CN108932468A - One kind being suitable for psychologic face recognition method - Google Patents

Abstract

The invention belongs to area of face recognition technology, disclose one kind and are suitable for psychologic face recognition method, carry out Image Acquisition to obtain high-definition image；It is identified from high-definition image and is partitioned into face image；The facial image of segmentation module segmentation is handled；Generate the feature vector to be identified of facial detail feature；The feature vector of known face is pre-stored with by face characteristic library module；All feature vectors in the feature vector to be identified of output and the feature database are subjected to matching primitives, export the corresponding identity result of the feature vector to be identified；Display acquisition image information and matching result information.The present invention constructs facial feature database by face characteristic library module, carries out Rapid matching by characteristic matching module, so as in the relatively limited system of computing resource real-time judge go out the identity of facial image to be identified, time-consuming short accuracy rate is high.

Description

One kind being suitable for psychologic face recognition method

Technical field

The invention belongs to area of face recognition technology, more particularly to one kind to be suitable for psychologic face recognition method.

Background technique

Currently, the prior art commonly used in the trade is such：

Face recognition --- face recognition is also known as recognition of face, face recognizing, face identification etc., and face recognition uses logical The video camera used is as identification information acquisition device.The face-image of identification object, department of computer science are obtained in a non-contact manner System completes identification process after being compared after obtaining image with database images.However, existing face recognition is to face-image Extract the identification accuracy inaccurately influenced to face；Simultaneously face recognition low efficiency, time-consuming.

Cognition and understanding to figure are the important foundations that the mankind obtain external information and judge and reflect.Wherein, The similitude of automatic identification figure is to realize the important technology for improving human vision cognitive efficiency and expanding intelligent cognitive domain One of.It is widely used in industrial technology, graph and image processing, pattern-recognition and artificial intelligence field, the daily life to us Work plays unknown profound influence.It is very necessary for developing a set of shape similarity identification technology.With computer digit Change is increasingly developed with graph technology, and the digitized processing efficiency of descriptive geometry characteristic information has also obtained large increase.Rationally, The support of efficient algorithm and environmental level, but also research has sufficient feasibility.

Existing shape similarity often has the least mean-square error and geometry of probability statistics algorithm, characteristic value with recognition methods The Weighted Average Algorithm etc. of external appearance characteristic necessary condition.Although achieving certain efficiency, there is also some shortcomings：Algorithm The matching of realization process and visual discrimination is not intuitive；Algorithm is complicated, causes data processing amount big, and operating cost is high；Algorithm Evenness analysis causes the variation of important geometrical characteristic in figure to the influence of overall similarity, and Stability and veracity is caused to be deposited In certain deviation.

In conclusion problem of the existing technology is：

Existing face recognition is inaccurate to facial image zooming-out, influences the identification accuracy to face；Face recognition simultaneously Low efficiency, time-consuming.

In existing figure segmentation, similarity often realizes that the matching of process and visual discrimination is not intuitive with recognition methods, Algorithm is complicated, causes data processing amount big, and operating cost is high, and Stability and veracity is caused to there are problems that certain deviation.

The data processing accuracy of the prior art is poor.

Summary of the invention

In view of the problems of the existing technology, the present invention provides one kind to be suitable for psychologic face recognition method.

The invention is realized in this way it is a kind of suitable for psychologic facial-recognition security systems, it is described suitable for psychologic Facial-recognition security systems include：

Image capture module is connect with image segmentation module, for carrying out Image Acquisition by high-definition camera to obtain High-definition image；

Image segmentation module is connect with image capture module, main control module, for receiving high-definition image, and from high definition figure It is identified as at and is partitioned into face image；

The mathematical model for establishing two figures establishes eigenmatrix corresponding with figure by the complete Vector Groups of description figure, Calculate the angle on adjacent both sides；Calculate the minimum distance between two figures；Enhancement processing to calculated result.

The side length of the mathematical model polygon of the foundation and adjacent angle are by one vector S of construction counterclockwise₁Indicate polygon Shape：

S₁=(l₁,α1,l_2,α₂…l_N-1,α_N-1,l_N,α_N)；

S₁There are mapping relations one by one with the polygon, indicates unrelated with corner initial order；

The complete Vector Groups have 2N vector S counterclockwise₁、S₂……S_2N-1、S_2NHave one with polygon One mapping relations constitute a complete Vector Groups of the polygon, are expressed as follows：

S₁=(l₁,α₁,l_2,α₂…l_N-1,α_N-1,l_N,α_N)；

S₂=(α₁,l₂, α₂…l_N-1,α_N-1,l_N,α_N,l₁)；

……

S_2N-1=(l_N,α_N,l₁,α₁,l₂, α₂…l_N-1,α_N-1)；

S_2N=(α_N,l₁,α₁,l₂, α₂…l_N-1,α_N-1,l_N)；

With matrix S_EIt indicates complete vector, and defines S_EFor the eigenmatrix of the polygon, S_EIt is expressed as follows：

Main control module, it is special with image capture module, image segmentation module, image processing module, feature generation module, face Library module, characteristic matching module, display module connection are levied, is worked normally for dispatching modules；

