₂＝arctan(y ₃-(y ₁+y ₂)/2,x ₃-(x ₁+x ₂)/2)×180/π；

Calculate face visual angle characteristic: d β _i=| β ₂-β ₁|.

S103: by the face visual angle characteristic d β of the facial image of two video camera shootings ₁, d β ₂compare, draw face visual angle characteristic difference Δ=| d β ₁-d β ₂|.

S104: by face visual angle characteristic difference Δ and predetermined threshold value T _Δcompare, as Δ > T _Δtime, be then judged as true, otherwise be false.

Embodiment 2: when video camera is odd number, for 5 video cameras, is with embodiment 1 distinctive points:

5 video cameras are arranged (preferably between 0-45 degree at same level height according to predetermined angle, predetermined angle in the present embodiment between adjacent two video cameras is 10 degree), and (5 video camera order labels are No. 1 video camera, No. 2 video cameras, No. 3 video cameras, No. 4 video cameras, No. 5 video cameras to take facial image, during shooting, face should just to being positioned at No. 3 middle video cameras, now, No. 1 video camera and No. 5 video camera symmetries, No. 2 video cameras and No. 4 video camera symmetries).

S101: respectively face feature point location is carried out to the facial image of different cameras shooting, has both determined each facial feature points, as the coordinate position at nose, the corners of the mouth, pupil center, canthus.

S102: select Liang Ge pupil center and nose as three facial feature points, wherein the first center coordinate of eye pupil is (x ₁, y ₁), the second center coordinate of eye pupil is (x ₂, y ₂), nose coordinate is (x ₃, y ₃), according to following formulae discovery face visual angle characteristic d β _i:

Calculate the angle of Liang Ge pupil center line and transverse axis: β ₁=arctan (y ₂-y ₁, x ₂-x ₁) × 180/ π;

Calculate nose and the line of interpupillary line central point and the angle of transverse axis:

β ₂＝arctan(y ₃-(y ₁+y ₂)/2,x ₃-(x ₁+x ₂)/2)×180/π；

Calculate face visual angle characteristic: d β _i=| β ₂-β ₁|.

S103: by the face visual angle characteristic d β of the facial image of 5 video camera shootings ₁, d β ₂, d β ₃, d β ₄, d β ₅the face visual angle characteristic of middle symmetrical seat in the plane compares, and show that face visual angle characteristic is poor in the present embodiment, in not wherein being calculated in without the visual angle characteristic of No. 3 video cameras of symmetrical seat in the plane.Only have 5 video cameras in the present embodiment, therefore symmetrical each other video camera is No. 1 video camera and No. 5 video cameras, No. 2 video cameras and No. 4 video cameras (during shooting video camera should relative to facial symmetry).

S104: by face visual angle characteristic difference Δ and predetermined threshold value T _Δcompare, as Δ > T _Δtime, be then judged as true, otherwise be false.

Embodiment 3: when video camera is even number, for 4 video cameras, symmetrical between two and form 2 symmetrical group relative to same axis between 4 video cameras, the video cameras of 2 symmetrical groups are in differentiated levels.Angle between two video cameras in different symmetrical group is different, if No. 1 video camera in the present embodiment and No. 4 video cameras are one symmetrical group, angle is between the two 30 degree, and No. 2 video cameras and No. 3 video cameras are one symmetrical group, and angle is between the two 45 degree; During shooting, face should just to the axis of symmetry of two pairs of video cameras.Be with embodiment 1 distinctive points:

S103: by the face visual angle characteristic d β of the facial image of 4 video camera shootings ₁, d β ₂, d β ₃, d β ₄the face visual angle characteristic of middle symmetrical seat in the plane compares, and show that face visual angle characteristic is poor in the present embodiment, $\mathrm{\Δ}=\frac{|{\mathrm{d\β}}_1-d{\mathrm{\β}}_4|+|d{\mathrm{\β}}_2-d{\mathrm{\β}}_3|}{2}.$

S104: by face visual angle characteristic difference Δ and predetermined threshold value T _Δcompare, as Δ > T _Δtime, be then judged as true, otherwise be false.

Embodiment 4: as shown in Figure 2 and Figure 3, in the present embodiment, except each step comprised as described in example 1 above (calculating referred to as face visual angle characteristic), also comprises the following step calculating face characteristic similarity S:

S201: extract textural characteristics value T from the facial image that each video camera is taken _i∈ R ^m, i is camera number, and the textural characteristics value of the facial image of the shooting of two video cameras in the present embodiment is respectively T ₁and T ₂m is intrinsic dimensionality, and R is set of real numbers.

S202: calculate each shot by camera facial image and and its facial image being in the shot by camera of symmetrical seat in the plane between human face similarity degree S _k, only have two video cameras in the present embodiment, therefore computing formula is:

$S_k=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{1,h}-T_{2,h})}^2}{T_{1,h}+T_{2,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{1,h}-T_{2,h})}^p}\end{array}\right.$

Wherein, T ₁, T ₂be respectively the textural characteristics value of two mutual symmetrical video cameras in position, because only have 2 video cameras,

Therefore the value of p is 1 or 2, P value when being 1, represents Euclidean distance, when P value is 2, and expression L1 distance.

S203: each face image similarity S calculated _kmean value, be this shooting human face similarity degree S, in the present embodiment, because only have 2 video cameras, therefore directly draw human face similarity degree S by S202.

S204: as S > T _s, be then judged as vacation, otherwise be true, wherein T _sfor default face characteristic similarity threshold.

In the present embodiment, because there is face characteristic Similarity Measure and the calculating of the face visual angle characteristic described in embodiment 1 simultaneously, therefore also comprise the step of comprehensive descision, both only had when two kinds are all judged as that true time is (namely as Δ > T _Δand simultaneously S≤T _stime), just judge that current face is as real human face, otherwise be judged to forge face.

Embodiment 5: when video camera is odd number, for 5 video cameras, except each step comprised as described in example 2 above (calculating referred to as face visual angle characteristic), also comprises the following step calculating face characteristic similarity S:

S201: the textural characteristics value of the facial image of the shooting of 5 video cameras in the present embodiment is respectively T ₁, T ₂, T ₃, T ₄, T ₅, m is intrinsic dimensionality, and R is set of real numbers.

S202: calculate each shot by camera facial image and and its facial image being in the shot by camera of symmetrical seat in the plane between human face similarity degree S _i5 video cameras are only had in the present embodiment, therefore symmetrical each other video camera is No. 1 video camera and No. 3 video cameras, No. 2 video cameras and No. 4 video cameras, and is positioned at No. 3 middle video cameras owing to not having symmetrical machine, and the image of therefore its shooting does not participate in Similarity Measure; Computing formula is:

$S_k=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{i,h}-T_{j,h})}^2}{T_{i,h}+T_{j,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{i,h}-T_{j,h})}^p}\end{array}\right.$

Wherein, T _i, T _jbe respectively the textural characteristics value of two mutual symmetrical video cameras in position, wherein the span of k be 1 to wherein:

$S_1=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{1,h}-T_{5,h})}^2}{T_{1,h}+T_{5,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{1,h}-T_{5,h})}^p}\end{array}\right.$

$S_2=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{2,h}-T_{4,h})}^2}{T_{2,h}+T_{4,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{2,h}-T_{4,h})}^p}\end{array}\right.$

The value of p is 1 or 2, P value when being 1, represents Euclidean distance, when P value is 2, and expression L1 distance.

S203: each face image similarity S calculated _kmean value, be this shooting human face similarity degree S, in the present embodiment, $S=\frac{S_1+S_2}{2}.$

S204: as S > T _s, be then judged as vacation, otherwise be true, wherein TS is for presetting face characteristic similarity threshold.

In the present embodiment, because there is the step of face characteristic Similarity Measure and the calculating of the face visual angle characteristic described in embodiment 2 simultaneously, therefore also comprise the step of comprehensive descision, both only had when two kinds are all judged as that true time is (namely as Δ > T _Δand simultaneously S≤T _stime), just judge that current face is as real human face, otherwise be judged to forge face.

Embodiment 6: when video camera is even number, for 4 video cameras, except each step comprised as described in example 3 above (calculating referred to as face visual angle characteristic), also comprises the following step calculating face characteristic similarity S:

S201: the textural characteristics value of the facial image of the shooting of 4 video cameras in the present embodiment is respectively T ₁, T ₂, T ₃, T ₄, m is intrinsic dimensionality, and R is set of real numbers.

S202: calculate each shot by camera facial image and and its facial image being in the shot by camera of symmetrical seat in the plane between human face similarity degree S _i, only have 4 video cameras in the present embodiment, therefore symmetrical each other video camera is No. 1 video camera and No. 4 video cameras, No. 2 video cameras and No. 3 video cameras, and the image of therefore its shooting does not participate in Similarity Measure; Computing formula is:

$S_k=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{i,h}-T_{j,h})}^2}{T_{i,h}+T_{j,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{i,h}-T_{j,h})}^p}\end{array}\right.$

Wherein, T _i, T _jbe respectively the textural characteristics value of two mutual symmetrical video cameras in position, wherein the span of k be 1 to wherein:

$S_1=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{1,h}-T_{4,h})}^2}{T_{1,h}+T_{4,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{1,h}-T_{4,h})}^p}\end{array}\right.$

$S_2=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{2,h}-T_{3,h})}^2}{T_{2,h}+T_{3,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{2,h}-T_{3,h})}^p}\end{array}\right.$

The value of p is 1 or 2, P value when being 1, represents Euclidean distance, when P value is 2, and expression L1 distance.