Image processing module is connect with main control module, is handled for the facial image to segmentation module segmentation；

Feature generation module, connect with main control module, generates facial detail feature for carrying out to facial image after processing Feature vector to be identified；

Face characteristic library module, connect with main control module, for being pre-stored with the feature vector of known face；

Characteristic matching module is connect with main control module, the feature to be identified for exporting the feature generation module All feature vectors in vector and the feature database carry out matching primitives, export the corresponding body of the feature vector to be identified Part result；

Matching primitives method includes：

Step 1: collecting N number of sample as training set X, sample mean m is found out using following formula：

Wherein, xi ∈ sample training collection X=(x1, x2 ..., xN)；

Step 2: finding out scatter matrix S：

Find out the eigenvalue λ i and corresponding feature vector ei of scatter matrix, wherein ei is principal component, by characteristic value from It arrives greatly and small is arranged successively λ 1, λ 2 ...；

P value is taken out, λ 1, λ 2 ..., λ p determine face space E=(e1, e2 ..., eP), on this face space, training sample In this X, the point that each element projects to the space is obtained by following formula：

X'i=Etxi, t=1,2 ..., N；

What is obtained by above formula is p dimensional vector by former vector after PCA dimensionality reduction；

Plurality of human faces identification is carried out using SRC face recognition algorithms, including：

The recognition result of each face of present frame is obtained to present frame Face datection and by coordinate sequence；It is each according to present frame The recognition result of a face calculates corresponding each face respectively adjacent n frame recognition result；The identity for counting each face, by surpassing The Unified Identity of more than half n/2 determines the final identity of target；

Wherein, calculate picture and face database to be identified it is of all categories between reconstruction error { r1, r2 ... rn }, r1<r2<…… <Rn, by obtained similarity value according toRule determine final recognition result；Wherein T1 is rate value, T1=0.6；

Display module is connect with main control module, for showing acquisition image information and matching result information.

Further, source figure and targeted graphical work pretreatment include in the figure:

Appropriate thresholding is set according to figure minimum containment rectangle length-width ratio, is filtered；

Thresholding is set according to side length each in the figure of source and the minimum value of perimeter ratio, removes the surpriseization part in targeted graphical；

Abbreviation processing is made to targeted graphical number of edges, makes that there is identical number of edges with source figure；

Abbreviation is handled：

1) to the source signal of the surpriseization part in removal targeted graphicalIt carries out low Energy pretreatment will that is, in each sampling instant pValue of the amplitude less than thresholding ε sets 0, It obtainsThe setting of thresholding ε is determined according to the average energy for receiving signal；

2) the time-frequency numeric field data of p moment (p=0,1,2, P-1) non-zero is found out, is used It indicates, whereinIndicate the response of p moment time-frequencyCorresponding frequency indices, right when non-zero The normalization pretreatment of these non-zeros, obtains pretreated vector b (p, q)=[b₁(p,q),b₂(p,q),,b_M(p,q) ]^T, wherein

3) using clustering algorithm estimate the jumping moment of each jump and respectively jump corresponding normalized mixed moment array to Amount, Hopping frequencies；It is right at p (p=0,1,2 ... the P-1) momentThe frequency values of expression are clustered, obtained cluster centre numberIndicate carrier frequency number existing for the p moment,A cluster centre then indicates the size of carrier frequency, uses respectively It indicates；To each sampling instant p (p=0,1,2 ... P-1), clustering algorithm pair is utilizedIt is clustered, it is same availableA cluster centre is usedIt indicates；To allIt averages and is rounded, obtain the estimation of source signal numberI.e.：

It finds outAt the time of, use p_hIt indicates, to the p of each section of continuous value_hIntermediate value is sought, is used

Indicate the l sections of p that are connected_hIntermediate value, thenIndicate the estimation at first of frequency hopping moment； It is obtained according to estimationp≠p_hAnd the 4th frequency hopping moment for estimating in step estimate it is each It jumps correspondingA hybrid matrix column vectorSpecifically formula is：

HereIt is corresponding to indicate that l is jumpedA mixing Matrix column vector estimated value；Estimate the corresponding carrier frequency of each jump, usesIt indicates that l is jumped to correspond to 'sA frequency estimation, calculation formula are as follows：

4) time-frequency domain frequency hopping source signal is estimated according to the normalization hybrid matrix column vector that estimation obtains；

5) the time-frequency domain frequency hopping source signal between different frequency hopping points is spliced；It is corresponding to estimate that l is jumpedA incidence Angle is usedIndicate that l jumps the corresponding incident angle of n-th of source signal,Calculation formula it is as follows：

Indicate that l jumps n-th of hybrid matrix column vector that estimation obtainsM-th of element, c indicate light Speed, i.e. v_c=3 × 10⁸Meter per second；Judge that l (l=2,3 ...) jumps the source signal of estimation and first and jump between the source signal of estimation Corresponding relationship, judgment formula is as follows：