S203: each face image similarity S calculated _kmean value, be this shooting human face similarity degree S, in the present embodiment, $S=\frac{S_1+S_2}{2}.$

S204: as S > T _s, be then judged as vacation, otherwise be true, wherein T _sfor default face characteristic similarity threshold.

In the present embodiment, because there is the step of face characteristic Similarity Measure and the calculating of the face visual angle characteristic described in embodiment 3 simultaneously, therefore also comprise the step of comprehensive descision, both only had when two kinds are all judged as that true time is (namely as Δ > T _Δand simultaneously S≤T _stime), just judge that current face is as real human face, otherwise be judged to forge face.

Embodiment 7: as shown in Figure 1, Figure 2, Figure 3 shows, in the present embodiment, compared with embodiment 4,5 or 6, difference is, the step of described comprehensive descision is: as α (Δ-T _Δ)+β (T _s-S) > T time, be judged to be real human face, otherwise be judged to be forge face, wherein α, β are constant, and meet alpha+beta=1; The threshold value that T pre-sets.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1., based on a face method for anti-counterfeit for multiple-camera, it is characterized in that, comprise following steps:

(1) be symmetrical arranged and take facial image by n platform video camera, n is more than 2 natural numbers;

(2) respectively face feature point location is carried out to the facial image of different cameras shooting, obtain the coordinate of each facial feature points;

(3) select any three not facial feature points on the same line, wherein fisrt feature point coordinate is (x ₁, y ₁), second feature point coordinate is (x ₂, y ₂), third feature point coordinate is (x ₃, y ₃), according to the face visual angle characteristic d β of each image of following formulae discovery _i:

β ₁＝arctan(y ₂-y ₁,x ₂-x ₁)×180/π；

β ₂＝arctan(y ₃-(y ₁+y ₂)/2,x ₃-(x ₁+x ₂)/2)×180/π；

dβ _i＝|β ₂-β ₁|；

(4) the face visual angle characteristic of the facial image taken by symmetrical camera compares, and show that face visual angle characteristic is poor

(5) by face visual angle characteristic difference Δ and predetermined threshold value T _Δcompare, as Δ > T _Δtime, be then judged as true, otherwise be false.

2., as claimed in claim 1 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, described facial feature points comprises nose, the corners of the mouth, pupil center, canthus.

3., as claimed in claim 2 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, in step (1-2), three facial feature points selected are respectively Liang Ge pupil center and nose.

4., as claimed in claim 1 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, also comprise and judge true and false step by face characteristic similarity S as follows:

(6) from the facial image that each video camera is taken, textural characteristics value T is extracted _i∈ R ^m, i is camera number, and m is intrinsic dimensionality, and R is set of real numbers;

(7) calculate each shot by camera facial image and and its facial image being in the shot by camera of symmetrical seat in the plane between human face similarity degree S _k, computing formula is:

$S_k=\left\{\begin{array}{c}\underset{h=1}{\overset{m}{\mathrm{\Σ}}}\frac{{(T_{i,h}-T_{j,h})}^2}{T_{i,h}+T_{j,h}}\\ \sqrt[p]{{\mathrm{\Σ}}_{h=1}^m{(T_{i,h}-T_{j,h})}^p}\end{array}\right.$

Wherein, T _i, T _jbe respectively the textural characteristics value of two mutual symmetrical video cameras in position, the span of k be 1 to the value of p is 1 or 2;

(8) each face image similarity S calculated _imean value, be this shooting human face similarity degree S;

(9) as S > T _s, be then judged as vacation, otherwise be true, wherein T _sfor default face characteristic similarity threshold.

5. as claimed in claim 4 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, in step (2-1), textural characteristics value obtains initial value for taking local binary patterns (LBP), gradient orientation histogram pattern (HOG) or Gabor filter patterns, and the initial value of acquisition is carried out dimensionality reduction or conversion process obtains through principal component analysis (PCA) or linear discriminant analysis.

6., as claimed in claim 4 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, as Δ > T _Δand S≤T _stime, be judged to be real human face, otherwise be judged to forge face.

7., as claimed in claim 4 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, as α (Δ-T _Δ)+β (T _s-S) > T time, be judged to be real human face, otherwise be judged to be forge face, wherein α, β are constant, and meet alpha+beta=1; The threshold value that T pre-sets.

8., as claimed in claim 1 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, have predetermined angle between adjacent camera in symmetrically arranged n platform video camera, this predetermined angle is 0-45 degree.

9., as claimed in claim 1 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, a described n video camera is in same level height.

10. as claimed in claim 1 based on the face method for anti-counterfeit of multiple-camera, it is characterized in that, when the quantity of described video camera is even number, symmetrical between two and form symmetrical group relative to same axis between video camera, the video camera of different symmetrical group is in differentiated levels;

Angle between two video cameras in different symmetrical group is identical or not identical.