Wherein m_n ^(l)Indicate that l jumps the m of estimation_n ^(l)A signal and n-th of signal of the first jump estimation belong to the same source Signal；By different frequency hopping point estimation to the signal for belonging to the same source signal be stitched together, as final time-frequency domain source Signal estimation, uses Y_nTime-frequency domain estimated value of n-th of the source signal of (p, q) expression on time frequency point (p, q), p=0,1,2 ..., P, q=0,1,2 ..., N_fft- 1, i.e.,：

6) according to source signal time-frequency domain estimated value, restore time domain frequency hopping source signal；To each sampling instant p (p=0,1, 2 ...) frequency domain data Y_n(p, q), q=0,1,2, N_fft- 1 is N_fftPoint IFFT transformation, obtain p sampling instant it is corresponding when Domain frequency hopping source signal, uses y_n(p,q_t)(q_t=0,1,2, N_fft- 1) it indicates；The time domain frequency hopping synthesizer letter obtained to above-mentioned all moment Number y_n(p,q_t) processing is merged, final time domain frequency hopping source signal estimation is obtained, specific formula is as follows：

Here K_c=N_fft/ C, C are the sampling number at Short Time Fourier Transform adding window interval, N_fftFor the length of FFT transform.

Further, the Euclidean distance of most like vector and maximum phase in source figure and targeted graphical eigenmatrix are obtained and is Number specifically includes：

Firstly, establishing the eigenmatrix P of source figure P and targeted graphical Q respectively counterclockwise_EAnd Q_E：

P_E=[P₁ ^-P₂ ^T…P_2N-1 ^T P_2N ^T]；

Q_E=[Q₁ ^T Q₂ ^T…Q_2N-1 ^T Q_2N ^T]；

Euclidean distance formula d (x, y) and included angle cosine formula sim (x, y) are as follows：

With d (x, y) and it is the basis sim (x, y), redefines two matrix Ds and S, make:

Find out the minimum value in D and S；

Eu is enabled respectively_e=min { D_ij, 1≤i≤j=2N；Sim_e=max { S_ij, 1≤i≤j=2N；

Then the eigenmatrix of needle directional structure vectorical structure figure P and Q, the above-mentioned calculation method of repetition find out two features in order again Minimum value Eu in matrix between most complete vector_cAnd Sim_c；

Finally enable Eu=min { Eu_e, Eu_c}；

Sim=min { Sim_e, Sim_c}；

Eu and Sim be two figure of P, Q correspond to most like vector Euclidean distance and it is maximum mutually and coefficient；

It is suitable for psychologic face recognition method another object of the present invention is to provide one kind to include the following steps：

Step 1 carries out Image Acquisition by image capture module to obtain high-definition image；By image segmentation module from It is identified at high-definition image and is partitioned into face image；

Step 2, main control module are dispatched image processing module and are handled the facial image of segmentation module segmentation；

Step 3 generates the feature vector to be identified of facial detail feature by feature generation module；Pass through face spy Sign library module is pre-stored with the feature vector of known face；

Step 4, the feature vector to be identified exported the feature generation module by characteristic matching module with it is described All feature vectors in feature database carry out matching primitives, export the corresponding identity result of the feature vector to be identified；

Step 5 shows acquisition image information and matching result information by display module.

Further, described image processing module processing method is as follows：

Firstly, image to be processed is carried out image segmentation by segmentation module, portrait image is obtained；

Secondly, carrying out Face datection to the portrait image, N number of characteristic point of the chin outline portion of face is obtained；

Then, one is formed according to multiple default key points in N number of characteristic point and the image to be processed to close Close region；

Finally, carrying out FIG pull handle to the portrait image according to the enclosed region, human face region image is obtained.

Advantages of the present invention and good effect are：

The present invention provides the portrait image that image processing module will reject image background, carries out face inspection to portrait image It surveys, obtains N number of characteristic point of the chin outline portion of face, the boundary of face and neck can be accurately divided, according to described N number of Multiple default key points in characteristic point and the image to be processed form an enclosed region, determine and need to reject in next step Range, according to the enclosed region to the portrait image carry out FIG pull handle, obtain human face region image, the face area Area image is accurately separated with the following body part of neck, thus, it is possible on the basis of existing green curtain scratches figure, accurately by people As the following body part of face and neck of image separates.Facial feature database is constructed by face characteristic library module simultaneously, By characteristic matching module carry out Rapid matching, so as in the relatively limited system of computing resource real-time judge go out wait know The identity of other facial image, time-consuming short accuracy rate are high.

The present invention extracts facial image eigenvector method, improves recognition of face degree to a certain extent, is conducive to image Acquisition and identification.

The present invention improves machine to the visual discrimination effect of face shape similarity, especially to manually being not easy to differentiate high phase Have very great help like the difficult point of degree figure；Test pattern effect has stronger stability and reliability；Detection time is short, and operation is high Effect, implementation result are at low cost.The present invention only inquires the side of figure, reduces data processing amount.The present invention passes through construction The eigenmatrix of figure chooses suitable decision criteria, and carries out multiple enhancement nonlinear transformation to eigenmatrix element, uses Most values, multi-standard weighted average establish Measurement of Similarity, it is efficient and have stronger stability to have reached algorithm.

Source figure and targeted graphical, which are made to pre-process, in the figure includes:

Appropriate thresholding is set according to figure minimum containment rectangle length-width ratio, is filtered；

Thresholding is set according to side length each in the figure of source and the minimum value of perimeter ratio, removes the surpriseization part in targeted graphical；

Abbreviation processing is made to targeted graphical number of edges, makes that there is identical number of edges with source figure；

Abbreviation is handled：

1) to the source signal of the surpriseization part in removal targeted graphicalIt carries out low Energy pretreatment will that is, in each sampling instant pValue of the amplitude less than thresholding ε sets 0, It obtainsThe setting of thresholding ε is determined according to the average energy for receiving signal；It can obtain It obtains and accurately handles data.

Detailed description of the invention

Fig. 1 is present invention implementation offer suitable for psychologic face recognition method flow chart.

Fig. 2 is present invention implementation offer suitable for psychologic facial-recognition security systems structural block diagram.

In figure：1, image capture module；2, image segmentation module；3, main control module；4, image processing module；5, feature is raw At module；6, face characteristic library module；7, characteristic matching module；8, display module.

Fig. 3 is present invention implementation offer under AutoCAD2002 environment, provides one group of totally 12 resolution chart.

Specific embodiment

In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to Limit the present invention.

With reference to the accompanying drawing and specific embodiment is further described application principle of the invention.

Include the following steps as shown in Figure 1, one kind provided by the invention is suitable for psychologic face recognition method：

S101 carries out Image Acquisition by image capture module to obtain high-definition image；By image segmentation module from height It is identified at clear image and is partitioned into face image；

S102, main control module are dispatched image processing module and are handled the facial image of segmentation module segmentation；

S103 generates the feature vector to be identified of facial detail feature by feature generation module；Pass through face characteristic Library module is pre-stored with the feature vector of known face；

S104, the feature vector to be identified for being exported the feature generation module by characteristic matching module and the spy All feature vectors levied in library carry out matching primitives, export the corresponding identity result of the feature vector to be identified；

S105 shows acquisition image information and matching result information by display module.

As shown in Fig. 2, provided by the invention include suitable for psychologic facial-recognition security systems：Image capture module 1, figure As segmentation module 2, main control module 3, image processing module 4, feature generation module 5, face characteristic library module 6, characteristic matching mould Block 7, display module 8.

Image capture module 1 is connect with image segmentation module 2, for carrying out Image Acquisition by high-definition camera to obtain Obtain high-definition image；

Image segmentation module 2 is connect with image capture module 1, main control module 3, for receiving high-definition image, and from high definition It is identified at image and is partitioned into face image；

Main control module 3, with image capture module 1, image segmentation module 2, image processing module 4, feature generation module 5, Face characteristic library module 6, characteristic matching module 7, display module 8 connect, and work normally for dispatching modules；

Image processing module 4 is connect with main control module 3, and the facial image for dividing to segmentation module 2 is handled；

Feature generation module 5 is connect with main control module 3, generates facial detail spy for carrying out to facial image after processing The feature vector to be identified of point；

Face characteristic library module 6 is connect with main control module 3, for being pre-stored with the feature vector of known face；

Characteristic matching module 7 is connect with main control module 3, the spy to be identified for exporting the feature generation module All feature vectors levied in vector and the feature database carry out matching primitives, and it is corresponding to export the feature vector to be identified Identity result；

Display module 8 is connect with main control module 3, for showing acquisition image information and matching result information.

4 processing method of image processing module provided by the invention is as follows：

Firstly, image to be processed is carried out image segmentation by segmentation module, portrait image is obtained；

Secondly, carrying out Face datection to the portrait image, N number of characteristic point of the chin outline portion of face is obtained；

Then, one is formed according to multiple default key points in N number of characteristic point and the image to be processed to close Close region；

Finally, carrying out FIG pull handle to the portrait image according to the enclosed region, human face region image is obtained.

Below with reference to concrete analysis, the invention will be further described.

Image segmentation module is connect with image capture module, main control module, for receiving high-definition image, and from high definition figure It is identified as at and is partitioned into face image；

The mathematical model for establishing two figures establishes eigenmatrix corresponding with figure by the complete Vector Groups of description figure, Calculate the angle on adjacent both sides；Calculate the minimum distance between two figures；Enhancement processing to calculated result.

The side length of the mathematical model polygon of the foundation and adjacent angle are by one vector S of construction counterclockwise₁Indicate polygon Shape：

S₁=(l₁,α1,l_2,α₂…l_N-1,α_N-1,l_N,α_N)；

S₁There are mapping relations one by one with the polygon, indicates unrelated with corner initial order；

The complete Vector Groups have 2N vector S counterclockwise₁、S₂……S_2N-1、S_2NHave one with polygon One mapping relations constitute a complete Vector Groups of the polygon, are expressed as follows：

S₁=(l₁,α₁,l₂, α₂…l_N-1,α_N-1,l_N,α_N)；

S₂=(α₁,l₂, α₂…l_N-1,α_N-1,l_N,α_N,l₁)；

……

S_2N-1=(l_N,α_N,l₁,α₁,l_2,α₂…l_N-1,α_N-1)；

S_2N=(α_N,l₁,α₁,l₂, α₂…l_N-1,α_N-1,l_N)；

With matrix S_EIt indicates complete vector, and defines S_EFor the eigenmatrix of the polygon, S_EIt is expressed as follows：

Main control module, it is special with image capture module, image segmentation module, image processing module, feature generation module, face Library module, characteristic matching module, display module connection are levied, is worked normally for dispatching modules；

Image processing module is connect with main control module, is handled for the facial image to segmentation module segmentation；

Feature generation module, connect with main control module, generates facial detail feature for carrying out to facial image after processing Feature vector to be identified；

Face characteristic library module, connect with main control module, for being pre-stored with the feature vector of known face；

Characteristic matching module is connect with main control module, the feature to be identified for exporting the feature generation module All feature vectors in vector and the feature database carry out matching primitives, export the corresponding body of the feature vector to be identified Part result；

Matching primitives method includes：

Step 1: collecting N number of sample as training set X, sample mean m is found out using following formula：

Wherein, xi ∈ sample training collection X=(x1, x2 ..., xN)；

Step 2: finding out scatter matrix S：

Find out the eigenvalue λ i and corresponding feature vector ei of scatter matrix, wherein ei is principal component, by characteristic value from It arrives greatly and small is arranged successively λ 1, λ 2 ...；

P value is taken out, λ 1, λ 2 ..., λ p determine face space E=(e1, e2 ..., eP), on this face space, training sample In this X, the point that each element projects to the space is obtained by following formula：

X'i=Etxi, t=1,2 ..., N；

What is obtained by above formula is p dimensional vector by former vector after PCA dimensionality reduction；

Plurality of human faces identification is carried out using SRC face recognition algorithms, including：

The recognition result of each face of present frame is obtained to present frame Face datection and by coordinate sequence；It is each according to present frame The recognition result of a face calculates corresponding each face respectively adjacent n frame recognition result；The identity for counting each face, by surpassing The Unified Identity of more than half n/2 determines the final identity of target；

Wherein, calculate picture and face database to be identified it is of all categories between reconstruction error { r1, r2 ... rn }, r1<r2<…… <Rn, by obtained similarity value according toRule determine final recognition result；Wherein T1 is rate value, T1=0.6；

Display module is connect with main control module, for showing acquisition image information and matching result information.

Source figure and targeted graphical, which are made to pre-process, in the figure includes:

Appropriate thresholding is set according to figure minimum containment rectangle length-width ratio, is filtered；

Thresholding is set according to side length each in the figure of source and the minimum value of perimeter ratio, removes the surpriseization part in targeted graphical；

Abbreviation processing is made to targeted graphical number of edges, makes that there is identical number of edges with source figure；

Abbreviation is handled：

1) to the source signal of the surpriseization part in removal targeted graphicalIt carries out low Energy pretreatment will that is, in each sampling instant pValue of the amplitude less than thresholding ε sets 0, It obtainsThe setting of thresholding ε is determined according to the average energy for receiving signal；

2) the time-frequency numeric field data of p moment (p=0,1,2, P-1) non-zero is found out, is used It indicates, whereinIndicate the response of p moment time-frequencyCorresponding frequency indices, right when non-zero The normalization pretreatment of these non-zeros, obtains pretreated vector b (p, q)=[b₁(p,q),b₂(p,q),,b_M(p,q) ]^T, wherein

3) using clustering algorithm estimate the jumping moment of each jump and respectively jump corresponding normalized mixed moment array to Amount, Hopping frequencies；It is right at p (p=0,1,2 ... the P-1) momentThe frequency values of expression are clustered, obtained cluster centre NumberIndicate carrier frequency number existing for the p moment,A cluster centre then indicates the size of carrier frequency, uses respectivelyIt indicates；To each sampling instant p (p=0,1,2 ... P-1), clustering algorithm pair is utilizedInto Row cluster, it is same availableA cluster centre is usedIt indicates；To allIt averages and is rounded, obtain To the estimation of source signal numberI.e.：

It finds outAt the time of, use p_hIt indicates, to the p of each section of continuous value_hIntermediate value is sought, is used

Indicate the l sections of p that are connected_hIntermediate value, thenIndicate the estimation at first of frequency hopping moment； It is obtained according to estimationp≠p_hAnd the 4th frequency hopping moment for estimating in step estimate it is each It jumps correspondingA hybrid matrix column vectorSpecifically formula is：

HereIt is corresponding to indicate that l is jumpedA mixing Matrix column vector estimated value；Estimate the corresponding carrier frequency of each jump, usesIt indicates that l is jumped to correspond to 'sA frequency estimation, calculation formula are as follows：

4) time-frequency domain frequency hopping source signal is estimated according to the normalization hybrid matrix column vector that estimation obtains；

5) the time-frequency domain frequency hopping source signal between different frequency hopping points is spliced；It is corresponding to estimate that l is jumpedA incidence Angle is usedIndicate that l jumps the corresponding incident angle of n-th of source signal,Calculation formula it is as follows：

Indicate that l jumps n-th of hybrid matrix column vector that estimation obtainsM-th of element, c indicate light Speed, i.e. v_c=3 × 10⁸Meter per second；Judge that l (l=2,3 ...) jumps the source signal of estimation and first and jump between the source signal of estimation Corresponding relationship, judgment formula is as follows：

Wherein m_n ^(l)Indicate that l jumps the m of estimation_n ^(l)A signal and n-th of signal of the first jump estimation belong to the same source Signal；By different frequency hopping point estimation to the signal for belonging to the same source signal be stitched together, as final time-frequency domain source Signal estimation, uses Y_nTime-frequency domain estimated value of n-th of the source signal of (p, q) expression on time frequency point (p, q), p=0,1,2 ..., P, q=0,1,2 ..., N_fft- 1, i.e.,：

6) according to source signal time-frequency domain estimated value, restore time domain frequency hopping source signal；To each sampling instant p (p=0,1, 2 ...) frequency domain data Y_n(p, q), q=0,1,2, N_fft- 1 is N_fftPoint IFFT transformation, obtain p sampling instant it is corresponding when Domain frequency hopping source signal, uses y_n(p,q_t)(q_t=0,1,2, N_fft- 1) it indicates；The time domain frequency hopping synthesizer letter obtained to above-mentioned all moment Number y_n(p,q_t) processing is merged, final time domain frequency hopping source signal estimation is obtained, specific formula is as follows：

Here K_c=N_fft/ C, C are the sampling number at Short Time Fourier Transform adding window interval, N_fftFor the length of FFT transform.

The Euclidean distance of most like vector and maximum phase and coefficient are specific in acquisition source figure and targeted graphical eigenmatrix Including：

Firstly, establishing the eigenmatrix P of source figure P and targeted graphical Q respectively counterclockwise_EAnd Q_E：

P_E=[P₁ ^T P₂ ^T…P_2N-1 ^T P_2N ^T]；

Q_E=[Q₁ ^T Q₂ ^T…Q_2N-1 ^T Q_2N ^T]；

Euclidean distance formula d (x, y) and included angle cosine formula sim (x, y) are as follows：

With d (x, y) and it is the basis sim (x, y), redefines two matrix Ds and S, make:

Find out the minimum value in D and S；

Eu is enabled respectively_e=min { D_ij, 1≤i≤j=2N；Sim_e=max { S_ij, 1≤i≤j=2N；

Then the eigenmatrix of needle directional structure vectorical structure figure P and Q, the above-mentioned calculation method of repetition find out two features in order again Minimum value Eu in matrix between most complete vector_cAnd Sim_c；

Finally enable Eu=min { Eu_e, Eu_c}；

Sim=min { Sim_e, Sim_c}；

Eu and Sim be two figure of P, Q correspond to most like vector Euclidean distance and it is maximum mutually and coefficient.

Below with reference to concrete analysis, the invention will be further described.

Multiple graphs similarity is calculated using Euclidean distance algorithm

Under AutoCAD2002 environment, one group of totally 12 resolution chart is provided, as shown in Figure 3.

The calculated result of Euclidean distance is as shown in table 1：

The Euclidean distance of target sample and master sample in table 1

Similarity value after 2 weighted average of table

Data are found out from two tables, and calculated result has local correction amendment before the calculated result ratio after weighting, compare Meet artificial vision's resolving effect.Table 1 calculates time-consuming 0.472s.

Result reflect calculated result of the present invention be it is reliable, in time and efficiently.The present invention can be used for The acquisition of target position in detecting and tracking in computer vision, according to existing template find in the picture one it is closest therewith Region.Then it follows always.Existing some algorithms such as BlobTracking, Meanshift, Camshift, particle filter Wave etc. is also all that the theory of this respect is needed to go to support.There are also the image retrievals for being on the one hand namely based on picture material, also It is usually to say to scheme inspection figure.For example it is set out in the image data base of magnanimity to your a certain individual most more matched therewith Image, this certain technology can may also be done so, and be several characteristic values by image abstraction, for example Trace is converted, image Hash Or Sift feature vector etc., come according to deposited in database these characteristic matchings return again to corresponding image improve effect Rate.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.

Claims (5)

1. one kind is suitable for psychologic facial-recognition security systems, which is characterized in that described to be suitable for psychologic face recognition system System includes：

Image capture module is connect with image segmentation module, for carrying out Image Acquisition by high-definition camera to obtain high definition Image；

Image segmentation module is connect with image capture module, main control module, for receiving high-definition image, and from high-definition image It identifies and is partitioned into face image；

The mathematical model for establishing two figures is established eigenmatrix corresponding with figure by the complete Vector Groups of description figure, is calculated The angle on adjacent both sides out；Calculate the minimum distance between two figures；Enhancement processing to calculated result.

The side length of the mathematical model polygon of the foundation and adjacent angle are by one vector S of construction counterclockwise₁Indicate polygon：

S₁=(l₁,α1,l₂, α₂…l_N-1,α_N-1,l_N,α_N)；

S₁There are mapping relations one by one with the polygon, indicates unrelated with corner initial order；

The complete Vector Groups have 2N vector S counterclockwise₁、S₂……S_2N-1、S_2NHave with polygon and maps one by one Relationship constitutes a complete Vector Groups of the polygon, is expressed as follows：

S₁=(l₁,α₁,l₂, α₂…l_N-1,α_N-1,l_N,α_N)；

S₂=(α₁,l_2,α₂…l_N-1,α_N-1,l_N,α_N,l₁)；

……

S_2N-1=(l_N,α_N,l₁,α₁,l_2,α₂…l_N-1,α_N-1)；

S_2N=(α_N,l₁,α₁,l₂, α₂…l_N-1,α_N-1,l_N)；

With matrix S_EIt indicates complete vector, and defines S_EFor the eigenmatrix of the polygon, S_EIt is expressed as follows：

Main control module, with image capture module, image segmentation module, image processing module, feature generation module, face characteristic library Module, characteristic matching module, display module connection, work normally for dispatching modules；

Image processing module is connect with main control module, is handled for the facial image to segmentation module segmentation；

Feature generation module, connect with main control module, for facial image after processing carry out generate facial detail feature to The feature vector of identification；

Face characteristic library module, connect with main control module, for being pre-stored with the feature vector of known face；

Characteristic matching module is connect with main control module, the feature vector to be identified for exporting the feature generation module Matching primitives are carried out with all feature vectors in the feature database, export the corresponding identity knot of the feature vector to be identified Fruit；

Matching primitives method includes：

Step 1: collecting N number of sample as training set X, sample mean m is found out using following formula：

Wherein, xi ∈ sample training collection X=(x1, x2 ..., xN)；

Step 2: finding out scatter matrix S：

Find out the eigenvalue λ i and corresponding feature vector ei of scatter matrix, wherein ei is principal component, by characteristic value from greatly to It is small to be arranged successively λ 1, λ 2 ...；

P value is taken out, λ 1, λ 2 ..., λ p determine face space E=(e1, e2 ..., eP), on this face space, training sample X In, the point that each element projects to the space is obtained by following formula：

X'i=Etxi, t=1,2 ..., N；

What is obtained by above formula is p dimensional vector by former vector after PCA dimensionality reduction；

Plurality of human faces identification is carried out using SRC face recognition algorithms, including：

The recognition result of each face of present frame is obtained to present frame Face datection and by coordinate sequence；According to each individual of present frame The recognition result of face calculates corresponding each face respectively adjacent n frame recognition result；The identity for counting each face, by being more than half The Unified Identity of number n/2 determines the final identity of target；

Wherein, calculate picture and face database to be identified it is of all categories between reconstruction error { r1, r2 ... rn }, r1<r2<……<Rn, By obtained similarity value according toRule determine final recognition result；Wherein T1 is Rate value, T1=0.6；

Display module is connect with main control module, for showing acquisition image information and matching result information.

2. being suitable for psychologic facial-recognition security systems as described in claim 1, which is characterized in that source figure in the figure Making pretreatment with targeted graphical includes:

Appropriate thresholding is set according to figure minimum containment rectangle length-width ratio, is filtered；

Thresholding is set according to side length each in the figure of source and the minimum value of perimeter ratio, removes the surpriseization part in targeted graphical；

Abbreviation processing is made to targeted graphical number of edges, makes that there is identical number of edges with source figure；

Abbreviation is handled：

1) to the source signal of the surpriseization part in removal targeted graphicalCarry out low energy Pretreatment will that is, in each sampling instant pValue of the amplitude less than thresholding ε sets 0, obtainsThe setting of thresholding ε is determined according to the average energy for receiving signal；

2) the time-frequency numeric field data of p moment (p=0,1,2 ... P-1) non-zero is found out, is used It indicates, whereinIndicate the response of p moment time-frequencyCorresponding frequency indices, right when non-zero The normalization pretreatment of these non-zeros, obtains pretreated vector b (p, q)=[b₁(p,q),b₂(p,q),…,b_M(p, q)]^T, wherein

3) jumping moment of each jump is estimated using clustering algorithm and respectively jump corresponding normalized hybrid matrix column vector, jump Frequent rate；It is right at p (p=0,1,2 ... the P-1) momentThe frequency values of expression are clustered, obtained cluster centre number Indicate carrier frequency number existing for the p moment,A cluster centre then indicates the size of carrier frequency, uses respectively It indicates；To each sampling instant p (p=0,1,2 ... P-1), clustering algorithm pair is utilizedIt is clustered, it is same availableA cluster centre is usedIt indicates；To allIt averages and is rounded, obtain the estimation of source signal numberI.e.：

It finds outAt the time of, use p_hIt indicates, to the p of each section of continuous value_hIntermediate value is sought, is usedL=1,2 ... table Show the l sections of p that are connected_hIntermediate value, thenIndicate the estimation at first of frequency hopping moment；It is obtained according to estimationp≠p_hAnd the 4th frequency hopping moment for estimating in step estimate it is each jump it is correspondingIt is a mixed Close matrix column vectorSpecifically formula is：

HereIt is corresponding to indicate that l is jumpedA hybrid matrix Column vector estimated value；Estimate the corresponding carrier frequency of each jump, usesIt is corresponding to indicate that l is jumped A frequency estimation, calculation formula are as follows：

4) time-frequency domain frequency hopping source signal is estimated according to the normalization hybrid matrix column vector that estimation obtains；

5) the time-frequency domain frequency hopping source signal between different frequency hopping points is spliced；It is corresponding to estimate that l is jumpedA incident angle, WithIndicate that l jumps the corresponding incident angle of n-th of source signal,Calculation formula it is as follows：

Indicate that l jumps n-th of hybrid matrix column vector that estimation obtainsM-th of element, c indicate the light velocity, i.e. v_c =3 × 10⁸Meter per second；It is corresponding between the source signal of estimation and the source signal of the first jump estimation to judge that l (l=2,3 ...) is jumped Relationship, judgment formula are as follows：

Wherein m_n ^(l)Indicate that l jumps the m of estimation_n ^(l)A signal and n-th of signal of the first jump estimation belong to the same source and believe Number；By different frequency hopping point estimation to the signal for belonging to the same source signal be stitched together, believe as final time-frequency domain source Number estimation, use Y_nTime-frequency domain estimated value of n-th of the source signal of (p, q) expression on time frequency point (p, q), p=0,1,2 ..., P, Q=0,1,2 ..., N_fft- 1, i.e.,：

6) according to source signal time-frequency domain estimated value, restore time domain frequency hopping source signal；To each sampling instant p (p=0,1,2 ...) Frequency domain data Y_n(p, q), q=0,1,2 ..., N_fft- 1 is N_fftThe IFFT transformation of point, obtains the corresponding time domain of p sampling instant Frequency hopping source signal, uses y_n(p,q_t)(q_t=0,1,2 ..., N_fft- 1) it indicates；The time domain frequency hopping synthesizer letter obtained to above-mentioned all moment Number y_n(p,q_t) processing is merged, final time domain frequency hopping source signal estimation is obtained, specific formula is as follows：

Here K_c=N_fft/ C, C are the sampling number at Short Time Fourier Transform adding window interval, N_fftFor the length of FFT transform.

3. being suitable for psychologic facial-recognition security systems as claimed in claim 2, which is characterized in that obtain source figure and target The Euclidean distance of most like vector and maximum phase and coefficient specifically include in graphic feature matrix：

Firstly, establishing the eigenmatrix P of source figure P and targeted graphical Q respectively counterclockwise_EAnd Q_E：

P_E=[P₁ ^T P₂ ^T … P_2N-1 ^T P_2N ^T]；

Q_E=[Q₁ ^T Q₂ ^T … Q_2N-1 ^T Q_2N ^T]；

Euclidean distance formula d (x, y) and included angle cosine formula sim (x, y) are as follows：

With d (x, y) and it is the basis sim (x, y), redefines two matrix Ds and S, make:

Find out the minimum value in D and S；

Eu is enabled respectively_e=min { D_ij, 1≤i≤j=2N；Sim_e=max { S_ij, 1≤i≤j=2N；

Then the eigenmatrix of needle directional structure vectorical structure figure P and Q, the above-mentioned calculation method of repetition find out two eigenmatrixes in order again In minimum value Eu between most complete vector_cAnd Sim_c；

Finally enable Eu=min { Eu_e, Eu_c}；

Sim=min { Sim_e, Sim_c}；

Eu and Sim be two figure of P, Q correspond to most like vector Euclidean distance and it is maximum mutually and coefficient.

4. a kind of be suitable for psychologic face recognition suitable for psychologic facial-recognition security systems as described in claim 1 Method, which is characterized in that described to include the following steps suitable for psychologic face recognition method：

Step 1 carries out Image Acquisition by image capture module to obtain high-definition image；By image segmentation module from high definition It is identified at image and is partitioned into face image；

Step 2, main control module are dispatched image processing module and are handled the facial image of segmentation module segmentation；

Step 3 generates the feature vector to be identified of facial detail feature by feature generation module；Pass through face characteristic library Module is pre-stored with the feature vector of known face；

Step 4, the feature vector to be identified and the feature for being exported the feature generation module by characteristic matching module All feature vectors in library carry out matching primitives, export the corresponding identity result of the feature vector to be identified；

Step 5 shows acquisition image information and matching result information by display module.

5. being suitable for psychologic face recognition method as claimed in claim 4, which is characterized in that described image processing module Processing method is as follows：

Firstly, image to be processed is carried out image segmentation by segmentation module, portrait image is obtained；

Secondly, carrying out Face datection to the portrait image, N number of characteristic point of the chin outline portion of face is obtained；

Then, a closed area is formed according to multiple default key points in N number of characteristic point and the image to be processed Domain；

Finally, carrying out FIG pull handle to the portrait image according to the enclosed region, human face region image is obtained.